Examine individual changes
This page allows you to examine the variables generated by the Edit Filter for an individual change.
Variables generated for this change
| Variable | Value |
|---|---|
Edit count of the user (user_editcount) | null |
Name of the user account (user_name) | '177.236.113.166' |
Age of the user account (user_age) | 0 |
Groups (including implicit) the user is in (user_groups) | [
0 => '*'
] |
Rights that the user has (user_rights) | [
0 => 'createaccount',
1 => 'read',
2 => 'edit',
3 => 'createtalk',
4 => 'writeapi',
5 => 'viewmywatchlist',
6 => 'editmywatchlist',
7 => 'viewmyprivateinfo',
8 => 'editmyprivateinfo',
9 => 'editmyoptions',
10 => 'abusefilter-log-detail',
11 => 'urlshortener-create-url',
12 => 'centralauth-merge',
13 => 'abusefilter-view',
14 => 'abusefilter-log',
15 => 'vipsscaler-test'
] |
Whether the user is editing from mobile app (user_app) | false |
Whether or not a user is editing through the mobile interface (user_mobile) | true |
Page ID (page_id) | 32472154 |
Page namespace (page_namespace) | 0 |
Page title without namespace (page_title) | 'Deep learning' |
Full page title (page_prefixedtitle) | 'Deep learning' |
Edit protection level of the page (page_restrictions_edit) | [] |
Last ten users to contribute to the page (page_recent_contributors) | [
0 => 'KH-1',
1 => 'Mjpayton',
2 => '2407:D000:A:54F3:65A6:47AC:84D0:EF34',
3 => '195.28.202.129',
4 => 'Telfyr',
5 => 'LearnMore',
6 => 'Joshuame13',
7 => 'ClueBot NG',
8 => '185.203.218.68',
9 => 'Bruce1ee'
] |
Page age in seconds (page_age) | 366413056 |
Action (action) | 'edit' |
Edit summary/reason (summary) | '' |
Old content model (old_content_model) | 'wikitext' |
New content model (new_content_model) | 'wikitext' |
Old page wikitext, before the edit (old_wikitext) | '{{Short description|Branch of machine learning}}
[[File:Deep Learning.jpg|alt=Representing images on multiple layers of abstraction in deep learning|thumb|upright=1.35|Representing images on multiple layers of abstraction in deep learning<ref>{{Cite journal|last1=Schulz|first1=Hannes|last2=Behnke|first2=Sven|date=1 November 2012|title=Deep Learning|journal=KI - Künstliche Intelligenz|language=en|volume=26|issue=4|pages=357–363|doi=10.1007/s13218-012-0198-z|s2cid=220523562|issn=1610-1987|url=https://www.semanticscholar.org/paper/51a80649d16a38d41dbd20472deb3bc9b61b59a0}}</ref>]]
{{machine learning|Artificial neural network}}
{{Artificial intelligence|Approaches}}
'''Deep learning''' is part of a broader family of [[machine learning]] methods based on [[artificial neural network]]s with [[representation learning]]. Learning can be [[Supervised learning|supervised]], [[Semi-supervised learning|semi-supervised]] or [[Unsupervised learning|unsupervised]].<ref name="NatureBengio">{{cite journal |last1=LeCun |first1= Yann|last2=Bengio |first2=Yoshua | last3=Hinton | first3= Geoffrey|s2cid=3074096 |year=2015 |title=Deep Learning |journal=Nature |volume=521 |issue=7553 |pages=436–444 |doi=10.1038/nature14539 |pmid=26017442|bibcode=2015Natur.521..436L }}</ref>
Deep-learning architectures such as [[#Deep_neural_networks|deep neural network]]s, [[deep belief network]]s, [[deep reinforcement learning]], [[recurrent neural networks]], [[convolutional neural networks]] and [[Transformer (machine learning model)|transformers]] have been applied to fields including [[computer vision]], [[speech recognition]], [[natural language processing]], [[machine translation]], [[bioinformatics]], [[drug design]], [[medical image analysis]], [[Climatology|climate science]], material inspection and [[board game]] programs, where they have produced results comparable to and in some cases surpassing human expert performance.<ref name=":9">{{Cite book |doi=10.1109/cvpr.2012.6248110 |isbn=978-1-4673-1228-8|arxiv=1202.2745|chapter=Multi-column deep neural networks for image classification|title=2012 IEEE Conference on Computer Vision and Pattern Recognition|pages=3642–3649|year=2012|last1=Ciresan|first1=D.|last2=Meier|first2=U.|last3=Schmidhuber|first3=J.|s2cid=2161592}}</ref><ref name="krizhevsky2012">{{cite journal|last1=Krizhevsky|first1=Alex|last2=Sutskever|first2=Ilya|last3=Hinton|first3=Geoffrey|date=2012|title=ImageNet Classification with Deep Convolutional Neural Networks|url=https://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf|journal=NIPS 2012: Neural Information Processing Systems, Lake Tahoe, Nevada|access-date=2017-05-24|archive-date=2017-01-10|archive-url=https://web.archive.org/web/20170110123024/http://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf|url-status=live}}</ref><ref>{{cite web |title=Google's AlphaGo AI wins three-match series against the world's best Go player |url=https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/ |website=TechCrunch |date=25 May 2017 |access-date=17 June 2018 |archive-date=17 June 2018 |archive-url=https://web.archive.org/web/20180617065807/https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/ |url-status=live }}</ref>
[[Artificial neural network]]s (ANNs) were inspired by information processing and distributed communication nodes in [[biological system]]s. ANNs have various differences from biological [[brain]]s. Specifically, artificial neural networks tend to be static and symbolic, while the biological brain of most living organisms is dynamic (plastic) and analog.<ref>{{Cite journal|last1=Marblestone|first1=Adam H.|last2=Wayne|first2=Greg|last3=Kording|first3=Konrad P.|s2cid=1994856|date=2016|title=Toward an Integration of Deep Learning and Neuroscience |journal=Frontiers in Computational Neuroscience |volume=10|pages=94|doi=10.3389/fncom.2016.00094 |pmc=5021692|pmid=27683554|bibcode=2016arXiv160603813M|arxiv=1606.03813|doi-access=free}}</ref><ref>{{cite arXiv|last1=Bengio|first1=Yoshua|last2=Lee|first2=Dong-Hyun|last3=Bornschein|first3=Jorg|last4=Mesnard|first4=Thomas|last5=Lin|first5=Zhouhan|date=13 February 2015|title=Towards Biologically Plausible Deep Learning|eprint=1502.04156|class=cs.LG}}</ref>
The adjective "deep" in deep learning refers to the use of multiple layers in the network. Early work showed that a linear [[perceptron]] cannot be a universal classifier, but that a network with a nonpolynomial activation function with one hidden layer of unbounded width can. Deep learning is a modern variation that is concerned with an unbounded number of layers of bounded size, which permits practical application and optimized implementation, while retaining theoretical universality under mild conditions. In deep learning the layers are also permitted to be heterogeneous and to deviate widely from biologically informed [[connectionism|connectionist]] models, for the sake of efficiency, trainability and understandability.
{{toclimit|3}}
== Definition ==
Deep learning is a class of [[machine learning]] [[algorithm]]s that<ref name="BOOK2014">{{cite journal|last2=Yu|first2=D.|year=2014|title=Deep Learning: Methods and Applications|url=http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf|journal=Foundations and Trends in Signal Processing|volume=7|issue=3–4|pages=1–199|doi=10.1561/2000000039|last1=Deng|first1=L.|access-date=2014-10-18|archive-date=2016-03-14|archive-url=https://web.archive.org/web/20160314152112/http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf|url-status=live}}</ref>{{rp|pages=199–200}} uses multiple layers to progressively extract higher-level features from the raw input. For example, in [[image processing]], lower layers may identify edges, while higher layers may identify the concepts relevant to a human such as digits or letters or faces.
== Overview ==
Most modern deep learning models are based on [[artificial neural network]]s, specifically [[convolutional neural network]]s (CNN)s, although they can also include [[propositional formula]]s or latent variables organized layer-wise in deep [[generative model]]s such as the nodes in [[deep belief network]]s and deep [[Boltzmann machine]]s.<ref name="BENGIODEEP">{{cite journal|last=Bengio|first=Yoshua|year=2009|title=Learning Deep Architectures for AI|url=http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20%282009%29.pdf|journal=Foundations and Trends in Machine Learning|volume=2|issue=1|pages=1–127|doi=10.1561/2200000006|citeseerx=10.1.1.701.9550|access-date=3 September 2015|archive-url=https://web.archive.org/web/20160304084250/http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf|archive-date=4 March 2016|url-status=dead}}</ref>
In deep learning, each level learns to transform its input data into a slightly more abstract and composite representation. In an image recognition application, the raw input may be a [[Matrix (mathematics)|matrix]] of pixels; the first representational layer may abstract the pixels and encode edges; the second layer may compose and encode arrangements of edges; the third layer may encode a nose and eyes; and the fourth layer may recognize that the image contains a face. Importantly, a deep learning process can learn which features to optimally place in which level ''on its own''. This does not eliminate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction.<ref name="BENGIO2012">{{cite journal|last2=Courville|first2=A.|last3=Vincent|first3=P.|year=2013|title=Representation Learning: A Review and New Perspectives|journal=IEEE Transactions on Pattern Analysis and Machine Intelligence|volume=35|issue=8|pages=1798–1828|arxiv=1206.5538|doi=10.1109/tpami.2013.50|pmid=23787338|last1=Bengio|first1=Y.|s2cid=393948}}</ref><ref>{{cite journal|last1=LeCun|first1=Yann|last2=Bengio|first2=Yoshua|last3=Hinton|first3=Geoffrey|s2cid=3074096|title=Deep learning|journal=Nature|date=28 May 2015|volume=521|issue=7553|pages=436–444|doi=10.1038/nature14539|pmid=26017442|bibcode=2015Natur.521..436L}}</ref>
The word "deep" in "deep learning" refers to the number of layers through which the data is transformed. More precisely, deep learning systems have a substantial ''credit assignment path'' (CAP) depth. The CAP is the chain of transformations from input to output. CAPs describe potentially causal connections between input and output. For a [[feedforward neural network]], the depth of the CAPs is that of the network and is the number of hidden layers plus one (as the output layer is also parameterized). For [[recurrent neural network]]s, in which a signal may propagate through a layer more than once, the CAP depth is potentially unlimited.<ref name="SCHIDHUB" /> No universally agreed-upon threshold of depth divides shallow learning from deep learning, but most researchers agree that deep learning involves CAP depth higher than 2. CAP of depth 2 has been shown to be a universal approximator in the sense that it can emulate any function.<ref>{{Cite book|url=https://books.google.com/books?id=9CqQDwAAQBAJ&pg=PA15|title=Human Behavior and Another Kind in Consciousness: Emerging Research and Opportunities: Emerging Research and Opportunities|last=Shigeki|first=Sugiyama|date=12 April 2019|publisher=IGI Global|isbn=978-1-5225-8218-2|language=en}}</ref> Beyond that, more layers do not add to the function approximator ability of the network. Deep models (CAP > 2) are able to extract better features than shallow models and hence, extra layers help in learning the features effectively.
Deep learning architectures can be constructed with a [[greedy algorithm|greedy]] layer-by-layer method.<ref name=BENGIO2007>{{cite conference | first1=Yoshua | last1=Bengio | first2=Pascal | last2=Lamblin | first3=Dan | last3=Popovici | first4=Hugo | last4=Larochelle | title=Greedy layer-wise training of deep networks | year=2007 | url=http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf | conference=Advances in neural information processing systems | pages=153–160 | access-date=2019-10-06 | archive-date=2019-10-20 | archive-url=https://web.archive.org/web/20191020195638/http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf | url-status=live }}</ref> Deep learning helps to disentangle these abstractions and pick out which features improve performance.<ref name="BENGIO2012" />
For [[supervised learning]] tasks, deep learning methods eliminate [[feature engineering]], by translating the data into compact intermediate representations akin to [[Principal Component Analysis|principal components]], and derive layered structures that remove redundancy in representation.
Deep learning algorithms can be applied to unsupervised learning tasks. This is an important benefit because unlabeled data are more abundant than the labeled data. Examples of deep structures that can be trained in an unsupervised manner are [[deep belief network]]s.<ref name="BENGIO2012" /><ref name="SCHOLARDBNS">{{cite journal | last1 = Hinton | first1 = G.E. | year = 2009| title = Deep belief networks | journal = Scholarpedia | volume = 4 | issue = 5| page = 5947 | doi=10.4249/scholarpedia.5947| bibcode = 2009SchpJ...4.5947H| doi-access = free }}</ref>
== Interpretations ==
Deep neural networks are generally interpreted in terms of the [[universal approximation theorem]]<ref name=cyb>{{cite journal | last1 = Cybenko | year = 1989 | title = Approximations by superpositions of sigmoidal functions | url = http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf | journal = [[Mathematics of Control, Signals, and Systems]] | volume = 2 | issue = 4 | pages = 303–314 | doi = 10.1007/bf02551274 | s2cid = 3958369 | url-status = dead | archive-url = https://web.archive.org/web/20151010204407/http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf | archive-date = 10 October 2015 }}</ref><ref name=horn>{{cite journal | last1 = Hornik | first1 = Kurt | year = 1991 | title = Approximation Capabilities of Multilayer Feedforward Networks | journal = Neural Networks | volume = 4 | issue = 2| pages = 251–257 | doi=10.1016/0893-6080(91)90009-t}}</ref><ref name="Haykin, Simon 1998">{{cite book|first=Simon S. |last=Haykin|title=Neural Networks: A Comprehensive Foundation|url={{google books |plainurl=y |id=bX4pAQAAMAAJ}}|year=1999|publisher=Prentice Hall|isbn=978-0-13-273350-2}}</ref><ref name="Hassoun, M. 1995 p. 48">{{cite book|first=Mohamad H. |last=Hassoun|title=Fundamentals of Artificial Neural Networks|url={{google books |plainurl=y |id=Otk32Y3QkxQC|page=48}}|year=1995|publisher=MIT Press|isbn=978-0-262-08239-6|page=48}}</ref><ref name=ZhouLu>Lu, Z., Pu, H., Wang, F., Hu, Z., & Wang, L. (2017). [http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width The Expressive Power of Neural Networks: A View from the Width] {{Webarchive|url=https://web.archive.org/web/20190213005539/http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width |date=2019-02-13 }}. Neural Information Processing Systems, 6231-6239.</ref> or [[Bayesian inference|probabilistic inference]].<ref>{{cite journal |last1=Orhan |first1=A. E. |last2=Ma |first2=W. J. |date=2017 |title=Efficient probabilistic inference in generic neural networks trained with non-probabilistic feedback |journal=Nature Communications |volume=8 |issue=1 |pages=138 | pmid=28743932 | doi=10.1038/s41467-017-00181-8|pmc=5527101 |bibcode=2017NatCo...8..138O | doi-access=free}}</ref><ref name="BOOK2014" /><ref name="BENGIO2012" /><ref name="SCHIDHUB">{{cite journal|last=Schmidhuber|first=J.|s2cid=11715509|year=2015|title=Deep Learning in Neural Networks: An Overview|journal=Neural Networks|volume=61|pages=85–117|arxiv=1404.7828|doi=10.1016/j.neunet.2014.09.003|pmid=25462637}}</ref><ref name = MURPHY>{{cite book|first=Kevin P. |last=Murphy|title=Machine Learning: A Probabilistic Perspective|url={{google books |plainurl=y |id=NZP6AQAAQBAJ}}|date=24 August 2012|publisher=MIT Press|isbn=978-0-262-01802-9}}</ref>
The classic universal approximation theorem concerns the capacity of [[feedforward neural networks]] with a single hidden layer of finite size to approximate [[continuous functions]].<ref name="cyb"/><ref name="horn"/><ref name="Haykin, Simon 1998"/><ref name="Hassoun, M. 1995 p. 48"/> In 1989, the first proof was published by [[George Cybenko]] for [[sigmoid function|sigmoid]] activation functions<ref name="cyb" /> and was generalised to feed-forward multi-layer architectures in 1991 by Kurt Hornik.<ref name="horn" /> Recent work also showed that universal approximation also holds for non-bounded activation functions such as the rectified linear unit.<ref name=sonoda17>{{cite journal | last1 = Sonoda | first1 = Sho | last2=Murata | first2=Noboru | s2cid = 12149203 | year = 2017 | title = Neural network with unbounded activation functions is universal approximator | journal = Applied and Computational Harmonic Analysis | volume = 43 | issue = 2 | pages = 233–268 | doi = 10.1016/j.acha.2015.12.005| arxiv = 1505.03654 }}</ref>
The universal approximation theorem for [[deep neural network]]s concerns the capacity of networks with bounded width but the depth is allowed to grow. Lu et al.<ref name=ZhouLu/> proved that if the width of a [[deep neural network]] with [[ReLU]] activation is strictly larger than the input dimension, then the network can approximate any [[Lebesgue integration|Lebesgue integrable function]]; If the width is smaller or equal to the input dimension, then a [[deep neural network]] is not a universal approximator.
The [[probabilistic]] interpretation<ref name="MURPHY" /> derives from the field of [[machine learning]]. It features inference,<ref name="BOOK2014" /><ref name="BENGIODEEP" /><ref name="BENGIO2012" /><ref name="SCHIDHUB" /><ref name="SCHOLARDBNS" /><ref name="MURPHY" /> as well as the [[optimization]] concepts of [[training]] and [[test (assessment)|testing]], related to fitting and [[generalization]], respectively. More specifically, the probabilistic interpretation considers the activation nonlinearity as a [[cumulative distribution function]].<ref name="MURPHY" /> The probabilistic interpretation led to the introduction of [[dropout (neural networks)|dropout]] as [[Regularization (mathematics)|regularizer]] in neural networks. The probabilistic interpretation was introduced by researchers including [[John Hopfield|Hopfield]], [[Bernard Widrow|Widrow]] and [[Kumpati S. Narendra|Narendra]] and popularized in surveys such as the one by [[Christopher Bishop|Bishop]].<ref name="prml">{{cite book|title=Pattern Recognition and Machine Learning|author=Bishop, Christopher M.|year=2006|publisher=Springer|url=http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf|isbn=978-0-387-31073-2|access-date=2017-08-06|archive-date=2017-01-11|archive-url=https://web.archive.org/web/20170111005101/http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf|url-status=live}}</ref>
== History ==
Some sources point out that [[Frank Rosenblatt]] developed and explored all of the basic ingredients of the deep learning systems of today.<ref name="Who Is the Father of Deep Learning?">{{cite book |chapter-url=https://ieeexplore.ieee.org/document/9070967 |chapter=Who Is the Father of Deep Learning? |publisher=IEEE |doi=10.1109/CSCI49370.2019.00067 |accessdate=31 May 2021|title=2019 International Conference on Computational Science and Computational Intelligence (CSCI) |year=2019 |last1=Tappert |first1=Charles C. |pages=343–348 |isbn=978-1-7281-5584-5 |s2cid=216043128 }}</ref> He described it in his book "Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms", published by Cornell Aeronautical Laboratory, Inc., Cornell University in 1962.
The first general, working learning algorithm for supervised, deep, feedforward, multilayer [[perceptron]]s was published by [[Alexey Ivakhnenko]] and Lapa in 1967.<ref name="ivak1965">{{cite book|first1=A. G. |last1=Ivakhnenko |first2=V. G. |last2=Lapa |title=Cybernetics and Forecasting Techniques|url={{google books |plainurl=y |id=rGFgAAAAMAAJ}}|year=1967|publisher=American Elsevier Publishing Co.|isbn=978-0-444-00020-0}}</ref> A 1971 paper described a deep network with eight layers trained by the [[group method of data handling]].<ref name="ivak1971">{{Cite journal|last=Ivakhnenko|first=Alexey|date=1971|title=Polynomial theory of complex systems|url=http://gmdh.net/articles/history/polynomial.pdf|journal=IEEE Transactions on Systems, Man and Cybernetics|pages=364–378|doi=10.1109/TSMC.1971.4308320|volume=SMC-1|issue=4|access-date=2019-11-05|archive-date=2017-08-29|archive-url=https://web.archive.org/web/20170829230621/http://www.gmdh.net/articles/history/polynomial.pdf|url-status=live}}</ref> Other deep learning working architectures, specifically those built for [[computer vision]], began with the [[Neocognitron]] introduced by [[Kunihiko Fukushima]] in 1980.<ref name="FUKU1980">{{cite journal | last1 = Fukushima | first1 = K. | year = 1980 | title = Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position | journal = Biol. Cybern. | volume = 36 | issue = 4| pages = 193–202 | doi=10.1007/bf00344251 | pmid=7370364| s2cid = 206775608 }}</ref>
The term ''Deep Learning'' was introduced to the machine learning community by [[Rina Dechter]] in 1986,<ref name="dechter1986">[[Rina Dechter]] (1986). Learning while searching in constraint-satisfaction problems. University of California, Computer Science Department, Cognitive Systems Laboratory.[https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems Online] {{Webarchive|url=https://web.archive.org/web/20160419054654/https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems |date=2016-04-19 }}</ref> and to [[Artificial Neural Networks|artificial neural networks]] by Igor Aizenberg and colleagues in 2000, in the context of [[Boolean network|Boolean]] threshold neurons.<ref name="aizenberg2000">Igor Aizenberg, Naum N. Aizenberg, Joos P.L. Vandewalle (2000). Multi-Valued and Universal Binary Neurons: Theory, Learning and Applications. Springer Science & Business Media.</ref><ref>Co-evolving recurrent neurons learn deep memory POMDPs. Proc. GECCO, Washington, D. C., pp. 1795-1802, ACM Press, New York, NY, USA, 2005.</ref>
In 1989, [[Yann LeCun]] et al. applied the standard [[backpropagation]] algorithm, which had been around as the reverse mode of [[automatic differentiation]] since 1970,<ref name="lin1970">[[Seppo Linnainmaa]] (1970). The representation of the cumulative rounding error of an algorithm as a Taylor expansion of the local rounding errors. Master's Thesis (in Finnish), Univ. Helsinki, 6-7.</ref><ref name="grie2012">{{Cite journal|last=Griewank|first=Andreas|date=2012|title=Who Invented the Reverse Mode of Differentiation?|url=http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf|journal=Documenta Mathematica|issue=Extra Volume ISMP|pages=389–400|access-date=11 June 2017|archive-url=https://web.archive.org/web/20170721211929/http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf|archive-date=21 July 2017|url-status=dead}}</ref><ref name="WERBOS1974">{{Cite journal|last=Werbos|first=P.|date=1974|title=Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences |url=https://www.researchgate.net/publication/35657389 |journal=Harvard University |access-date=12 June 2017}}</ref><ref name="werbos1982">{{Cite book|chapter-url=ftp://ftp.idsia.ch/pub/juergen/habilitation.pdf|title=System modeling and optimization|last=Werbos|first=Paul|publisher=Springer|year=1982|pages=762–770|chapter=Applications of advances in nonlinear sensitivity analysis}}</ref> to a deep neural network with the purpose of [[Handwriting recognition|recognizing handwritten ZIP code]]s on mail. While the algorithm worked, training required 3 days.<ref name="LECUN1989">LeCun ''et al.'', "Backpropagation Applied to Handwritten Zip Code Recognition," ''Neural Computation'', 1, pp. 541–551, 1989.</ref>
Independently in 1988, Wei Zhang et al. applied the backpropagation algorithm to a convolutional neural network (a simplified Neocognitron by keeping only the convolutional interconnections between the image feature layers and the last fully connected layer) for alphabets recognition and also proposed an implementation of the CNN with an optical computing system.<ref name="wz1988">{{cite journal |last=Zhang |first=Wei |date=1988 |title=Shift-invariant pattern recognition neural network and its optical architecture |url=https://drive.google.com/file/d/1nN_5odSG_QVae54EsQN_qSz-0ZsX6wA0/view?usp=sharing |journal=Proceedings of Annual Conference of the Japan Society of Applied Physics}}</ref><ref name="wz1990">{{cite journal |last=Zhang |first=Wei |date=1990 |title=Parallel distributed processing model with local space-invariant interconnections and its optical architecture |url=https://drive.google.com/file/d/0B65v6Wo67Tk5ODRzZmhSR29VeDg/view?usp=sharing |journal=Applied Optics |volume=29 |issue=32 |pages=4790–7 |doi=10.1364/AO.29.004790 |pmid=20577468 |bibcode=1990ApOpt..29.4790Z}}</ref> Subsequently, Wei Zhang, et al. modified the model by removing the last fully connected layer and applied it for medical image object segmentation in 1991<ref>{{cite journal |last=Zhang |first=Wei |date=1991 |title=Image processing of human corneal endothelium based on a learning network |url=https://drive.google.com/file/d/0B65v6Wo67Tk5cm5DTlNGd0NPUmM/view?usp=sharing |journal=Applied Optics |volume=30 |issue=29 |pages=4211–7 |doi=10.1364/AO.30.004211 |pmid=20706526 |bibcode=1991ApOpt..30.4211Z}}</ref> and breast cancer detection in mammograms in 1994.<ref>{{cite journal |last=Zhang |first=Wei |date=1994 |title=Computerized detection of clustered microcalcifications in digital mammograms using a shift-invariant artificial neural network |url=https://drive.google.com/file/d/0B65v6Wo67Tk5Ml9qeW5nQ3poVTQ/view?usp=sharing |journal=Medical Physics |volume=21 |issue=4 |pages=517–24 |doi=10.1118/1.597177 |pmid=8058017 |bibcode=1994MedPh..21..517Z}}</ref>
In 1994, André de Carvalho, together with Mike Fairhurst and David Bisset, published experimental results of a multi-layer boolean neural network, also known as a weightless neural network, composed of a 3-layers self-organising feature extraction neural network module (SOFT) followed by a multi-layer classification neural network module (GSN), which were independently trained. Each layer in the feature extraction module extracted features with growing complexity regarding the previous layer.<ref>{{Cite journal |title=An integrated Boolean neural network for pattern classification |journal=Pattern Recognition Letters |date=8 August 1994 |pages=807–813 |volume=15 |issue=8 |doi=10.1016/0167-8655(94)90009-4 |first=Andre C. L. F. |last1=de Carvalho |first2 = Mike C. |last2=Fairhurst |first3=David |last3 = Bisset|bibcode=1994PaReL..15..807D }}</ref>
In 1995, [[Brendan Frey]] demonstrated that it was possible to train (over two days) a network containing six fully connected layers and several hundred hidden units using the [[wake-sleep algorithm]], co-developed with [[Peter Dayan]] and [[Geoffrey Hinton|Hinton]].<ref>{{Cite journal|title = The wake-sleep algorithm for unsupervised neural networks |journal = Science|date = 26 May 1995|pages = 1158–1161|volume = 268|issue = 5214|doi = 10.1126/science.7761831|pmid = 7761831|first1 = Geoffrey E.|last1 = Hinton|first2 = Peter|last2 = Dayan|first3 = Brendan J.|last3 = Frey|first4 = Radford|last4 = Neal|bibcode = 1995Sci...268.1158H}}</ref> Many factors contribute to the slow speed, including the [[vanishing gradient problem]] analyzed in 1991 by [[Sepp Hochreiter]].<ref name="HOCH1991">S. Hochreiter., "[http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf Untersuchungen zu dynamischen neuronalen Netzen] {{Webarchive|url=https://web.archive.org/web/20150306075401/http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf |date=2015-03-06 }}," ''Diploma thesis. Institut f. Informatik, Technische Univ. Munich. Advisor: J. Schmidhuber'', 1991.</ref><ref name="HOCH2001">{{cite book|chapter-url={{google books |plainurl=y |id=NWOcMVA64aAC}}|title=A Field Guide to Dynamical Recurrent Networks|last=Hochreiter|first=S.|display-authors=etal|date=15 January 2001|publisher=John Wiley & Sons|isbn=978-0-7803-5369-5|chapter=Gradient flow in recurrent nets: the difficulty of learning long-term dependencies|editor-last2=Kremer|editor-first2=Stefan C.|editor-first1=John F.|editor-last1=Kolen}}</ref>
Since 1997, Sven Behnke extended the feed-forward hierarchical convolutional approach in the Neural Abstraction Pyramid<ref>{{cite book
| last = Behnke | first = Sven
| doi = 10.1007/b11963
| isbn = 3-540-40722-7
| publisher = Springer
| series = Lecture Notes in Computer Science
| title = Hierarchical Neural Networks for Image Interpretation
| volume = 2766
| year = 2003| s2cid = 1304548
}}</ref> by lateral and backward connections in order to flexibly incorporate context into decisions and iteratively resolve local ambiguities.
Simpler models that use task-specific handcrafted features such as [[Gabor filter]]s and [[support vector machine]]s (SVMs) were a popular choice in the 1990s and 2000s, because of [[artificial neural network]]'s (ANN) computational cost and a lack of understanding of how the brain wires its biological networks.
Both shallow and deep learning (e.g., recurrent nets) of ANNs have been explored for many years.<ref>{{Cite journal|last1=Morgan|first1=Nelson|last2=Bourlard |first2=Hervé |last3=Renals |first3=Steve |last4=Cohen |first4=Michael|last5=Franco |first5=Horacio |date=1 August 1993 |title=Hybrid neural network/hidden markov model systems for continuous speech recognition |journal=International Journal of Pattern Recognition and Artificial Intelligence|volume=07|issue=4|pages=899–916|doi=10.1142/s0218001493000455|issn=0218-0014}}</ref><ref name="Robinson1992">{{Cite journal|last=Robinson|first=T.|author-link=Tony Robinson (speech recognition)|date=1992|title=A real-time recurrent error propagation network word recognition system|url=http://dl.acm.org/citation.cfm?id=1895720|journal=ICASSP|pages=617–620|isbn=9780780305328|series=Icassp'92|access-date=2017-06-12|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123135/https://dl.acm.org/doi/10.5555/1895550.1895720|url-status=live}}</ref><ref>{{Cite journal|last1=Waibel|first1=A.|last2=Hanazawa|first2=T.|last3=Hinton|first3=G.|last4=Shikano|first4=K.|last5=Lang|first5=K. J.|date=March 1989|title=Phoneme recognition using time-delay neural networks|journal=IEEE Transactions on Acoustics, Speech, and Signal Processing|volume=37|issue=3|pages=328–339|doi=10.1109/29.21701|issn=0096-3518|hdl=10338.dmlcz/135496|url=http://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf|access-date=2019-09-24|archive-date=2021-04-27|archive-url=https://web.archive.org/web/20210427001446/https://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf|url-status=live}}</ref> These methods never outperformed non-uniform internal-handcrafting Gaussian [[mixture model]]/[[Hidden Markov model]] (GMM-HMM) technology based on generative models of speech trained discriminatively.<ref name="Baker2009">{{cite journal | last1 = Baker | first1 = J. | last2 = Deng | first2 = Li | last3 = Glass | first3 = Jim | last4 = Khudanpur | first4 = S. | last5 = Lee | first5 = C.-H. | last6 = Morgan | first6 = N. | last7 = O'Shaughnessy | first7 = D. | year = 2009 | title = Research Developments and Directions in Speech Recognition and Understanding, Part 1 | journal = IEEE Signal Processing Magazine | volume = 26 | issue = 3| pages = 75–80 | doi=10.1109/msp.2009.932166| bibcode = 2009ISPM...26...75B | hdl = 1721.1/51891 | s2cid = 357467 }}</ref> Key difficulties have been analyzed, including gradient diminishing<ref name="HOCH1991" /> and weak temporal correlation structure in neural predictive models.<ref name="Bengio1991">{{Cite web|url=https://www.researchgate.net/publication/41229141|title=Artificial Neural Networks and their Application to Speech/Sequence Recognition|last=Bengio|first=Y.|date=1991|publisher=McGill University Ph.D. thesis|access-date=2017-06-12|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123131/https://www.researchgate.net/publication/41229141_Artificial_neural_networks_and_their_application_to_sequence_recognition|url-status=live}}</ref><ref name="Deng1994">{{cite journal | last1 = Deng | first1 = L. | last2 = Hassanein | first2 = K. | last3 = Elmasry | first3 = M. | year = 1994 | title = Analysis of correlation structure for a neural predictive model with applications to speech recognition | journal = Neural Networks | volume = 7 | issue = 2| pages = 331–339 | doi=10.1016/0893-6080(94)90027-2}}</ref> Additional difficulties were the lack of training data and limited computing power.
Most [[speech recognition]] researchers moved away from neural nets to pursue generative modeling. An exception was at [[SRI International]] in the late 1990s. Funded by the US government's [[National Security Agency|NSA]] and [[DARPA]], SRI studied deep neural networks in speech and [[speaker recognition]]. The speaker recognition team led by [[Larry Heck]] reported significant success with deep neural networks in speech processing in the 1998 [[National Institute of Standards and Technology]] Speaker Recognition evaluation.<ref name="Doddington2000">{{cite journal | last1 = Doddington | first1 = G. | last2 = Przybocki | first2 = M. | last3 = Martin | first3 = A. | last4 = Reynolds | first4 = D. | year = 2000 | title = The NIST speaker recognition evaluation ± Overview, methodology, systems, results, perspective | journal = Speech Communication | volume = 31 | issue = 2| pages = 225–254 | doi=10.1016/S0167-6393(99)00080-1}}</ref> The SRI deep neural network was then deployed in the Nuance Verifier, representing the first major industrial application of deep learning.<ref name="Heck2000">{{cite journal | last1 = Heck | first1 = L. | last2 = Konig | first2 = Y. | last3 = Sonmez | first3 = M. | last4 = Weintraub | first4 = M. | year = 2000 | title = Robustness to Telephone Handset Distortion in Speaker Recognition by Discriminative Feature Design | journal = Speech Communication | volume = 31 | issue = 2| pages = 181–192 | doi=10.1016/s0167-6393(99)00077-1}}</ref>
The principle of elevating "raw" features over hand-crafted optimization was first explored successfully in the architecture of deep autoencoder on the "raw" spectrogram or linear filter-bank features in the late 1990s,<ref name="Heck2000" /> showing its superiority over the Mel-Cepstral features that contain stages of fixed transformation from spectrograms. The raw features of speech, [[waveform]]s, later produced excellent larger-scale results.<ref>{{Cite web|url=https://www.researchgate.net/publication/266030526|title=Acoustic Modeling with Deep Neural Networks Using Raw Time Signal for LVCSR (PDF Download Available)|website=ResearchGate|access-date=14 June 2017|archive-date=9 May 2021|archive-url=https://web.archive.org/web/20210509123218/https://www.researchgate.net/publication/266030526_Acoustic_Modeling_with_Deep_Neural_Networks_Using_Raw_Time_Signal_for_LVCSR|url-status=live}}</ref>
Many aspects of speech recognition were taken over by a deep learning method called [[long short-term memory]] (LSTM), a recurrent neural network published by Hochreiter and [[Jürgen Schmidhuber|Schmidhuber]] in 1997.<ref name=":0">{{Cite journal|last1=Hochreiter|first1=Sepp|last2=Schmidhuber|first2=Jürgen|s2cid=1915014|date=1 November 1997|title=Long Short-Term Memory|journal=Neural Computation|volume=9|issue=8|pages=1735–1780|doi=10.1162/neco.1997.9.8.1735|issn=0899-7667|pmid=9377276}}</ref> LSTM [[Recurrent neural network|RNN]]s avoid the vanishing gradient problem and can learn "Very Deep Learning" tasks<ref name="SCHIDHUB" /> that require memories of events that happened thousands of discrete time steps before, which is important for speech. In 2003, LSTM started to become competitive with traditional speech recognizers on certain tasks.<ref name="graves2003">{{Cite web|url=Ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf|title=Biologically Plausible Speech Recognition with LSTM Neural Nets|last1=Graves|first1=Alex|last2=Eck|first2=Douglas|date=2003|website=1st Intl. Workshop on Biologically Inspired Approaches to Advanced Information Technology, Bio-ADIT 2004, Lausanne, Switzerland|pages=175–184|last3=Beringer|first3=Nicole|last4=Schmidhuber|first4=Jürgen|access-date=2016-04-09|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123139/ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf|url-status=live}}</ref> Later it was combined with connectionist temporal classification (CTC)<ref name=":1">{{Cite journal|last1=Graves|first1=Alex|last2=Fernández|first2=Santiago|last3=Gomez|first3=Faustino|date=2006|title=Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks|journal=Proceedings of the International Conference on Machine Learning, ICML 2006|pages=369–376|citeseerx=10.1.1.75.6306}}</ref> in stacks of LSTM RNNs.<ref name="fernandez2007keyword">Santiago Fernandez, Alex Graves, and Jürgen Schmidhuber (2007). [https://mediatum.ub.tum.de/doc/1289941/file.pdf An application of recurrent neural networks to discriminative keyword spotting] {{Webarchive|url=https://web.archive.org/web/20181118164457/https://mediatum.ub.tum.de/doc/1289941/file.pdf |date=2018-11-18 }}. Proceedings of ICANN (2), pp. 220–229.</ref> In 2015, Google's speech recognition reportedly experienced a dramatic performance jump of 49% through CTC-trained LSTM, which they made available through [[Google Voice Search]].<ref name="sak2015">{{Cite web|url=http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html|title=Google voice search: faster and more accurate|last1=Sak|first1=Haşim|last2=Senior|first2=Andrew|date=September 2015|last3=Rao|first3=Kanishka|last4=Beaufays|first4=Françoise|last5=Schalkwyk|first5=Johan|access-date=2016-04-09|archive-date=2016-03-09|archive-url=https://web.archive.org/web/20160309191532/http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html|url-status=live}}</ref>
In 2006, publications by [[Geoffrey Hinton|Geoff Hinton]], [[Russ Salakhutdinov|Ruslan Salakhutdinov]], Osindero and [[Yee Whye Teh|Teh]]<ref>{{Cite journal|last=Hinton|first=Geoffrey E.|date=1 October 2007|title=Learning multiple layers of representation|url=http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3|journal=Trends in Cognitive Sciences|volume=11|issue=10|pages=428–434|doi=10.1016/j.tics.2007.09.004|issn=1364-6613|pmid=17921042|s2cid=15066318|access-date=12 June 2017|archive-date=11 October 2013|archive-url=https://web.archive.org/web/20131011071435/http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3|url-status=live}}</ref><ref name=hinton06>{{Cite journal | last1 = Hinton | first1 = G. E. | author-link1 = Geoff Hinton | last2 = Osindero | first2 = S. | last3 = Teh | first3 = Y. W. | doi = 10.1162/neco.2006.18.7.1527 | title = A Fast Learning Algorithm for Deep Belief Nets | journal = [[Neural Computation (journal)|Neural Computation]] | volume = 18 | issue = 7 | pages = 1527–1554 | year = 2006 | pmid = 16764513 | s2cid = 2309950 | url = http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf | access-date = 2011-07-20 | archive-date = 2015-12-23 | archive-url = https://web.archive.org/web/20151223164129/http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf | url-status = live }}</ref><ref name=bengio2012>{{cite arXiv |last=Bengio |first=Yoshua |author-link=Yoshua Bengio |eprint=1206.5533 |title=Practical recommendations for gradient-based training of deep architectures |class=cs.LG|year=2012 }}</ref> showed how a many-layered [[feedforward neural network]] could be effectively pre-trained one layer at a time, treating each layer in turn as an unsupervised [[restricted Boltzmann machine]], then fine-tuning it using supervised [[backpropagation]].<ref name="HINTON2007">G. E. Hinton., "[http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf Learning multiple layers of representation] {{Webarchive|url=https://web.archive.org/web/20180522112408/http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf |date=2018-05-22 }}," ''Trends in Cognitive Sciences'', 11, pp. 428–434, 2007.</ref> The papers referred to ''learning'' for ''deep belief nets.''
Deep learning is part of state-of-the-art systems in various disciplines, particularly computer vision and [[automatic speech recognition]] (ASR). Results on commonly used evaluation sets such as [[TIMIT]] (ASR) and [[MNIST database|MNIST]] ([[image classification]]), as well as a range of large-vocabulary speech recognition tasks have steadily improved.<ref name="HintonDengYu2012" /><ref>{{Cite book |doi=10.1109/ICCCI50826.2021.9402569|isbn=978-1-7281-5875-4|chapter=Non-linear frequency warping using constant-Q transformation for speech emotion recognition|title=2021 International Conference on Computer Communication and Informatics (ICCCI)|pages=1–4|year=2021|last1=Singh|first1=Premjeet|last2=Saha|first2=Goutam|last3=Sahidullah|first3=Md|arxiv=2102.04029|s2cid=231846518}}</ref> [[Convolutional neural network]]s (CNNs) were superseded for ASR by CTC<ref name=":1" /> for [[LSTM]].<ref name=":0" /><ref name="sak2015" /><ref name="sak2014">{{Cite web|url=https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf|title=Long Short-Term Memory recurrent neural network architectures for large scale acoustic modeling|last1=Sak|first1=Hasim|last2=Senior|first2=Andrew|date=2014|last3=Beaufays|first3=Francoise|archive-url=https://web.archive.org/web/20180424203806/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf|archive-date=24 April 2018|url-status=dead}}</ref><ref name="liwu2015">{{cite arXiv |eprint=1410.4281|last1=Li|first1=Xiangang|title=Constructing Long Short-Term Memory based Deep Recurrent Neural Networks for Large Vocabulary Speech Recognition|last2=Wu|first2=Xihong|class=cs.CL|year=2014}}</ref><ref name="zen2015">{{Cite web|url=https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf|title=Unidirectional Long Short-Term Memory Recurrent Neural Network with Recurrent Output Layer for Low-Latency Speech Synthesis|last1=Zen|first1=Heiga|last2=Sak|first2=Hasim|date=2015|website=Google.com|publisher=ICASSP|pages=4470–4474|access-date=2017-06-13|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123113/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf|url-status=live}}</ref> but are more successful in computer vision.
The impact of deep learning in industry began in the early 2000s, when CNNs already processed an estimated 10% to 20% of all the checks written in the US, according to Yann LeCun.<ref name="lecun2016slides">[[Yann LeCun]] (2016). Slides on Deep Learning [https://indico.cern.ch/event/510372/ Online] {{Webarchive|url=https://web.archive.org/web/20160423021403/https://indico.cern.ch/event/510372/ |date=2016-04-23 }}</ref> Industrial applications of deep learning to large-scale speech recognition started around 2010.
The 2009 NIPS Workshop on Deep Learning for Speech Recognition was motivated by the limitations of deep generative models of speech, and the possibility that given more capable hardware and large-scale data sets that deep neural nets (DNN) might become practical. It was believed that pre-training DNNs using generative models of deep belief nets (DBN) would overcome the main difficulties of neural nets. However, it was discovered that replacing pre-training with large amounts of training data for straightforward [[backpropagation]] when using DNNs with large, context-dependent output layers produced error rates dramatically lower than then-state-of-the-art Gaussian mixture model (GMM)/Hidden Markov Model (HMM) and also than more-advanced generative model-based systems.<ref name="HintonDengYu2012">{{cite journal | last1 = Hinton | first1 = G. | last2 = Deng | first2 = L. | last3 = Yu | first3 = D. | last4 = Dahl | first4 = G. | last5 = Mohamed | first5 = A. | last6 = Jaitly | first6 = N. | last7 = Senior | first7 = A. | last8 = Vanhoucke | first8 = V. | last9 = Nguyen | first9 = P. | last10 = Sainath | first10 = T. | last11 = Kingsbury | first11 = B. | year = 2012 | title = Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups| journal = IEEE Signal Processing Magazine | volume = 29 | issue = 6| pages = 82–97 | doi=10.1109/msp.2012.2205597 | bibcode = 2012ISPM...29...82H| s2cid = 206485943 }}</ref> The nature of the recognition errors produced by the two types of systems was characteristically different,<ref name="ReferenceICASSP2013" /> offering technical insights into how to integrate deep learning into the existing highly efficient, run-time speech decoding system deployed by all major speech recognition systems.<ref name="BOOK2014" /><ref name="ReferenceA">{{cite book|last2=Deng|first2=L.|date=2014|title=Automatic Speech Recognition: A Deep Learning Approach (Publisher: Springer)|url={{google books |plainurl=y |id=rUBTBQAAQBAJ}}|isbn=978-1-4471-5779-3|last1=Yu|first1=D.}}</ref><ref>{{cite web|title=Deng receives prestigious IEEE Technical Achievement Award - Microsoft Research|url=https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/|website=Microsoft Research|date=3 December 2015|access-date=16 March 2018|archive-date=16 March 2018|archive-url=https://web.archive.org/web/20180316084821/https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/|url-status=live}}</ref> Analysis around 2009–2010, contrasting the GMM (and other generative speech models) vs. DNN models, stimulated early industrial investment in deep learning for speech recognition,<ref name="ReferenceICASSP2013" /> eventually leading to pervasive and dominant use in that industry. That analysis was done with comparable performance (less than 1.5% in error rate) between discriminative DNNs and generative models.<ref name="HintonDengYu2012" /><ref name="ReferenceICASSP2013">{{cite journal|last2=Hinton|first2=G.|last3=Kingsbury|first3=B.|date=2013|title=New types of deep neural network learning for speech recognition and related applications: An overview (ICASSP)|url=https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf|last1=Deng|first1=L.|journal=|access-date=2017-06-12|archive-date=2017-09-26|archive-url=https://web.archive.org/web/20170926190920/https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf|url-status=live}}</ref><ref name="interspeech2014Keynote">{{Cite web|url=https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing|title=Keynote talk: 'Achievements and Challenges of Deep Learning - From Speech Analysis and Recognition To Language and Multimodal Processing'|last=Li|first=Deng|date=September 2014|website=Interspeech|access-date=2017-06-12|archive-date=2017-09-26|archive-url=https://web.archive.org/web/20170926190732/https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing|url-status=live}}</ref>
In 2010, researchers extended deep learning from TIMIT to large vocabulary speech recognition, by adopting large output layers of the DNN based on context-dependent HMM states constructed by [[decision tree]]s.<ref name="Roles2010">{{cite journal|last1=Yu|first1=D.|last2=Deng|first2=L.|date=2010|title=Roles of Pre-Training and Fine-Tuning in Context-Dependent DBN-HMMs for Real-World Speech Recognition|url=https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/|journal=NIPS Workshop on Deep Learning and Unsupervised Feature Learning|access-date=2017-06-14|archive-date=2017-10-12|archive-url=https://web.archive.org/web/20171012095148/https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/|url-status=live}}</ref><ref>{{Cite journal|last1=Seide|first1=F.|last2=Li|first2=G.|last3=Yu|first3=D.|date=2011|title=Conversational speech transcription using context-dependent deep neural networks|url=https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks|journal=Interspeech|pages=437–440|doi=10.21437/Interspeech.2011-169|access-date=2017-06-14|archive-date=2017-10-12|archive-url=https://web.archive.org/web/20171012095522/https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks/|url-status=live}}</ref><ref>{{Cite journal|last1=Deng|first1=Li|last2=Li|first2=Jinyu|last3=Huang|first3=Jui-Ting|last4=Yao|first4=Kaisheng|last5=Yu|first5=Dong|last6=Seide|first6=Frank|last7=Seltzer|first7=Mike|last8=Zweig|first8=Geoff|last9=He|first9=Xiaodong|date=1 May 2013|title=Recent Advances in Deep Learning for Speech Research at Microsoft|url=https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/|journal=Microsoft Research|access-date=14 June 2017|archive-date=12 October 2017|archive-url=https://web.archive.org/web/20171012044053/https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/|url-status=live}}</ref><ref name="ReferenceA" />
Advances in hardware have driven renewed interest in deep learning. In 2009, [[Nvidia]] was involved in what was called the “big bang” of deep learning, “as deep-learning neural networks were trained with Nvidia [[graphics processing unit]]s (GPUs).”<ref>{{cite web|url=https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/|title=Nvidia CEO bets big on deep learning and VR|date=5 April 2016|publisher=[[Venture Beat]]|access-date=21 April 2017|archive-date=25 November 2020|archive-url=https://web.archive.org/web/20201125202428/https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/|url-status=live}}</ref> That year, [[Andrew Ng]] determined that GPUs could increase the speed of deep-learning systems by about 100 times.<ref>{{cite news|url=https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not|title=From not working to neural networking|newspaper=[[The Economist]]|access-date=2017-08-26|archive-date=2016-12-31|archive-url=https://web.archive.org/web/20161231203934/https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not|url-status=live}}</ref> In particular, GPUs are well-suited for the matrix/vector computations involved in machine learning.<ref name="jung2004">{{cite journal | last1 = Oh | first1 = K.-S. | last2 = Jung | first2 = K. | year = 2004 | title = GPU implementation of neural networks | journal = Pattern Recognition | volume = 37 | issue = 6| pages = 1311–1314 | doi=10.1016/j.patcog.2004.01.013| bibcode = 2004PatRe..37.1311O }}</ref><ref>"[https://www.academia.edu/40135801 A Survey of Techniques for Optimizing Deep Learning on GPUs] {{Webarchive|url=https://web.archive.org/web/20210509123120/https://www.academia.edu/40135801/A_Survey_of_Techniques_for_Optimizing_Deep_Learning_on_GPUs |date=2021-05-09 }}", S. Mittal and S. Vaishay, Journal of Systems Architecture, 2019</ref><ref name="chellapilla2006">{{Citation | first1 = Kumar | last1 = Chellapilla | first2 = Sidd | last2 = Puri | first3 = Patrice | last3 = Simard | title = High performance convolutional neural networks for document processing | url = https://hal.inria.fr/inria-00112631/document | date = 2006 | access-date = 2021-02-14 | archive-date = 2020-05-18 | archive-url = https://web.archive.org/web/20200518193413/https://hal.inria.fr/inria-00112631/document | url-status = live }}</ref> GPUs speed up training algorithms by orders of magnitude, reducing running times from weeks to days.<ref name=":3">{{Cite journal|last1=Cireşan|first1=Dan Claudiu|last2=Meier|first2=Ueli|last3=Gambardella|first3=Luca Maria|last4=Schmidhuber|first4=Jürgen|date=21 September 2010|title=Deep, Big, Simple Neural Nets for Handwritten Digit Recognition|journal=Neural Computation|volume=22|issue=12|pages=3207–3220|doi=10.1162/neco_a_00052|pmid=20858131|issn=0899-7667|arxiv=1003.0358|s2cid=1918673}}</ref><ref>{{Cite journal|last1=Raina|first1=Rajat|last2=Madhavan|first2=Anand|last3=Ng|first3=Andrew Y.|s2cid=392458|date=2009|title=Large-scale Deep Unsupervised Learning Using Graphics Processors|journal=Proceedings of the 26th Annual International Conference on Machine Learning|series=ICML '09|location=New York, NY, USA|publisher=ACM|pages=873–880|doi=10.1145/1553374.1553486|isbn=9781605585161|citeseerx=10.1.1.154.372}}</ref> Further, specialized hardware and algorithm optimizations can be used for efficient processing of deep learning models.<ref name="sze2017">{{cite arXiv
|title= Efficient Processing of Deep Neural Networks: A Tutorial and Survey
|last1=Sze |first1=Vivienne|author1-link=Vivienne Sze
|last2=Chen |first2=Yu-Hsin
|last3=Yang |first3=Tien-Ju
|last4=Emer |first4=Joel
|eprint=1703.09039
|year=2017
|class=cs.CV }}</ref>
=== Deep learning revolution ===
[[File:AI-ML-DL.svg|thumb|How deep learning is a subset of machine learning and how machine learning is a subset of artificial intelligence (AI)]]
In 2012, a team led by George E. Dahl won the "Merck Molecular Activity Challenge" using multi-task deep neural networks to predict the [[biomolecular target]] of one drug.<ref name="MERCK2012">{{Cite web|url=https://kaggle.com/c/MerckActivity|title=Merck Molecular Activity Challenge|website=kaggle.com|access-date=2020-07-16|archive-date=2020-07-16|archive-url=https://web.archive.org/web/20200716190808/https://www.kaggle.com/c/MerckActivity|url-status=live}}</ref><ref name=":5">{{Cite web|url=http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions|title=Multi-task Neural Networks for QSAR Predictions {{!}} Data Science Association|website=www.datascienceassn.org|access-date=14 June 2017|archive-date=30 April 2017|archive-url=https://web.archive.org/web/20170430142049/http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions|url-status=live}}</ref> In 2014, Hochreiter's group used deep learning to detect off-target and toxic effects of environmental chemicals in nutrients, household products and drugs and won the "Tox21 Data Challenge" of [[NIH]], [[FDA]] and [[National Center for Advancing Translational Sciences|NCATS]].<ref name="TOX21">"Toxicology in the 21st century Data Challenge"</ref><ref name="TOX21Data">{{cite web|url=https://tripod.nih.gov/tox21/challenge/leaderboard.jsp|title=NCATS Announces Tox21 Data Challenge Winners|access-date=2015-03-05|archive-date=2015-09-08|archive-url=https://web.archive.org/web/20150908025122/https://tripod.nih.gov/tox21/challenge/leaderboard.jsp|url-status=live}}</ref><ref name=":11">{{cite web|url=http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html|title=NCATS Announces Tox21 Data Challenge Winners|archive-url=https://web.archive.org/web/20150228225709/http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html|archive-date=28 February 2015|url-status=dead|access-date=5 March 2015}}</ref>
Significant additional impacts in image or object recognition were felt from 2011 to 2012. Although CNNs trained by [[backpropagation]] had been around for decades, and GPU implementations of NNs for years, including CNNs, fast implementations of CNNs on GPUs were needed to progress on computer vision.<ref name="jung2004" /><ref name="chellapilla2006" /><ref name="LECUN1989" /><ref name=":6">{{Cite journal|last1=Ciresan|first1=D. C.|last2=Meier|first2=U.|last3=Masci|first3=J.|last4=Gambardella|first4=L. M.|last5=Schmidhuber|first5=J.|date=2011|title=Flexible, High Performance Convolutional Neural Networks for Image Classification|url=http://ijcai.org/papers11/Papers/IJCAI11-210.pdf|journal=International Joint Conference on Artificial Intelligence|doi=10.5591/978-1-57735-516-8/ijcai11-210|access-date=2017-06-13|archive-date=2014-09-29|archive-url=https://web.archive.org/web/20140929094040/http://ijcai.org/papers11/Papers/IJCAI11-210.pdf|url-status=live}}</ref><ref name="SCHIDHUB" /> In 2011, this approach achieved for the first time superhuman performance in a visual pattern recognition contest. Also in 2011, it won the ICDAR Chinese handwriting contest, and in May 2012, it won the ISBI image segmentation contest.<ref name=":8">{{Cite book|url=http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf|title=Advances in Neural Information Processing Systems 25|last1=Ciresan|first1=Dan|last2=Giusti|first2=Alessandro|last3=Gambardella|first3=Luca M.|last4=Schmidhuber|first4=Juergen|date=2012|publisher=Curran Associates, Inc.|editor-last=Pereira|editor-first=F.|pages=2843–2851|editor-last2=Burges|editor-first2=C. J. C.|editor-last3=Bottou|editor-first3=L.|editor-last4=Weinberger|editor-first4=K. Q.|access-date=2017-06-13|archive-date=2017-08-09|archive-url=https://web.archive.org/web/20170809081713/http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf|url-status=live}}</ref> Until 2011, CNNs did not play a major role at computer vision conferences, but in June 2012, a paper by Ciresan et al. at the leading conference CVPR<ref name=":9" /> showed how max-pooling CNNs on GPU can dramatically improve many vision benchmark records. In October 2012, a similar system by Krizhevsky et al.<ref name="krizhevsky2012" /> won the large-scale [[ImageNet competition]] by a significant margin over shallow machine learning methods. In November 2012, Ciresan et al.'s system also won the ICPR contest on analysis of large medical images for cancer detection, and in the following year also the MICCAI Grand Challenge on the same topic.<ref name="ciresan2013miccai">{{Cite journal|last1=Ciresan|first1=D.|last2=Giusti|first2=A.|last3=Gambardella|first3=L.M.|last4=Schmidhuber|first4=J.|date=2013|title=Mitosis Detection in Breast Cancer Histology Images using Deep Neural Networks|journal=Proceedings MICCAI|volume=7908|issue=Pt 2|pages=411–418|doi=10.1007/978-3-642-40763-5_51|pmid=24579167|series=Lecture Notes in Computer Science|isbn=978-3-642-38708-1}}</ref> In 2013 and 2014, the error rate on the ImageNet task using deep learning was further reduced, following a similar trend in large-scale speech recognition.
Image classification was then extended to the more challenging task of [[Automatic image annotation|generating descriptions]] (captions) for images, often as a combination of CNNs and LSTMs.<ref name="1411.4555">{{cite arXiv |eprint=1411.4555|last1=Vinyals|first1=Oriol|title=Show and Tell: A Neural Image Caption Generator|last2=Toshev|first2=Alexander|last3=Bengio|first3=Samy|last4=Erhan|first4=Dumitru|class=cs.CV|year=2014}}.</ref><ref name="1411.4952">{{cite arXiv |eprint=1411.4952|last1=Fang|first1=Hao|title=From Captions to Visual Concepts and Back|last2=Gupta|first2=Saurabh|last3=Iandola|first3=Forrest|last4=Srivastava|first4=Rupesh|last5=Deng|first5=Li|last6=Dollár|first6=Piotr|last7=Gao|first7=Jianfeng|last8=He|first8=Xiaodong|last9=Mitchell|first9=Margaret|last10=Platt|first10=John C|last11=Lawrence Zitnick|first11=C|last12=Zweig|first12=Geoffrey|class=cs.CV|year=2014}}.</ref><ref name="1411.2539">{{cite arXiv |eprint=1411.2539|last1=Kiros|first1=Ryan|title=Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models|last2=Salakhutdinov|first2=Ruslan|last3=Zemel|first3=Richard S|class=cs.LG|year=2014}}.</ref>
Some researchers state that the October 2012 ImageNet victory anchored the start of a "deep learning revolution" that has transformed the AI industry.<ref>{{cite news|title=Why Deep Learning Is Suddenly Changing Your Life|url=http://fortune.com/ai-artificial-intelligence-deep-machine-learning/|access-date=13 April 2018|work=Fortune|date=2016|archive-date=14 April 2018|archive-url=https://web.archive.org/web/20180414031925/http://fortune.com/ai-artificial-intelligence-deep-machine-learning/|url-status=live}}</ref>
In March 2019, [[Yoshua Bengio]], [[Geoffrey Hinton]] and [[Yann LeCun]] were awarded the [[Turing Award]] for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing.
== Neural networks ==
=== Artificial neural networks ===
{{Main|Artificial neural network}}
'''Artificial neural networks''' ('''ANNs''') or '''[[Connectionism|connectionist]] systems''' are computing systems inspired by the [[biological neural network]]s that constitute animal brains. Such systems learn (progressively improve their ability) to do tasks by considering examples, generally without task-specific programming. For example, in image recognition, they might learn to identify images that contain cats by analyzing example images that have been manually [[Labeled data|labeled]] as "cat" or "no cat" and using the analytic results to identify cats in other images. They have found most use in applications difficult to express with a traditional computer algorithm using [[rule-based programming]].
An ANN is based on a collection of connected units called [[artificial neuron]]s, (analogous to biological neurons in a [[Brain|biological brain]]). Each connection ([[synapse]]) between neurons can transmit a signal to another neuron. The receiving (postsynaptic) neuron can process the signal(s) and then signal downstream neurons connected to it. Neurons may have state, generally represented by [[real numbers]], typically between 0 and 1. Neurons and synapses may also have a weight that varies as learning proceeds, which can increase or decrease the strength of the signal that it sends downstream.
Typically, neurons are organized in layers. Different layers may perform different kinds of transformations on their inputs. Signals travel from the first (input), to the last (output) layer, possibly after traversing the layers multiple times.
The original goal of the neural network approach was to solve problems in the same way that a human brain would. Over time, attention focused on matching specific mental abilities, leading to deviations from biology such as [[backpropagation]], or passing information in the reverse direction and adjusting the network to reflect that information.
Neural networks have been used on a variety of tasks, including computer vision, [[speech recognition]], [[machine translation]], [[social network]] filtering, [[general game playing|playing board and video games]] and medical diagnosis.
As of 2017, neural networks typically have a few thousand to a few million units and millions of connections. Despite this number being several order of magnitude less than the number of neurons on a human brain, these networks can perform many tasks at a level beyond that of humans (e.g., recognizing faces, or playing "Go"<ref>{{Cite journal|last1=Silver|first1=David|last2=Huang|first2=Aja|last3=Maddison|first3=Chris J.|last4=Guez|first4=Arthur|last5=Sifre|first5=Laurent|last6=Driessche|first6=George van den|last7=Schrittwieser|first7=Julian|last8=Antonoglou|first8=Ioannis|last9=Panneershelvam|first9=Veda|s2cid=515925|date=January 2016|title=Mastering the game of Go with deep neural networks and tree search|journal=Nature|volume=529|issue=7587|pages=484–489|doi=10.1038/nature16961|issn=1476-4687|pmid=26819042|bibcode=2016Natur.529..484S}}</ref> ).
=== Deep neural networks ===
A deep neural network (DNN) is an [[artificial neural network]] (ANN) with multiple layers between the input and output layers.<ref name="BENGIODEEP" /><ref name="SCHIDHUB" /> There are different types of neural networks but they always consist of the same components: neurons, synapses, weights, biases, and functions.<ref name="Nokkada">{{citation |title=A Guide to Deep Learning and Neural Networks |url=https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks |access-date=2020-11-16 |archive-date=2020-11-02 |archive-url=https://web.archive.org/web/20201102151103/https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks |url-status=live }}</ref> These components as a whole function similarly to a human brain, and can be trained like any other ML algorithm.{{Citation needed|date=November 2020}}
For example, a DNN that is trained to recognize dog breeds will go over the given image and calculate the probability that the dog in the image is a certain breed. The user can review the results and select which probabilities the network should display (above a certain threshold, etc.) and return the proposed label. Each mathematical manipulation as such is considered a layer,{{citation needed|date=March 2022}} and complex DNN have many layers, hence the name "deep" networks.
DNNs can model complex non-linear relationships. DNN architectures generate compositional models where the object is expressed as a layered composition of [[Primitive data type|primitives]].<ref>{{Cite journal|last1=Szegedy|first1=Christian|last2=Toshev|first2=Alexander|last3=Erhan|first3=Dumitru|date=2013|title=Deep neural networks for object detection|url=https://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection|journal=Advances in Neural Information Processing Systems|pages=2553–2561|access-date=2017-06-13|archive-date=2017-06-29|archive-url=https://web.archive.org/web/20170629172111/http://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection|url-status=live}}</ref> The extra layers enable composition of features from lower layers, potentially modeling complex data with fewer units than a similarly performing shallow network.<ref name="BENGIODEEP" /> For instance, it was proved that sparse [[multivariate polynomial]]s are exponentially easier to approximate with DNNs than with shallow networks.<ref>{{cite conference|last1=Rolnick|first1=David|last2=Tegmark|first2=Max|date=2018|title=The power of deeper networks for expressing natural functions|url=https://openreview.net/pdf?id=SyProzZAW|conference=ICLR 2018|book-title=International Conference on Learning Representations|access-date=2021-01-05|archive-date=2021-01-07|archive-url=https://web.archive.org/web/20210107183647/https://openreview.net/pdf?id=SyProzZAW|url-status=live}}</ref>
Deep architectures include many variants of a few basic approaches. Each architecture has found success in specific domains. It is not always possible to compare the performance of multiple architectures, unless they have been evaluated on the same data sets.
DNNs are typically feedforward networks in which data flows from the input layer to the output layer without looping back. At first, the DNN creates a map of virtual neurons and assigns random numerical values, or "weights", to connections between them. The weights and inputs are multiplied and return an output between 0 and 1. If the network did not accurately recognize a particular pattern, an algorithm would adjust the weights.<ref>{{Cite news|last=Hof|first=Robert D.|title=Is Artificial Intelligence Finally Coming into Its Own?|work=MIT Technology Review|url=https://www.technologyreview.com/s/513696/deep-learning/|access-date=10 July 2018|archive-url=https://web.archive.org/web/20190331092832/https://www.technologyreview.com/s/513696/deep-learning/|archive-date=31 March 2019}}</ref> That way the algorithm can make certain parameters more influential, until it determines the correct mathematical manipulation to fully process the data.
[[Recurrent neural networks]] (RNNs), in which data can flow in any direction, are used for applications such as [[language model]]ing.<ref name="gers2001">{{cite journal|last1=Gers|first1=Felix A.|last2=Schmidhuber|first2=Jürgen|year=2001|title=LSTM Recurrent Networks Learn Simple Context Free and Context Sensitive Languages|url=http://elartu.tntu.edu.ua/handle/lib/30719|journal=IEEE Transactions on Neural Networks|volume=12|issue=6|pages=1333–1340|doi=10.1109/72.963769|pmid=18249962|access-date=2020-02-25|archive-date=2020-01-26|archive-url=https://web.archive.org/web/20200126045722/http://elartu.tntu.edu.ua/handle/lib/30719|url-status=live}}</ref><ref name="NIPS2014"/><ref name="vinyals2016">{{cite arXiv |eprint=1602.02410|last1=Jozefowicz|first1=Rafal|title=Exploring the Limits of Language Modeling|last2=Vinyals|first2=Oriol|last3=Schuster|first3=Mike|last4=Shazeer|first4=Noam|last5=Wu|first5=Yonghui|class=cs.CL|year=2016}}</ref><ref name="gillick2015">{{cite arXiv |eprint=1512.00103|last1=Gillick|first1=Dan|title=Multilingual Language Processing from Bytes|last2=Brunk|first2=Cliff|last3=Vinyals|first3=Oriol|last4=Subramanya|first4=Amarnag|class=cs.CL|year=2015}}</ref><ref name="MIKO2010">{{Cite journal|last=Mikolov|first=T.|display-authors=etal|date=2010|title=Recurrent neural network based language model|url=http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf|journal=Interspeech|pages=1045–1048|doi=10.21437/Interspeech.2010-343|access-date=2017-06-13|archive-date=2017-05-16|archive-url=https://web.archive.org/web/20170516181940/http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf|url-status=live}}</ref> Long short-term memory is particularly effective for this use.<ref name=":0" /><ref name=":10">{{Cite web|url=https://www.researchgate.net/publication/220320057|title=Learning Precise Timing with LSTM Recurrent Networks (PDF Download Available)|website=ResearchGate|access-date=13 June 2017|archive-date=9 May 2021|archive-url=https://web.archive.org/web/20210509123147/https://www.researchgate.net/publication/220320057_Learning_Precise_Timing_with_LSTM_Recurrent_Networks|url-status=live}}</ref>
[[Convolutional neural network|Convolutional deep neural networks (CNNs)]] are used in computer vision.<ref name="LECUN86">{{cite journal |last1=LeCun |first1=Y. |display-authors=etal |year= 1998|title=Gradient-based learning applied to document recognition |journal=Proceedings of the IEEE |volume=86 |issue=11 |pages=2278–2324 |doi=10.1109/5.726791|s2cid=14542261 |url=http://elartu.tntu.edu.ua/handle/lib/38369 }}</ref> CNNs also have been applied to [[acoustic model]]ing for automatic speech recognition (ASR).<ref name=":2">{{Cite book |doi=10.1109/icassp.2013.6639347|isbn=978-1-4799-0356-6|chapter=Deep convolutional neural networks for LVCSR|title=2013 IEEE International Conference on Acoustics, Speech and Signal Processing|pages=8614–8618|year=2013|last1=Sainath|first1=Tara N.|last2=Mohamed|first2=Abdel-Rahman|last3=Kingsbury|first3=Brian|last4=Ramabhadran|first4=Bhuvana|s2cid=13816461}}</ref>
==== Challenges ====
As with ANNs, many issues can arise with naively trained DNNs. Two common issues are [[overfitting]] and computation time.
DNNs are prone to overfitting because of the added layers of abstraction, which allow them to model rare dependencies in the training data. [[Regularization (mathematics)|Regularization]] methods such as Ivakhnenko's unit pruning<ref name="ivak1971"/> or [[weight decay]] (<math> \ell_2 </math>-regularization) or [[sparse matrix|sparsity]] (<math> \ell_1 </math>-regularization) can be applied during training to combat overfitting.<ref>{{Cite book |doi=10.1109/icassp.2013.6639349|isbn=978-1-4799-0356-6|arxiv=1212.0901|citeseerx=10.1.1.752.9151|chapter=Advances in optimizing recurrent networks|title=2013 IEEE International Conference on Acoustics, Speech and Signal Processing|pages=8624–8628|year=2013|last1=Bengio|first1=Yoshua|last2=Boulanger-Lewandowski|first2=Nicolas|last3=Pascanu|first3=Razvan|s2cid=12485056}}</ref> Alternatively dropout regularization randomly omits units from the hidden layers during training. This helps to exclude rare dependencies.<ref name="DAHL2013">{{Cite journal|last=Dahl|first=G.|display-authors=etal|date=2013|title=Improving DNNs for LVCSR using rectified linear units and dropout|url=http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf|journal=ICASSP|access-date=2017-06-13|archive-date=2017-08-12|archive-url=https://web.archive.org/web/20170812140509/http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf|url-status=live}}</ref> Finally, data can be augmented via methods such as cropping and rotating such that smaller training sets can be increased in size to reduce the chances of overfitting.<ref>{{Cite web|url=https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation|title=Data Augmentation - deeplearning.ai {{!}} Coursera|website=Coursera|access-date=30 November 2017|archive-date=1 December 2017|archive-url=https://web.archive.org/web/20171201032606/https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation|url-status=live}}</ref>
DNNs must consider many training parameters, such as the size (number of layers and number of units per layer), the [[learning rate]], and initial weights. [[Hyperparameter optimization#Grid search|Sweeping through the parameter space]] for optimal parameters may not be feasible due to the cost in time and computational resources. Various tricks, such as batching (computing the gradient on several training examples at once rather than individual examples)<ref name="RBMTRAIN">{{Cite journal|last=Hinton|first=G. E.|date=2010|title=A Practical Guide to Training Restricted Boltzmann Machines|url=https://www.researchgate.net/publication/221166159|journal=Tech. Rep. UTML TR 2010-003|access-date=2017-06-13|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123211/https://www.researchgate.net/publication/221166159_A_brief_introduction_to_Weightless_Neural_Systems|url-status=live}}</ref> speed up computation. Large processing capabilities of many-core architectures (such as GPUs or the Intel Xeon Phi) have produced significant speedups in training, because of the suitability of such processing architectures for the matrix and vector computations.<ref>{{cite book|last1=You|first1=Yang|title=Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis on - SC '17|pages=1–12|last2=Buluç|first2=Aydın|last3=Demmel|first3=James|chapter=Scaling deep learning on GPU and knights landing clusters|chapter-url=https://dl.acm.org/citation.cfm?doid=3126908.3126912|publisher=SC '17, ACM|date=November 2017|access-date=5 March 2018|doi=10.1145/3126908.3126912|isbn=9781450351140|s2cid=8869270|url=http://www.escholarship.org/uc/item/6ch40821|archive-date=29 July 2020|archive-url=https://web.archive.org/web/20200729133850/https://escholarship.org/uc/item/6ch40821|url-status=live}}</ref><ref>{{cite journal|last1=Viebke|first1=André|last2=Memeti|first2=Suejb|last3=Pllana|first3=Sabri|last4=Abraham|first4=Ajith|s2cid=14135321|title=CHAOS: a parallelization scheme for training convolutional neural networks on Intel Xeon Phi|journal=The Journal of Supercomputing|volume=75|pages=197–227|doi=10.1007/s11227-017-1994-x|arxiv=1702.07908|bibcode=2017arXiv170207908V|year=2019}}</ref>
Alternatively, engineers may look for other types of neural networks with more straightforward and convergent training algorithms. CMAC ([[cerebellar model articulation controller]]) is one such kind of neural network. It doesn't require learning rates or randomized initial weights for CMAC. The training process can be guaranteed to converge in one step with a new batch of data, and the computational complexity of the training algorithm is linear with respect to the number of neurons involved.<ref name=Qin1>Ting Qin, et al. "A learning algorithm of CMAC based on RLS." Neural Processing Letters 19.1 (2004): 49-61.</ref><ref name=Qin2>Ting Qin, et al. "[http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf Continuous CMAC-QRLS and its systolic array] {{Webarchive|url=https://web.archive.org/web/20181118122850/http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf |date=2018-11-18 }}." Neural Processing Letters 22.1 (2005): 1-16.</ref>
== Hardware ==
Since the 2010s, advances in both machine learning algorithms and [[computer hardware]] have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.<ref>{{cite web|last1=Research|first1=AI|title=Deep Neural Networks for Acoustic Modeling in Speech Recognition|url=http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/|website=airesearch.com|access-date=23 October 2015|date=23 October 2015|archive-date=1 February 2016|archive-url=https://web.archive.org/web/20160201033801/http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/|url-status=live}}</ref> By 2019, graphic processing units ([[GPU]]s), often with AI-specific enhancements, had displaced CPUs as the dominant method of training large-scale commercial cloud AI.<ref>{{cite news |title=GPUs Continue to Dominate the AI Accelerator Market for Now |url=https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475 |access-date=11 June 2020 |work=InformationWeek |date=December 2019 |language=en |archive-date=10 June 2020 |archive-url=https://web.archive.org/web/20200610094310/https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475 |url-status=live }}</ref> [[OpenAI]] estimated the hardware computation used in the largest deep learning projects from AlexNet (2012) to AlphaZero (2017), and found a 300,000-fold increase in the amount of computation required, with a doubling-time trendline of 3.4 months.<ref>{{cite news |last1=Ray |first1=Tiernan |title=AI is changing the entire nature of computation |url=https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/ |access-date=11 June 2020 |work=ZDNet |date=2019 |language=en |archive-date=25 May 2020 |archive-url=https://web.archive.org/web/20200525144635/https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/ |url-status=live }}</ref><ref>{{cite web |title=AI and Compute |url=https://openai.com/blog/ai-and-compute/ |website=OpenAI |access-date=11 June 2020 |language=en |date=16 May 2018 |archive-date=17 June 2020 |archive-url=https://web.archive.org/web/20200617200602/https://openai.com/blog/ai-and-compute/ |url-status=live }}</ref>
Special [[electronic circuit]]s called [[deep learning processor]]s were designed to speed up deep learning algorithms. Deep learning processors include neural processing units (NPUs) in [[Huawei]] cellphones<ref>{{Cite web|url=https://consumer.huawei.com/en/press/news/2017/ifa2017-kirin970/|title=HUAWEI Reveals the Future of Mobile AI at IFA 2017 | HUAWEI Latest News | HUAWEI Global|website=consumer.huawei.com}}</ref> and [[cloud computing]] servers such as [[tensor processing unit]]s (TPU) in the [[Google Cloud Platform]].<ref>{{Cite journal|last1=P|first1=JouppiNorman|last2=YoungCliff|last3=PatilNishant|last4=PattersonDavid|last5=AgrawalGaurav|last6=BajwaRaminder|last7=BatesSarah|last8=BhatiaSuresh|last9=BodenNan|last10=BorchersAl|last11=BoyleRick|date=2017-06-24|title=In-Datacenter Performance Analysis of a Tensor Processing Unit|journal=ACM SIGARCH Computer Architecture News|volume=45|issue=2|pages=1–12|language=EN|doi=10.1145/3140659.3080246|doi-access=free}}</ref> [[Cerebras|Cerebras Systems]] has also built a dedicated system to handle large deep learning models, the CS-2, based on the largest processor in the industry, the second-generation Wafer Scale Engine (WSE-2).<ref>{{Cite web |last=Woodie |first=Alex |date=2021-11-01 |title=Cerebras Hits the Accelerator for Deep Learning Workloads |url=https://www.datanami.com/2021/11/01/cerebras-hits-the-accelerator-for-deep-learning-workloads/ |access-date=2022-08-03 |website=Datanami}}</ref><ref>{{Cite web |date=2021-04-20 |title=Cerebras launches new AI supercomputing processor with 2.6 trillion transistors |url=https://venturebeat.com/2021/04/20/cerebras-systems-launches-new-ai-supercomputing-processor-with-2-6-trillion-transistors/ |access-date=2022-08-03 |website=VentureBeat |language=en-US}}</ref>
Atomically thin [[semiconductors]] are considered promising for energy-efficient deep learning hardware where the same basic device structure is used for both logic operations and data storage.
In 2020, Marega et al. published experiments with a large-area active channel material for developing logic-in-memory devices and circuits based on [[floating-gate]] [[field-effect transistor]]s (FGFETs).<ref name="atomthin">{{cite journal|title=Logic-in-memory based on an atomically thin semiconductor|year=2020|doi=10.1038/s41586-020-2861-0|last1=Marega|first1=Guilherme Migliato|last2=Zhao|first2=Yanfei|last3=Avsar|first3=Ahmet|last4=Wang|first4=Zhenyu|last5=Tripati|first5=Mukesh|last6=Radenovic|first6=Aleksandra|last7=Kis|first7=Anras|journal=Nature|volume=587|issue=2|pages=72–77|pmid=33149289|pmc=7116757|bibcode=2020Natur.587...72M }}</ref>
In 2021, J. Feldmann et al. proposed an integrated [[photonic]] [[hardware accelerator]] for parallel convolutional processing.<ref name="photonic">{{cite journal |title=Parallel convolutional processing using an integrated photonic tensor |year=2021 |doi=10.1038/s41586-020-03070-1 |last1=Feldmann |first1=J. |last2=Youngblood|first2=N. |last3=Karpov |first3=M. | last4=Gehring |first4=H. | display-authors=3 | journal=Nature |volume=589 |issue=2 |pages=52–58|pmid=33408373 |arxiv=2002.00281 |s2cid=211010976 }}</ref> The authors identify two key advantages of integrated photonics over its electronic counterparts: (1) massively parallel data transfer through [[wavelength]] division [[multiplexing]] in conjunction with [[frequency comb]]s, and (2) extremely high data modulation speeds.<ref name="photonic"/> Their system can execute trillions of multiply-accumulate operations per second, indicating the potential of [[Photonic integrated circuit|integrated]] [[photonics]] in data-heavy AI applications.<ref name="photonic"/>
== Applications ==
=== Automatic speech recognition ===
{{Main|Speech recognition}}
Large-scale automatic speech recognition is the first and most convincing successful case of deep learning. LSTM RNNs can learn "Very Deep Learning" tasks<ref name="SCHIDHUB"/> that involve multi-second intervals containing speech events separated by thousands of discrete time steps, where one time step corresponds to about 10 ms. LSTM with forget gates<ref name=":10" /> is competitive with traditional speech recognizers on certain tasks.<ref name="graves2003"/>
The initial success in speech recognition was based on small-scale recognition tasks based on TIMIT. The data set contains 630 speakers from eight major [[dialect]]s of [[American English]], where each speaker reads 10 sentences.<ref name="LDCTIMIT">''TIMIT Acoustic-Phonetic Continuous Speech Corpus'' Linguistic Data Consortium, Philadelphia.</ref> Its small size lets many configurations be tried. More importantly, the TIMIT task concerns phone-sequence recognition, which, unlike word-sequence recognition, allows weak phone [[bigram]] language models. This lets the strength of the acoustic modeling aspects of speech recognition be more easily analyzed. The error rates listed below, including these early results and measured as percent phone error rates (PER), have been summarized since 1991.
{| class="wikitable"
|-
! Method !! Percent phone<br>error rate (PER) (%)
|-
| Randomly Initialized RNN<ref>{{cite journal |last1=Robinson |first1=Tony |author-link=Tony Robinson (speech recognition)|title=Several Improvements to a Recurrent Error Propagation Network Phone Recognition System |journal=Cambridge University Engineering Department Technical Report |date=30 September 1991 |volume=CUED/F-INFENG/TR82 |doi=10.13140/RG.2.2.15418.90567 }}</ref>|| 26.1
|-
| Bayesian Triphone GMM-HMM || 25.6
|-
| Hidden Trajectory (Generative) Model|| 24.8
|-
| Monophone Randomly Initialized DNN|| 23.4
|-
| Monophone DBN-DNN|| 22.4
|-
| Triphone GMM-HMM with BMMI Training|| 21.7
|-
| Monophone DBN-DNN on fbank || 20.7
|-
| Convolutional DNN<ref name="CNN-2014">{{cite journal|last1=Abdel-Hamid|first1=O.|title=Convolutional Neural Networks for Speech Recognition|journal=IEEE/ACM Transactions on Audio, Speech, and Language Processing|date=2014|volume=22|issue=10|pages=1533–1545|doi=10.1109/taslp.2014.2339736|s2cid=206602362|display-authors=etal|url=https://zenodo.org/record/891433|access-date=2018-04-20|archive-date=2020-09-22|archive-url=https://web.archive.org/web/20200922180719/https://zenodo.org/record/891433|url-status=live}}</ref>|| 20.0
|-
| Convolutional DNN w. Heterogeneous Pooling|| 18.7
|-
| Ensemble DNN/CNN/RNN<ref name="EnsembleDL">{{cite journal|last2=Platt|first2=J.|date=2014|title=Ensemble Deep Learning for Speech Recognition|journal=Proc. Interspeech|last1=Deng|first1=L.|s2cid=15641618}}</ref>|| 18.3
|-
| Bidirectional LSTM|| 17.8
|-
| Hierarchical Convolutional Deep Maxout Network<ref name="HCDMM">{{cite journal|last1=Tóth|first1=Laszló|date=2015|title=Phone Recognition with Hierarchical Convolutional Deep Maxout Networks|journal=EURASIP Journal on Audio, Speech, and Music Processing|volume=2015|doi=10.1186/s13636-015-0068-3|s2cid=217950236|url=http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf|access-date=2019-04-01|archive-date=2020-09-24|archive-url=https://web.archive.org/web/20200924085514/http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf|url-status=live}}</ref> || 16.5
|}
The debut of DNNs for speaker recognition in the late 1990s and speech recognition around 2009-2011 and of LSTM around 2003–2007, accelerated progress in eight major areas:<ref name="BOOK2014" /><ref name="interspeech2014Keynote" /><ref name="ReferenceA" />
* Scale-up/out and accelerated DNN training and decoding
* Sequence discriminative training
* Feature processing by deep models with solid understanding of the underlying mechanisms
* Adaptation of DNNs and related deep models
* [[Multi-task learning|Multi-task]] and [[transfer learning]] by DNNs and related deep models
* [[Convolutional neural network|CNNs]] and how to design them to best exploit [[domain knowledge]] of speech
* [[Recurrent neural network|RNN]] and its rich LSTM variants
* Other types of deep models including tensor-based models and integrated deep generative/discriminative models.
All major commercial speech recognition systems (e.g., Microsoft [[Cortana (software)|Cortana]], [[Xbox]], [[Skype Translator]], [[Amazon Alexa]], [[Google Now]], [[Siri|Apple Siri]], [[Baidu]] and [[IFlytek|iFlyTek]] voice search, and a range of [[Nuance Communications|Nuance]] speech products, etc.) are based on deep learning.<ref name=BOOK2014 /><ref>{{Cite magazine|url=https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/|title=How Skype Used AI to Build Its Amazing New Language Translator {{!}} WIRED|magazine=Wired|access-date=14 June 2017|date=17 December 2014|last1=McMillan|first1=Robert|archive-date=8 June 2017|archive-url=https://web.archive.org/web/20170608062106/https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/|url-status=live}}</ref><ref name="Baidu">{{cite arXiv |eprint=1412.5567|last1=Hannun|first1=Awni|title=Deep Speech: Scaling up end-to-end speech recognition|last2=Case|first2=Carl|last3=Casper|first3=Jared|last4=Catanzaro|first4=Bryan|last5=Diamos|first5=Greg|last6=Elsen|first6=Erich|last7=Prenger|first7=Ryan|last8=Satheesh|first8=Sanjeev|last9=Sengupta|first9=Shubho|last10=Coates|first10=Adam|last11=Ng|first11=Andrew Y|class=cs.CL|year=2014}}</ref>
=== Image recognition ===
{{Main|Computer vision}}
A common evaluation set for image classification is the MNIST database data set. MNIST is composed of handwritten digits and includes 60,000 training examples and 10,000 test examples. As with TIMIT, its small size lets users test multiple configurations. A comprehensive list of results on this set is available.<ref name="YANNMNIST">{{cite web|url=http://yann.lecun.com/exdb/mnist/.|title=MNIST handwritten digit database, Yann LeCun, Corinna Cortes and Chris Burges|website=yann.lecun.com|access-date=2014-01-28|archive-date=2014-01-13|archive-url=https://web.archive.org/web/20140113175237/http://yann.lecun.com/exdb/mnist/|url-status=live}}</ref>
Deep learning-based image recognition has become "superhuman", producing more accurate results than human contestants. This first occurred in 2011 in recognition of traffic signs, and in 2014, with recognition of human faces.<ref name=":7">{{Cite journal|last1=Cireşan|first1=Dan|last2=Meier|first2=Ueli|last3=Masci|first3=Jonathan|last4=Schmidhuber|first4=Jürgen|date=August 2012|title=Multi-column deep neural network for traffic sign classification|journal=Neural Networks|series=Selected Papers from IJCNN 2011|volume=32|pages=333–338|doi=10.1016/j.neunet.2012.02.023|pmid=22386783|citeseerx=10.1.1.226.8219}}</ref><ref name=surpass1>{{cite arXiv|title=Surpassing Human Level Face Recognition|author1=Chaochao Lu |author2= Xiaoou Tang |year=2014 |class=cs.CV |eprint=1404.3840 }}</ref>
Deep learning-trained vehicles now interpret 360° camera views.<ref>[http://www.technologyreview.com/news/533936/nvidia-demos-a-car-computer-trained-with-deep-learning/ Nvidia Demos a Car Computer Trained with "Deep Learning"] (6 January 2015), David Talbot, ''[[MIT Technology Review]]''</ref> Another example is Facial Dysmorphology Novel Analysis (FDNA) used to analyze cases of human malformation connected to a large database of genetic syndromes.
=== Visual art processing ===
Closely related to the progress that has been made in image recognition is the increasing application of deep learning techniques to various visual art tasks. DNNs have proven themselves capable, for example, of
*identifying the style period of a given painting<ref name = art1/><ref name = art2/>
*[[Neural Style Transfer]]{{snd}} capturing the style of a given artwork and applying it in a visually pleasing manner to an arbitrary photograph or video<ref name = art1/><ref name = art2/>
*generating striking imagery based on random visual input fields.<ref name = art1 >{{cite journal |author1=G. W. Smith|author2=Frederic Fol Leymarie|date=10 April 2017|title=The Machine as Artist: An Introduction|journal=Arts|volume=6|issue=4|pages=5|doi=10.3390/arts6020005|doi-access=free}}</ref><ref name = art2>{{cite journal |author=Blaise Agüera y Arcas|date=29 September 2017|title=Art in the Age of Machine Intelligence|journal=Arts|volume=6|issue=4|pages=18|doi=10.3390/arts6040018|doi-access=free}}</ref>
=== Natural language processing ===
{{Main|Natural language processing}}
Neural networks have been used for implementing language models since the early 2000s.<ref name="gers2001" /> LSTM helped to improve machine translation and language modeling.<ref name="NIPS2014" /><ref name="vinyals2016" /><ref name="gillick2015" />
Other key techniques in this field are negative sampling<ref name="GoldbergLevy2014">{{cite arXiv|last1=Goldberg|first1=Yoav|last2=Levy|first2=Omar|title=word2vec Explained: Deriving Mikolov et al.'s Negative-Sampling Word-Embedding Method|eprint=1402.3722|class=cs.CL|year=2014}}</ref> and [[word embedding]]. Word embedding, such as ''[[word2vec]]'', can be thought of as a representational layer in a deep learning architecture that transforms an atomic word into a positional representation of the word relative to other words in the dataset; the position is represented as a point in a [[vector space]]. Using word embedding as an RNN input layer allows the network to parse sentences and phrases using an effective compositional vector grammar. A compositional vector grammar can be thought of as [[probabilistic context free grammar]] (PCFG) implemented by an RNN.<ref name="SocherManning2014">{{cite web|last1=Socher|first1=Richard|last2=Manning|first2=Christopher|title=Deep Learning for NLP|url=http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf|access-date=26 October 2014|archive-date=6 July 2014|archive-url=https://web.archive.org/web/20140706040227/http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf|url-status=live}}</ref> Recursive auto-encoders built atop word embeddings can assess sentence similarity and detect paraphrasing.<ref name="SocherManning2014" /> Deep neural architectures provide the best results for [[Statistical parsing|constituency parsing]],<ref>{{Cite journal|url = http://aclweb.org/anthology/P/P13/P13-1045.pdf|title = Parsing With Compositional Vector Grammars|last1 = Socher|first1 = Richard|date = 2013|journal = Proceedings of the ACL 2013 Conference|last2 = Bauer|first2 = John|last3 = Manning|first3 = Christopher|last4 = Ng|first4 = Andrew|access-date = 2014-09-03|archive-date = 2014-11-27|archive-url = https://web.archive.org/web/20141127005912/http://www.aclweb.org/anthology/P/P13/P13-1045.pdf|url-status = live}}</ref> [[sentiment analysis]],<ref>{{Cite journal|url = http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf|title = Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank|last = Socher|first = Richard|date = 2013|journal = |access-date = 2014-09-03|archive-date = 2016-12-28|archive-url = https://web.archive.org/web/20161228100300/http://nlp.stanford.edu/%7Esocherr/EMNLP2013_RNTN.pdf|url-status = live}}</ref> information retrieval,<ref>{{Cite journal|last1=Shen|first1=Yelong|last2=He|first2=Xiaodong|last3=Gao|first3=Jianfeng|last4=Deng|first4=Li|last5=Mesnil|first5=Gregoire|date=1 November 2014|title=A Latent Semantic Model with Convolutional-Pooling Structure for Information Retrieval|url=https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/|journal=Microsoft Research|access-date=14 June 2017|archive-date=27 October 2017|archive-url=https://web.archive.org/web/20171027050418/https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/|url-status=live}}</ref><ref>{{Cite journal|last1=Huang|first1=Po-Sen|last2=He|first2=Xiaodong|last3=Gao|first3=Jianfeng|last4=Deng|first4=Li|last5=Acero|first5=Alex|last6=Heck|first6=Larry|date=1 October 2013|title=Learning Deep Structured Semantic Models for Web Search using Clickthrough Data|url=https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/|journal=Microsoft Research|access-date=14 June 2017|archive-date=27 October 2017|archive-url=https://web.archive.org/web/20171027050414/https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/|url-status=live}}</ref> spoken language understanding,<ref name="IEEE-TASL2015">{{cite journal | last1 = Mesnil | first1 = G. | last2 = Dauphin | first2 = Y. | last3 = Yao | first3 = K. | last4 = Bengio | first4 = Y. | last5 = Deng | first5 = L. | last6 = Hakkani-Tur | first6 = D. | last7 = He | first7 = X. | last8 = Heck | first8 = L. | last9 = Tur | first9 = G. | last10 = Yu | first10 = D. | last11 = Zweig | first11 = G. | s2cid = 1317136 | year = 2015 | title = Using recurrent neural networks for slot filling in spoken language understanding | journal = IEEE Transactions on Audio, Speech, and Language Processing | volume = 23 | issue = 3| pages = 530–539 | doi=10.1109/taslp.2014.2383614}}</ref> machine translation,<ref name="NIPS2014">{{Cite journal|last1=Sutskever|first1=L.|last2=Vinyals|first2=O.|last3=Le|first3=Q.|date=2014|title=Sequence to Sequence Learning with Neural Networks|url=https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf|journal=Proc. NIPS|bibcode=2014arXiv1409.3215S|arxiv=1409.3215|access-date=2017-06-13|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123145/https://papers.nips.cc/paper/2014/file/a14ac55a4f27472c5d894ec1c3c743d2-Paper.pdf|url-status=live}}</ref><ref name="auto">{{Cite journal|last1=Gao|first1=Jianfeng|last2=He|first2=Xiaodong|last3=Yih|first3=Scott Wen-tau|last4=Deng|first4=Li|date=1 June 2014|title=Learning Continuous Phrase Representations for Translation Modeling|url=https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/|journal=Microsoft Research|access-date=14 June 2017|archive-date=27 October 2017|archive-url=https://web.archive.org/web/20171027050403/https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/|url-status=live}}</ref> contextual entity linking,<ref name="auto"/> writing style recognition,<ref name="BROC2017">{{Cite journal |doi = 10.1002/dac.3259|title = Authorship verification using deep belief network systems|journal = International Journal of Communication Systems|volume = 30|issue = 12|pages = e3259|year = 2017|last1 = Brocardo|first1 = Marcelo Luiz|last2 = Traore|first2 = Issa|last3 = Woungang|first3 = Isaac|last4 = Obaidat|first4 = Mohammad S.| s2cid=40745740 }}</ref> Text classification and others.<ref>{{Cite news|url=https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/|title=Deep Learning for Natural Language Processing: Theory and Practice (CIKM2014 Tutorial) - Microsoft Research|work=Microsoft Research|access-date=14 June 2017|archive-date=13 March 2017|archive-url=https://web.archive.org/web/20170313184253/https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/|url-status=live}}</ref>
Recent developments generalize [[word embedding]] to [[sentence embedding]].
[[Google Translate]] (GT) uses a large end-to-end [[long short-term memory]] (LSTM) network.<ref name="GT_Turovsky_2016">{{cite web|url=https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/|title=Found in translation: More accurate, fluent sentences in Google Translate|last=Turovsky|first=Barak|date=15 November 2016|website=The Keyword Google Blog|access-date=23 March 2017|archive-date=7 April 2017|archive-url=https://web.archive.org/web/20170407071226/https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/|url-status=live}}</ref><ref name="googleblog_GNMT_2016">{{cite web|url=https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html|title=Zero-Shot Translation with Google's Multilingual Neural Machine Translation System|last1=Schuster|first1=Mike|last2=Johnson|first2=Melvin|date=22 November 2016|website=Google Research Blog|access-date=23 March 2017|last3=Thorat|first3=Nikhil|archive-date=10 July 2017|archive-url=https://web.archive.org/web/20170710183732/https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html|url-status=live}}</ref><ref name="GoogleTranslate">{{cite arXiv |eprint=1609.08144|last1=Wu|first1=Yonghui|title=Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation|last2=Schuster|first2=Mike|last3=Chen|first3=Zhifeng|last4=Le|first4=Quoc V|last5=Norouzi|first5=Mohammad|last6=Macherey|first6=Wolfgang|last7=Krikun|first7=Maxim|last8=Cao|first8=Yuan|last9=Gao|first9=Qin|last10=Macherey|first10=Klaus|last11=Klingner|first11=Jeff|last12=Shah|first12=Apurva|last13=Johnson|first13=Melvin|last14=Liu|first14=Xiaobing|last15=Kaiser|first15=Łukasz|last16=Gouws|first16=Stephan|last17=Kato|first17=Yoshikiyo|last18=Kudo|first18=Taku|last19=Kazawa|first19=Hideto|last20=Stevens|first20=Keith|last21=Kurian|first21=George|last22=Patil|first22=Nishant|last23=Wang|first23=Wei|last24=Young|first24=Cliff|last25=Smith|first25=Jason|last26=Riesa|first26=Jason|last27=Rudnick|first27=Alex|last28=Vinyals|first28=Oriol|last29=Corrado|first29=Greg|last30=Hughes|first30=Macduff|display-authors=29|class=cs.CL|year=2016}}</ref><ref name="WiredGoogleTranslate">{{cite magazine |title=An Infusion of AI Makes Google Translate More Powerful Than Ever |first=Cade |last=Metz |magazine=[[Wired (magazine)|Wired]] |date=27 September 2016 |url=https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/ |access-date=12 October 2017 |archive-date=8 November 2020 |archive-url=https://web.archive.org/web/20201108101324/https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/ |url-status=live }}</ref> [[Google Neural Machine Translation|Google Neural Machine Translation (GNMT)]] uses an [[example-based machine translation]] method in which the system "learns from millions of examples."<ref name="googleblog_GNMT_2016" /> It translates "whole sentences at a time, rather than pieces. Google Translate supports over one hundred languages.<ref name="googleblog_GNMT_2016" /> The network encodes the "semantics of the sentence rather than simply memorizing phrase-to-phrase translations".<ref name="googleblog_GNMT_2016" /><ref name="Biotet">{{cite web|url=http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf|title=MT on and for the Web|last1=Boitet|first1=Christian|last2=Blanchon|first2=Hervé|date=2010|access-date=1 December 2016|last3=Seligman|first3=Mark|last4=Bellynck|first4=Valérie|archive-date=29 March 2017|archive-url=https://web.archive.org/web/20170329125916/http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf|url-status=dead}}</ref> GT uses English as an intermediate between most language pairs.<ref name="Biotet" />
=== Drug discovery and toxicology ===
{{For|more information|Drug discovery|Toxicology}}
A large percentage of candidate drugs fail to win regulatory approval. These failures are caused by insufficient efficacy (on-target effect), undesired interactions (off-target effects), or unanticipated [[Toxicity|toxic effects]].<ref name="ARROWSMITH2013">{{Cite journal
| pmid = 23903212
| year = 2013
| last1 = Arrowsmith
| first1 = J
| title = Trial watch: Phase II and phase III attrition rates 2011-2012
| journal = Nature Reviews Drug Discovery
| volume = 12
| issue = 8
| pages = 569
| last2 = Miller
| first2 = P
| s2cid = 20246434
| doi = 10.1038/nrd4090
}}</ref><ref name="VERBIEST2015">{{Cite journal
| pmid = 25582842
| year = 2015
| last1 = Verbist
| first1 = B
| title = Using transcriptomics to guide lead optimization in drug discovery projects: Lessons learned from the QSTAR project
| journal = Drug Discovery Today
| last2 = Klambauer
| first2 = G
| last3 = Vervoort
| first3 = L
| last4 = Talloen
| first4 = W
| last5 = The Qstar
| first5 = Consortium
| last6 = Shkedy
| first6 = Z
| last7 = Thas
| first7 = O
| last8 = Bender
| first8 = A
| last9 = Göhlmann
| first9 = H. W.
| last10 = Hochreiter
| first10 = S
| doi = 10.1016/j.drudis.2014.12.014
| volume=20
| issue = 5
| pages=505–513
| doi-access = free
}}</ref> Research has explored use of deep learning to predict the [[biomolecular target]]s,<ref name="MERCK2012" /><ref name=":5" /> [[off-target]]s, and [[Toxicity|toxic effects]] of environmental chemicals in nutrients, household products and drugs.<ref name="TOX21" /><ref name="TOX21Data" /><ref name=":11" />
AtomNet is a deep learning system for structure-based [[Drug design|rational drug design]].<ref>{{cite arXiv|title = AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery|eprint= 1510.02855|date = 9 October 2015|first1 = Izhar|last1 = Wallach|first2 = Michael|last2 = Dzamba|first3 = Abraham|last3 = Heifets|class= cs.LG}}</ref> AtomNet was used to predict novel candidate biomolecules for disease targets such as the [[Ebola virus]]<ref>{{Cite news |title = Toronto startup has a faster way to discover effective medicines |url = https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/ |website = The Globe and Mail |access-date = 9 November 2015 |archive-date = 20 October 2015 |archive-url = https://web.archive.org/web/20151020040115/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/ |url-status = live }}</ref> and [[multiple sclerosis]].<ref>{{Cite web|title = Startup Harnesses Supercomputers to Seek Cures|url = http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/|website = KQED Future of You|access-date = 9 November 2015|archive-date = 24 December 2015|archive-url = https://web.archive.org/web/20151224104721/http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/|url-status = live}}</ref><ref>{{Cite web|url=https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/|title=Toronto startup has a faster way to discover effective medicines|website=[[The Globe and Mail]]|access-date=2017-08-26|archive-date=2015-12-25|archive-url=https://web.archive.org/web/20151225162547/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5d%20and%20multiple%20sclerosis%20%5b/|url-status=live}}</ref>
In 2017 [[graph neural network]]s were used for the first time to predict various properties of molecules in a large toxicology data set.<ref>{{cite arXiv|last1=Gilmer|first1=Justin|last2=Schoenholz|first2=Samuel S.|last3=Riley|first3=Patrick F.|last4=Vinyals|first4=Oriol|last5=Dahl|first5=George E.|date=2017-06-12|title=Neural Message Passing for Quantum Chemistry|class=cs.LG|eprint=1704.01212}}</ref> In 2019, generative neural networks were used to produce molecules that were validated experimentally all the way into mice.<ref>{{cite journal |last1=Zhavoronkov |first1=Alex|s2cid=201716327|date=2019|title=Deep learning enables rapid identification of potent DDR1 kinase inhibitors |journal=Nature Biotechnology |volume=37|issue=9|pages=1038–1040|doi=10.1038/s41587-019-0224-x |pmid=31477924}}</ref><ref>{{cite magazine |last1=Gregory |first1=Barber |title=A Molecule Designed By AI Exhibits 'Druglike' Qualities |url=https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/ |magazine=Wired |access-date=2019-09-05 |archive-date=2020-04-30 |archive-url=https://web.archive.org/web/20200430143244/https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/ |url-status=live }}</ref>
=== Customer relationship management ===
{{Main|Customer relationship management}}
[[Deep reinforcement learning]] has been used to approximate the value of possible [[direct marketing]] actions, defined in terms of [[RFM (customer value)|RFM]] variables. The estimated value function was shown to have a natural interpretation as [[customer lifetime value]].<ref>{{cite arXiv|last=Tkachenko |first=Yegor |title=Autonomous CRM Control via CLV Approximation with Deep Reinforcement Learning in Discrete and Continuous Action Space |date=8 April 2015 |eprint=1504.01840|class=cs.LG }}</ref>
=== Recommendation systems ===
{{Main|Recommender system}}
Recommendation systems have used deep learning to extract meaningful features for a latent factor model for content-based music and journal recommendations.<ref>{{Cite book|url=http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf|title=Advances in Neural Information Processing Systems 26|last1=van den Oord|first1=Aaron|last2=Dieleman|first2=Sander|last3=Schrauwen|first3=Benjamin|date=2013|publisher=Curran Associates, Inc.|editor-last=Burges|editor-first=C. J. C.|pages=2643–2651|editor-last2=Bottou|editor-first2=L.|editor-last3=Welling|editor-first3=M.|editor-last4=Ghahramani|editor-first4=Z.|editor-last5=Weinberger|editor-first5=K. Q.|access-date=2017-06-14|archive-date=2017-05-16|archive-url=https://web.archive.org/web/20170516185259/http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf|url-status=live}}</ref><ref>{{cite journal | last1 = Feng | first1 = X.Y. | last2 = Zhang | first2 = H. | last3 = Ren | first3 = Y.J. | last4 = Shang | first4 = P.H. | last5 = Zhu | first5 = Y. | last6 = Liang | first6 = Y.C. | last7 = Guan | first7 = R.C. | last8 = Xu | first8 = D. | year = 2019 | title = The Deep Learning–Based Recommender System "Pubmender" for Choosing a Biomedical Publication Venue: Development and Validation Study | journal = [[Journal of Medical Internet Research]] | volume = 21 | issue = 5| page = e12957 | doi = 10.2196/12957 | pmid = 31127715 | pmc = 6555124 }}</ref> Multi-view deep learning has been applied for learning user preferences from multiple domains.<ref>{{Cite journal|last1=Elkahky|first1=Ali Mamdouh|last2=Song|first2=Yang|last3=He|first3=Xiaodong|date=1 May 2015|title=A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems|url=https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/|journal=Microsoft Research|access-date=14 June 2017|archive-date=25 January 2018|archive-url=https://web.archive.org/web/20180125134534/https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/|url-status=live}}</ref> The model uses a hybrid collaborative and content-based approach and enhances recommendations in multiple tasks.
=== Bioinformatics ===
{{Main|Bioinformatics}}
An [[autoencoder]] ANN was used in [[bioinformatics]], to predict [[Gene Ontology|gene ontology]] annotations and gene-function relationships.<ref>{{cite book|title=Deep Autoencoder Neural Networks for Gene Ontology Annotation Predictions|first1=Davide|last1=Chicco|first2=Peter|last2=Sadowski|first3=Pierre|last3=Baldi|s2cid=207217210|date=1 January 2014|publisher=ACM|pages=533–540|doi=10.1145/2649387.2649442|journal=Proceedings of the 5th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics - BCB '14|isbn=9781450328944|hdl=11311/964622|url=http://dl.acm.org/citation.cfm?id=2649442|access-date=23 November 2015|archive-date=9 May 2021|archive-url=https://web.archive.org/web/20210509123140/https://dl.acm.org/doi/10.1145/2649387.2649442|url-status=live}}</ref>
In medical informatics, deep learning was used to predict sleep quality based on data from wearables<ref>{{Cite journal|last=Sathyanarayana|first=Aarti|s2cid=3821594|date=1 January 2016|title=Sleep Quality Prediction From Wearable Data Using Deep Learning|journal=JMIR mHealth and uHealth|volume=4|issue=4|doi=10.2196/mhealth.6562|pmid=27815231|pmc=5116102|pages=e125}}</ref> and predictions of health complications from [[electronic health record]] data.<ref>{{Cite journal|last1=Choi|first1=Edward|last2=Schuetz|first2=Andy|last3=Stewart|first3=Walter F.|last4=Sun|first4=Jimeng|date=13 August 2016|title=Using recurrent neural network models for early detection of heart failure onset|journal=Journal of the American Medical Informatics Association|volume=24|issue=2|pages=361–370|doi=10.1093/jamia/ocw112|issn=1067-5027|pmid=27521897|pmc=5391725}}</ref>
=== Medical image analysis ===
Deep learning has been shown to produce competitive results in medical application such as cancer cell classification, lesion detection, organ segmentation and image enhancement.<ref>{{Cite journal|last1=Litjens|first1=Geert|last2=Kooi|first2=Thijs|last3=Bejnordi|first3=Babak Ehteshami|last4=Setio|first4=Arnaud Arindra Adiyoso|last5=Ciompi|first5=Francesco|last6=Ghafoorian|first6=Mohsen|last7=van der Laak|first7=Jeroen A.W.M.|last8=van Ginneken|first8=Bram|last9=Sánchez|first9=Clara I.|s2cid=2088679|date=December 2017|title=A survey on deep learning in medical image analysis|journal=Medical Image Analysis|volume=42|pages=60–88|doi=10.1016/j.media.2017.07.005|pmid=28778026|arxiv=1702.05747|bibcode=2017arXiv170205747L}}</ref><ref>{{Cite book|doi=10.1109/ICCVW.2017.18|isbn=9781538610343|chapter=Deep Convolutional Neural Networks for Detecting Cellular Changes Due to Malignancy|title=2017 IEEE International Conference on Computer Vision Workshops (ICCVW)|pages=82–89|year=2017|last1=Forslid|first1=Gustav|last2=Wieslander|first2=Hakan|last3=Bengtsson|first3=Ewert|last4=Wahlby|first4=Carolina|last5=Hirsch|first5=Jan-Michael|last6=Stark|first6=Christina Runow|last7=Sadanandan|first7=Sajith Kecheril|s2cid=4728736|chapter-url=http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-326160|access-date=2019-11-12|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123157/https://d1bxh8uas1mnw7.cloudfront.net/assets/embed.js|url-status=live}}</ref> Modern deep learning tools demonstrate the high accuracy of detecting various diseases and the helpfulness of their use by specialists to improve the diagnosis efficiency.<ref>{{Cite journal |last1=Dong |first1=Xin |last2=Zhou |first2=Yizhao |last3=Wang |first3=Lantian |last4=Peng |first4=Jingfeng |last5=Lou |first5=Yanbo |last6=Fan |first6=Yiqun |date=2020 |title=Liver Cancer Detection Using Hybridized Fully Convolutional Neural Network Based on Deep Learning Framework |url=https://ieeexplore.ieee.org/document/9130662 |journal=IEEE Access |volume=8 |pages=129889–129898 |doi=10.1109/ACCESS.2020.3006362 |s2cid=220733699 |issn=2169-3536}}</ref><ref>{{Cite journal |last1=Lyakhov |first1=Pavel Alekseevich |last2=Lyakhova |first2=Ulyana Alekseevna |last3=Nagornov |first3=Nikolay Nikolaevich |date=2022-04-03 |title=System for the Recognizing of Pigmented Skin Lesions with Fusion and Analysis of Heterogeneous Data Based on a Multimodal Neural Network |journal=Cancers |language=en |volume=14 |issue=7 |pages=1819 |doi=10.3390/cancers14071819 |pmid=35406591 |pmc=8997449 |issn=2072-6694|doi-access=free }}</ref>
=== Mobile advertising ===
Finding the appropriate mobile audience for [[mobile advertising]] is always challenging, since many data points must be considered and analyzed before a target segment can be created and used in ad serving by any ad server.<ref>{{cite book |doi=10.1109/CSCITA.2017.8066548 |isbn=978-1-5090-4381-1|chapter=Predicting the popularity of instagram posts for a lifestyle magazine using deep learning|title=2017 2nd International Conference on Communication Systems, Computing and IT Applications (CSCITA)|pages=174–177|year=2017|last1=De|first1=Shaunak|last2=Maity|first2=Abhishek|last3=Goel|first3=Vritti|last4=Shitole|first4=Sanjay|last5=Bhattacharya|first5=Avik|s2cid=35350962}}</ref> Deep learning has been used to interpret large, many-dimensioned advertising datasets. Many data points are collected during the request/serve/click internet advertising cycle. This information can form the basis of machine learning to improve ad selection.
=== Image restoration ===
Deep learning has been successfully applied to [[inverse problems]] such as [[denoising]], [[super-resolution]], [[inpainting]], and [[film colorization]].<ref>{{Cite web|url=https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/|title=Colorizing and Restoring Old Images with Deep Learning|date=13 November 2018|website=FloydHub Blog|language=en|access-date=11 October 2019|archive-date=11 October 2019|archive-url=https://web.archive.org/web/20191011162814/https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/|url-status=live}}</ref> These applications include learning methods such as "Shrinkage Fields for Effective Image Restoration"<ref>{{cite conference |url= http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf |first1= Uwe |last1= Schmidt |first2= Stefan |last2= Roth |conference= Computer Vision and Pattern Recognition (CVPR), 2014 IEEE Conference on |title= Shrinkage Fields for Effective Image Restoration |access-date= 2018-01-01 |archive-date= 2018-01-02 |archive-url= https://web.archive.org/web/20180102013217/http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf |url-status= live }}</ref> which trains on an image dataset, and [[Deep Image Prior]], which trains on the image that needs restoration.
=== Financial fraud detection ===
Deep learning is being successfully applied to financial [[fraud detection]], tax evasion detection,<ref>{{cite journal
|first1=Christos |last1=Kleanthous
|first2=Sotirios |last2=Chatzis
|title=Gated Mixture Variational Autoencoders for Value Added Tax audit case selection
|journal=Knowledge-Based Systems
|volume=188
|year=2020
|page=105048
|doi=10.1016/j.knosys.2019.105048
|s2cid=204092079
}}</ref> and anti-money laundering.<ref>{{cite journal
|first=Tomasz
|last=Czech
|title=Deep learning: the next frontier for money laundering detection
|url=https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/
|journal=Global Banking and Finance Review
|date=28 June 2018
|access-date=2018-07-15
|archive-date=2018-11-16
|archive-url=https://web.archive.org/web/20181116082711/https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/
|url-status=live
}}</ref>
=== Military ===
The United States Department of Defense applied deep learning to train robots in new tasks through observation.<ref name=":12">{{Cite web|url=https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php|title=Army researchers develop new algorithms to train robots|website=EurekAlert!|access-date=29 August 2018|archive-date=28 August 2018|archive-url=https://web.archive.org/web/20180828035608/https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php|url-status=live}}</ref>
=== Partial differential equations ===
Physics informed neural networks have been used to solve [[partial differential equation]]s in both forward and inverse problems in a data driven manner.<ref>{{Cite journal|last1=Raissi|first1=M.|last2=Perdikaris|first2=P.|last3=Karniadakis|first3=G. E.|date=2019-02-01|title=Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations|url=https://www.sciencedirect.com/science/article/pii/S0021999118307125|journal=Journal of Computational Physics|language=en|volume=378|pages=686–707|doi=10.1016/j.jcp.2018.10.045|bibcode=2019JCoPh.378..686R|osti=1595805|s2cid=57379996|issn=0021-9991}}</ref> One example is the reconstructing fluid flow governed by the [[Navier–Stokes equations|Navier-Stokes equations]]. Using physics informed neural networks does not require the often expensive mesh generation that conventional [[Computational fluid dynamics|CFD]] methods relies on.<ref>{{Cite journal|last1=Mao|first1=Zhiping|last2=Jagtap|first2=Ameya D.|last3=Karniadakis|first3=George Em|date=2020-03-01|title=Physics-informed neural networks for high-speed flows|url=https://www.sciencedirect.com/science/article/pii/S0045782519306814|journal=Computer Methods in Applied Mechanics and Engineering|language=en|volume=360|pages=112789|doi=10.1016/j.cma.2019.112789|bibcode=2020CMAME.360k2789M|s2cid=212755458|issn=0045-7825}}</ref><ref>{{Cite journal|last1=Raissi|first1=Maziar|last2=Yazdani|first2=Alireza|last3=Karniadakis|first3=George Em|date=2020-02-28|title=Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations|journal=Science|volume=367|issue=6481|pages=1026–1030|doi=10.1126/science.aaw4741|pmc=7219083|pmid=32001523|bibcode=2020Sci...367.1026R}}</ref>
=== Image Reconstruction ===
Image reconstruction is the reconstruction of the underlying images from the image-related measurements. Several works showed the better and superior performance of the deep learning methods compared to analytical methods for various applications, e.g., spectral imaging <ref>{{Cite journal |last1=Oktem |first1=Figen S. |last2=Kar |first2=Oğuzhan Fatih |last3=Bezek |first3=Can Deniz |last4=Kamalabadi |first4=Farzad |date=2021 |title=High-Resolution Multi-Spectral Imaging With Diffractive Lenses and Learned Reconstruction |url=https://ieeexplore.ieee.org/document/9415140 |journal=IEEE Transactions on Computational Imaging |volume=7 |pages=489–504 |doi=10.1109/TCI.2021.3075349 |arxiv=2008.11625 |s2cid=235340737 |issn=2333-9403}}</ref> and ultrasound imaging.<ref>{{Cite journal |last1=Bernhardt |first1=Melanie |last2=Vishnevskiy |first2=Valery |last3=Rau |first3=Richard |last4=Goksel |first4=Orcun |date=December 2020 |title=Training Variational Networks With Multidomain Simulations: Speed-of-Sound Image Reconstruction |url=https://ieeexplore.ieee.org/abstract/document/9144249?casa_token=NLVQbGR8h-8AAAAA:rz3cvMmiplVSRCGTDouUZkIeribqQKaMtK9t0MxlwKBMRwaTJL1onKOwkEk0qW_fFJCfd2ejOg |journal=IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control |volume=67 |issue=12 |pages=2584–2594 |doi=10.1109/TUFFC.2020.3010186 |pmid=32746211 |arxiv=2006.14395 |s2cid=220055785 |issn=1525-8955}}</ref>
<big>'''Epigenetic clock'''</big>
For more information, see [[Ageing clock|Epigenetic clock]].
An '''epigenetic clock''' is a [[Biomarkers of aging|biochemical test]] that can be used to measure age. Galkin et al. used deep [[Neural network|neural networks]] to train an epigenetic aging clock of unprecedented accuracy using >6,000 blood samples. The clock uses information from 1000 CpG sites and predicts people with certain conditions older than healthy controls: [[Inflammatory bowel disease|IBD]], [[Dementia|frontotemporal dementia]], ovarian cancer, obesity. The aging clock is planned to be released for public use in 2021 by an [[Insilico Medicine]] spinoff company Deep Longevity.
== Relation to human cognitive and brain development ==
Deep learning is closely related to a class of theories of [[brain development]] (specifically, neocortical development) proposed by [[cognitive neuroscientist]]s in the early 1990s.<ref name="UTGOFF">{{cite journal | last1 = Utgoff | first1 = P. E. | last2 = Stracuzzi | first2 = D. J. | s2cid = 1119517 | year = 2002 | title = Many-layered learning | journal = Neural Computation | volume = 14 | issue = 10| pages = 2497–2529 | doi=10.1162/08997660260293319| pmid = 12396572 }}</ref><ref name="ELMAN">{{cite book|url={{google books |plainurl=y |id=vELaRu_MrwoC}}|title=Rethinking Innateness: A Connectionist Perspective on Development|last=Elman|first=Jeffrey L.|publisher=MIT Press|year=1998|isbn=978-0-262-55030-7}}</ref><ref name="SHRAGER">{{cite journal | last1 = Shrager | first1 = J. | last2 = Johnson | first2 = MH | year = 1996 | title = Dynamic plasticity influences the emergence of function in a simple cortical array | journal = Neural Networks | volume = 9 | issue = 7| pages = 1119–1129 | doi=10.1016/0893-6080(96)00033-0| pmid = 12662587 }}</ref><ref name="QUARTZ">{{cite journal | last1 = Quartz | first1 = SR | last2 = Sejnowski | first2 = TJ | year = 1997 | title = The neural basis of cognitive development: A constructivist manifesto | journal = Behavioral and Brain Sciences | volume = 20 | issue = 4| pages = 537–556 | doi=10.1017/s0140525x97001581| pmid = 10097006 | citeseerx = 10.1.1.41.7854 | s2cid = 5818342 }}</ref> These developmental theories were instantiated in computational models, making them predecessors of deep learning systems. These developmental models share the property that various proposed learning dynamics in the brain (e.g., a wave of [[nerve growth factor]]) support the [[self-organization]] somewhat analogous to the neural networks utilized in deep learning models. Like the [[neocortex]], neural networks employ a hierarchy of layered filters in which each layer considers information from a prior layer (or the operating environment), and then passes its output (and possibly the original input), to other layers. This process yields a self-organizing stack of [[transducer]]s, well-tuned to their operating environment. A 1995 description stated, "...the infant's brain seems to organize itself under the influence of waves of so-called trophic-factors ... different regions of the brain become connected sequentially, with one layer of tissue maturing before another and so on until the whole brain is mature."<ref name="BLAKESLEE">S. Blakeslee., "In brain's early growth, timetable may be critical," ''The New York Times, Science Section'', pp. B5–B6, 1995.</ref>
A variety of approaches have been used to investigate the plausibility of deep learning models from a neurobiological perspective. On the one hand, several variants of the [[backpropagation]] algorithm have been proposed in order to increase its processing realism.<ref>{{Cite journal|last1=Mazzoni|first1=P.|last2=Andersen|first2=R. A.|last3=Jordan|first3=M. I.|date=15 May 1991|title=A more biologically plausible learning rule for neural networks.|journal=Proceedings of the National Academy of Sciences|volume=88|issue=10|pages=4433–4437|doi=10.1073/pnas.88.10.4433|issn=0027-8424|pmid=1903542|pmc=51674|bibcode=1991PNAS...88.4433M|doi-access=free}}</ref><ref>{{Cite journal|last=O'Reilly|first=Randall C.|s2cid=2376781|date=1 July 1996|title=Biologically Plausible Error-Driven Learning Using Local Activation Differences: The Generalized Recirculation Algorithm|journal=Neural Computation|volume=8|issue=5|pages=895–938|doi=10.1162/neco.1996.8.5.895|issn=0899-7667}}</ref> Other researchers have argued that unsupervised forms of deep learning, such as those based on hierarchical [[generative model]]s and [[deep belief network]]s, may be closer to biological reality.<ref>{{Cite journal|last1=Testolin|first1=Alberto|last2=Zorzi|first2=Marco|s2cid=9868901|date=2016|title=Probabilistic Models and Generative Neural Networks: Towards an Unified Framework for Modeling Normal and Impaired Neurocognitive Functions|journal=Frontiers in Computational Neuroscience|volume=10|pages=73|doi=10.3389/fncom.2016.00073|pmid=27468262|pmc=4943066|issn=1662-5188|doi-access=free}}</ref><ref>{{Cite journal|last1=Testolin|first1=Alberto|last2=Stoianov|first2=Ivilin|last3=Zorzi|first3=Marco|s2cid=24504018|date=September 2017|title=Letter perception emerges from unsupervised deep learning and recycling of natural image features|journal=Nature Human Behaviour|volume=1|issue=9|pages=657–664|doi=10.1038/s41562-017-0186-2|pmid=31024135|issn=2397-3374}}</ref> In this respect, generative neural network models have been related to neurobiological evidence about sampling-based processing in the cerebral cortex.<ref>{{Cite journal|last1=Buesing|first1=Lars|last2=Bill|first2=Johannes|last3=Nessler|first3=Bernhard|last4=Maass|first4=Wolfgang|s2cid=7504633|date=3 November 2011|title=Neural Dynamics as Sampling: A Model for Stochastic Computation in Recurrent Networks of Spiking Neurons|journal=PLOS Computational Biology|volume=7|issue=11|pages=e1002211|doi=10.1371/journal.pcbi.1002211|pmid=22096452|pmc=3207943|issn=1553-7358|bibcode=2011PLSCB...7E2211B}}</ref>
Although a systematic comparison between the human brain organization and the neuronal encoding in deep networks has not yet been established, several analogies have been reported. For example, the computations performed by deep learning units could be similar to those of actual neurons<ref>{{Cite journal|last1=Cash|first1=S.|last2=Yuste|first2=R.|date=February 1999|title=Linear summation of excitatory inputs by CA1 pyramidal neurons|journal=Neuron|volume=22|issue=2|pages=383–394|issn=0896-6273|pmid=10069343|doi=10.1016/s0896-6273(00)81098-3|s2cid=14663106}}</ref> and neural populations.<ref>{{Cite journal|date=1 August 2004|title=Sparse coding of sensory inputs|journal=Current Opinion in Neurobiology|volume=14|issue=4|pages=481–487|doi=10.1016/j.conb.2004.07.007|pmid=15321069|issn=0959-4388 | last1 = Olshausen | first1 = B | last2 = Field | first2 = D|s2cid=16560320}}</ref> Similarly, the representations developed by deep learning models are similar to those measured in the primate visual system<ref>{{Cite journal|last1=Yamins|first1=Daniel L K|last2=DiCarlo|first2=James J|s2cid=16970545|date=March 2016|title=Using goal-driven deep learning models to understand sensory cortex|journal=Nature Neuroscience|volume=19|issue=3|pages=356–365|doi=10.1038/nn.4244|pmid=26906502|issn=1546-1726}}</ref> both at the single-unit<ref>{{Cite journal|last1=Zorzi|first1=Marco|last2=Testolin|first2=Alberto|s2cid=39281431|date=19 February 2018|title=An emergentist perspective on the origin of number sense|journal=Phil. Trans. R. Soc. B|volume=373|issue=1740|pages=20170043|doi=10.1098/rstb.2017.0043|issn=0962-8436|pmid=29292348|pmc=5784047}}</ref> and at the population<ref>{{Cite journal|last1=Güçlü|first1=Umut|last2=van Gerven|first2=Marcel A. J.|date=8 July 2015|title=Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream|journal=Journal of Neuroscience|volume=35|issue=27|pages=10005–10014|doi=10.1523/jneurosci.5023-14.2015|pmid=26157000|pmc=6605414|arxiv=1411.6422}}</ref> levels.
== Commercial activity ==
[[Facebook]]'s AI lab performs tasks such as [[Automatic image annotation|automatically tagging uploaded pictures]] with the names of the people in them.<ref name="METZ2013">{{cite magazine |first=C. |last=Metz |title=Facebook's 'Deep Learning' Guru Reveals the Future of AI |url=https://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/ |magazine=Wired |date=12 December 2013 |access-date=26 August 2017 |archive-date=28 March 2014 |archive-url=https://web.archive.org/web/20140328071226/http://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/ |url-status=live }}</ref>
Google's [[DeepMind Technologies]] developed a system capable of learning how to play [[Atari]] video games using only pixels as data input. In 2015 they demonstrated their [[AlphaGo]] system, which learned the game of [[Go (game)|Go]] well enough to beat a professional Go player.<ref>{{Cite journal|title = Google AI algorithm masters ancient game of Go|url = http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234|journal = Nature|year = 2016|doi = 10.1038/529445a|access-date = 30 January 2016|archive-date = 2 May 2019|archive-url = https://web.archive.org/web/20190502200837/http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234|url-status = live|last1 = Gibney|first1 = Elizabeth|volume = 529|issue = 7587|pages = 445–446|pmid = 26819021|bibcode = 2016Natur.529..445G|s2cid = 4460235}}</ref><ref>{{Cite journal|title = Mastering the game of Go with deep neural networks and tree search|journal = [[Nature (journal)|Nature]]| issn= 0028-0836|pages = 484–489|volume = 529|issue = 7587|doi = 10.1038/nature16961|pmid = 26819042|first1 = David|last1 = Silver|author-link1=David Silver (programmer)|first2 = Aja|last2 = Huang|author-link2=Aja Huang|first3 = Chris J.|last3 = Maddison|first4 = Arthur|last4 = Guez|first5 = Laurent|last5 = Sifre|first6 = George van den|last6 = Driessche|first7 = Julian|last7 = Schrittwieser|first8 = Ioannis|last8 = Antonoglou|first9 = Veda|last9 = Panneershelvam|first10= Marc|last10= Lanctot|first11= Sander|last11= Dieleman|first12=Dominik|last12= Grewe|first13= John|last13= Nham|first14= Nal|last14= Kalchbrenner|first15= Ilya|last15= Sutskever|author-link15=Ilya Sutskever|first16= Timothy|last16= Lillicrap|first17= Madeleine|last17= Leach|first18= Koray|last18= Kavukcuoglu|first19= Thore|last19= Graepel|first20= Demis |last20=Hassabis|s2cid = 515925|author-link20=Demis Hassabis|date= 28 January 2016|bibcode = 2016Natur.529..484S}}{{closed access}}</ref><ref>{{Cite web|title = A Google DeepMind Algorithm Uses Deep Learning and More to Master the Game of Go {{!}} MIT Technology Review|url = http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/|website = MIT Technology Review|access-date = 30 January 2016|archive-date = 1 February 2016|archive-url = https://web.archive.org/web/20160201140636/http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/|url-status = dead}}</ref> [[Google Translate]] uses a neural network to translate between more than 100 languages.
In 2017, Covariant.ai was launched, which focuses on integrating deep learning into factories.<ref>{{Cite news|url=https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html|title=A.I. Researchers Leave Elon Musk Lab to Begin Robotics Start-Up|first=Cade|last=Metz|newspaper=The New York Times|date=6 November 2017|access-date=5 July 2019|archive-date=7 July 2019|archive-url=https://web.archive.org/web/20190707161547/https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html|url-status=live}}</ref>
As of 2008,<ref>{{Cite journal|title=TAMER: Training an Agent Manually via Evaluative Reinforcement|author1=Bradley Knox, W.|author2=Stone, Peter|year=2008|journal=2008 7th IEEE International Conference on Development and Learning|pages = 292–297|doi=10.1109/devlrn.2008.4640845|isbn = 978-1-4244-2661-4|s2cid = 5613334}}</ref> researchers at [[University of Texas at Austin|The University of Texas at Austin]] (UT) developed a machine learning framework called Training an Agent Manually via Evaluative Reinforcement, or TAMER, which proposed new methods for robots or computer programs to learn how to perform tasks by interacting with a human instructor.<ref name=":12" /> First developed as TAMER, a new algorithm called Deep TAMER was later introduced in 2018 during a collaboration between [[U.S. Army Research Laboratory]] (ARL) and UT researchers. Deep TAMER used deep learning to provide a robot the ability to learn new tasks through observation.<ref name=":12" /> Using Deep TAMER, a robot learned a task with a human trainer, watching video streams or observing a human perform a task in-person. The robot later practiced the task with the help of some coaching from the trainer, who provided feedback such as “good job” and “bad job.”<ref>{{Cite web|url=https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner|title=Talk to the Algorithms: AI Becomes a Faster Learner|website=governmentciomedia.com|access-date=29 August 2018|archive-date=28 August 2018|archive-url=https://web.archive.org/web/20180828001727/https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner|url-status=live}}</ref>
== Criticism and comment ==
Deep learning has attracted both criticism and comment, in some cases from outside the field of computer science.
=== Theory ===
{{see also|Explainable artificial intelligence}}
A main criticism concerns the lack of theory surrounding some methods.<ref>{{Cite web|url=https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1|title=In defense of skepticism about deep learning|last=Marcus|first=Gary|date=14 January 2018|website=Gary Marcus|access-date=11 October 2018|archive-date=12 October 2018|archive-url=https://web.archive.org/web/20181012035405/https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1|url-status=live}}</ref> Learning in the most common deep architectures is implemented using well-understood gradient descent. However, the theory surrounding other algorithms, such as contrastive divergence is less clear.{{citation needed|date=July 2016}} (e.g., Does it converge? If so, how fast? What is it approximating?) Deep learning methods are often looked at as a [[black box]], with most confirmations done empirically, rather than theoretically.<ref name="Knight 2017">{{cite web | last=Knight | first=Will | title=DARPA is funding projects that will try to open up AI's black boxes | website=MIT Technology Review | date=14 March 2017 | url=https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/ | access-date=2 November 2017 | archive-date=4 November 2019 | archive-url=https://web.archive.org/web/20191104033107/https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/ | url-status=live }}</ref>
Others point out that deep learning should be looked at as a step towards realizing strong AI, not as an all-encompassing solution. Despite the power of deep learning methods, they still lack much of the functionality needed for realizing this goal entirely. Research psychologist Gary Marcus noted:<blockquote>"Realistically, deep learning is only part of the larger challenge of building intelligent machines. Such techniques lack ways of representing [[causality|causal relationships]] (...) have no obvious ways of performing [[inference|logical inferences]], and they are also still a long way from integrating abstract knowledge, such as information about what objects are, what they are for, and how they are typically used. The most powerful A.I. systems, like [[Watson (computer)|Watson]] (...) use techniques like deep learning as just one element in a very complicated ensemble of techniques, ranging from the statistical technique of [[Bayesian inference]] to [[deductive reasoning]]."<ref>{{cite magazine|url=https://www.newyorker.com/|title=Is "Deep Learning" a Revolution in Artificial Intelligence?|last=Marcus|first=Gary|date=November 25, 2012|magazine=The New Yorker|access-date=2017-06-14|archive-date=2009-11-27|archive-url=https://web.archive.org/web/20091127184826/http://www.newyorker.com/|url-status=live}}</ref></blockquote>
In further reference to the idea that artistic sensitivity might be inherent in relatively low levels of the cognitive hierarchy, a published series of graphic representations of the internal states of deep (20-30 layers) neural networks attempting to discern within essentially random data the images on which they were trained<ref>{{cite web |url=http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html |author1=Alexander Mordvintsev |author2=Christopher Olah |author3=Mike Tyka |date=17 June 2015 |title=Inceptionism: Going Deeper into Neural Networks |publisher=Google Research Blog |access-date=20 June 2015 |archive-date=3 July 2015 |archive-url=https://web.archive.org/web/20150703064823/http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html |url-status=live }}</ref> demonstrate a visual appeal: the original research notice received well over 1,000 comments, and was the subject of what was for a time the most frequently accessed article on ''[[The Guardian]]'s''<ref>{{cite news|url=https://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep|title=Yes, androids do dream of electric sheep|date=18 June 2015|newspaper=The Guardian|author=Alex Hern|access-date=20 June 2015|archive-date=19 June 2015|archive-url=https://web.archive.org/web/20150619200845/http://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep|url-status=live}}</ref> website.
=== Errors ===
Some deep learning architectures display problematic behaviors,<ref name=goertzel>{{cite web |first=Ben |last=Goertzel |title=Are there Deep Reasons Underlying the Pathologies of Today's Deep Learning Algorithms? |year=2015 |url=http://goertzel.org/DeepLearning_v1.pdf |access-date=2015-05-10 |archive-date=2015-05-13 |archive-url=https://web.archive.org/web/20150513053107/http://goertzel.org/DeepLearning_v1.pdf |url-status=live }}</ref> such as confidently classifying unrecognizable images as belonging to a familiar category of ordinary images (2014)<ref>{{cite arXiv |eprint=1412.1897|last1=Nguyen|first1=Anh|title=Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images|last2=Yosinski|first2=Jason|last3=Clune|first3=Jeff|class=cs.CV|year=2014}}</ref> and misclassifying minuscule perturbations of correctly classified images (2013).<ref>{{cite arXiv |eprint=1312.6199|last1=Szegedy|first1=Christian|title=Intriguing properties of neural networks|last2=Zaremba|first2=Wojciech|last3=Sutskever|first3=Ilya|last4=Bruna|first4=Joan|last5=Erhan|first5=Dumitru|last6=Goodfellow|first6=Ian|last7=Fergus|first7=Rob|class=cs.CV|year=2013}}</ref> [[Ben Goertzel|Goertzel]] hypothesized that these behaviors are due to limitations in their internal representations and that these limitations would inhibit integration into heterogeneous multi-component [[artificial general intelligence]] (AGI) architectures.<ref name="goertzel" /> These issues may possibly be addressed by deep learning architectures that internally form states homologous to image-grammar<ref>{{cite journal | last1 = Zhu | first1 = S.C. | last2 = Mumford | first2 = D. | year = 2006| title = A stochastic grammar of images | journal = Found. Trends Comput. Graph. Vis. | volume = 2 | issue = 4| pages = 259–362 | doi = 10.1561/0600000018| citeseerx = 10.1.1.681.2190 }}</ref> decompositions of observed entities and events.<ref name="goertzel"/> [[Grammar induction|Learning a grammar]] (visual or linguistic) from training data would be equivalent to restricting the system to [[commonsense reasoning]] that operates on concepts in terms of grammatical [[Production (computer science)|production rules]] and is a basic goal of both human language acquisition<ref>Miller, G. A., and N. Chomsky. "Pattern conception." Paper for Conference on pattern detection, University of Michigan. 1957.</ref> and [[artificial intelligence]] (AI).<ref>{{cite web |first=Jason |last=Eisner |title=Deep Learning of Recursive Structure: Grammar Induction |url=http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/ |access-date=2015-05-10 |archive-date=2017-12-30 |archive-url=https://web.archive.org/web/20171230010335/http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/ |url-status=dead }}</ref>
=== Cyber threat ===
As deep learning moves from the lab into the world, research and experience show that artificial neural networks are vulnerable to hacks and deception.<ref>{{Cite web|url=https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425|title=Hackers Have Already Started to Weaponize Artificial Intelligence|website=Gizmodo|date=11 September 2017 |access-date=11 October 2019|archive-date=11 October 2019|archive-url=https://web.archive.org/web/20191011162231/https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425|url-status=live}}</ref> By identifying patterns that these systems use to function, attackers can modify inputs to ANNs in such a way that the ANN finds a match that human observers would not recognize. For example, an attacker can make subtle changes to an image such that the ANN finds a match even though the image looks to a human nothing like the search target. Such manipulation is termed an “adversarial attack.”<ref>{{Cite web|url=https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/|title=How hackers can force AI to make dumb mistakes|date=18 June 2018|website=The Daily Dot|language=en|access-date=11 October 2019|archive-date=11 October 2019|archive-url=https://web.archive.org/web/20191011162230/https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/|url-status=live}}</ref>
In 2016 researchers used one ANN to doctor images in trial and error fashion, identify another's focal points and thereby generate images that deceived it. The modified images looked no different to human eyes. Another group showed that printouts of doctored images then photographed successfully tricked an image classification system.<ref name=":4">{{Cite news|url=https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change|title=AI Is Easy to Fool—Why That Needs to Change|date=10 October 2017|work=Singularity Hub|access-date=11 October 2017|archive-date=11 October 2017|archive-url=https://web.archive.org/web/20171011233017/https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change/|url-status=live}}</ref> One defense is reverse image search, in which a possible fake image is submitted to a site such as [[TinEye]] that can then find other instances of it. A refinement is to search using only parts of the image, to identify images from which that piece may have been taken'''.'''<ref>{{Cite journal|last=Gibney|first=Elizabeth|title=The scientist who spots fake videos|url=https://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784|journal=Nature|doi=10.1038/nature.2017.22784|year=2017|access-date=2017-10-11|archive-date=2017-10-10|archive-url=https://web.archive.org/web/20171010011017/http://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784|url-status=live}}</ref>
Another group showed that certain [[Psychedelic art|psychedelic]] spectacles could fool a [[facial recognition system]] into thinking ordinary people were celebrities, potentially allowing one person to impersonate another. In 2017 researchers added stickers to [[stop sign]]s and caused an ANN to misclassify them.<ref name=":4" />
ANNs can however be further trained to detect attempts at deception, potentially leading attackers and defenders into an arms race similar to the kind that already defines the [[malware]] defense industry. ANNs have been trained to defeat ANN-based anti-malware software by repeatedly attacking a defense with malware that was continually altered by a [[genetic algorithm]] until it tricked the anti-malware while retaining its ability to damage the target.<ref name=":4" />
In 2016, another group demonstrated that certain sounds could make the [[Google Now]] voice command system open a particular web address, and hypothesized that this could "serve as a stepping stone for further attacks (e.g., opening a web page hosting drive-by malware)."<ref name=":4" />
In “data poisoning,” false data is continually smuggled into a machine learning system's training set to prevent it from achieving mastery.<ref name=":4" />
=== Reliance on human [[microwork]] ===
{{More citations needed|section|date=April 2021}}
Most Deep Learning systems rely on training and verification data that is generated and/or annotated by humans.<ref>{{Cite journal |last=Tubaro |first=Paola |date=2020 |title=Whose intelligence is artificial intelligence? |url=https://hal.science/hal-03029735 |journal=Global Dialogue |language=en |pages=38}}</ref> It has been argued in [[Media studies|media philosophy]] that not only low-paid [[Clickworkers|clickwork]] (e.g. on [[Amazon Mechanical Turk]]) is regularly deployed for this purpose, but also implicit forms of human [[microwork]] that are often not recognized as such.<ref name=":13">{{Cite journal|last=Mühlhoff|first=Rainer|date=6 November 2019|title=Human-aided artificial intelligence: Or, how to run large computations in human brains? Toward a media sociology of machine learning|journal=New Media & Society|language=en|volume=22|issue=10|pages=1868–1884|doi=10.1177/1461444819885334|s2cid=209363848|issn=1461-4448|url=https://depositonce.tu-berlin.de/handle/11303/12510}}</ref> The philosopher Rainer Mühlhoff distinguishes five types of "machinic capture" of human microwork to generate training data: (1) [[gamification]] (the embedding of annotation or computation tasks in the flow of a game), (2) "trapping and tracking" (e.g. [[CAPTCHA]]s for image recognition or click-tracking on Google [[Search engine results page|search results pages]]), (3) exploitation of social motivations (e.g. [[Tag (Facebook)|tagging faces]] on [[Facebook]] to obtain labeled facial images), (4) [[information mining]] (e.g. by leveraging [[Quantified self|quantified-self]] devices such as [[activity tracker]]s) and (5) [[Clickworkers|clickwork]].<ref name=":13" />
Mühlhoff argues that in most commercial end-user applications of Deep Learning such as [[DeepFace|Facebook's face recognition system]], the need for training data does not stop once an ANN is trained. Rather, there is a continued demand for human-generated verification data to constantly calibrate and update the ANN. For this purpose Facebook introduced the feature that once a user is automatically recognized in an image, they receive a notification. They can choose whether of not they like to be publicly labeled on the image, or tell Facebook that it is not them in the picture.<ref>{{Cite magazine|url=https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/|title=Facebook Can Now Find Your Face, Even When It's Not Tagged|magazine=Wired|access-date=22 November 2019|language=en|issn=1059-1028|archive-date=10 August 2019|archive-url=https://web.archive.org/web/20190810223940/https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/|url-status=live}}</ref> This user interface is a mechanism to generate "a constant stream of verification data"<ref name=":13" /> to further train the network in real-time. As Mühlhoff argues, involvement of human users to generate training and verification data is so typical for most commercial end-user applications of Deep Learning that such systems may be referred to as "human-aided artificial intelligence".<ref name=":13" />
== See also ==
* [[Applications of artificial intelligence]]
* [[Comparison of deep learning software]]
* [[Compressed sensing]]
* [[Differentiable programming]]
* [[Echo state network]]
* [[List of artificial intelligence projects]]
* [[Liquid state machine]]
* [[List of datasets for machine-learning research]]
* [[Reservoir computing]]
* [[Scale space#Deep learning and scale space|Scale space and deep learning]]
* [[Sparse coding]]
== References ==
{{Reflist|30em}}
== Further reading ==
{{refbegin}}
* {{cite book
|title=Deep Learning
|year=2016
|first1=Ian
|last1=Goodfellow
|author-link1=Ian Goodfellow
|first2=Yoshua
|last2=Bengio
|author-link2=Yoshua Bengio
|first3=Aaron
|last3=Courville
|publisher=MIT Press
|url=http://www.deeplearningbook.org
|isbn=978-0-26203561-3
|postscript=, introductory textbook.
|access-date=2021-05-09
|archive-date=2016-04-16
|archive-url=https://web.archive.org/web/20160416111010/http://www.deeplearningbook.org/
|url-status=live
}}
{{Prone to spam|date=June 2015}}<!-- {{No more links}}
Please be cautious about adding more external links.
Wikipedia is not a collection of links and should not be used for advertising.
Excessive or inappropriate links will be removed.
See [[Wikipedia:External links]] and [[Wikipedia:Spam]] for details.
If there are already suitable links, propose additions or replacements on
the article's talk page, or submit your link to the relevant category at
DMOZ (dmoz.org) and link there using {{Dmoz}}.
-->
{{refend}}
{{Differentiable computing}}
[[Category:Deep learning| ]]
[[Category:Artificial neural networks]]
[[Category:Emerging technologies]]' |
New page wikitext, after the edit (new_wikitext) | 'https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt
== Definition ==
Deep learning is a class of [[machine learning]] [[algorithm]]s that<ref name="BOOK2014">{{cite journal|last2=Yu|first2=D.|year=2014|title=Deep Learning: Methods and Applications|url=http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf|journal=Foundations and Trends in Signal Processing|volume=7|issue=3–4|pages=1–199|doi=10.1561/2000000039|last1=Deng|first1=L.|access-date=2014-10-18|archive-date=2016-03-14|archive-url=https://web.archive.org/web/20160314152112/http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf|url-status=live}}</ref>{{rp|pages=199–200}} uses multiple layers to progressively extract higher-level features from the raw input. For example, in [[image processing]], lower layers may identify edges, while higher layers may identify the concepts relevant to a human such as digits or letters or faces.
== Overview ==
Most modern deep learning models are based on [[artificial neural network]]s, specifically [[convolutional neural network]]s (CNN)s, although they can also include [[propositional formula]]s or latent variables organized layer-wise in deep [[generative model]]s such as the nodes in [[deep belief network]]s and deep [[Boltzmann machine]]s.<ref name="BENGIODEEP">{{cite journal|last=Bengio|first=Yoshua|year=2009|title=Learning Deep Architectures for AI|url=http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20%282009%29.pdf|journal=Foundations and Trends in Machine Learning|volume=2|issue=1|pages=1–127|doi=10.1561/2200000006|citeseerx=10.1.1.701.9550|access-date=3 September 2015|archive-url=https://web.archive.org/web/20160304084250/http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf|archive-date=4 March 2016|url-status=dead}}</ref>
In deep learning, each level learns to transform its input data into a slightly more abstract and composite representation. In an image recognition application, the raw input may be a [[Matrix (mathematics)|matrix]] of pixels; the first representational layer may abstract the pixels and encode edges; the second layer may compose and encode arrangements of edges; the third layer may encode a nose and eyes; and the fourth layer may recognize that the image contains a face. Importantly, a deep learning process can learn which features to optimally place in which level ''on its own''. This does not eliminate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction.<ref name="BENGIO2012">{{cite journal|last2=Courville|first2=A.|last3=Vincent|first3=P.|year=2013|title=Representation Learning: A Review and New Perspectives|journal=IEEE Transactions on Pattern Analysis and Machine Intelligence|volume=35|issue=8|pages=1798–1828|arxiv=1206.5538|doi=10.1109/tpami.2013.50|pmid=23787338|last1=Bengio|first1=Y.|s2cid=393948}}</ref><ref>{{cite journal|last1=LeCun|first1=Yann|last2=Bengio|first2=Yoshua|last3=Hinton|first3=Geoffrey|s2cid=3074096|title=Deep learning|journal=Nature|date=28 May 2015|volume=521|issue=7553|pages=436–444|doi=10.1038/nature14539|pmid=26017442|bibcode=2015Natur.521..436L}}</ref>
The word "deep" in "deep learning" refers to the number of layers through which the data is transformed. More precisely, deep learning systems have a substantial ''credit assignment path'' (CAP) depth. The CAP is the chain of transformations from input to output. CAPs describe potentially causal connections between input and output. For a [[feedforward neural network]], the depth of the CAPs is that of the network and is the number of hidden layers plus one (as the output layer is also parameterized). For [[recurrent neural network]]s, in which a signal may propagate through a layer more than once, the CAP depth is potentially unlimited.<ref name="SCHIDHUB" /> No universally agreed-upon threshold of depth divides shallow learning from deep learning, but most researchers agree that deep learning involves CAP depth higher than 2. CAP of depth 2 has been shown to be a universal approximator in the sense that it can emulate any function.<ref>{{Cite book|url=https://books.google.com/books?id=9CqQDwAAQBAJ&pg=PA15|title=Human Behavior and Another Kind in Consciousness: Emerging Research and Opportunities: Emerging Research and Opportunities|last=Shigeki|first=Sugiyama|date=12 April 2019|publisher=IGI Global|isbn=978-1-5225-8218-2|language=en}}</ref> Beyond that, more layers do not add to the function approximator ability of the network. Deep models (CAP > 2) are able to extract better features than shallow models and hence, extra layers help in learning the features effectively.
Deep learning architectures can be constructed with a [[greedy algorithm|greedy]] layer-by-layer method.<ref name=BENGIO2007>{{cite conference | first1=Yoshua | last1=Bengio | first2=Pascal | last2=Lamblin | first3=Dan | last3=Popovici | first4=Hugo | last4=Larochelle | title=Greedy layer-wise training of deep networks | year=2007 | url=http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf | conference=Advances in neural information processing systems | pages=153–160 | access-date=2019-10-06 | archive-date=2019-10-20 | archive-url=https://web.archive.org/web/20191020195638/http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf | url-status=live }}</ref> Deep learning helps to disentangle these abstractions and pick out which features improve performance.<ref name="BENGIO2012" />
For [[supervised learning]] tasks, deep learning methods eliminate [[feature engineering]], by translating the data into compact intermediate representations akin to [[Principal Component Analysis|principal components]], and derive layered structures that remove redundancy in representation.
Deep learning algorithms can be applied to unsupervised learning tasks. This is an important benefit because unlabeled data are more abundant than the labeled data. Examples of deep structures that can be trained in an unsupervised manner are [[deep belief network]]s.<ref name="BENGIO2012" /><ref name="SCHOLARDBNS">{{cite journal | last1 = Hinton | first1 = G.E. | year = 2009| title = Deep belief networks | journal = Scholarpedia | volume = 4 | issue = 5| page = 5947 | doi=10.4249/scholarpedia.5947| bibcode = 2009SchpJ...4.5947H| doi-access = free }}</ref>
== Interpretations ==
Deep neural networks are generally interpreted in terms of the [[universal approximation theorem]]<ref name=cyb>{{cite journal | last1 = Cybenko | year = 1989 | title = Approximations by superpositions of sigmoidal functions | url = http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf | journal = [[Mathematics of Control, Signals, and Systems]] | volume = 2 | issue = 4 | pages = 303–314 | doi = 10.1007/bf02551274 | s2cid = 3958369 | url-status = dead | archive-url = https://web.archive.org/web/20151010204407/http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf | archive-date = 10 October 2015 }}</ref><ref name=horn>{{cite journal | last1 = Hornik | first1 = Kurt | year = 1991 | title = Approximation Capabilities of Multilayer Feedforward Networks | journal = Neural Networks | volume = 4 | issue = 2| pages = 251–257 | doi=10.1016/0893-6080(91)90009-t}}</ref><ref name="Haykin, Simon 1998">{{cite book|first=Simon S. |last=Haykin|title=Neural Networks: A Comprehensive Foundation|url={{google books |plainurl=y |id=bX4pAQAAMAAJ}}|year=1999|publisher=Prentice Hall|isbn=978-0-13-273350-2}}</ref><ref name="Hassoun, M. 1995 p. 48">{{cite book|first=Mohamad H. |last=Hassoun|title=Fundamentals of Artificial Neural Networks|url={{google books |plainurl=y |id=Otk32Y3QkxQC|page=48}}|year=1995|publisher=MIT Press|isbn=978-0-262-08239-6|page=48}}</ref><ref name=ZhouLu>Lu, Z., Pu, H., Wang, F., Hu, Z., & Wang, L. (2017). [http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width The Expressive Power of Neural Networks: A View from the Width] {{Webarchive|url=https://web.archive.org/web/20190213005539/http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width |date=2019-02-13 }}. Neural Information Processing Systems, 6231-6239.</ref> or [[Bayesian inference|probabilistic inference]].<ref>{{cite journal |last1=Orhan |first1=A. E. |last2=Ma |first2=W. J. |date=2017 |title=Efficient probabilistic inference in generic neural networks trained with non-probabilistic feedback |journal=Nature Communications |volume=8 |issue=1 |pages=138 | pmid=28743932 | doi=10.1038/s41467-017-00181-8|pmc=5527101 |bibcode=2017NatCo...8..138O | doi-access=free}}</ref><ref name="BOOK2014" /><ref name="BENGIO2012" /><ref name="SCHIDHUB">{{cite journal|last=Schmidhuber|first=J.|s2cid=11715509|year=2015|title=Deep Learning in Neural Networks: An Overview|journal=Neural Networks|volume=61|pages=85–117|arxiv=1404.7828|doi=10.1016/j.neunet.2014.09.003|pmid=25462637}}</ref><ref name = MURPHY>{{cite book|first=Kevin P. |last=Murphy|title=Machine Learning: A Probabilistic Perspective|url={{google books |plainurl=y |id=NZP6AQAAQBAJ}}|date=24 August 2012|publisher=MIT Press|isbn=978-0-262-01802-9}}</ref>
The classic universal approximation theorem concerns the capacity of [[feedforward neural networks]] with a single hidden layer of finite size to approximate [[continuous functions]].<ref name="cyb"/><ref name="horn"/><ref name="Haykin, Simon 1998"/><ref name="Hassoun, M. 1995 p. 48"/> In 1989, the first proof was published by [[George Cybenko]] for [[sigmoid function|sigmoid]] activation functions<ref name="cyb" /> and was generalised to feed-forward multi-layer architectures in 1991 by Kurt Hornik.<ref name="horn" /> Recent work also showed that universal approximation also holds for non-bounded activation functions such as the rectified linear unit.<ref name=sonoda17>{{cite journal | last1 = Sonoda | first1 = Sho | last2=Murata | first2=Noboru | s2cid = 12149203 | year = 2017 | title = Neural network with unbounded activation functions is universal approximator | journal = Applied and Computational Harmonic Analysis | volume = 43 | issue = 2 | pages = 233–268 | doi = 10.1016/j.acha.2015.12.005| arxiv = 1505.03654 }}</ref>
The universal approximation theorem for [[deep neural network]]s concerns the capacity of networks with bounded width but the depth is allowed to grow. Lu et al.<ref name=ZhouLu/> proved that if the width of a [[deep neural network]] with [[ReLU]] activation is strictly larger than the input dimension, then the network can approximate any [[Lebesgue integration|Lebesgue integrable function]]; If the width is smaller or equal to the input dimension, then a [[deep neural network]] is not a universal approximator.
The [[probabilistic]] interpretation<ref name="MURPHY" /> derives from the field of [[machine learning]]. It features inference,<ref name="BOOK2014" /><ref name="BENGIODEEP" /><ref name="BENGIO2012" /><ref name="SCHIDHUB" /><ref name="SCHOLARDBNS" /><ref name="MURPHY" /> as well as the [[optimization]] concepts of [[training]] and [[test (assessment)|testing]], related to fitting and [[generalization]], respectively. More specifically, the probabilistic interpretation considers the activation nonlinearity as a [[cumulative distribution function]].<ref name="MURPHY" /> The probabilistic interpretation led to the introduction of [[dropout (neural networks)|dropout]] as [[Regularization (mathematics)|regularizer]] in neural networks. The probabilistic interpretation was introduced by researchers including [[John Hopfield|Hopfield]], [[Bernard Widrow|Widrow]] and [[Kumpati S. Narendra|Narendra]] and popularized in surveys such as the one by [[Christopher Bishop|Bishop]].<ref name="prml">{{cite book|title=Pattern Recognition and Machine Learning|author=Bishop, Christopher M.|year=2006|publisher=Springer|url=http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf|isbn=978-0-387-31073-2|access-date=2017-08-06|archive-date=2017-01-11|archive-url=https://web.archive.org/web/20170111005101/http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf|url-status=live}}</ref>
== History ==
Some sources point out that [[Frank Rosenblatt]] developed and explored all of the basic ingredients of the deep learning systems of today.<ref name="Who Is the Father of Deep Learning?">{{cite book |chapter-url=https://ieeexplore.ieee.org/document/9070967 |chapter=Who Is the Father of Deep Learning? |publisher=IEEE |doi=10.1109/CSCI49370.2019.00067 |accessdate=31 May 2021|title=2019 International Conference on Computational Science and Computational Intelligence (CSCI) |year=2019 |last1=Tappert |first1=Charles C. |pages=343–348 |isbn=978-1-7281-5584-5 |s2cid=216043128 }}</ref> He described it in his book "Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms", published by Cornell Aeronautical Laboratory, Inc., Cornell University in 1962.
The first general, working learning algorithm for supervised, deep, feedforward, multilayer [[perceptron]]s was published by [[Alexey Ivakhnenko]] and Lapa in 1967.<ref name="ivak1965">{{cite book|first1=A. G. |last1=Ivakhnenko |first2=V. G. |last2=Lapa |title=Cybernetics and Forecasting Techniques|url={{google books |plainurl=y |id=rGFgAAAAMAAJ}}|year=1967|publisher=American Elsevier Publishing Co.|isbn=978-0-444-00020-0}}</ref> A 1971 paper described a deep network with eight layers trained by the [[group method of data handling]].<ref name="ivak1971">{{Cite journal|last=Ivakhnenko|first=Alexey|date=1971|title=Polynomial theory of complex systems|url=http://gmdh.net/articles/history/polynomial.pdf|journal=IEEE Transactions on Systems, Man and Cybernetics|pages=364–378|doi=10.1109/TSMC.1971.4308320|volume=SMC-1|issue=4|access-date=2019-11-05|archive-date=2017-08-29|archive-url=https://web.archive.org/web/20170829230621/http://www.gmdh.net/articles/history/polynomial.pdf|url-status=live}}</ref> Other deep learning working architectures, specifically those built for [[computer vision]], began with the [[Neocognitron]] introduced by [[Kunihiko Fukushima]] in 1980.<ref name="FUKU1980">{{cite journal | last1 = Fukushima | first1 = K. | year = 1980 | title = Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position | journal = Biol. Cybern. | volume = 36 | issue = 4| pages = 193–202 | doi=10.1007/bf00344251 | pmid=7370364| s2cid = 206775608 }}</ref>
The term ''Deep Learning'' was introduced to the machine learning community by [[Rina Dechter]] in 1986,<ref name="dechter1986">[[Rina Dechter]] (1986). Learning while searching in constraint-satisfaction problems. University of California, Computer Science Department, Cognitive Systems Laboratory.[https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems Online] {{Webarchive|url=https://web.archive.org/web/20160419054654/https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems |date=2016-04-19 }}</ref> and to [[Artificial Neural Networks|artificial neural networks]] by Igor Aizenberg and colleagues in 2000, in the context of [[Boolean network|Boolean]] threshold neurons.<ref name="aizenberg2000">Igor Aizenberg, Naum N. Aizenberg, Joos P.L. Vandewalle (2000). Multi-Valued and Universal Binary Neurons: Theory, Learning and Applications. Springer Science & Business Media.</ref><ref>Co-evolving recurrent neurons learn deep memory POMDPs. Proc. GECCO, Washington, D. C., pp. 1795-1802, ACM Press, New York, NY, USA, 2005.</ref>
In 1989, [[Yann LeCun]] et al. applied the standard [[backpropagation]] algorithm, which had been around as the reverse mode of [[automatic differentiation]] since 1970,<ref name="lin1970">[[Seppo Linnainmaa]] (1970). The representation of the cumulative rounding error of an algorithm as a Taylor expansion of the local rounding errors. Master's Thesis (in Finnish), Univ. Helsinki, 6-7.</ref><ref name="grie2012">{{Cite journal|last=Griewank|first=Andreas|date=2012|title=Who Invented the Reverse Mode of Differentiation?|url=http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf|journal=Documenta Mathematica|issue=Extra Volume ISMP|pages=389–400|access-date=11 June 2017|archive-url=https://web.archive.org/web/20170721211929/http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf|archive-date=21 July 2017|url-status=dead}}</ref><ref name="WERBOS1974">{{Cite journal|last=Werbos|first=P.|date=1974|title=Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences |url=https://www.researchgate.net/publication/35657389 |journal=Harvard University |access-date=12 June 2017}}</ref><ref name="werbos1982">{{Cite book|chapter-url=ftp://ftp.idsia.ch/pub/juergen/habilitation.pdf|title=System modeling and optimization|last=Werbos|first=Paul|publisher=Springer|year=1982|pages=762–770|chapter=Applications of advances in nonlinear sensitivity analysis}}</ref> to a deep neural network with the purpose of [[Handwriting recognition|recognizing handwritten ZIP code]]s on mail. While the algorithm worked, training required 3 days.<ref name="LECUN1989">LeCun ''et al.'', "Backpropagation Applied to Handwritten Zip Code Recognition," ''Neural Computation'', 1, pp. 541–551, 1989.</ref>
Independently in 1988, Wei Zhang et al. applied the backpropagation algorithm to a convolutional neural network (a simplified Neocognitron by keeping only the convolutional interconnections between the image feature layers and the last fully connected layer) for alphabets recognition and also proposed an implementation of the CNN with an optical computing system.<ref name="wz1988">{{cite journal |last=Zhang |first=Wei |date=1988 |title=Shift-invariant pattern recognition neural network and its optical architecture |url=https://drive.google.com/file/d/1nN_5odSG_QVae54EsQN_qSz-0ZsX6wA0/view?usp=sharing |journal=Proceedings of Annual Conference of the Japan Society of Applied Physics}}</ref><ref name="wz1990">{{cite journal |last=Zhang |first=Wei |date=1990 |title=Parallel distributed processing model with local space-invariant interconnections and its optical architecture |url=https://drive.google.com/file/d/0B65v6Wo67Tk5ODRzZmhSR29VeDg/view?usp=sharing |journal=Applied Optics |volume=29 |issue=32 |pages=4790–7 |doi=10.1364/AO.29.004790 |pmid=20577468 |bibcode=1990ApOpt..29.4790Z}}</ref> Subsequently, Wei Zhang, et al. modified the model by removing the last fully connected layer and applied it for medical image object segmentation in 1991<ref>{{cite journal |last=Zhang |first=Wei |date=1991 |title=Image processing of human corneal endothelium based on a learning network |url=https://drive.google.com/file/d/0B65v6Wo67Tk5cm5DTlNGd0NPUmM/view?usp=sharing |journal=Applied Optics |volume=30 |issue=29 |pages=4211–7 |doi=10.1364/AO.30.004211 |pmid=20706526 |bibcode=1991ApOpt..30.4211Z}}</ref> and breast cancer detection in mammograms in 1994.<ref>{{cite journal |last=Zhang |first=Wei |date=1994 |title=Computerized detection of clustered microcalcifications in digital mammograms using a shift-invariant artificial neural network |url=https://drive.google.com/file/d/0B65v6Wo67Tk5Ml9qeW5nQ3poVTQ/view?usp=sharing |journal=Medical Physics |volume=21 |issue=4 |pages=517–24 |doi=10.1118/1.597177 |pmid=8058017 |bibcode=1994MedPh..21..517Z}}</ref>
In 1994, André de Carvalho, together with Mike Fairhurst and David Bisset, published experimental results of a multi-layer boolean neural network, also known as a weightless neural network, composed of a 3-layers self-organising feature extraction neural network module (SOFT) followed by a multi-layer classification neural network module (GSN), which were independently trained. Each layer in the feature extraction module extracted features with growing complexity regarding the previous layer.<ref>{{Cite journal |title=An integrated Boolean neural network for pattern classification |journal=Pattern Recognition Letters |date=8 August 1994 |pages=807–813 |volume=15 |issue=8 |doi=10.1016/0167-8655(94)90009-4 |first=Andre C. L. F. |last1=de Carvalho |first2 = Mike C. |last2=Fairhurst |first3=David |last3 = Bisset|bibcode=1994PaReL..15..807D }}</ref>
In 1995, [[Brendan Frey]] demonstrated that it was possible to train (over two days) a network containing six fully connected layers and several hundred hidden units using the [[wake-sleep algorithm]], co-developed with [[Peter Dayan]] and [[Geoffrey Hinton|Hinton]].<ref>{{Cite journal|title = The wake-sleep algorithm for unsupervised neural networks |journal = Science|date = 26 May 1995|pages = 1158–1161|volume = 268|issue = 5214|doi = 10.1126/science.7761831|pmid = 7761831|first1 = Geoffrey E.|last1 = Hinton|first2 = Peter|last2 = Dayan|first3 = Brendan J.|last3 = Frey|first4 = Radford|last4 = Neal|bibcode = 1995Sci...268.1158H}}</ref> Many factors contribute to the slow speed, including the [[vanishing gradient problem]] analyzed in 1991 by [[Sepp Hochreiter]].<ref name="HOCH1991">S. Hochreiter., "[http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf Untersuchungen zu dynamischen neuronalen Netzen] {{Webarchive|url=https://web.archive.org/web/20150306075401/http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf |date=2015-03-06 }}," ''Diploma thesis. Institut f. Informatik, Technische Univ. Munich. Advisor: J. Schmidhuber'', 1991.</ref><ref name="HOCH2001">{{cite book|chapter-url={{google books |plainurl=y |id=NWOcMVA64aAC}}|title=A Field Guide to Dynamical Recurrent Networks|last=Hochreiter|first=S.|display-authors=etal|date=15 January 2001|publisher=John Wiley & Sons|isbn=978-0-7803-5369-5|chapter=Gradient flow in recurrent nets: the difficulty of learning long-term dependencies|editor-last2=Kremer|editor-first2=Stefan C.|editor-first1=John F.|editor-last1=Kolen}}</ref>
Since 1997, Sven Behnke extended the feed-forward hierarchical convolutional approach in the Neural Abstraction Pyramid<ref>{{cite book
| last = Behnke | first = Sven
| doi = 10.1007/b11963
| isbn = 3-540-40722-7
| publisher = Springer
| series = Lecture Notes in Computer Science
| title = Hierarchical Neural Networks for Image Interpretation
| volume = 2766
| year = 2003| s2cid = 1304548
}}</ref> by lateral and backward connections in order to flexibly incorporate context into decisions and iteratively resolve local ambiguities.
Simpler models that use task-specific handcrafted features such as [[Gabor filter]]s and [[support vector machine]]s (SVMs) were a popular choice in the 1990s and 2000s, because of [[artificial neural network]]'s (ANN) computational cost and a lack of understanding of how the brain wires its biological networks.
Both shallow and deep learning (e.g., recurrent nets) of ANNs have been explored for many years.<ref>{{Cite journal|last1=Morgan|first1=Nelson|last2=Bourlard |first2=Hervé |last3=Renals |first3=Steve |last4=Cohen |first4=Michael|last5=Franco |first5=Horacio |date=1 August 1993 |title=Hybrid neural network/hidden markov model systems for continuous speech recognition |journal=International Journal of Pattern Recognition and Artificial Intelligence|volume=07|issue=4|pages=899–916|doi=10.1142/s0218001493000455|issn=0218-0014}}</ref><ref name="Robinson1992">{{Cite journal|last=Robinson|first=T.|author-link=Tony Robinson (speech recognition)|date=1992|title=A real-time recurrent error propagation network word recognition system|url=http://dl.acm.org/citation.cfm?id=1895720|journal=ICASSP|pages=617–620|isbn=9780780305328|series=Icassp'92|access-date=2017-06-12|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123135/https://dl.acm.org/doi/10.5555/1895550.1895720|url-status=live}}</ref><ref>{{Cite journal|last1=Waibel|first1=A.|last2=Hanazawa|first2=T.|last3=Hinton|first3=G.|last4=Shikano|first4=K.|last5=Lang|first5=K. J.|date=March 1989|title=Phoneme recognition using time-delay neural networks|journal=IEEE Transactions on Acoustics, Speech, and Signal Processing|volume=37|issue=3|pages=328–339|doi=10.1109/29.21701|issn=0096-3518|hdl=10338.dmlcz/135496|url=http://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf|access-date=2019-09-24|archive-date=2021-04-27|archive-url=https://web.archive.org/web/20210427001446/https://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf|url-status=live}}</ref> These methods never outperformed non-uniform internal-handcrafting Gaussian [[mixture model]]/[[Hidden Markov model]] (GMM-HMM) technology based on generative models of speech trained discriminatively.<ref name="Baker2009">{{cite journal | last1 = Baker | first1 = J. | last2 = Deng | first2 = Li | last3 = Glass | first3 = Jim | last4 = Khudanpur | first4 = S. | last5 = Lee | first5 = C.-H. | last6 = Morgan | first6 = N. | last7 = O'Shaughnessy | first7 = D. | year = 2009 | title = Research Developments and Directions in Speech Recognition and Understanding, Part 1 | journal = IEEE Signal Processing Magazine | volume = 26 | issue = 3| pages = 75–80 | doi=10.1109/msp.2009.932166| bibcode = 2009ISPM...26...75B | hdl = 1721.1/51891 | s2cid = 357467 }}</ref> Key difficulties have been analyzed, including gradient diminishing<ref name="HOCH1991" /> and weak temporal correlation structure in neural predictive models.<ref name="Bengio1991">{{Cite web|url=https://www.researchgate.net/publication/41229141|title=Artificial Neural Networks and their Application to Speech/Sequence Recognition|last=Bengio|first=Y.|date=1991|publisher=McGill University Ph.D. thesis|access-date=2017-06-12|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123131/https://www.researchgate.net/publication/41229141_Artificial_neural_networks_and_their_application_to_sequence_recognition|url-status=live}}</ref><ref name="Deng1994">{{cite journal | last1 = Deng | first1 = L. | last2 = Hassanein | first2 = K. | last3 = Elmasry | first3 = M. | year = 1994 | title = Analysis of correlation structure for a neural predictive model with applications to speech recognition | journal = Neural Networks | volume = 7 | issue = 2| pages = 331–339 | doi=10.1016/0893-6080(94)90027-2}}</ref> Additional difficulties were the lack of training data and limited computing power.
Most [[speech recognition]] researchers moved away from neural nets to pursue generative modeling. An exception was at [[SRI International]] in the late 1990s. Funded by the US government's [[National Security Agency|NSA]] and [[DARPA]], SRI studied deep neural networks in speech and [[speaker recognition]]. The speaker recognition team led by [[Larry Heck]] reported significant success with deep neural networks in speech processing in the 1998 [[National Institute of Standards and Technology]] Speaker Recognition evaluation.<ref name="Doddington2000">{{cite journal | last1 = Doddington | first1 = G. | last2 = Przybocki | first2 = M. | last3 = Martin | first3 = A. | last4 = Reynolds | first4 = D. | year = 2000 | title = The NIST speaker recognition evaluation ± Overview, methodology, systems, results, perspective | journal = Speech Communication | volume = 31 | issue = 2| pages = 225–254 | doi=10.1016/S0167-6393(99)00080-1}}</ref> The SRI deep neural network was then deployed in the Nuance Verifier, representing the first major industrial application of deep learning.<ref name="Heck2000">{{cite journal | last1 = Heck | first1 = L. | last2 = Konig | first2 = Y. | last3 = Sonmez | first3 = M. | last4 = Weintraub | first4 = M. | year = 2000 | title = Robustness to Telephone Handset Distortion in Speaker Recognition by Discriminative Feature Design | journal = Speech Communication | volume = 31 | issue = 2| pages = 181–192 | doi=10.1016/s0167-6393(99)00077-1}}</ref>
The principle of elevating "raw" features over hand-crafted optimization was first explored successfully in the architecture of deep autoencoder on the "raw" spectrogram or linear filter-bank features in the late 1990s,<ref name="Heck2000" /> showing its superiority over the Mel-Cepstral features that contain stages of fixed transformation from spectrograms. The raw features of speech, [[waveform]]s, later produced excellent larger-scale results.<ref>{{Cite web|url=https://www.researchgate.net/publication/266030526|title=Acoustic Modeling with Deep Neural Networks Using Raw Time Signal for LVCSR (PDF Download Available)|website=ResearchGate|access-date=14 June 2017|archive-date=9 May 2021|archive-url=https://web.archive.org/web/20210509123218/https://www.researchgate.net/publication/266030526_Acoustic_Modeling_with_Deep_Neural_Networks_Using_Raw_Time_Signal_for_LVCSR|url-status=live}}</ref>
Many aspects of speech recognition were taken over by a deep learning method called [[long short-term memory]] (LSTM), a recurrent neural network published by Hochreiter and [[Jürgen Schmidhuber|Schmidhuber]] in 1997.<ref name=":0">{{Cite journal|last1=Hochreiter|first1=Sepp|last2=Schmidhuber|first2=Jürgen|s2cid=1915014|date=1 November 1997|title=Long Short-Term Memory|journal=Neural Computation|volume=9|issue=8|pages=1735–1780|doi=10.1162/neco.1997.9.8.1735|issn=0899-7667|pmid=9377276}}</ref> LSTM [[Recurrent neural network|RNN]]s avoid the vanishing gradient problem and can learn "Very Deep Learning" tasks<ref name="SCHIDHUB" /> that require memories of events that happened thousands of discrete time steps before, which is important for speech. In 2003, LSTM started to become competitive with traditional speech recognizers on certain tasks.<ref name="graves2003">{{Cite web|url=Ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf|title=Biologically Plausible Speech Recognition with LSTM Neural Nets|last1=Graves|first1=Alex|last2=Eck|first2=Douglas|date=2003|website=1st Intl. Workshop on Biologically Inspired Approaches to Advanced Information Technology, Bio-ADIT 2004, Lausanne, Switzerland|pages=175–184|last3=Beringer|first3=Nicole|last4=Schmidhuber|first4=Jürgen|access-date=2016-04-09|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123139/ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf|url-status=live}}</ref> Later it was combined with connectionist temporal classification (CTC)<ref name=":1">{{Cite journal|last1=Graves|first1=Alex|last2=Fernández|first2=Santiago|last3=Gomez|first3=Faustino|date=2006|title=Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks|journal=Proceedings of the International Conference on Machine Learning, ICML 2006|pages=369–376|citeseerx=10.1.1.75.6306}}</ref> in stacks of LSTM RNNs.<ref name="fernandez2007keyword">Santiago Fernandez, Alex Graves, and Jürgen Schmidhuber (2007). [https://mediatum.ub.tum.de/doc/1289941/file.pdf An application of recurrent neural networks to discriminative keyword spotting] {{Webarchive|url=https://web.archive.org/web/20181118164457/https://mediatum.ub.tum.de/doc/1289941/file.pdf |date=2018-11-18 }}. Proceedings of ICANN (2), pp. 220–229.</ref> In 2015, Google's speech recognition reportedly experienced a dramatic performance jump of 49% through CTC-trained LSTM, which they made available through [[Google Voice Search]].<ref name="sak2015">{{Cite web|url=http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html|title=Google voice search: faster and more accurate|last1=Sak|first1=Haşim|last2=Senior|first2=Andrew|date=September 2015|last3=Rao|first3=Kanishka|last4=Beaufays|first4=Françoise|last5=Schalkwyk|first5=Johan|access-date=2016-04-09|archive-date=2016-03-09|archive-url=https://web.archive.org/web/20160309191532/http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html|url-status=live}}</ref>
In 2006, publications by [[Geoffrey Hinton|Geoff Hinton]], [[Russ Salakhutdinov|Ruslan Salakhutdinov]], Osindero and [[Yee Whye Teh|Teh]]<ref>{{Cite journal|last=Hinton|first=Geoffrey E.|date=1 October 2007|title=Learning multiple layers of representation|url=http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3|journal=Trends in Cognitive Sciences|volume=11|issue=10|pages=428–434|doi=10.1016/j.tics.2007.09.004|issn=1364-6613|pmid=17921042|s2cid=15066318|access-date=12 June 2017|archive-date=11 October 2013|archive-url=https://web.archive.org/web/20131011071435/http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3|url-status=live}}</ref><ref name=hinton06>{{Cite journal | last1 = Hinton | first1 = G. E. | author-link1 = Geoff Hinton | last2 = Osindero | first2 = S. | last3 = Teh | first3 = Y. W. | doi = 10.1162/neco.2006.18.7.1527 | title = A Fast Learning Algorithm for Deep Belief Nets | journal = [[Neural Computation (journal)|Neural Computation]] | volume = 18 | issue = 7 | pages = 1527–1554 | year = 2006 | pmid = 16764513 | s2cid = 2309950 | url = http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf | access-date = 2011-07-20 | archive-date = 2015-12-23 | archive-url = https://web.archive.org/web/20151223164129/http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf | url-status = live }}</ref><ref name=bengio2012>{{cite arXiv |last=Bengio |first=Yoshua |author-link=Yoshua Bengio |eprint=1206.5533 |title=Practical recommendations for gradient-based training of deep architectures |class=cs.LG|year=2012 }}</ref> showed how a many-layered [[feedforward neural network]] could be effectively pre-trained one layer at a time, treating each layer in turn as an unsupervised [[restricted Boltzmann machine]], then fine-tuning it using supervised [[backpropagation]].<ref name="HINTON2007">G. E. Hinton., "[http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf Learning multiple layers of representation] {{Webarchive|url=https://web.archive.org/web/20180522112408/http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf |date=2018-05-22 }}," ''Trends in Cognitive Sciences'', 11, pp. 428–434, 2007.</ref> The papers referred to ''learning'' for ''deep belief nets.''
Deep learning is part of state-of-the-art systems in various disciplines, particularly computer vision and [[automatic speech recognition]] (ASR). Results on commonly used evaluation sets such as [[TIMIT]] (ASR) and [[MNIST database|MNIST]] ([[image classification]]), as well as a range of large-vocabulary speech recognition tasks have steadily improved.<ref name="HintonDengYu2012" /><ref>{{Cite book |doi=10.1109/ICCCI50826.2021.9402569|isbn=978-1-7281-5875-4|chapter=Non-linear frequency warping using constant-Q transformation for speech emotion recognition|title=2021 International Conference on Computer Communication and Informatics (ICCCI)|pages=1–4|year=2021|last1=Singh|first1=Premjeet|last2=Saha|first2=Goutam|last3=Sahidullah|first3=Md|arxiv=2102.04029|s2cid=231846518}}</ref> [[Convolutional neural network]]s (CNNs) were superseded for ASR by CTC<ref name=":1" /> for [[LSTM]].<ref name=":0" /><ref name="sak2015" /><ref name="sak2014">{{Cite web|url=https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf|title=Long Short-Term Memory recurrent neural network architectures for large scale acoustic modeling|last1=Sak|first1=Hasim|last2=Senior|first2=Andrew|date=2014|last3=Beaufays|first3=Francoise|archive-url=https://web.archive.org/web/20180424203806/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf|archive-date=24 April 2018|url-status=dead}}</ref><ref name="liwu2015">{{cite arXiv |eprint=1410.4281|last1=Li|first1=Xiangang|title=Constructing Long Short-Term Memory based Deep Recurrent Neural Networks for Large Vocabulary Speech Recognition|last2=Wu|first2=Xihong|class=cs.CL|year=2014}}</ref><ref name="zen2015">{{Cite web|url=https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf|title=Unidirectional Long Short-Term Memory Recurrent Neural Network with Recurrent Output Layer for Low-Latency Speech Synthesis|last1=Zen|first1=Heiga|last2=Sak|first2=Hasim|date=2015|website=Google.com|publisher=ICASSP|pages=4470–4474|access-date=2017-06-13|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123113/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf|url-status=live}}</ref> but are more successful in computer vision.
The impact of deep learning in industry began in the early 2000s, when CNNs already processed an estimated 10% to 20% of all the checks written in the US, according to Yann LeCun.<ref name="lecun2016slides">[[Yann LeCun]] (2016). Slides on Deep Learning [https://indico.cern.ch/event/510372/ Online] {{Webarchive|url=https://web.archive.org/web/20160423021403/https://indico.cern.ch/event/510372/ |date=2016-04-23 }}</ref> Industrial applications of deep learning to large-scale speech recognition started around 2010.
The 2009 NIPS Workshop on Deep Learning for Speech Recognition was motivated by the limitations of deep generative models of speech, and the possibility that given more capable hardware and large-scale data sets that deep neural nets (DNN) might become practical. It was believed that pre-training DNNs using generative models of deep belief nets (DBN) would overcome the main difficulties of neural nets. However, it was discovered that replacing pre-training with large amounts of training data for straightforward [[backpropagation]] when using DNNs with large, context-dependent output layers produced error rates dramatically lower than then-state-of-the-art Gaussian mixture model (GMM)/Hidden Markov Model (HMM) and also than more-advanced generative model-based systems.<ref name="HintonDengYu2012">{{cite journal | last1 = Hinton | first1 = G. | last2 = Deng | first2 = L. | last3 = Yu | first3 = D. | last4 = Dahl | first4 = G. | last5 = Mohamed | first5 = A. | last6 = Jaitly | first6 = N. | last7 = Senior | first7 = A. | last8 = Vanhoucke | first8 = V. | last9 = Nguyen | first9 = P. | last10 = Sainath | first10 = T. | last11 = Kingsbury | first11 = B. | year = 2012 | title = Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups| journal = IEEE Signal Processing Magazine | volume = 29 | issue = 6| pages = 82–97 | doi=10.1109/msp.2012.2205597 | bibcode = 2012ISPM...29...82H| s2cid = 206485943 }}</ref> The nature of the recognition errors produced by the two types of systems was characteristically different,<ref name="ReferenceICASSP2013" /> offering technical insights into how to integrate deep learning into the existing highly efficient, run-time speech decoding system deployed by all major speech recognition systems.<ref name="BOOK2014" /><ref name="ReferenceA">{{cite book|last2=Deng|first2=L.|date=2014|title=Automatic Speech Recognition: A Deep Learning Approach (Publisher: Springer)|url={{google books |plainurl=y |id=rUBTBQAAQBAJ}}|isbn=978-1-4471-5779-3|last1=Yu|first1=D.}}</ref><ref>{{cite web|title=Deng receives prestigious IEEE Technical Achievement Award - Microsoft Research|url=https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/|website=Microsoft Research|date=3 December 2015|access-date=16 March 2018|archive-date=16 March 2018|archive-url=https://web.archive.org/web/20180316084821/https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/|url-status=live}}</ref> Analysis around 2009–2010, contrasting the GMM (and other generative speech models) vs. DNN models, stimulated early industrial investment in deep learning for speech recognition,<ref name="ReferenceICASSP2013" /> eventually leading to pervasive and dominant use in that industry. That analysis was done with comparable performance (less than 1.5% in error rate) between discriminative DNNs and generative models.<ref name="HintonDengYu2012" /><ref name="ReferenceICASSP2013">{{cite journal|last2=Hinton|first2=G.|last3=Kingsbury|first3=B.|date=2013|title=New types of deep neural network learning for speech recognition and related applications: An overview (ICASSP)|url=https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf|last1=Deng|first1=L.|journal=|access-date=2017-06-12|archive-date=2017-09-26|archive-url=https://web.archive.org/web/20170926190920/https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf|url-status=live}}</ref><ref name="interspeech2014Keynote">{{Cite web|url=https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing|title=Keynote talk: 'Achievements and Challenges of Deep Learning - From Speech Analysis and Recognition To Language and Multimodal Processing'|last=Li|first=Deng|date=September 2014|website=Interspeech|access-date=2017-06-12|archive-date=2017-09-26|archive-url=https://web.archive.org/web/20170926190732/https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing|url-status=live}}</ref>
In 2010, researchers extended deep learning from TIMIT to large vocabulary speech recognition, by adopting large output layers of the DNN based on context-dependent HMM states constructed by [[decision tree]]s.<ref name="Roles2010">{{cite journal|last1=Yu|first1=D.|last2=Deng|first2=L.|date=2010|title=Roles of Pre-Training and Fine-Tuning in Context-Dependent DBN-HMMs for Real-World Speech Recognition|url=https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/|journal=NIPS Workshop on Deep Learning and Unsupervised Feature Learning|access-date=2017-06-14|archive-date=2017-10-12|archive-url=https://web.archive.org/web/20171012095148/https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/|url-status=live}}</ref><ref>{{Cite journal|last1=Seide|first1=F.|last2=Li|first2=G.|last3=Yu|first3=D.|date=2011|title=Conversational speech transcription using context-dependent deep neural networks|url=https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks|journal=Interspeech|pages=437–440|doi=10.21437/Interspeech.2011-169|access-date=2017-06-14|archive-date=2017-10-12|archive-url=https://web.archive.org/web/20171012095522/https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks/|url-status=live}}</ref><ref>{{Cite journal|last1=Deng|first1=Li|last2=Li|first2=Jinyu|last3=Huang|first3=Jui-Ting|last4=Yao|first4=Kaisheng|last5=Yu|first5=Dong|last6=Seide|first6=Frank|last7=Seltzer|first7=Mike|last8=Zweig|first8=Geoff|last9=He|first9=Xiaodong|date=1 May 2013|title=Recent Advances in Deep Learning for Speech Research at Microsoft|url=https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/|journal=Microsoft Research|access-date=14 June 2017|archive-date=12 October 2017|archive-url=https://web.archive.org/web/20171012044053/https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/|url-status=live}}</ref><ref name="ReferenceA" />
Advances in hardware have driven renewed interest in deep learning. In 2009, [[Nvidia]] was involved in what was called the “big bang” of deep learning, “as deep-learning neural networks were trained with Nvidia [[graphics processing unit]]s (GPUs).”<ref>{{cite web|url=https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/|title=Nvidia CEO bets big on deep learning and VR|date=5 April 2016|publisher=[[Venture Beat]]|access-date=21 April 2017|archive-date=25 November 2020|archive-url=https://web.archive.org/web/20201125202428/https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/|url-status=live}}</ref> That year, [[Andrew Ng]] determined that GPUs could increase the speed of deep-learning systems by about 100 times.<ref>{{cite news|url=https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not|title=From not working to neural networking|newspaper=[[The Economist]]|access-date=2017-08-26|archive-date=2016-12-31|archive-url=https://web.archive.org/web/20161231203934/https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not|url-status=live}}</ref> In particular, GPUs are well-suited for the matrix/vector computations involved in machine learning.<ref name="jung2004">{{cite journal | last1 = Oh | first1 = K.-S. | last2 = Jung | first2 = K. | year = 2004 | title = GPU implementation of neural networks | journal = Pattern Recognition | volume = 37 | issue = 6| pages = 1311–1314 | doi=10.1016/j.patcog.2004.01.013| bibcode = 2004PatRe..37.1311O }}</ref><ref>"[https://www.academia.edu/40135801 A Survey of Techniques for Optimizing Deep Learning on GPUs] {{Webarchive|url=https://web.archive.org/web/20210509123120/https://www.academia.edu/40135801/A_Survey_of_Techniques_for_Optimizing_Deep_Learning_on_GPUs |date=2021-05-09 }}", S. Mittal and S. Vaishay, Journal of Systems Architecture, 2019</ref><ref name="chellapilla2006">{{Citation | first1 = Kumar | last1 = Chellapilla | first2 = Sidd | last2 = Puri | first3 = Patrice | last3 = Simard | title = High performance convolutional neural networks for document processing | url = https://hal.inria.fr/inria-00112631/document | date = 2006 | access-date = 2021-02-14 | archive-date = 2020-05-18 | archive-url = https://web.archive.org/web/20200518193413/https://hal.inria.fr/inria-00112631/document | url-status = live }}</ref> GPUs speed up training algorithms by orders of magnitude, reducing running times from weeks to days.<ref name=":3">{{Cite journal|last1=Cireşan|first1=Dan Claudiu|last2=Meier|first2=Ueli|last3=Gambardella|first3=Luca Maria|last4=Schmidhuber|first4=Jürgen|date=21 September 2010|title=Deep, Big, Simple Neural Nets for Handwritten Digit Recognition|journal=Neural Computation|volume=22|issue=12|pages=3207–3220|doi=10.1162/neco_a_00052|pmid=20858131|issn=0899-7667|arxiv=1003.0358|s2cid=1918673}}</ref><ref>{{Cite journal|last1=Raina|first1=Rajat|last2=Madhavan|first2=Anand|last3=Ng|first3=Andrew Y.|s2cid=392458|date=2009|title=Large-scale Deep Unsupervised Learning Using Graphics Processors|journal=Proceedings of the 26th Annual International Conference on Machine Learning|series=ICML '09|location=New York, NY, USA|publisher=ACM|pages=873–880|doi=10.1145/1553374.1553486|isbn=9781605585161|citeseerx=10.1.1.154.372}}</ref> Further, specialized hardware and algorithm optimizations can be used for efficient processing of deep learning models.<ref name="sze2017">{{cite arXiv
|title= Efficient Processing of Deep Neural Networks: A Tutorial and Survey
|last1=Sze |first1=Vivienne|author1-link=Vivienne Sze
|last2=Chen |first2=Yu-Hsin
|last3=Yang |first3=Tien-Ju
|last4=Emer |first4=Joel
|eprint=1703.09039
|year=2017
|class=cs.CV }}</ref>
=== Deep learning revolution ===
[[File:AI-ML-DL.svg|thumb|How deep learning is a subset of machine learning and how machine learning is a subset of artificial intelligence (AI)]]
In 2012, a team led by George E. Dahl won the "Merck Molecular Activity Challenge" using multi-task deep neural networks to predict the [[biomolecular target]] of one drug.<ref name="MERCK2012">{{Cite web|url=https://kaggle.com/c/MerckActivity|title=Merck Molecular Activity Challenge|website=kaggle.com|access-date=2020-07-16|archive-date=2020-07-16|archive-url=https://web.archive.org/web/20200716190808/https://www.kaggle.com/c/MerckActivity|url-status=live}}</ref><ref name=":5">{{Cite web|url=http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions|title=Multi-task Neural Networks for QSAR Predictions {{!}} Data Science Association|website=www.datascienceassn.org|access-date=14 June 2017|archive-date=30 April 2017|archive-url=https://web.archive.org/web/20170430142049/http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions|url-status=live}}</ref> In 2014, Hochreiter's group used deep learning to detect off-target and toxic effects of environmental chemicals in nutrients, household products and drugs and won the "Tox21 Data Challenge" of [[NIH]], [[FDA]] and [[National Center for Advancing Translational Sciences|NCATS]].<ref name="TOX21">"Toxicology in the 21st century Data Challenge"</ref><ref name="TOX21Data">{{cite web|url=https://tripod.nih.gov/tox21/challenge/leaderboard.jsp|title=NCATS Announces Tox21 Data Challenge Winners|access-date=2015-03-05|archive-date=2015-09-08|archive-url=https://web.archive.org/web/20150908025122/https://tripod.nih.gov/tox21/challenge/leaderboard.jsp|url-status=live}}</ref><ref name=":11">{{cite web|url=http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html|title=NCATS Announces Tox21 Data Challenge Winners|archive-url=https://web.archive.org/web/20150228225709/http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html|archive-date=28 February 2015|url-status=dead|access-date=5 March 2015}}</ref>
Significant additional impacts in image or object recognition were felt from 2011 to 2012. Although CNNs trained by [[backpropagation]] had been around for decades, and GPU implementations of NNs for years, including CNNs, fast implementations of CNNs on GPUs were needed to progress on computer vision.<ref name="jung2004" /><ref name="chellapilla2006" /><ref name="LECUN1989" /><ref name=":6">{{Cite journal|last1=Ciresan|first1=D. C.|last2=Meier|first2=U.|last3=Masci|first3=J.|last4=Gambardella|first4=L. M.|last5=Schmidhuber|first5=J.|date=2011|title=Flexible, High Performance Convolutional Neural Networks for Image Classification|url=http://ijcai.org/papers11/Papers/IJCAI11-210.pdf|journal=International Joint Conference on Artificial Intelligence|doi=10.5591/978-1-57735-516-8/ijcai11-210|access-date=2017-06-13|archive-date=2014-09-29|archive-url=https://web.archive.org/web/20140929094040/http://ijcai.org/papers11/Papers/IJCAI11-210.pdf|url-status=live}}</ref><ref name="SCHIDHUB" /> In 2011, this approach achieved for the first time superhuman performance in a visual pattern recognition contest. Also in 2011, it won the ICDAR Chinese handwriting contest, and in May 2012, it won the ISBI image segmentation contest.<ref name=":8">{{Cite book|url=http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf|title=Advances in Neural Information Processing Systems 25|last1=Ciresan|first1=Dan|last2=Giusti|first2=Alessandro|last3=Gambardella|first3=Luca M.|last4=Schmidhuber|first4=Juergen|date=2012|publisher=Curran Associates, Inc.|editor-last=Pereira|editor-first=F.|pages=2843–2851|editor-last2=Burges|editor-first2=C. J. C.|editor-last3=Bottou|editor-first3=L.|editor-last4=Weinberger|editor-first4=K. Q.|access-date=2017-06-13|archive-date=2017-08-09|archive-url=https://web.archive.org/web/20170809081713/http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf|url-status=live}}</ref> Until 2011, CNNs did not play a major role at computer vision conferences, but in June 2012, a paper by Ciresan et al. at the leading conference CVPR<ref name=":9" /> showed how max-pooling CNNs on GPU can dramatically improve many vision benchmark records. In October 2012, a similar system by Krizhevsky et al.<ref name="krizhevsky2012" /> won the large-scale [[ImageNet competition]] by a significant margin over shallow machine learning methods. In November 2012, Ciresan et al.'s system also won the ICPR contest on analysis of large medical images for cancer detection, and in the following year also the MICCAI Grand Challenge on the same topic.<ref name="ciresan2013miccai">{{Cite journal|last1=Ciresan|first1=D.|last2=Giusti|first2=A.|last3=Gambardella|first3=L.M.|last4=Schmidhuber|first4=J.|date=2013|title=Mitosis Detection in Breast Cancer Histology Images using Deep Neural Networks|journal=Proceedings MICCAI|volume=7908|issue=Pt 2|pages=411–418|doi=10.1007/978-3-642-40763-5_51|pmid=24579167|series=Lecture Notes in Computer Science|isbn=978-3-642-38708-1}}</ref> In 2013 and 2014, the error rate on the ImageNet task using deep learning was further reduced, following a similar trend in large-scale speech recognition.
Image classification was then extended to the more challenging task of [[Automatic image annotation|generating descriptions]] (captions) for images, often as a combination of CNNs and LSTMs.<ref name="1411.4555">{{cite arXiv |eprint=1411.4555|last1=Vinyals|first1=Oriol|title=Show and Tell: A Neural Image Caption Generator|last2=Toshev|first2=Alexander|last3=Bengio|first3=Samy|last4=Erhan|first4=Dumitru|class=cs.CV|year=2014}}.</ref><ref name="1411.4952">{{cite arXiv |eprint=1411.4952|last1=Fang|first1=Hao|title=From Captions to Visual Concepts and Back|last2=Gupta|first2=Saurabh|last3=Iandola|first3=Forrest|last4=Srivastava|first4=Rupesh|last5=Deng|first5=Li|last6=Dollár|first6=Piotr|last7=Gao|first7=Jianfeng|last8=He|first8=Xiaodong|last9=Mitchell|first9=Margaret|last10=Platt|first10=John C|last11=Lawrence Zitnick|first11=C|last12=Zweig|first12=Geoffrey|class=cs.CV|year=2014}}.</ref><ref name="1411.2539">{{cite arXiv |eprint=1411.2539|last1=Kiros|first1=Ryan|title=Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models|last2=Salakhutdinov|first2=Ruslan|last3=Zemel|first3=Richard S|class=cs.LG|year=2014}}.</ref>
Some researchers state that the October 2012 ImageNet victory anchored the start of a "deep learning revolution" that has transformed the AI industry.<ref>{{cite news|title=Why Deep Learning Is Suddenly Changing Your Life|url=http://fortune.com/ai-artificial-intelligence-deep-machine-learning/|access-date=13 April 2018|work=Fortune|date=2016|archive-date=14 April 2018|archive-url=https://web.archive.org/web/20180414031925/http://fortune.com/ai-artificial-intelligence-deep-machine-learning/|url-status=live}}</ref>
In March 2019, [[Yoshua Bengio]], [[Geoffrey Hinton]] and [[Yann LeCun]] were awarded the [[Turing Award]] for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing.
== Neural networks ==
=== Artificial neural networks ===
{{Main|Artificial neural network}}
'''Artificial neural networks''' ('''ANNs''') or '''[[Connectionism|connectionist]] systems''' are computing systems inspired by the [[biological neural network]]s that constitute animal brains. Such systems learn (progressively improve their ability) to do tasks by considering examples, generally without task-specific programming. For example, in image recognition, they might learn to identify images that contain cats by analyzing example images that have been manually [[Labeled data|labeled]] as "cat" or "no cat" and using the analytic results to identify cats in other images. They have found most use in applications difficult to express with a traditional computer algorithm using [[rule-based programming]].
An ANN is based on a collection of connected units called [[artificial neuron]]s, (analogous to biological neurons in a [[Brain|biological brain]]). Each connection ([[synapse]]) between neurons can transmit a signal to another neuron. The receiving (postsynaptic) neuron can process the signal(s) and then signal downstream neurons connected to it. Neurons may have state, generally represented by [[real numbers]], typically between 0 and 1. Neurons and synapses may also have a weight that varies as learning proceeds, which can increase or decrease the strength of the signal that it sends downstream.
Typically, neurons are organized in layers. Different layers may perform different kinds of transformations on their inputs. Signals travel from the first (input), to the last (output) layer, possibly after traversing the layers multiple times.
The original goal of the neural network approach was to solve problems in the same way that a human brain would. Over time, attention focused on matching specific mental abilities, leading to deviations from biology such as [[backpropagation]], or passing information in the reverse direction and adjusting the network to reflect that information.
Neural networks have been used on a variety of tasks, including computer vision, [[speech recognition]], [[machine translation]], [[social network]] filtering, [[general game playing|playing board and video games]] and medical diagnosis.
As of 2017, neural networks typically have a few thousand to a few million units and millions of connections. Despite this number being several order of magnitude less than the number of neurons on a human brain, these networks can perform many tasks at a level beyond that of humans (e.g., recognizing faces, or playing "Go"<ref>{{Cite journal|last1=Silver|first1=David|last2=Huang|first2=Aja|last3=Maddison|first3=Chris J.|last4=Guez|first4=Arthur|last5=Sifre|first5=Laurent|last6=Driessche|first6=George van den|last7=Schrittwieser|first7=Julian|last8=Antonoglou|first8=Ioannis|last9=Panneershelvam|first9=Veda|s2cid=515925|date=January 2016|title=Mastering the game of Go with deep neural networks and tree search|journal=Nature|volume=529|issue=7587|pages=484–489|doi=10.1038/nature16961|issn=1476-4687|pmid=26819042|bibcode=2016Natur.529..484S}}</ref> ).
=== Deep neural networks ===
A deep neural network (DNN) is an [[artificial neural network]] (ANN) with multiple layers between the input and output layers.<ref name="BENGIODEEP" /><ref name="SCHIDHUB" /> There are different types of neural networks but they always consist of the same components: neurons, synapses, weights, biases, and functions.<ref name="Nokkada">{{citation |title=A Guide to Deep Learning and Neural Networks |url=https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks |access-date=2020-11-16 |archive-date=2020-11-02 |archive-url=https://web.archive.org/web/20201102151103/https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks |url-status=live }}</ref> These components as a whole function similarly to a human brain, and can be trained like any other ML algorithm.{{Citation needed|date=November 2020}}
For example, a DNN that is trained to recognize dog breeds will go over the given image and calculate the probability that the dog in the image is a certain breed. The user can review the results and select which probabilities the network should display (above a certain threshold, etc.) and return the proposed label. Each mathematical manipulation as such is considered a layer,{{citation needed|date=March 2022}} and complex DNN have many layers, hence the name "deep" networks.
DNNs can model complex non-linear relationships. DNN architectures generate compositional models where the object is expressed as a layered composition of [[Primitive data type|primitives]].<ref>{{Cite journal|last1=Szegedy|first1=Christian|last2=Toshev|first2=Alexander|last3=Erhan|first3=Dumitru|date=2013|title=Deep neural networks for object detection|url=https://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection|journal=Advances in Neural Information Processing Systems|pages=2553–2561|access-date=2017-06-13|archive-date=2017-06-29|archive-url=https://web.archive.org/web/20170629172111/http://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection|url-status=live}}</ref> The extra layers enable composition of features from lower layers, potentially modeling complex data with fewer units than a similarly performing shallow network.<ref name="BENGIODEEP" /> For instance, it was proved that sparse [[multivariate polynomial]]s are exponentially easier to approximate with DNNs than with shallow networks.<ref>{{cite conference|last1=Rolnick|first1=David|last2=Tegmark|first2=Max|date=2018|title=The power of deeper networks for expressing natural functions|url=https://openreview.net/pdf?id=SyProzZAW|conference=ICLR 2018|book-title=International Conference on Learning Representations|access-date=2021-01-05|archive-date=2021-01-07|archive-url=https://web.archive.org/web/20210107183647/https://openreview.net/pdf?id=SyProzZAW|url-status=live}}</ref>
Deep architectures include many variants of a few basic approaches. Each architecture has found success in specific domains. It is not always possible to compare the performance of multiple architectures, unless they have been evaluated on the same data sets.
DNNs are typically feedforward networks in which data flows from the input layer to the output layer without looping back. At first, the DNN creates a map of virtual neurons and assigns random numerical values, or "weights", to connections between them. The weights and inputs are multiplied and return an output between 0 and 1. If the network did not accurately recognize a particular pattern, an algorithm would adjust the weights.<ref>{{Cite news|last=Hof|first=Robert D.|title=Is Artificial Intelligence Finally Coming into Its Own?|work=MIT Technology Review|url=https://www.technologyreview.com/s/513696/deep-learning/|access-date=10 July 2018|archive-url=https://web.archive.org/web/20190331092832/https://www.technologyreview.com/s/513696/deep-learning/|archive-date=31 March 2019}}</ref> That way the algorithm can make certain parameters more influential, until it determines the correct mathematical manipulation to fully process the data.
[[Recurrent neural networks]] (RNNs), in which data can flow in any direction, are used for applications such as [[language model]]ing.<ref name="gers2001">{{cite journal|last1=Gers|first1=Felix A.|last2=Schmidhuber|first2=Jürgen|year=2001|title=LSTM Recurrent Networks Learn Simple Context Free and Context Sensitive Languages|url=http://elartu.tntu.edu.ua/handle/lib/30719|journal=IEEE Transactions on Neural Networks|volume=12|issue=6|pages=1333–1340|doi=10.1109/72.963769|pmid=18249962|access-date=2020-02-25|archive-date=2020-01-26|archive-url=https://web.archive.org/web/20200126045722/http://elartu.tntu.edu.ua/handle/lib/30719|url-status=live}}</ref><ref name="NIPS2014"/><ref name="vinyals2016">{{cite arXiv |eprint=1602.02410|last1=Jozefowicz|first1=Rafal|title=Exploring the Limits of Language Modeling|last2=Vinyals|first2=Oriol|last3=Schuster|first3=Mike|last4=Shazeer|first4=Noam|last5=Wu|first5=Yonghui|class=cs.CL|year=2016}}</ref><ref name="gillick2015">{{cite arXiv |eprint=1512.00103|last1=Gillick|first1=Dan|title=Multilingual Language Processing from Bytes|last2=Brunk|first2=Cliff|last3=Vinyals|first3=Oriol|last4=Subramanya|first4=Amarnag|class=cs.CL|year=2015}}</ref><ref name="MIKO2010">{{Cite journal|last=Mikolov|first=T.|display-authors=etal|date=2010|title=Recurrent neural network based language model|url=http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf|journal=Interspeech|pages=1045–1048|doi=10.21437/Interspeech.2010-343|access-date=2017-06-13|archive-date=2017-05-16|archive-url=https://web.archive.org/web/20170516181940/http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf|url-status=live}}</ref> Long short-term memory is particularly effective for this use.<ref name=":0" /><ref name=":10">{{Cite web|url=https://www.researchgate.net/publication/220320057|title=Learning Precise Timing with LSTM Recurrent Networks (PDF Download Available)|website=ResearchGate|access-date=13 June 2017|archive-date=9 May 2021|archive-url=https://web.archive.org/web/20210509123147/https://www.researchgate.net/publication/220320057_Learning_Precise_Timing_with_LSTM_Recurrent_Networks|url-status=live}}</ref>
[[Convolutional neural network|Convolutional deep neural networks (CNNs)]] are used in computer vision.<ref name="LECUN86">{{cite journal |last1=LeCun |first1=Y. |display-authors=etal |year= 1998|title=Gradient-based learning applied to document recognition |journal=Proceedings of the IEEE |volume=86 |issue=11 |pages=2278–2324 |doi=10.1109/5.726791|s2cid=14542261 |url=http://elartu.tntu.edu.ua/handle/lib/38369 }}</ref> CNNs also have been applied to [[acoustic model]]ing for automatic speech recognition (ASR).<ref name=":2">{{Cite book |doi=10.1109/icassp.2013.6639347|isbn=978-1-4799-0356-6|chapter=Deep convolutional neural networks for LVCSR|title=2013 IEEE International Conference on Acoustics, Speech and Signal Processing|pages=8614–8618|year=2013|last1=Sainath|first1=Tara N.|last2=Mohamed|first2=Abdel-Rahman|last3=Kingsbury|first3=Brian|last4=Ramabhadran|first4=Bhuvana|s2cid=13816461}}</ref>
==== Challenges ====
As with ANNs, many issues can arise with naively trained DNNs. Two common issues are [[overfitting]] and computation time.
DNNs are prone to overfitting because of the added layers of abstraction, which allow them to model rare dependencies in the training data. [[Regularization (mathematics)|Regularization]] methods such as Ivakhnenko's unit pruning<ref name="ivak1971"/> or [[weight decay]] (<math> \ell_2 </math>-regularization) or [[sparse matrix|sparsity]] (<math> \ell_1 </math>-regularization) can be applied during training to combat overfitting.<ref>{{Cite book |doi=10.1109/icassp.2013.6639349|isbn=978-1-4799-0356-6|arxiv=1212.0901|citeseerx=10.1.1.752.9151|chapter=Advances in optimizing recurrent networks|title=2013 IEEE International Conference on Acoustics, Speech and Signal Processing|pages=8624–8628|year=2013|last1=Bengio|first1=Yoshua|last2=Boulanger-Lewandowski|first2=Nicolas|last3=Pascanu|first3=Razvan|s2cid=12485056}}</ref> Alternatively dropout regularization randomly omits units from the hidden layers during training. This helps to exclude rare dependencies.<ref name="DAHL2013">{{Cite journal|last=Dahl|first=G.|display-authors=etal|date=2013|title=Improving DNNs for LVCSR using rectified linear units and dropout|url=http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf|journal=ICASSP|access-date=2017-06-13|archive-date=2017-08-12|archive-url=https://web.archive.org/web/20170812140509/http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf|url-status=live}}</ref> Finally, data can be augmented via methods such as cropping and rotating such that smaller training sets can be increased in size to reduce the chances of overfitting.<ref>{{Cite web|url=https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation|title=Data Augmentation - deeplearning.ai {{!}} Coursera|website=Coursera|access-date=30 November 2017|archive-date=1 December 2017|archive-url=https://web.archive.org/web/20171201032606/https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation|url-status=live}}</ref>
DNNs must consider many training parameters, such as the size (number of layers and number of units per layer), the [[learning rate]], and initial weights. [[Hyperparameter optimization#Grid search|Sweeping through the parameter space]] for optimal parameters may not be feasible due to the cost in time and computational resources. Various tricks, such as batching (computing the gradient on several training examples at once rather than individual examples)<ref name="RBMTRAIN">{{Cite journal|last=Hinton|first=G. E.|date=2010|title=A Practical Guide to Training Restricted Boltzmann Machines|url=https://www.researchgate.net/publication/221166159|journal=Tech. Rep. UTML TR 2010-003|access-date=2017-06-13|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123211/https://www.researchgate.net/publication/221166159_A_brief_introduction_to_Weightless_Neural_Systems|url-status=live}}</ref> speed up computation. Large processing capabilities of many-core architectures (such as GPUs or the Intel Xeon Phi) have produced significant speedups in training, because of the suitability of such processing architectures for the matrix and vector computations.<ref>{{cite book|last1=You|first1=Yang|title=Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis on - SC '17|pages=1–12|last2=Buluç|first2=Aydın|last3=Demmel|first3=James|chapter=Scaling deep learning on GPU and knights landing clusters|chapter-url=https://dl.acm.org/citation.cfm?doid=3126908.3126912|publisher=SC '17, ACM|date=November 2017|access-date=5 March 2018|doi=10.1145/3126908.3126912|isbn=9781450351140|s2cid=8869270|url=http://www.escholarship.org/uc/item/6ch40821|archive-date=29 July 2020|archive-url=https://web.archive.org/web/20200729133850/https://escholarship.org/uc/item/6ch40821|url-status=live}}</ref><ref>{{cite journal|last1=Viebke|first1=André|last2=Memeti|first2=Suejb|last3=Pllana|first3=Sabri|last4=Abraham|first4=Ajith|s2cid=14135321|title=CHAOS: a parallelization scheme for training convolutional neural networks on Intel Xeon Phi|journal=The Journal of Supercomputing|volume=75|pages=197–227|doi=10.1007/s11227-017-1994-x|arxiv=1702.07908|bibcode=2017arXiv170207908V|year=2019}}</ref>
Alternatively, engineers may look for other types of neural networks with more straightforward and convergent training algorithms. CMAC ([[cerebellar model articulation controller]]) is one such kind of neural network. It doesn't require learning rates or randomized initial weights for CMAC. The training process can be guaranteed to converge in one step with a new batch of data, and the computational complexity of the training algorithm is linear with respect to the number of neurons involved.<ref name=Qin1>Ting Qin, et al. "A learning algorithm of CMAC based on RLS." Neural Processing Letters 19.1 (2004): 49-61.</ref><ref name=Qin2>Ting Qin, et al. "[http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf Continuous CMAC-QRLS and its systolic array] {{Webarchive|url=https://web.archive.org/web/20181118122850/http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf |date=2018-11-18 }}." Neural Processing Letters 22.1 (2005): 1-16.</ref>
== Hardware ==
Since the 2010s, advances in both machine learning algorithms and [[computer hardware]] have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.<ref>{{cite web|last1=Research|first1=AI|title=Deep Neural Networks for Acoustic Modeling in Speech Recognition|url=http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/|website=airesearch.com|access-date=23 October 2015|date=23 October 2015|archive-date=1 February 2016|archive-url=https://web.archive.org/web/20160201033801/http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/|url-status=live}}</ref> By 2019, graphic processing units ([[GPU]]s), often with AI-specific enhancements, had displaced CPUs as the dominant method of training large-scale commercial cloud AI.<ref>{{cite news |title=GPUs Continue to Dominate the AI Accelerator Market for Now |url=https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475 |access-date=11 June 2020 |work=InformationWeek |date=December 2019 |language=en |archive-date=10 June 2020 |archive-url=https://web.archive.org/web/20200610094310/https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475 |url-status=live }}</ref> [[OpenAI]] estimated the hardware computation used in the largest deep learning projects from AlexNet (2012) to AlphaZero (2017), and found a 300,000-fold increase in the amount of computation required, with a doubling-time trendline of 3.4 months.<ref>{{cite news |last1=Ray |first1=Tiernan |title=AI is changing the entire nature of computation |url=https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/ |access-date=11 June 2020 |work=ZDNet |date=2019 |language=en |archive-date=25 May 2020 |archive-url=https://web.archive.org/web/20200525144635/https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/ |url-status=live }}</ref><ref>{{cite web |title=AI and Compute |url=https://openai.com/blog/ai-and-compute/ |website=OpenAI |access-date=11 June 2020 |language=en |date=16 May 2018 |archive-date=17 June 2020 |archive-url=https://web.archive.org/web/20200617200602/https://openai.com/blog/ai-and-compute/ |url-status=live }}</ref>
Special [[electronic circuit]]s called [[deep learning processor]]s were designed to speed up deep learning algorithms. Deep learning processors include neural processing units (NPUs) in [[Huawei]] cellphones<ref>{{Cite web|url=https://consumer.huawei.com/en/press/news/2017/ifa2017-kirin970/|title=HUAWEI Reveals the Future of Mobile AI at IFA 2017 | HUAWEI Latest News | HUAWEI Global|website=consumer.huawei.com}}</ref> and [[cloud computing]] servers such as [[tensor processing unit]]s (TPU) in the [[Google Cloud Platform]].<ref>{{Cite journal|last1=P|first1=JouppiNorman|last2=YoungCliff|last3=PatilNishant|last4=PattersonDavid|last5=AgrawalGaurav|last6=BajwaRaminder|last7=BatesSarah|last8=BhatiaSuresh|last9=BodenNan|last10=BorchersAl|last11=BoyleRick|date=2017-06-24|title=In-Datacenter Performance Analysis of a Tensor Processing Unit|journal=ACM SIGARCH Computer Architecture News|volume=45|issue=2|pages=1–12|language=EN|doi=10.1145/3140659.3080246|doi-access=free}}</ref> [[Cerebras|Cerebras Systems]] has also built a dedicated system to handle large deep learning models, the CS-2, based on the largest processor in the industry, the second-generation Wafer Scale Engine (WSE-2).<ref>{{Cite web |last=Woodie |first=Alex |date=2021-11-01 |title=Cerebras Hits the Accelerator for Deep Learning Workloads |url=https://www.datanami.com/2021/11/01/cerebras-hits-the-accelerator-for-deep-learning-workloads/ |access-date=2022-08-03 |website=Datanami}}</ref><ref>{{Cite web |date=2021-04-20 |title=Cerebras launches new AI supercomputing processor with 2.6 trillion transistors |url=https://venturebeat.com/2021/04/20/cerebras-systems-launches-new-ai-supercomputing-processor-with-2-6-trillion-transistors/ |access-date=2022-08-03 |website=VentureBeat |language=en-US}}</ref>
Atomically thin [[semiconductors]] are considered promising for energy-efficient deep learning hardware where the same basic device structure is used for both logic operations and data storage.
In 2020, Marega et al. published experiments with a large-area active channel material for developing logic-in-memory devices and circuits based on [[floating-gate]] [[field-effect transistor]]s (FGFETs).<ref name="atomthin">{{cite journal|title=Logic-in-memory based on an atomically thin semiconductor|year=2020|doi=10.1038/s41586-020-2861-0|last1=Marega|first1=Guilherme Migliato|last2=Zhao|first2=Yanfei|last3=Avsar|first3=Ahmet|last4=Wang|first4=Zhenyu|last5=Tripati|first5=Mukesh|last6=Radenovic|first6=Aleksandra|last7=Kis|first7=Anras|journal=Nature|volume=587|issue=2|pages=72–77|pmid=33149289|pmc=7116757|bibcode=2020Natur.587...72M }}</ref>
In 2021, J. Feldmann et al. proposed an integrated [[photonic]] [[hardware accelerator]] for parallel convolutional processing.<ref name="photonic">{{cite journal |title=Parallel convolutional processing using an integrated photonic tensor |year=2021 |doi=10.1038/s41586-020-03070-1 |last1=Feldmann |first1=J. |last2=Youngblood|first2=N. |last3=Karpov |first3=M. | last4=Gehring |first4=H. | display-authors=3 | journal=Nature |volume=589 |issue=2 |pages=52–58|pmid=33408373 |arxiv=2002.00281 |s2cid=211010976 }}</ref> The authors identify two key advantages of integrated photonics over its electronic counterparts: (1) massively parallel data transfer through [[wavelength]] division [[multiplexing]] in conjunction with [[frequency comb]]s, and (2) extremely high data modulation speeds.<ref name="photonic"/> Their system can execute trillions of multiply-accumulate operations per second, indicating the potential of [[Photonic integrated circuit|integrated]] [[photonics]] in data-heavy AI applications.<ref name="photonic"/>
== Applications ==
=== Automatic speech recognition ===
{{Main|Speech recognition}}
Large-scale automatic speech recognition is the first and most convincing successful case of deep learning. LSTM RNNs can learn "Very Deep Learning" tasks<ref name="SCHIDHUB"/> that involve multi-second intervals containing speech events separated by thousands of discrete time steps, where one time step corresponds to about 10 ms. LSTM with forget gates<ref name=":10" /> is competitive with traditional speech recognizers on certain tasks.<ref name="graves2003"/>
The initial success in speech recognition was based on small-scale recognition tasks based on TIMIT. The data set contains 630 speakers from eight major [[dialect]]s of [[American English]], where each speaker reads 10 sentences.<ref name="LDCTIMIT">''TIMIT Acoustic-Phonetic Continuous Speech Corpus'' Linguistic Data Consortium, Philadelphia.</ref> Its small size lets many configurations be tried. More importantly, the TIMIT task concerns phone-sequence recognition, which, unlike word-sequence recognition, allows weak phone [[bigram]] language models. This lets the strength of the acoustic modeling aspects of speech recognition be more easily analyzed. The error rates listed below, including these early results and measured as percent phone error rates (PER), have been summarized since 1991.
{| class="wikitable"
|-
! Method !! Percent phone<br>error rate (PER) (%)
|-
| Randomly Initialized RNN<ref>{{cite journal |last1=Robinson |first1=Tony |author-link=Tony Robinson (speech recognition)|title=Several Improvements to a Recurrent Error Propagation Network Phone Recognition System |journal=Cambridge University Engineering Department Technical Report |date=30 September 1991 |volume=CUED/F-INFENG/TR82 |doi=10.13140/RG.2.2.15418.90567 }}</ref>|| 26.1
|-
| Bayesian Triphone GMM-HMM || 25.6
|-
| Hidden Trajectory (Generative) Model|| 24.8
|-
| Monophone Randomly Initialized DNN|| 23.4
|-
| Monophone DBN-DNN|| 22.4
|-
| Triphone GMM-HMM with BMMI Training|| 21.7
|-
| Monophone DBN-DNN on fbank || 20.7
|-
| Convolutional DNN<ref name="CNN-2014">{{cite journal|last1=Abdel-Hamid|first1=O.|title=Convolutional Neural Networks for Speech Recognition|journal=IEEE/ACM Transactions on Audio, Speech, and Language Processing|date=2014|volume=22|issue=10|pages=1533–1545|doi=10.1109/taslp.2014.2339736|s2cid=206602362|display-authors=etal|url=https://zenodo.org/record/891433|access-date=2018-04-20|archive-date=2020-09-22|archive-url=https://web.archive.org/web/20200922180719/https://zenodo.org/record/891433|url-status=live}}</ref>|| 20.0
|-
| Convolutional DNN w. Heterogeneous Pooling|| 18.7
|-
| Ensemble DNN/CNN/RNN<ref name="EnsembleDL">{{cite journal|last2=Platt|first2=J.|date=2014|title=Ensemble Deep Learning for Speech Recognition|journal=Proc. Interspeech|last1=Deng|first1=L.|s2cid=15641618}}</ref>|| 18.3
|-
| Bidirectional LSTM|| 17.8
|-
| Hierarchical Convolutional Deep Maxout Network<ref name="HCDMM">{{cite journal|last1=Tóth|first1=Laszló|date=2015|title=Phone Recognition with Hierarchical Convolutional Deep Maxout Networks|journal=EURASIP Journal on Audio, Speech, and Music Processing|volume=2015|doi=10.1186/s13636-015-0068-3|s2cid=217950236|url=http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf|access-date=2019-04-01|archive-date=2020-09-24|archive-url=https://web.archive.org/web/20200924085514/http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf|url-status=live}}</ref> || 16.5
|}
The debut of DNNs for speaker recognition in the late 1990s and speech recognition around 2009-2011 and of LSTM around 2003–2007, accelerated progress in eight major areas:<ref name="BOOK2014" /><ref name="interspeech2014Keynote" /><ref name="ReferenceA" />
* Scale-up/out and accelerated DNN training and decoding
* Sequence discriminative training
* Feature processing by deep models with solid understanding of the underlying mechanisms
* Adaptation of DNNs and related deep models
* [[Multi-task learning|Multi-task]] and [[transfer learning]] by DNNs and related deep models
* [[Convolutional neural network|CNNs]] and how to design them to best exploit [[domain knowledge]] of speech
* [[Recurrent neural network|RNN]] and its rich LSTM variants
* Other types of deep models including tensor-based models and integrated deep generative/discriminative models.
All major commercial speech recognition systems (e.g., Microsoft [[Cortana (software)|Cortana]], [[Xbox]], [[Skype Translator]], [[Amazon Alexa]], [[Google Now]], [[Siri|Apple Siri]], [[Baidu]] and [[IFlytek|iFlyTek]] voice search, and a range of [[Nuance Communications|Nuance]] speech products, etc.) are based on deep learning.<ref name=BOOK2014 /><ref>{{Cite magazine|url=https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/|title=How Skype Used AI to Build Its Amazing New Language Translator {{!}} WIRED|magazine=Wired|access-date=14 June 2017|date=17 December 2014|last1=McMillan|first1=Robert|archive-date=8 June 2017|archive-url=https://web.archive.org/web/20170608062106/https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/|url-status=live}}</ref><ref name="Baidu">{{cite arXiv |eprint=1412.5567|last1=Hannun|first1=Awni|title=Deep Speech: Scaling up end-to-end speech recognition|last2=Case|first2=Carl|last3=Casper|first3=Jared|last4=Catanzaro|first4=Bryan|last5=Diamos|first5=Greg|last6=Elsen|first6=Erich|last7=Prenger|first7=Ryan|last8=Satheesh|first8=Sanjeev|last9=Sengupta|first9=Shubho|last10=Coates|first10=Adam|last11=Ng|first11=Andrew Y|class=cs.CL|year=2014}}</ref>
=== Image recognition ===
{{Main|Computer vision}}
A common evaluation set for image classification is the MNIST database data set. MNIST is composed of handwritten digits and includes 60,000 training examples and 10,000 test examples. As with TIMIT, its small size lets users test multiple configurations. A comprehensive list of results on this set is available.<ref name="YANNMNIST">{{cite web|url=http://yann.lecun.com/exdb/mnist/.|title=MNIST handwritten digit database, Yann LeCun, Corinna Cortes and Chris Burges|website=yann.lecun.com|access-date=2014-01-28|archive-date=2014-01-13|archive-url=https://web.archive.org/web/20140113175237/http://yann.lecun.com/exdb/mnist/|url-status=live}}</ref>
Deep learning-based image recognition has become "superhuman", producing more accurate results than human contestants. This first occurred in 2011 in recognition of traffic signs, and in 2014, with recognition of human faces.<ref name=":7">{{Cite journal|last1=Cireşan|first1=Dan|last2=Meier|first2=Ueli|last3=Masci|first3=Jonathan|last4=Schmidhuber|first4=Jürgen|date=August 2012|title=Multi-column deep neural network for traffic sign classification|journal=Neural Networks|series=Selected Papers from IJCNN 2011|volume=32|pages=333–338|doi=10.1016/j.neunet.2012.02.023|pmid=22386783|citeseerx=10.1.1.226.8219}}</ref><ref name=surpass1>{{cite arXiv|title=Surpassing Human Level Face Recognition|author1=Chaochao Lu |author2= Xiaoou Tang |year=2014 |class=cs.CV |eprint=1404.3840 }}</ref>
Deep learning-trained vehicles now interpret 360° camera views.<ref>[http://www.technologyreview.com/news/533936/nvidia-demos-a-car-computer-trained-with-deep-learning/ Nvidia Demos a Car Computer Trained with "Deep Learning"] (6 January 2015), David Talbot, ''[[MIT Technology Review]]''</ref> Another example is Facial Dysmorphology Novel Analysis (FDNA) used to analyze cases of human malformation connected to a large database of genetic syndromes.
=== Visual art processing ===
Closely related to the progress that has been made in image recognition is the increasing application of deep learning techniques to various visual art tasks. DNNs have proven themselves capable, for example, of
*identifying the style period of a given painting<ref name = art1/><ref name = art2/>
*[[Neural Style Transfer]]{{snd}} capturing the style of a given artwork and applying it in a visually pleasing manner to an arbitrary photograph or video<ref name = art1/><ref name = art2/>
*generating striking imagery based on random visual input fields.<ref name = art1 >{{cite journal |author1=G. W. Smith|author2=Frederic Fol Leymarie|date=10 April 2017|title=The Machine as Artist: An Introduction|journal=Arts|volume=6|issue=4|pages=5|doi=10.3390/arts6020005|doi-access=free}}</ref><ref name = art2>{{cite journal |author=Blaise Agüera y Arcas|date=29 September 2017|title=Art in the Age of Machine Intelligence|journal=Arts|volume=6|issue=4|pages=18|doi=10.3390/arts6040018|doi-access=free}}</ref>
=== Natural language processing ===
{{Main|Natural language processing}}
Neural networks have been used for implementing language models since the early 2000s.<ref name="gers2001" /> LSTM helped to improve machine translation and language modeling.<ref name="NIPS2014" /><ref name="vinyals2016" /><ref name="gillick2015" />
Other key techniques in this field are negative sampling<ref name="GoldbergLevy2014">{{cite arXiv|last1=Goldberg|first1=Yoav|last2=Levy|first2=Omar|title=word2vec Explained: Deriving Mikolov et al.'s Negative-Sampling Word-Embedding Method|eprint=1402.3722|class=cs.CL|year=2014}}</ref> and [[word embedding]]. Word embedding, such as ''[[word2vec]]'', can be thought of as a representational layer in a deep learning architecture that transforms an atomic word into a positional representation of the word relative to other words in the dataset; the position is represented as a point in a [[vector space]]. Using word embedding as an RNN input layer allows the network to parse sentences and phrases using an effective compositional vector grammar. A compositional vector grammar can be thought of as [[probabilistic context free grammar]] (PCFG) implemented by an RNN.<ref name="SocherManning2014">{{cite web|last1=Socher|first1=Richard|last2=Manning|first2=Christopher|title=Deep Learning for NLP|url=http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf|access-date=26 October 2014|archive-date=6 July 2014|archive-url=https://web.archive.org/web/20140706040227/http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf|url-status=live}}</ref> Recursive auto-encoders built atop word embeddings can assess sentence similarity and detect paraphrasing.<ref name="SocherManning2014" /> Deep neural architectures provide the best results for [[Statistical parsing|constituency parsing]],<ref>{{Cite journal|url = http://aclweb.org/anthology/P/P13/P13-1045.pdf|title = Parsing With Compositional Vector Grammars|last1 = Socher|first1 = Richard|date = 2013|journal = Proceedings of the ACL 2013 Conference|last2 = Bauer|first2 = John|last3 = Manning|first3 = Christopher|last4 = Ng|first4 = Andrew|access-date = 2014-09-03|archive-date = 2014-11-27|archive-url = https://web.archive.org/web/20141127005912/http://www.aclweb.org/anthology/P/P13/P13-1045.pdf|url-status = live}}</ref> [[sentiment analysis]],<ref>{{Cite journal|url = http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf|title = Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank|last = Socher|first = Richard|date = 2013|journal = |access-date = 2014-09-03|archive-date = 2016-12-28|archive-url = https://web.archive.org/web/20161228100300/http://nlp.stanford.edu/%7Esocherr/EMNLP2013_RNTN.pdf|url-status = live}}</ref> information retrieval,<ref>{{Cite journal|last1=Shen|first1=Yelong|last2=He|first2=Xiaodong|last3=Gao|first3=Jianfeng|last4=Deng|first4=Li|last5=Mesnil|first5=Gregoire|date=1 November 2014|title=A Latent Semantic Model with Convolutional-Pooling Structure for Information Retrieval|url=https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/|journal=Microsoft Research|access-date=14 June 2017|archive-date=27 October 2017|archive-url=https://web.archive.org/web/20171027050418/https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/|url-status=live}}</ref><ref>{{Cite journal|last1=Huang|first1=Po-Sen|last2=He|first2=Xiaodong|last3=Gao|first3=Jianfeng|last4=Deng|first4=Li|last5=Acero|first5=Alex|last6=Heck|first6=Larry|date=1 October 2013|title=Learning Deep Structured Semantic Models for Web Search using Clickthrough Data|url=https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/|journal=Microsoft Research|access-date=14 June 2017|archive-date=27 October 2017|archive-url=https://web.archive.org/web/20171027050414/https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/|url-status=live}}</ref> spoken language understanding,<ref name="IEEE-TASL2015">{{cite journal | last1 = Mesnil | first1 = G. | last2 = Dauphin | first2 = Y. | last3 = Yao | first3 = K. | last4 = Bengio | first4 = Y. | last5 = Deng | first5 = L. | last6 = Hakkani-Tur | first6 = D. | last7 = He | first7 = X. | last8 = Heck | first8 = L. | last9 = Tur | first9 = G. | last10 = Yu | first10 = D. | last11 = Zweig | first11 = G. | s2cid = 1317136 | year = 2015 | title = Using recurrent neural networks for slot filling in spoken language understanding | journal = IEEE Transactions on Audio, Speech, and Language Processing | volume = 23 | issue = 3| pages = 530–539 | doi=10.1109/taslp.2014.2383614}}</ref> machine translation,<ref name="NIPS2014">{{Cite journal|last1=Sutskever|first1=L.|last2=Vinyals|first2=O.|last3=Le|first3=Q.|date=2014|title=Sequence to Sequence Learning with Neural Networks|url=https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf|journal=Proc. NIPS|bibcode=2014arXiv1409.3215S|arxiv=1409.3215|access-date=2017-06-13|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123145/https://papers.nips.cc/paper/2014/file/a14ac55a4f27472c5d894ec1c3c743d2-Paper.pdf|url-status=live}}</ref><ref name="auto">{{Cite journal|last1=Gao|first1=Jianfeng|last2=He|first2=Xiaodong|last3=Yih|first3=Scott Wen-tau|last4=Deng|first4=Li|date=1 June 2014|title=Learning Continuous Phrase Representations for Translation Modeling|url=https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/|journal=Microsoft Research|access-date=14 June 2017|archive-date=27 October 2017|archive-url=https://web.archive.org/web/20171027050403/https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/|url-status=live}}</ref> contextual entity linking,<ref name="auto"/> writing style recognition,<ref name="BROC2017">{{Cite journal |doi = 10.1002/dac.3259|title = Authorship verification using deep belief network systems|journal = International Journal of Communication Systems|volume = 30|issue = 12|pages = e3259|year = 2017|last1 = Brocardo|first1 = Marcelo Luiz|last2 = Traore|first2 = Issa|last3 = Woungang|first3 = Isaac|last4 = Obaidat|first4 = Mohammad S.| s2cid=40745740 }}</ref> Text classification and others.<ref>{{Cite news|url=https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/|title=Deep Learning for Natural Language Processing: Theory and Practice (CIKM2014 Tutorial) - Microsoft Research|work=Microsoft Research|access-date=14 June 2017|archive-date=13 March 2017|archive-url=https://web.archive.org/web/20170313184253/https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/|url-status=live}}</ref>
Recent developments generalize [[word embedding]] to [[sentence embedding]].
[[Google Translate]] (GT) uses a large end-to-end [[long short-term memory]] (LSTM) network.<ref name="GT_Turovsky_2016">{{cite web|url=https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/|title=Found in translation: More accurate, fluent sentences in Google Translate|last=Turovsky|first=Barak|date=15 November 2016|website=The Keyword Google Blog|access-date=23 March 2017|archive-date=7 April 2017|archive-url=https://web.archive.org/web/20170407071226/https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/|url-status=live}}</ref><ref name="googleblog_GNMT_2016">{{cite web|url=https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html|title=Zero-Shot Translation with Google's Multilingual Neural Machine Translation System|last1=Schuster|first1=Mike|last2=Johnson|first2=Melvin|date=22 November 2016|website=Google Research Blog|access-date=23 March 2017|last3=Thorat|first3=Nikhil|archive-date=10 July 2017|archive-url=https://web.archive.org/web/20170710183732/https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html|url-status=live}}</ref><ref name="GoogleTranslate">{{cite arXiv |eprint=1609.08144|last1=Wu|first1=Yonghui|title=Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation|last2=Schuster|first2=Mike|last3=Chen|first3=Zhifeng|last4=Le|first4=Quoc V|last5=Norouzi|first5=Mohammad|last6=Macherey|first6=Wolfgang|last7=Krikun|first7=Maxim|last8=Cao|first8=Yuan|last9=Gao|first9=Qin|last10=Macherey|first10=Klaus|last11=Klingner|first11=Jeff|last12=Shah|first12=Apurva|last13=Johnson|first13=Melvin|last14=Liu|first14=Xiaobing|last15=Kaiser|first15=Łukasz|last16=Gouws|first16=Stephan|last17=Kato|first17=Yoshikiyo|last18=Kudo|first18=Taku|last19=Kazawa|first19=Hideto|last20=Stevens|first20=Keith|last21=Kurian|first21=George|last22=Patil|first22=Nishant|last23=Wang|first23=Wei|last24=Young|first24=Cliff|last25=Smith|first25=Jason|last26=Riesa|first26=Jason|last27=Rudnick|first27=Alex|last28=Vinyals|first28=Oriol|last29=Corrado|first29=Greg|last30=Hughes|first30=Macduff|display-authors=29|class=cs.CL|year=2016}}</ref><ref name="WiredGoogleTranslate">{{cite magazine |title=An Infusion of AI Makes Google Translate More Powerful Than Ever |first=Cade |last=Metz |magazine=[[Wired (magazine)|Wired]] |date=27 September 2016 |url=https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/ |access-date=12 October 2017 |archive-date=8 November 2020 |archive-url=https://web.archive.org/web/20201108101324/https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/ |url-status=live }}</ref> [[Google Neural Machine Translation|Google Neural Machine Translation (GNMT)]] uses an [[example-based machine translation]] method in which the system "learns from millions of examples."<ref name="googleblog_GNMT_2016" /> It translates "whole sentences at a time, rather than pieces. Google Translate supports over one hundred languages.<ref name="googleblog_GNMT_2016" /> The network encodes the "semantics of the sentence rather than simply memorizing phrase-to-phrase translations".<ref name="googleblog_GNMT_2016" /><ref name="Biotet">{{cite web|url=http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf|title=MT on and for the Web|last1=Boitet|first1=Christian|last2=Blanchon|first2=Hervé|date=2010|access-date=1 December 2016|last3=Seligman|first3=Mark|last4=Bellynck|first4=Valérie|archive-date=29 March 2017|archive-url=https://web.archive.org/web/20170329125916/http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf|url-status=dead}}</ref> GT uses English as an intermediate between most language pairs.<ref name="Biotet" />
=== Drug discovery and toxicology ===
{{For|more information|Drug discovery|Toxicology}}
A large percentage of candidate drugs fail to win regulatory approval. These failures are caused by insufficient efficacy (on-target effect), undesired interactions (off-target effects), or unanticipated [[Toxicity|toxic effects]].<ref name="ARROWSMITH2013">{{Cite journal
| pmid = 23903212
| year = 2013
| last1 = Arrowsmith
| first1 = J
| title = Trial watch: Phase II and phase III attrition rates 2011-2012
| journal = Nature Reviews Drug Discovery
| volume = 12
| issue = 8
| pages = 569
| last2 = Miller
| first2 = P
| s2cid = 20246434
| doi = 10.1038/nrd4090
}}</ref><ref name="VERBIEST2015">{{Cite journal
| pmid = 25582842
| year = 2015
| last1 = Verbist
| first1 = B
| title = Using transcriptomics to guide lead optimization in drug discovery projects: Lessons learned from the QSTAR project
| journal = Drug Discovery Today
| last2 = Klambauer
| first2 = G
| last3 = Vervoort
| first3 = L
| last4 = Talloen
| first4 = W
| last5 = The Qstar
| first5 = Consortium
| last6 = Shkedy
| first6 = Z
| last7 = Thas
| first7 = O
| last8 = Bender
| first8 = A
| last9 = Göhlmann
| first9 = H. W.
| last10 = Hochreiter
| first10 = S
| doi = 10.1016/j.drudis.2014.12.014
| volume=20
| issue = 5
| pages=505–513
| doi-access = free
}}</ref> Research has explored use of deep learning to predict the [[biomolecular target]]s,<ref name="MERCK2012" /><ref name=":5" /> [[off-target]]s, and [[Toxicity|toxic effects]] of environmental chemicals in nutrients, household products and drugs.<ref name="TOX21" /><ref name="TOX21Data" /><ref name=":11" />
AtomNet is a deep learning system for structure-based [[Drug design|rational drug design]].<ref>{{cite arXiv|title = AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery|eprint= 1510.02855|date = 9 October 2015|first1 = Izhar|last1 = Wallach|first2 = Michael|last2 = Dzamba|first3 = Abraham|last3 = Heifets|class= cs.LG}}</ref> AtomNet was used to predict novel candidate biomolecules for disease targets such as the [[Ebola virus]]<ref>{{Cite news |title = Toronto startup has a faster way to discover effective medicines |url = https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/ |website = The Globe and Mail |access-date = 9 November 2015 |archive-date = 20 October 2015 |archive-url = https://web.archive.org/web/20151020040115/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/ |url-status = live }}</ref> and [[multiple sclerosis]].<ref>{{Cite web|title = Startup Harnesses Supercomputers to Seek Cures|url = http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/|website = KQED Future of You|access-date = 9 November 2015|archive-date = 24 December 2015|archive-url = https://web.archive.org/web/20151224104721/http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/|url-status = live}}</ref><ref>{{Cite web|url=https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/|title=Toronto startup has a faster way to discover effective medicines|website=[[The Globe and Mail]]|access-date=2017-08-26|archive-date=2015-12-25|archive-url=https://web.archive.org/web/20151225162547/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5d%20and%20multiple%20sclerosis%20%5b/|url-status=live}}</ref>
In 2017 [[graph neural network]]s were used for the first time to predict various properties of molecules in a large toxicology data set.<ref>{{cite arXiv|last1=Gilmer|first1=Justin|last2=Schoenholz|first2=Samuel S.|last3=Riley|first3=Patrick F.|last4=Vinyals|first4=Oriol|last5=Dahl|first5=George E.|date=2017-06-12|title=Neural Message Passing for Quantum Chemistry|class=cs.LG|eprint=1704.01212}}</ref> In 2019, generative neural networks were used to produce molecules that were validated experimentally all the way into mice.<ref>{{cite journal |last1=Zhavoronkov |first1=Alex|s2cid=201716327|date=2019|title=Deep learning enables rapid identification of potent DDR1 kinase inhibitors |journal=Nature Biotechnology |volume=37|issue=9|pages=1038–1040|doi=10.1038/s41587-019-0224-x |pmid=31477924}}</ref><ref>{{cite magazine |last1=Gregory |first1=Barber |title=A Molecule Designed By AI Exhibits 'Druglike' Qualities |url=https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/ |magazine=Wired |access-date=2019-09-05 |archive-date=2020-04-30 |archive-url=https://web.archive.org/web/20200430143244/https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/ |url-status=live }}</ref>
=== Customer relationship management ===
{{Main|Customer relationship management}}
[[Deep reinforcement learning]] has been used to approximate the value of possible [[direct marketing]] actions, defined in terms of [[RFM (customer value)|RFM]] variables. The estimated value function was shown to have a natural interpretation as [[customer lifetime value]].<ref>{{cite arXiv|last=Tkachenko |first=Yegor |title=Autonomous CRM Control via CLV Approximation with Deep Reinforcement Learning in Discrete and Continuous Action Space |date=8 April 2015 |eprint=1504.01840|class=cs.LG }}</ref>
=== Recommendation systems ===
{{Main|Recommender system}}
Recommendation systems have used deep learning to extract meaningful features for a latent factor model for content-based music and journal recommendations.<ref>{{Cite book|url=http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf|title=Advances in Neural Information Processing Systems 26|last1=van den Oord|first1=Aaron|last2=Dieleman|first2=Sander|last3=Schrauwen|first3=Benjamin|date=2013|publisher=Curran Associates, Inc.|editor-last=Burges|editor-first=C. J. C.|pages=2643–2651|editor-last2=Bottou|editor-first2=L.|editor-last3=Welling|editor-first3=M.|editor-last4=Ghahramani|editor-first4=Z.|editor-last5=Weinberger|editor-first5=K. Q.|access-date=2017-06-14|archive-date=2017-05-16|archive-url=https://web.archive.org/web/20170516185259/http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf|url-status=live}}</ref><ref>{{cite journal | last1 = Feng | first1 = X.Y. | last2 = Zhang | first2 = H. | last3 = Ren | first3 = Y.J. | last4 = Shang | first4 = P.H. | last5 = Zhu | first5 = Y. | last6 = Liang | first6 = Y.C. | last7 = Guan | first7 = R.C. | last8 = Xu | first8 = D. | year = 2019 | title = The Deep Learning–Based Recommender System "Pubmender" for Choosing a Biomedical Publication Venue: Development and Validation Study | journal = [[Journal of Medical Internet Research]] | volume = 21 | issue = 5| page = e12957 | doi = 10.2196/12957 | pmid = 31127715 | pmc = 6555124 }}</ref> Multi-view deep learning has been applied for learning user preferences from multiple domains.<ref>{{Cite journal|last1=Elkahky|first1=Ali Mamdouh|last2=Song|first2=Yang|last3=He|first3=Xiaodong|date=1 May 2015|title=A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems|url=https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/|journal=Microsoft Research|access-date=14 June 2017|archive-date=25 January 2018|archive-url=https://web.archive.org/web/20180125134534/https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/|url-status=live}}</ref> The model uses a hybrid collaborative and content-based approach and enhances recommendations in multiple tasks.
=== Bioinformatics ===
{{Main|Bioinformatics}}
An [[autoencoder]] ANN was used in [[bioinformatics]], to predict [[Gene Ontology|gene ontology]] annotations and gene-function relationships.<ref>{{cite book|title=Deep Autoencoder Neural Networks for Gene Ontology Annotation Predictions|first1=Davide|last1=Chicco|first2=Peter|last2=Sadowski|first3=Pierre|last3=Baldi|s2cid=207217210|date=1 January 2014|publisher=ACM|pages=533–540|doi=10.1145/2649387.2649442|journal=Proceedings of the 5th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics - BCB '14|isbn=9781450328944|hdl=11311/964622|url=http://dl.acm.org/citation.cfm?id=2649442|access-date=23 November 2015|archive-date=9 May 2021|archive-url=https://web.archive.org/web/20210509123140/https://dl.acm.org/doi/10.1145/2649387.2649442|url-status=live}}</ref>
In medical informatics, deep learning was used to predict sleep quality based on data from wearables<ref>{{Cite journal|last=Sathyanarayana|first=Aarti|s2cid=3821594|date=1 January 2016|title=Sleep Quality Prediction From Wearable Data Using Deep Learning|journal=JMIR mHealth and uHealth|volume=4|issue=4|doi=10.2196/mhealth.6562|pmid=27815231|pmc=5116102|pages=e125}}</ref> and predictions of health complications from [[electronic health record]] data.<ref>{{Cite journal|last1=Choi|first1=Edward|last2=Schuetz|first2=Andy|last3=Stewart|first3=Walter F.|last4=Sun|first4=Jimeng|date=13 August 2016|title=Using recurrent neural network models for early detection of heart failure onset|journal=Journal of the American Medical Informatics Association|volume=24|issue=2|pages=361–370|doi=10.1093/jamia/ocw112|issn=1067-5027|pmid=27521897|pmc=5391725}}</ref>
=== Medical image analysis ===
Deep learning has been shown to produce competitive results in medical application such as cancer cell classification, lesion detection, organ segmentation and image enhancement.<ref>{{Cite journal|last1=Litjens|first1=Geert|last2=Kooi|first2=Thijs|last3=Bejnordi|first3=Babak Ehteshami|last4=Setio|first4=Arnaud Arindra Adiyoso|last5=Ciompi|first5=Francesco|last6=Ghafoorian|first6=Mohsen|last7=van der Laak|first7=Jeroen A.W.M.|last8=van Ginneken|first8=Bram|last9=Sánchez|first9=Clara I.|s2cid=2088679|date=December 2017|title=A survey on deep learning in medical image analysis|journal=Medical Image Analysis|volume=42|pages=60–88|doi=10.1016/j.media.2017.07.005|pmid=28778026|arxiv=1702.05747|bibcode=2017arXiv170205747L}}</ref><ref>{{Cite book|doi=10.1109/ICCVW.2017.18|isbn=9781538610343|chapter=Deep Convolutional Neural Networks for Detecting Cellular Changes Due to Malignancy|title=2017 IEEE International Conference on Computer Vision Workshops (ICCVW)|pages=82–89|year=2017|last1=Forslid|first1=Gustav|last2=Wieslander|first2=Hakan|last3=Bengtsson|first3=Ewert|last4=Wahlby|first4=Carolina|last5=Hirsch|first5=Jan-Michael|last6=Stark|first6=Christina Runow|last7=Sadanandan|first7=Sajith Kecheril|s2cid=4728736|chapter-url=http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-326160|access-date=2019-11-12|archive-date=2021-05-09|archive-url=https://web.archive.org/web/20210509123157/https://d1bxh8uas1mnw7.cloudfront.net/assets/embed.js|url-status=live}}</ref> Modern deep learning tools demonstrate the high accuracy of detecting various diseases and the helpfulness of their use by specialists to improve the diagnosis efficiency.<ref>{{Cite journal |last1=Dong |first1=Xin |last2=Zhou |first2=Yizhao |last3=Wang |first3=Lantian |last4=Peng |first4=Jingfeng |last5=Lou |first5=Yanbo |last6=Fan |first6=Yiqun |date=2020 |title=Liver Cancer Detection Using Hybridized Fully Convolutional Neural Network Based on Deep Learning Framework |url=https://ieeexplore.ieee.org/document/9130662 |journal=IEEE Access |volume=8 |pages=129889–129898 |doi=10.1109/ACCESS.2020.3006362 |s2cid=220733699 |issn=2169-3536}}</ref><ref>{{Cite journal |last1=Lyakhov |first1=Pavel Alekseevich |last2=Lyakhova |first2=Ulyana Alekseevna |last3=Nagornov |first3=Nikolay Nikolaevich |date=2022-04-03 |title=System for the Recognizing of Pigmented Skin Lesions with Fusion and Analysis of Heterogeneous Data Based on a Multimodal Neural Network |journal=Cancers |language=en |volume=14 |issue=7 |pages=1819 |doi=10.3390/cancers14071819 |pmid=35406591 |pmc=8997449 |issn=2072-6694|doi-access=free }}</ref>
=== Mobile advertising ===
Finding the appropriate mobile audience for [[mobile advertising]] is always challenging, since many data points must be considered and analyzed before a target segment can be created and used in ad serving by any ad server.<ref>{{cite book |doi=10.1109/CSCITA.2017.8066548 |isbn=978-1-5090-4381-1|chapter=Predicting the popularity of instagram posts for a lifestyle magazine using deep learning|title=2017 2nd International Conference on Communication Systems, Computing and IT Applications (CSCITA)|pages=174–177|year=2017|last1=De|first1=Shaunak|last2=Maity|first2=Abhishek|last3=Goel|first3=Vritti|last4=Shitole|first4=Sanjay|last5=Bhattacharya|first5=Avik|s2cid=35350962}}</ref> Deep learning has been used to interpret large, many-dimensioned advertising datasets. Many data points are collected during the request/serve/click internet advertising cycle. This information can form the basis of machine learning to improve ad selection.
=== Image restoration ===
Deep learning has been successfully applied to [[inverse problems]] such as [[denoising]], [[super-resolution]], [[inpainting]], and [[film colorization]].<ref>{{Cite web|url=https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/|title=Colorizing and Restoring Old Images with Deep Learning|date=13 November 2018|website=FloydHub Blog|language=en|access-date=11 October 2019|archive-date=11 October 2019|archive-url=https://web.archive.org/web/20191011162814/https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/|url-status=live}}</ref> These applications include learning methods such as "Shrinkage Fields for Effective Image Restoration"<ref>{{cite conference |url= http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf |first1= Uwe |last1= Schmidt |first2= Stefan |last2= Roth |conference= Computer Vision and Pattern Recognition (CVPR), 2014 IEEE Conference on |title= Shrinkage Fields for Effective Image Restoration |access-date= 2018-01-01 |archive-date= 2018-01-02 |archive-url= https://web.archive.org/web/20180102013217/http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf |url-status= live }}</ref> which trains on an image dataset, and [[Deep Image Prior]], which trains on the image that needs restoration.
=== Financial fraud detection ===
Deep learning is being successfully applied to financial [[fraud detection]], tax evasion detection,<ref>{{cite journal
|first1=Christos |last1=Kleanthous
|first2=Sotirios |last2=Chatzis
|title=Gated Mixture Variational Autoencoders for Value Added Tax audit case selection
|journal=Knowledge-Based Systems
|volume=188
|year=2020
|page=105048
|doi=10.1016/j.knosys.2019.105048
|s2cid=204092079
}}</ref> and anti-money laundering.<ref>{{cite journal
|first=Tomasz
|last=Czech
|title=Deep learning: the next frontier for money laundering detection
|url=https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/
|journal=Global Banking and Finance Review
|date=28 June 2018
|access-date=2018-07-15
|archive-date=2018-11-16
|archive-url=https://web.archive.org/web/20181116082711/https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/
|url-status=live
}}</ref>
=== Military ===
The United States Department of Defense applied deep learning to train robots in new tasks through observation.<ref name=":12">{{Cite web|url=https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php|title=Army researchers develop new algorithms to train robots|website=EurekAlert!|access-date=29 August 2018|archive-date=28 August 2018|archive-url=https://web.archive.org/web/20180828035608/https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php|url-status=live}}</ref>
=== Partial differential equations ===
Physics informed neural networks have been used to solve [[partial differential equation]]s in both forward and inverse problems in a data driven manner.<ref>{{Cite journal|last1=Raissi|first1=M.|last2=Perdikaris|first2=P.|last3=Karniadakis|first3=G. E.|date=2019-02-01|title=Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations|url=https://www.sciencedirect.com/science/article/pii/S0021999118307125|journal=Journal of Computational Physics|language=en|volume=378|pages=686–707|doi=10.1016/j.jcp.2018.10.045|bibcode=2019JCoPh.378..686R|osti=1595805|s2cid=57379996|issn=0021-9991}}</ref> One example is the reconstructing fluid flow governed by the [[Navier–Stokes equations|Navier-Stokes equations]]. Using physics informed neural networks does not require the often expensive mesh generation that conventional [[Computational fluid dynamics|CFD]] methods relies on.<ref>{{Cite journal|last1=Mao|first1=Zhiping|last2=Jagtap|first2=Ameya D.|last3=Karniadakis|first3=George Em|date=2020-03-01|title=Physics-informed neural networks for high-speed flows|url=https://www.sciencedirect.com/science/article/pii/S0045782519306814|journal=Computer Methods in Applied Mechanics and Engineering|language=en|volume=360|pages=112789|doi=10.1016/j.cma.2019.112789|bibcode=2020CMAME.360k2789M|s2cid=212755458|issn=0045-7825}}</ref><ref>{{Cite journal|last1=Raissi|first1=Maziar|last2=Yazdani|first2=Alireza|last3=Karniadakis|first3=George Em|date=2020-02-28|title=Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations|journal=Science|volume=367|issue=6481|pages=1026–1030|doi=10.1126/science.aaw4741|pmc=7219083|pmid=32001523|bibcode=2020Sci...367.1026R}}</ref>
=== Image Reconstruction ===
Image reconstruction is the reconstruction of the underlying images from the image-related measurements. Several works showed the better and superior performance of the deep learning methods compared to analytical methods for various applications, e.g., spectral imaging <ref>{{Cite journal |last1=Oktem |first1=Figen S. |last2=Kar |first2=Oğuzhan Fatih |last3=Bezek |first3=Can Deniz |last4=Kamalabadi |first4=Farzad |date=2021 |title=High-Resolution Multi-Spectral Imaging With Diffractive Lenses and Learned Reconstruction |url=https://ieeexplore.ieee.org/document/9415140 |journal=IEEE Transactions on Computational Imaging |volume=7 |pages=489–504 |doi=10.1109/TCI.2021.3075349 |arxiv=2008.11625 |s2cid=235340737 |issn=2333-9403}}</ref> and ultrasound imaging.<ref>{{Cite journal |last1=Bernhardt |first1=Melanie |last2=Vishnevskiy |first2=Valery |last3=Rau |first3=Richard |last4=Goksel |first4=Orcun |date=December 2020 |title=Training Variational Networks With Multidomain Simulations: Speed-of-Sound Image Reconstruction |url=https://ieeexplore.ieee.org/abstract/document/9144249?casa_token=NLVQbGR8h-8AAAAA:rz3cvMmiplVSRCGTDouUZkIeribqQKaMtK9t0MxlwKBMRwaTJL1onKOwkEk0qW_fFJCfd2ejOg |journal=IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control |volume=67 |issue=12 |pages=2584–2594 |doi=10.1109/TUFFC.2020.3010186 |pmid=32746211 |arxiv=2006.14395 |s2cid=220055785 |issn=1525-8955}}</ref>
<big>'''Epigenetic clock'''</big>
For more information, see [[Ageing clock|Epigenetic clock]].
An '''epigenetic clock''' is a [[Biomarkers of aging|biochemical test]] that can be used to measure age. Galkin et al. used deep [[Neural network|neural networks]] to train an epigenetic aging clock of unprecedented accuracy using >6,000 blood samples. The clock uses information from 1000 CpG sites and predicts people with certain conditions older than healthy controls: [[Inflammatory bowel disease|IBD]], [[Dementia|frontotemporal dementia]], ovarian cancer, obesity. The aging clock is planned to be released for public use in 2021 by an [[Insilico Medicine]] spinoff company Deep Longevity.
== Relation to human cognitive and brain development ==
Deep learning is closely related to a class of theories of [[brain development]] (specifically, neocortical development) proposed by [[cognitive neuroscientist]]s in the early 1990s.<ref name="UTGOFF">{{cite journal | last1 = Utgoff | first1 = P. E. | last2 = Stracuzzi | first2 = D. J. | s2cid = 1119517 | year = 2002 | title = Many-layered learning | journal = Neural Computation | volume = 14 | issue = 10| pages = 2497–2529 | doi=10.1162/08997660260293319| pmid = 12396572 }}</ref><ref name="ELMAN">{{cite book|url={{google books |plainurl=y |id=vELaRu_MrwoC}}|title=Rethinking Innateness: A Connectionist Perspective on Development|last=Elman|first=Jeffrey L.|publisher=MIT Press|year=1998|isbn=978-0-262-55030-7}}</ref><ref name="SHRAGER">{{cite journal | last1 = Shrager | first1 = J. | last2 = Johnson | first2 = MH | year = 1996 | title = Dynamic plasticity influences the emergence of function in a simple cortical array | journal = Neural Networks | volume = 9 | issue = 7| pages = 1119–1129 | doi=10.1016/0893-6080(96)00033-0| pmid = 12662587 }}</ref><ref name="QUARTZ">{{cite journal | last1 = Quartz | first1 = SR | last2 = Sejnowski | first2 = TJ | year = 1997 | title = The neural basis of cognitive development: A constructivist manifesto | journal = Behavioral and Brain Sciences | volume = 20 | issue = 4| pages = 537–556 | doi=10.1017/s0140525x97001581| pmid = 10097006 | citeseerx = 10.1.1.41.7854 | s2cid = 5818342 }}</ref> These developmental theories were instantiated in computational models, making them predecessors of deep learning systems. These developmental models share the property that various proposed learning dynamics in the brain (e.g., a wave of [[nerve growth factor]]) support the [[self-organization]] somewhat analogous to the neural networks utilized in deep learning models. Like the [[neocortex]], neural networks employ a hierarchy of layered filters in which each layer considers information from a prior layer (or the operating environment), and then passes its output (and possibly the original input), to other layers. This process yields a self-organizing stack of [[transducer]]s, well-tuned to their operating environment. A 1995 description stated, "...the infant's brain seems to organize itself under the influence of waves of so-called trophic-factors ... different regions of the brain become connected sequentially, with one layer of tissue maturing before another and so on until the whole brain is mature."<ref name="BLAKESLEE">S. Blakeslee., "In brain's early growth, timetable may be critical," ''The New York Times, Science Section'', pp. B5–B6, 1995.</ref>
A variety of approaches have been used to investigate the plausibility of deep learning models from a neurobiological perspective. On the one hand, several variants of the [[backpropagation]] algorithm have been proposed in order to increase its processing realism.<ref>{{Cite journal|last1=Mazzoni|first1=P.|last2=Andersen|first2=R. A.|last3=Jordan|first3=M. I.|date=15 May 1991|title=A more biologically plausible learning rule for neural networks.|journal=Proceedings of the National Academy of Sciences|volume=88|issue=10|pages=4433–4437|doi=10.1073/pnas.88.10.4433|issn=0027-8424|pmid=1903542|pmc=51674|bibcode=1991PNAS...88.4433M|doi-access=free}}</ref><ref>{{Cite journal|last=O'Reilly|first=Randall C.|s2cid=2376781|date=1 July 1996|title=Biologically Plausible Error-Driven Learning Using Local Activation Differences: The Generalized Recirculation Algorithm|journal=Neural Computation|volume=8|issue=5|pages=895–938|doi=10.1162/neco.1996.8.5.895|issn=0899-7667}}</ref> Other researchers have argued that unsupervised forms of deep learning, such as those based on hierarchical [[generative model]]s and [[deep belief network]]s, may be closer to biological reality.<ref>{{Cite journal|last1=Testolin|first1=Alberto|last2=Zorzi|first2=Marco|s2cid=9868901|date=2016|title=Probabilistic Models and Generative Neural Networks: Towards an Unified Framework for Modeling Normal and Impaired Neurocognitive Functions|journal=Frontiers in Computational Neuroscience|volume=10|pages=73|doi=10.3389/fncom.2016.00073|pmid=27468262|pmc=4943066|issn=1662-5188|doi-access=free}}</ref><ref>{{Cite journal|last1=Testolin|first1=Alberto|last2=Stoianov|first2=Ivilin|last3=Zorzi|first3=Marco|s2cid=24504018|date=September 2017|title=Letter perception emerges from unsupervised deep learning and recycling of natural image features|journal=Nature Human Behaviour|volume=1|issue=9|pages=657–664|doi=10.1038/s41562-017-0186-2|pmid=31024135|issn=2397-3374}}</ref> In this respect, generative neural network models have been related to neurobiological evidence about sampling-based processing in the cerebral cortex.<ref>{{Cite journal|last1=Buesing|first1=Lars|last2=Bill|first2=Johannes|last3=Nessler|first3=Bernhard|last4=Maass|first4=Wolfgang|s2cid=7504633|date=3 November 2011|title=Neural Dynamics as Sampling: A Model for Stochastic Computation in Recurrent Networks of Spiking Neurons|journal=PLOS Computational Biology|volume=7|issue=11|pages=e1002211|doi=10.1371/journal.pcbi.1002211|pmid=22096452|pmc=3207943|issn=1553-7358|bibcode=2011PLSCB...7E2211B}}</ref>
Although a systematic comparison between the human brain organization and the neuronal encoding in deep networks has not yet been established, several analogies have been reported. For example, the computations performed by deep learning units could be similar to those of actual neurons<ref>{{Cite journal|last1=Cash|first1=S.|last2=Yuste|first2=R.|date=February 1999|title=Linear summation of excitatory inputs by CA1 pyramidal neurons|journal=Neuron|volume=22|issue=2|pages=383–394|issn=0896-6273|pmid=10069343|doi=10.1016/s0896-6273(00)81098-3|s2cid=14663106}}</ref> and neural populations.<ref>{{Cite journal|date=1 August 2004|title=Sparse coding of sensory inputs|journal=Current Opinion in Neurobiology|volume=14|issue=4|pages=481–487|doi=10.1016/j.conb.2004.07.007|pmid=15321069|issn=0959-4388 | last1 = Olshausen | first1 = B | last2 = Field | first2 = D|s2cid=16560320}}</ref> Similarly, the representations developed by deep learning models are similar to those measured in the primate visual system<ref>{{Cite journal|last1=Yamins|first1=Daniel L K|last2=DiCarlo|first2=James J|s2cid=16970545|date=March 2016|title=Using goal-driven deep learning models to understand sensory cortex|journal=Nature Neuroscience|volume=19|issue=3|pages=356–365|doi=10.1038/nn.4244|pmid=26906502|issn=1546-1726}}</ref> both at the single-unit<ref>{{Cite journal|last1=Zorzi|first1=Marco|last2=Testolin|first2=Alberto|s2cid=39281431|date=19 February 2018|title=An emergentist perspective on the origin of number sense|journal=Phil. Trans. R. Soc. B|volume=373|issue=1740|pages=20170043|doi=10.1098/rstb.2017.0043|issn=0962-8436|pmid=29292348|pmc=5784047}}</ref> and at the population<ref>{{Cite journal|last1=Güçlü|first1=Umut|last2=van Gerven|first2=Marcel A. J.|date=8 July 2015|title=Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream|journal=Journal of Neuroscience|volume=35|issue=27|pages=10005–10014|doi=10.1523/jneurosci.5023-14.2015|pmid=26157000|pmc=6605414|arxiv=1411.6422}}</ref> levels.
== Commercial activity ==
[[Facebook]]'s AI lab performs tasks such as [[Automatic image annotation|automatically tagging uploaded pictures]] with the names of the people in them.<ref name="METZ2013">{{cite magazine |first=C. |last=Metz |title=Facebook's 'Deep Learning' Guru Reveals the Future of AI |url=https://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/ |magazine=Wired |date=12 December 2013 |access-date=26 August 2017 |archive-date=28 March 2014 |archive-url=https://web.archive.org/web/20140328071226/http://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/ |url-status=live }}</ref>
Google's [[DeepMind Technologies]] developed a system capable of learning how to play [[Atari]] video games using only pixels as data input. In 2015 they demonstrated their [[AlphaGo]] system, which learned the game of [[Go (game)|Go]] well enough to beat a professional Go player.<ref>{{Cite journal|title = Google AI algorithm masters ancient game of Go|url = http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234|journal = Nature|year = 2016|doi = 10.1038/529445a|access-date = 30 January 2016|archive-date = 2 May 2019|archive-url = https://web.archive.org/web/20190502200837/http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234|url-status = live|last1 = Gibney|first1 = Elizabeth|volume = 529|issue = 7587|pages = 445–446|pmid = 26819021|bibcode = 2016Natur.529..445G|s2cid = 4460235}}</ref><ref>{{Cite journal|title = Mastering the game of Go with deep neural networks and tree search|journal = [[Nature (journal)|Nature]]| issn= 0028-0836|pages = 484–489|volume = 529|issue = 7587|doi = 10.1038/nature16961|pmid = 26819042|first1 = David|last1 = Silver|author-link1=David Silver (programmer)|first2 = Aja|last2 = Huang|author-link2=Aja Huang|first3 = Chris J.|last3 = Maddison|first4 = Arthur|last4 = Guez|first5 = Laurent|last5 = Sifre|first6 = George van den|last6 = Driessche|first7 = Julian|last7 = Schrittwieser|first8 = Ioannis|last8 = Antonoglou|first9 = Veda|last9 = Panneershelvam|first10= Marc|last10= Lanctot|first11= Sander|last11= Dieleman|first12=Dominik|last12= Grewe|first13= John|last13= Nham|first14= Nal|last14= Kalchbrenner|first15= Ilya|last15= Sutskever|author-link15=Ilya Sutskever|first16= Timothy|last16= Lillicrap|first17= Madeleine|last17= Leach|first18= Koray|last18= Kavukcuoglu|first19= Thore|last19= Graepel|first20= Demis |last20=Hassabis|s2cid = 515925|author-link20=Demis Hassabis|date= 28 January 2016|bibcode = 2016Natur.529..484S}}{{closed access}}</ref><ref>{{Cite web|title = A Google DeepMind Algorithm Uses Deep Learning and More to Master the Game of Go {{!}} MIT Technology Review|url = http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/|website = MIT Technology Review|access-date = 30 January 2016|archive-date = 1 February 2016|archive-url = https://web.archive.org/web/20160201140636/http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/|url-status = dead}}</ref> [[Google Translate]] uses a neural network to translate between more than 100 languages.
In 2017, Covariant.ai was launched, which focuses on integrating deep learning into factories.<ref>{{Cite news|url=https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html|title=A.I. Researchers Leave Elon Musk Lab to Begin Robotics Start-Up|first=Cade|last=Metz|newspaper=The New York Times|date=6 November 2017|access-date=5 July 2019|archive-date=7 July 2019|archive-url=https://web.archive.org/web/20190707161547/https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html|url-status=live}}</ref>
As of 2008,<ref>{{Cite journal|title=TAMER: Training an Agent Manually via Evaluative Reinforcement|author1=Bradley Knox, W.|author2=Stone, Peter|year=2008|journal=2008 7th IEEE International Conference on Development and Learning|pages = 292–297|doi=10.1109/devlrn.2008.4640845|isbn = 978-1-4244-2661-4|s2cid = 5613334}}</ref> researchers at [[University of Texas at Austin|The University of Texas at Austin]] (UT) developed a machine learning framework called Training an Agent Manually via Evaluative Reinforcement, or TAMER, which proposed new methods for robots or computer programs to learn how to perform tasks by interacting with a human instructor.<ref name=":12" /> First developed as TAMER, a new algorithm called Deep TAMER was later introduced in 2018 during a collaboration between [[U.S. Army Research Laboratory]] (ARL) and UT researchers. Deep TAMER used deep learning to provide a robot the ability to learn new tasks through observation.<ref name=":12" /> Using Deep TAMER, a robot learned a task with a human trainer, watching video streams or observing a human perform a task in-person. The robot later practiced the task with the help of some coaching from the trainer, who provided feedback such as “good job” and “bad job.”<ref>{{Cite web|url=https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner|title=Talk to the Algorithms: AI Becomes a Faster Learner|website=governmentciomedia.com|access-date=29 August 2018|archive-date=28 August 2018|archive-url=https://web.archive.org/web/20180828001727/https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner|url-status=live}}</ref>
== Criticism and comment ==
Deep learning has attracted both criticism and comment, in some cases from outside the field of computer science.
=== Theory ===
{{see also|Explainable artificial intelligence}}
A main criticism concerns the lack of theory surrounding some methods.<ref>{{Cite web|url=https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1|title=In defense of skepticism about deep learning|last=Marcus|first=Gary|date=14 January 2018|website=Gary Marcus|access-date=11 October 2018|archive-date=12 October 2018|archive-url=https://web.archive.org/web/20181012035405/https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1|url-status=live}}</ref> Learning in the most common deep architectures is implemented using well-understood gradient descent. However, the theory surrounding other algorithms, such as contrastive divergence is less clear.{{citation needed|date=July 2016}} (e.g., Does it converge? If so, how fast? What is it approximating?) Deep learning methods are often looked at as a [[black box]], with most confirmations done empirically, rather than theoretically.<ref name="Knight 2017">{{cite web | last=Knight | first=Will | title=DARPA is funding projects that will try to open up AI's black boxes | website=MIT Technology Review | date=14 March 2017 | url=https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/ | access-date=2 November 2017 | archive-date=4 November 2019 | archive-url=https://web.archive.org/web/20191104033107/https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/ | url-status=live }}</ref>
Others point out that deep learning should be looked at as a step towards realizing strong AI, not as an all-encompassing solution. Despite the power of deep learning methods, they still lack much of the functionality needed for realizing this goal entirely. Research psychologist Gary Marcus noted:<blockquote>"Realistically, deep learning is only part of the larger challenge of building intelligent machines. Such techniques lack ways of representing [[causality|causal relationships]] (...) have no obvious ways of performing [[inference|logical inferences]], and they are also still a long way from integrating abstract knowledge, such as information about what objects are, what they are for, and how they are typically used. The most powerful A.I. systems, like [[Watson (computer)|Watson]] (...) use techniques like deep learning as just one element in a very complicated ensemble of techniques, ranging from the statistical technique of [[Bayesian inference]] to [[deductive reasoning]]."<ref>{{cite magazine|url=https://www.newyorker.com/|title=Is "Deep Learning" a Revolution in Artificial Intelligence?|last=Marcus|first=Gary|date=November 25, 2012|magazine=The New Yorker|access-date=2017-06-14|archive-date=2009-11-27|archive-url=https://web.archive.org/web/20091127184826/http://www.newyorker.com/|url-status=live}}</ref></blockquote>
In further reference to the idea that artistic sensitivity might be inherent in relatively low levels of the cognitive hierarchy, a published series of graphic representations of the internal states of deep (20-30 layers) neural networks attempting to discern within essentially random data the images on which they were trained<ref>{{cite web |url=http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html |author1=Alexander Mordvintsev |author2=Christopher Olah |author3=Mike Tyka |date=17 June 2015 |title=Inceptionism: Going Deeper into Neural Networks |publisher=Google Research Blog |access-date=20 June 2015 |archive-date=3 July 2015 |archive-url=https://web.archive.org/web/20150703064823/http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html |url-status=live }}</ref> demonstrate a visual appeal: the original research notice received well over 1,000 comments, and was the subject of what was for a time the most frequently accessed article on ''[[The Guardian]]'s''<ref>{{cite news|url=https://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep|title=Yes, androids do dream of electric sheep|date=18 June 2015|newspaper=The Guardian|author=Alex Hern|access-date=20 June 2015|archive-date=19 June 2015|archive-url=https://web.archive.org/web/20150619200845/http://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep|url-status=live}}</ref> website.
=== Errors ===
Some deep learning architectures display problematic behaviors,<ref name=goertzel>{{cite web |first=Ben |last=Goertzel |title=Are there Deep Reasons Underlying the Pathologies of Today's Deep Learning Algorithms? |year=2015 |url=http://goertzel.org/DeepLearning_v1.pdf |access-date=2015-05-10 |archive-date=2015-05-13 |archive-url=https://web.archive.org/web/20150513053107/http://goertzel.org/DeepLearning_v1.pdf |url-status=live }}</ref> such as confidently classifying unrecognizable images as belonging to a familiar category of ordinary images (2014)<ref>{{cite arXiv |eprint=1412.1897|last1=Nguyen|first1=Anh|title=Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images|last2=Yosinski|first2=Jason|last3=Clune|first3=Jeff|class=cs.CV|year=2014}}</ref> and misclassifying minuscule perturbations of correctly classified images (2013).<ref>{{cite arXiv |eprint=1312.6199|last1=Szegedy|first1=Christian|title=Intriguing properties of neural networks|last2=Zaremba|first2=Wojciech|last3=Sutskever|first3=Ilya|last4=Bruna|first4=Joan|last5=Erhan|first5=Dumitru|last6=Goodfellow|first6=Ian|last7=Fergus|first7=Rob|class=cs.CV|year=2013}}</ref> [[Ben Goertzel|Goertzel]] hypothesized that these behaviors are due to limitations in their internal representations and that these limitations would inhibit integration into heterogeneous multi-component [[artificial general intelligence]] (AGI) architectures.<ref name="goertzel" /> These issues may possibly be addressed by deep learning architectures that internally form states homologous to image-grammar<ref>{{cite journal | last1 = Zhu | first1 = S.C. | last2 = Mumford | first2 = D. | year = 2006| title = A stochastic grammar of images | journal = Found. Trends Comput. Graph. Vis. | volume = 2 | issue = 4| pages = 259–362 | doi = 10.1561/0600000018| citeseerx = 10.1.1.681.2190 }}</ref> decompositions of observed entities and events.<ref name="goertzel"/> [[Grammar induction|Learning a grammar]] (visual or linguistic) from training data would be equivalent to restricting the system to [[commonsense reasoning]] that operates on concepts in terms of grammatical [[Production (computer science)|production rules]] and is a basic goal of both human language acquisition<ref>Miller, G. A., and N. Chomsky. "Pattern conception." Paper for Conference on pattern detection, University of Michigan. 1957.</ref> and [[artificial intelligence]] (AI).<ref>{{cite web |first=Jason |last=Eisner |title=Deep Learning of Recursive Structure: Grammar Induction |url=http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/ |access-date=2015-05-10 |archive-date=2017-12-30 |archive-url=https://web.archive.org/web/20171230010335/http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/ |url-status=dead }}</ref>
=== Cyber threat ===
As deep learning moves from the lab into the world, research and experience show that artificial neural networks are vulnerable to hacks and deception.<ref>{{Cite web|url=https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425|title=Hackers Have Already Started to Weaponize Artificial Intelligence|website=Gizmodo|date=11 September 2017 |access-date=11 October 2019|archive-date=11 October 2019|archive-url=https://web.archive.org/web/20191011162231/https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425|url-status=live}}</ref> By identifying patterns that these systems use to function, attackers can modify inputs to ANNs in such a way that the ANN finds a match that human observers would not recognize. For example, an attacker can make subtle changes to an image such that the ANN finds a match even though the image looks to a human nothing like the search target. Such manipulation is termed an “adversarial attack.”<ref>{{Cite web|url=https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/|title=How hackers can force AI to make dumb mistakes|date=18 June 2018|website=The Daily Dot|language=en|access-date=11 October 2019|archive-date=11 October 2019|archive-url=https://web.archive.org/web/20191011162230/https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/|url-status=live}}</ref>
In 2016 researchers used one ANN to doctor images in trial and error fashion, identify another's focal points and thereby generate images that deceived it. The modified images looked no different to human eyes. Another group showed that printouts of doctored images then photographed successfully tricked an image classification system.<ref name=":4">{{Cite news|url=https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change|title=AI Is Easy to Fool—Why That Needs to Change|date=10 October 2017|work=Singularity Hub|access-date=11 October 2017|archive-date=11 October 2017|archive-url=https://web.archive.org/web/20171011233017/https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change/|url-status=live}}</ref> One defense is reverse image search, in which a possible fake image is submitted to a site such as [[TinEye]] that can then find other instances of it. A refinement is to search using only parts of the image, to identify images from which that piece may have been taken'''.'''<ref>{{Cite journal|last=Gibney|first=Elizabeth|title=The scientist who spots fake videos|url=https://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784|journal=Nature|doi=10.1038/nature.2017.22784|year=2017|access-date=2017-10-11|archive-date=2017-10-10|archive-url=https://web.archive.org/web/20171010011017/http://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784|url-status=live}}</ref>
Another group showed that certain [[Psychedelic art|psychedelic]] spectacles could fool a [[facial recognition system]] into thinking ordinary people were celebrities, potentially allowing one person to impersonate another. In 2017 researchers added stickers to [[stop sign]]s and caused an ANN to misclassify them.<ref name=":4" />
ANNs can however be further trained to detect attempts at deception, potentially leading attackers and defenders into an arms race similar to the kind that already defines the [[malware]] defense industry. ANNs have been trained to defeat ANN-based anti-malware software by repeatedly attacking a defense with malware that was continually altered by a [[genetic algorithm]] until it tricked the anti-malware while retaining its ability to damage the target.<ref name=":4" />
In 2016, another group demonstrated that certain sounds could make the [[Google Now]] voice command system open a particular web address, and hypothesized that this could "serve as a stepping stone for further attacks (e.g., opening a web page hosting drive-by malware)."<ref name=":4" />
In “data poisoning,” false data is continually smuggled into a machine learning system's training set to prevent it from achieving mastery.<ref name=":4" />
=== Reliance on human [[microwork]] ===
{{More citations needed|section|date=April 2021}}
Most Deep Learning systems rely on training and verification data that is generated and/or annotated by humans.<ref>{{Cite journal |last=Tubaro |first=Paola |date=2020 |title=Whose intelligence is artificial intelligence? |url=https://hal.science/hal-03029735 |journal=Global Dialogue |language=en |pages=38}}</ref> It has been argued in [[Media studies|media philosophy]] that not only low-paid [[Clickworkers|clickwork]] (e.g. on [[Amazon Mechanical Turk]]) is regularly deployed for this purpose, but also implicit forms of human [[microwork]] that are often not recognized as such.<ref name=":13">{{Cite journal|last=Mühlhoff|first=Rainer|date=6 November 2019|title=Human-aided artificial intelligence: Or, how to run large computations in human brains? Toward a media sociology of machine learning|journal=New Media & Society|language=en|volume=22|issue=10|pages=1868–1884|doi=10.1177/1461444819885334|s2cid=209363848|issn=1461-4448|url=https://depositonce.tu-berlin.de/handle/11303/12510}}</ref> The philosopher Rainer Mühlhoff distinguishes five types of "machinic capture" of human microwork to generate training data: (1) [[gamification]] (the embedding of annotation or computation tasks in the flow of a game), (2) "trapping and tracking" (e.g. [[CAPTCHA]]s for image recognition or click-tracking on Google [[Search engine results page|search results pages]]), (3) exploitation of social motivations (e.g. [[Tag (Facebook)|tagging faces]] on [[Facebook]] to obtain labeled facial images), (4) [[information mining]] (e.g. by leveraging [[Quantified self|quantified-self]] devices such as [[activity tracker]]s) and (5) [[Clickworkers|clickwork]].<ref name=":13" />
Mühlhoff argues that in most commercial end-user applications of Deep Learning such as [[DeepFace|Facebook's face recognition system]], the need for training data does not stop once an ANN is trained. Rather, there is a continued demand for human-generated verification data to constantly calibrate and update the ANN. For this purpose Facebook introduced the feature that once a user is automatically recognized in an image, they receive a notification. They can choose whether of not they like to be publicly labeled on the image, or tell Facebook that it is not them in the picture.<ref>{{Cite magazine|url=https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/|title=Facebook Can Now Find Your Face, Even When It's Not Tagged|magazine=Wired|access-date=22 November 2019|language=en|issn=1059-1028|archive-date=10 August 2019|archive-url=https://web.archive.org/web/20190810223940/https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/|url-status=live}}</ref> This user interface is a mechanism to generate "a constant stream of verification data"<ref name=":13" /> to further train the network in real-time. As Mühlhoff argues, involvement of human users to generate training and verification data is so typical for most commercial end-user applications of Deep Learning that such systems may be referred to as "human-aided artificial intelligence".<ref name=":13" />
== See also ==
* [[Applications of artificial intelligence]]
* [[Comparison of deep learning software]]
* [[Compressed sensing]]
* [[Differentiable programming]]
* [[Echo state network]]
* [[List of artificial intelligence projects]]
* [[Liquid state machine]]
* [[List of datasets for machine-learning research]]
* [[Reservoir computing]]
* [[Scale space#Deep learning and scale space|Scale space and deep learning]]
* [[Sparse coding]]
== References ==
{{Reflist|30em}}
== Further reading ==
{{refbegin}}
* {{cite book
|title=Deep Learning
|year=2016
|first1=Ian
|last1=Goodfellow
|author-link1=Ian Goodfellow
|first2=Yoshua
|last2=Bengio
|author-link2=Yoshua Bengio
|first3=Aaron
|last3=Courville
|publisher=MIT Press
|url=http://www.deeplearningbook.org
|isbn=978-0-26203561-3
|postscript=, introductory textbook.
|access-date=2021-05-09
|archive-date=2016-04-16
|archive-url=https://web.archive.org/web/20160416111010/http://www.deeplearningbook.org/
|url-status=live
}}
{{Prone to spam|date=June 2015}}<!-- {{No more links}}
Please be cautious about adding more external links.
Wikipedia is not a collection of links and should not be used for advertising.
Excessive or inappropriate links will be removed.
See [[Wikipedia:External links]] and [[Wikipedia:Spam]] for details.
If there are already suitable links, propose additions or replacements on
the article's talk page, or submit your link to the relevant category at
DMOZ (dmoz.org) and link there using {{Dmoz}}.
-->
{{refend}}
{{Differentiable computing}}
[[Category:Deep learning| ]]
[[Category:Artificial neural networks]]
[[Category:Emerging technologies]]' |
Unified diff of changes made by edit (edit_diff) | '@@ -1,16 +1,3 @@
-{{Short description|Branch of machine learning}}
-[[File:Deep Learning.jpg|alt=Representing images on multiple layers of abstraction in deep learning|thumb|upright=1.35|Representing images on multiple layers of abstraction in deep learning<ref>{{Cite journal|last1=Schulz|first1=Hannes|last2=Behnke|first2=Sven|date=1 November 2012|title=Deep Learning|journal=KI - Künstliche Intelligenz|language=en|volume=26|issue=4|pages=357–363|doi=10.1007/s13218-012-0198-z|s2cid=220523562|issn=1610-1987|url=https://www.semanticscholar.org/paper/51a80649d16a38d41dbd20472deb3bc9b61b59a0}}</ref>]]
-{{machine learning|Artificial neural network}}
-{{Artificial intelligence|Approaches}}
-
-'''Deep learning''' is part of a broader family of [[machine learning]] methods based on [[artificial neural network]]s with [[representation learning]]. Learning can be [[Supervised learning|supervised]], [[Semi-supervised learning|semi-supervised]] or [[Unsupervised learning|unsupervised]].<ref name="NatureBengio">{{cite journal |last1=LeCun |first1= Yann|last2=Bengio |first2=Yoshua | last3=Hinton | first3= Geoffrey|s2cid=3074096 |year=2015 |title=Deep Learning |journal=Nature |volume=521 |issue=7553 |pages=436–444 |doi=10.1038/nature14539 |pmid=26017442|bibcode=2015Natur.521..436L }}</ref>
-
-Deep-learning architectures such as [[#Deep_neural_networks|deep neural network]]s, [[deep belief network]]s, [[deep reinforcement learning]], [[recurrent neural networks]], [[convolutional neural networks]] and [[Transformer (machine learning model)|transformers]] have been applied to fields including [[computer vision]], [[speech recognition]], [[natural language processing]], [[machine translation]], [[bioinformatics]], [[drug design]], [[medical image analysis]], [[Climatology|climate science]], material inspection and [[board game]] programs, where they have produced results comparable to and in some cases surpassing human expert performance.<ref name=":9">{{Cite book |doi=10.1109/cvpr.2012.6248110 |isbn=978-1-4673-1228-8|arxiv=1202.2745|chapter=Multi-column deep neural networks for image classification|title=2012 IEEE Conference on Computer Vision and Pattern Recognition|pages=3642–3649|year=2012|last1=Ciresan|first1=D.|last2=Meier|first2=U.|last3=Schmidhuber|first3=J.|s2cid=2161592}}</ref><ref name="krizhevsky2012">{{cite journal|last1=Krizhevsky|first1=Alex|last2=Sutskever|first2=Ilya|last3=Hinton|first3=Geoffrey|date=2012|title=ImageNet Classification with Deep Convolutional Neural Networks|url=https://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf|journal=NIPS 2012: Neural Information Processing Systems, Lake Tahoe, Nevada|access-date=2017-05-24|archive-date=2017-01-10|archive-url=https://web.archive.org/web/20170110123024/http://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf|url-status=live}}</ref><ref>{{cite web |title=Google's AlphaGo AI wins three-match series against the world's best Go player |url=https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/ |website=TechCrunch |date=25 May 2017 |access-date=17 June 2018 |archive-date=17 June 2018 |archive-url=https://web.archive.org/web/20180617065807/https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/ |url-status=live }}</ref>
-
-[[Artificial neural network]]s (ANNs) were inspired by information processing and distributed communication nodes in [[biological system]]s. ANNs have various differences from biological [[brain]]s. Specifically, artificial neural networks tend to be static and symbolic, while the biological brain of most living organisms is dynamic (plastic) and analog.<ref>{{Cite journal|last1=Marblestone|first1=Adam H.|last2=Wayne|first2=Greg|last3=Kording|first3=Konrad P.|s2cid=1994856|date=2016|title=Toward an Integration of Deep Learning and Neuroscience |journal=Frontiers in Computational Neuroscience |volume=10|pages=94|doi=10.3389/fncom.2016.00094 |pmc=5021692|pmid=27683554|bibcode=2016arXiv160603813M|arxiv=1606.03813|doi-access=free}}</ref><ref>{{cite arXiv|last1=Bengio|first1=Yoshua|last2=Lee|first2=Dong-Hyun|last3=Bornschein|first3=Jorg|last4=Mesnard|first4=Thomas|last5=Lin|first5=Zhouhan|date=13 February 2015|title=Towards Biologically Plausible Deep Learning|eprint=1502.04156|class=cs.LG}}</ref>
-
-The adjective "deep" in deep learning refers to the use of multiple layers in the network. Early work showed that a linear [[perceptron]] cannot be a universal classifier, but that a network with a nonpolynomial activation function with one hidden layer of unbounded width can. Deep learning is a modern variation that is concerned with an unbounded number of layers of bounded size, which permits practical application and optimized implementation, while retaining theoretical universality under mild conditions. In deep learning the layers are also permitted to be heterogeneous and to deviate widely from biologically informed [[connectionism|connectionist]] models, for the sake of efficiency, trainability and understandability.
-
-{{toclimit|3}}
+https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt
== Definition ==
' |
New page size (new_size) | 146040 |
Old page size (old_size) | 150965 |
Size change in edit (edit_delta) | -4925 |
Lines added in edit (added_lines) | [
0 => 'https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt'
] |
Lines removed in edit (removed_lines) | [
0 => '{{Short description|Branch of machine learning}}',
1 => '[[File:Deep Learning.jpg|alt=Representing images on multiple layers of abstraction in deep learning|thumb|upright=1.35|Representing images on multiple layers of abstraction in deep learning<ref>{{Cite journal|last1=Schulz|first1=Hannes|last2=Behnke|first2=Sven|date=1 November 2012|title=Deep Learning|journal=KI - Künstliche Intelligenz|language=en|volume=26|issue=4|pages=357–363|doi=10.1007/s13218-012-0198-z|s2cid=220523562|issn=1610-1987|url=https://www.semanticscholar.org/paper/51a80649d16a38d41dbd20472deb3bc9b61b59a0}}</ref>]]',
2 => '{{machine learning|Artificial neural network}}',
3 => '{{Artificial intelligence|Approaches}}',
4 => '',
5 => ''''Deep learning''' is part of a broader family of [[machine learning]] methods based on [[artificial neural network]]s with [[representation learning]]. Learning can be [[Supervised learning|supervised]], [[Semi-supervised learning|semi-supervised]] or [[Unsupervised learning|unsupervised]].<ref name="NatureBengio">{{cite journal |last1=LeCun |first1= Yann|last2=Bengio |first2=Yoshua | last3=Hinton | first3= Geoffrey|s2cid=3074096 |year=2015 |title=Deep Learning |journal=Nature |volume=521 |issue=7553 |pages=436–444 |doi=10.1038/nature14539 |pmid=26017442|bibcode=2015Natur.521..436L }}</ref>',
6 => '',
7 => 'Deep-learning architectures such as [[#Deep_neural_networks|deep neural network]]s, [[deep belief network]]s, [[deep reinforcement learning]], [[recurrent neural networks]], [[convolutional neural networks]] and [[Transformer (machine learning model)|transformers]] have been applied to fields including [[computer vision]], [[speech recognition]], [[natural language processing]], [[machine translation]], [[bioinformatics]], [[drug design]], [[medical image analysis]], [[Climatology|climate science]], material inspection and [[board game]] programs, where they have produced results comparable to and in some cases surpassing human expert performance.<ref name=":9">{{Cite book |doi=10.1109/cvpr.2012.6248110 |isbn=978-1-4673-1228-8|arxiv=1202.2745|chapter=Multi-column deep neural networks for image classification|title=2012 IEEE Conference on Computer Vision and Pattern Recognition|pages=3642–3649|year=2012|last1=Ciresan|first1=D.|last2=Meier|first2=U.|last3=Schmidhuber|first3=J.|s2cid=2161592}}</ref><ref name="krizhevsky2012">{{cite journal|last1=Krizhevsky|first1=Alex|last2=Sutskever|first2=Ilya|last3=Hinton|first3=Geoffrey|date=2012|title=ImageNet Classification with Deep Convolutional Neural Networks|url=https://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf|journal=NIPS 2012: Neural Information Processing Systems, Lake Tahoe, Nevada|access-date=2017-05-24|archive-date=2017-01-10|archive-url=https://web.archive.org/web/20170110123024/http://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf|url-status=live}}</ref><ref>{{cite web |title=Google's AlphaGo AI wins three-match series against the world's best Go player |url=https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/ |website=TechCrunch |date=25 May 2017 |access-date=17 June 2018 |archive-date=17 June 2018 |archive-url=https://web.archive.org/web/20180617065807/https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/ |url-status=live }}</ref>',
8 => '',
9 => '[[Artificial neural network]]s (ANNs) were inspired by information processing and distributed communication nodes in [[biological system]]s. ANNs have various differences from biological [[brain]]s. Specifically, artificial neural networks tend to be static and symbolic, while the biological brain of most living organisms is dynamic (plastic) and analog.<ref>{{Cite journal|last1=Marblestone|first1=Adam H.|last2=Wayne|first2=Greg|last3=Kording|first3=Konrad P.|s2cid=1994856|date=2016|title=Toward an Integration of Deep Learning and Neuroscience |journal=Frontiers in Computational Neuroscience |volume=10|pages=94|doi=10.3389/fncom.2016.00094 |pmc=5021692|pmid=27683554|bibcode=2016arXiv160603813M|arxiv=1606.03813|doi-access=free}}</ref><ref>{{cite arXiv|last1=Bengio|first1=Yoshua|last2=Lee|first2=Dong-Hyun|last3=Bornschein|first3=Jorg|last4=Mesnard|first4=Thomas|last5=Lin|first5=Zhouhan|date=13 February 2015|title=Towards Biologically Plausible Deep Learning|eprint=1502.04156|class=cs.LG}}</ref>',
10 => '',
11 => 'The adjective "deep" in deep learning refers to the use of multiple layers in the network. Early work showed that a linear [[perceptron]] cannot be a universal classifier, but that a network with a nonpolynomial activation function with one hidden layer of unbounded width can. Deep learning is a modern variation that is concerned with an unbounded number of layers of bounded size, which permits practical application and optimized implementation, while retaining theoretical universality under mild conditions. In deep learning the layers are also permitted to be heterogeneous and to deviate widely from biologically informed [[connectionism|connectionist]] models, for the sake of efficiency, trainability and understandability.',
12 => '',
13 => '{{toclimit|3}}'
] |
All external links added in the edit (added_links) | [
0 => 'https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt'
] |
All external links removed in the edit (removed_links) | [
0 => 'https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/',
1 => 'https://ui.adsabs.harvard.edu/abs/2016arXiv160603813M',
2 => 'https://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf',
3 => 'https://pubmed.ncbi.nlm.nih.gov/27683554',
4 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5021692',
5 => 'https://web.archive.org/web/20170110123024/http://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf',
6 => 'https://web.archive.org/web/20180617065807/https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/',
7 => 'https://arxiv.org/abs/1202.2745',
8 => 'https://arxiv.org/abs/1502.04156',
9 => 'https://arxiv.org/abs/1606.03813',
10 => 'https://doi.org/10.1007%2Fs13218-012-0198-z',
11 => 'https://doi.org/10.1109%2Fcvpr.2012.6248110',
12 => 'https://doi.org/10.3389%2Ffncom.2016.00094',
13 => 'https://api.semanticscholar.org/CorpusID:1994856',
14 => 'https://api.semanticscholar.org/CorpusID:2161592',
15 => 'https://api.semanticscholar.org/CorpusID:220523562',
16 => 'https://www.semanticscholar.org/paper/51a80649d16a38d41dbd20472deb3bc9b61b59a0',
17 => 'https://www.worldcat.org/issn/1610-1987'
] |
All external links in the new text (all_links) | [
0 => 'http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf',
1 => 'https://doi.org/10.1561%2F2000000039',
2 => 'https://web.archive.org/web/20160314152112/http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf',
3 => 'https://web.archive.org/web/20160304084250/http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf',
4 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.701.9550',
5 => 'https://doi.org/10.1561%2F2200000006',
6 => 'http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf',
7 => 'https://arxiv.org/abs/1206.5538',
8 => 'https://doi.org/10.1109%2Ftpami.2013.50',
9 => 'https://pubmed.ncbi.nlm.nih.gov/23787338',
10 => 'https://api.semanticscholar.org/CorpusID:393948',
11 => 'https://ui.adsabs.harvard.edu/abs/2015Natur.521..436L',
12 => 'https://doi.org/10.1038%2Fnature14539',
13 => 'https://pubmed.ncbi.nlm.nih.gov/26017442',
14 => 'https://api.semanticscholar.org/CorpusID:3074096',
15 => 'https://arxiv.org/abs/1404.7828',
16 => 'https://doi.org/10.1016%2Fj.neunet.2014.09.003',
17 => 'https://pubmed.ncbi.nlm.nih.gov/25462637',
18 => 'https://api.semanticscholar.org/CorpusID:11715509',
19 => 'https://books.google.com/books?id=9CqQDwAAQBAJ&pg=PA15',
20 => 'http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf',
21 => 'https://web.archive.org/web/20191020195638/http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf',
22 => 'https://doi.org/10.4249%2Fscholarpedia.5947',
23 => 'https://ui.adsabs.harvard.edu/abs/2009SchpJ...4.5947H',
24 => 'https://web.archive.org/web/20151010204407/http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf',
25 => 'https://doi.org/10.1007%2Fbf02551274',
26 => 'https://api.semanticscholar.org/CorpusID:3958369',
27 => 'http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf',
28 => 'https://doi.org/10.1016%2F0893-6080(91)90009-t',
29 => 'https://books.google.com/books?id=bX4pAQAAMAAJ',
30 => 'https://books.google.com/books?id=Otk32Y3QkxQC&pg=PA48',
31 => 'http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width',
32 => 'https://web.archive.org/web/20190213005539/http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width',
33 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5527101',
34 => 'https://ui.adsabs.harvard.edu/abs/2017NatCo...8..138O',
35 => 'https://doi.org/10.1038%2Fs41467-017-00181-8',
36 => 'https://pubmed.ncbi.nlm.nih.gov/28743932',
37 => 'https://books.google.com/books?id=NZP6AQAAQBAJ',
38 => 'https://arxiv.org/abs/1505.03654',
39 => 'https://doi.org/10.1016%2Fj.acha.2015.12.005',
40 => 'https://api.semanticscholar.org/CorpusID:12149203',
41 => 'http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf',
42 => 'https://web.archive.org/web/20170111005101/http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf',
43 => 'https://ieeexplore.ieee.org/document/9070967',
44 => 'https://doi.org/10.1109%2FCSCI49370.2019.00067',
45 => 'https://api.semanticscholar.org/CorpusID:216043128',
46 => 'https://books.google.com/books?id=rGFgAAAAMAAJ',
47 => 'http://gmdh.net/articles/history/polynomial.pdf',
48 => 'https://doi.org/10.1109%2FTSMC.1971.4308320',
49 => 'https://web.archive.org/web/20170829230621/http://www.gmdh.net/articles/history/polynomial.pdf',
50 => 'https://doi.org/10.1007%2Fbf00344251',
51 => 'https://pubmed.ncbi.nlm.nih.gov/7370364',
52 => 'https://api.semanticscholar.org/CorpusID:206775608',
53 => 'https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems',
54 => 'https://web.archive.org/web/20160419054654/https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems',
55 => 'https://web.archive.org/web/20170721211929/http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf',
56 => 'http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf',
57 => 'https://www.researchgate.net/publication/35657389',
58 => 'ftp://ftp.idsia.ch/pub/juergen/habilitation.pdf',
59 => 'https://drive.google.com/file/d/1nN_5odSG_QVae54EsQN_qSz-0ZsX6wA0/view?usp=sharing',
60 => 'https://drive.google.com/file/d/0B65v6Wo67Tk5ODRzZmhSR29VeDg/view?usp=sharing',
61 => 'https://ui.adsabs.harvard.edu/abs/1990ApOpt..29.4790Z',
62 => 'https://doi.org/10.1364%2FAO.29.004790',
63 => 'https://pubmed.ncbi.nlm.nih.gov/20577468',
64 => 'https://drive.google.com/file/d/0B65v6Wo67Tk5cm5DTlNGd0NPUmM/view?usp=sharing',
65 => 'https://ui.adsabs.harvard.edu/abs/1991ApOpt..30.4211Z',
66 => 'https://doi.org/10.1364%2FAO.30.004211',
67 => 'https://pubmed.ncbi.nlm.nih.gov/20706526',
68 => 'https://drive.google.com/file/d/0B65v6Wo67Tk5Ml9qeW5nQ3poVTQ/view?usp=sharing',
69 => 'https://ui.adsabs.harvard.edu/abs/1994MedPh..21..517Z',
70 => 'https://doi.org/10.1118%2F1.597177',
71 => 'https://pubmed.ncbi.nlm.nih.gov/8058017',
72 => 'https://ui.adsabs.harvard.edu/abs/1994PaReL..15..807D',
73 => 'https://doi.org/10.1016%2F0167-8655(94)90009-4',
74 => 'https://ui.adsabs.harvard.edu/abs/1995Sci...268.1158H',
75 => 'https://doi.org/10.1126%2Fscience.7761831',
76 => 'https://pubmed.ncbi.nlm.nih.gov/7761831',
77 => 'http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf',
78 => 'https://web.archive.org/web/20150306075401/http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf',
79 => 'https://books.google.com/books?id=NWOcMVA64aAC',
80 => 'https://doi.org/10.1007%2Fb11963',
81 => 'https://api.semanticscholar.org/CorpusID:1304548',
82 => 'https://doi.org/10.1142%2Fs0218001493000455',
83 => 'https://www.worldcat.org/issn/0218-0014',
84 => 'http://dl.acm.org/citation.cfm?id=1895720',
85 => 'https://web.archive.org/web/20210509123135/https://dl.acm.org/doi/10.5555/1895550.1895720',
86 => 'http://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf',
87 => 'https://doi.org/10.1109%2F29.21701',
88 => 'https://hdl.handle.net/10338.dmlcz%2F135496',
89 => 'https://www.worldcat.org/issn/0096-3518',
90 => 'https://web.archive.org/web/20210427001446/https://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf',
91 => 'https://ui.adsabs.harvard.edu/abs/2009ISPM...26...75B',
92 => 'https://doi.org/10.1109%2Fmsp.2009.932166',
93 => 'https://hdl.handle.net/1721.1%2F51891',
94 => 'https://api.semanticscholar.org/CorpusID:357467',
95 => 'https://www.researchgate.net/publication/41229141',
96 => 'https://web.archive.org/web/20210509123131/https://www.researchgate.net/publication/41229141_Artificial_neural_networks_and_their_application_to_sequence_recognition',
97 => 'https://doi.org/10.1016%2F0893-6080(94)90027-2',
98 => 'https://doi.org/10.1016%2FS0167-6393(99)00080-1',
99 => 'https://doi.org/10.1016%2Fs0167-6393(99)00077-1',
100 => 'https://www.researchgate.net/publication/266030526',
101 => 'https://web.archive.org/web/20210509123218/https://www.researchgate.net/publication/266030526_Acoustic_Modeling_with_Deep_Neural_Networks_Using_Raw_Time_Signal_for_LVCSR',
102 => 'https://doi.org/10.1162%2Fneco.1997.9.8.1735',
103 => 'https://www.worldcat.org/issn/0899-7667',
104 => 'https://pubmed.ncbi.nlm.nih.gov/9377276',
105 => 'https://api.semanticscholar.org/CorpusID:1915014',
106 => 'Ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf',
107 => 'https://web.archive.org/web/20210509123139/ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf',
108 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.75.6306',
109 => 'https://mediatum.ub.tum.de/doc/1289941/file.pdf',
110 => 'https://web.archive.org/web/20181118164457/https://mediatum.ub.tum.de/doc/1289941/file.pdf',
111 => 'http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html',
112 => 'https://web.archive.org/web/20160309191532/http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html',
113 => 'http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3',
114 => 'https://doi.org/10.1016%2Fj.tics.2007.09.004',
115 => 'https://www.worldcat.org/issn/1364-6613',
116 => 'https://pubmed.ncbi.nlm.nih.gov/17921042',
117 => 'https://api.semanticscholar.org/CorpusID:15066318',
118 => 'https://web.archive.org/web/20131011071435/http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3',
119 => 'http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf',
120 => 'https://doi.org/10.1162%2Fneco.2006.18.7.1527',
121 => 'https://pubmed.ncbi.nlm.nih.gov/16764513',
122 => 'https://api.semanticscholar.org/CorpusID:2309950',
123 => 'https://web.archive.org/web/20151223164129/http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf',
124 => 'https://arxiv.org/abs/1206.5533',
125 => '//arxiv.org/archive/cs.LG',
126 => 'http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf',
127 => 'https://web.archive.org/web/20180522112408/http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf',
128 => 'https://ui.adsabs.harvard.edu/abs/2012ISPM...29...82H',
129 => 'https://doi.org/10.1109%2Fmsp.2012.2205597',
130 => 'https://api.semanticscholar.org/CorpusID:206485943',
131 => 'https://arxiv.org/abs/2102.04029',
132 => 'https://doi.org/10.1109%2FICCCI50826.2021.9402569',
133 => 'https://api.semanticscholar.org/CorpusID:231846518',
134 => 'https://web.archive.org/web/20180424203806/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf',
135 => 'https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf',
136 => 'https://arxiv.org/abs/1410.4281',
137 => '//arxiv.org/archive/cs.CL',
138 => 'https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf',
139 => 'https://web.archive.org/web/20210509123113/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf',
140 => 'https://indico.cern.ch/event/510372/',
141 => 'https://web.archive.org/web/20160423021403/https://indico.cern.ch/event/510372/',
142 => 'https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf',
143 => 'https://web.archive.org/web/20170926190920/https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf',
144 => 'https://books.google.com/books?id=rUBTBQAAQBAJ',
145 => 'https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/',
146 => 'https://web.archive.org/web/20180316084821/https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/',
147 => 'https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing',
148 => 'https://web.archive.org/web/20170926190732/https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing',
149 => 'https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/',
150 => 'https://web.archive.org/web/20171012095148/https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/',
151 => 'https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks',
152 => 'https://doi.org/10.21437%2FInterspeech.2011-169',
153 => 'https://web.archive.org/web/20171012095522/https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks/',
154 => 'https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/',
155 => 'https://web.archive.org/web/20171012044053/https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/',
156 => 'https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/',
157 => 'https://web.archive.org/web/20201125202428/https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/',
158 => 'https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not',
159 => 'https://web.archive.org/web/20161231203934/https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not',
160 => 'https://ui.adsabs.harvard.edu/abs/2004PatRe..37.1311O',
161 => 'https://doi.org/10.1016%2Fj.patcog.2004.01.013',
162 => 'https://www.academia.edu/40135801',
163 => 'https://web.archive.org/web/20210509123120/https://www.academia.edu/40135801/A_Survey_of_Techniques_for_Optimizing_Deep_Learning_on_GPUs',
164 => 'https://hal.inria.fr/inria-00112631/document',
165 => 'https://web.archive.org/web/20200518193413/https://hal.inria.fr/inria-00112631/document',
166 => 'https://arxiv.org/abs/1003.0358',
167 => 'https://doi.org/10.1162%2Fneco_a_00052',
168 => 'https://pubmed.ncbi.nlm.nih.gov/20858131',
169 => 'https://api.semanticscholar.org/CorpusID:1918673',
170 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.154.372',
171 => 'https://doi.org/10.1145%2F1553374.1553486',
172 => 'https://api.semanticscholar.org/CorpusID:392458',
173 => 'https://arxiv.org/abs/1703.09039',
174 => '//arxiv.org/archive/cs.CV',
175 => 'https://kaggle.com/c/MerckActivity',
176 => 'https://web.archive.org/web/20200716190808/https://www.kaggle.com/c/MerckActivity',
177 => 'http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions',
178 => 'https://web.archive.org/web/20170430142049/http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions',
179 => 'https://tripod.nih.gov/tox21/challenge/leaderboard.jsp',
180 => 'https://web.archive.org/web/20150908025122/https://tripod.nih.gov/tox21/challenge/leaderboard.jsp',
181 => 'https://web.archive.org/web/20150228225709/http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html',
182 => 'http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html',
183 => 'http://ijcai.org/papers11/Papers/IJCAI11-210.pdf',
184 => 'https://doi.org/10.5591%2F978-1-57735-516-8%2Fijcai11-210',
185 => 'https://web.archive.org/web/20140929094040/http://ijcai.org/papers11/Papers/IJCAI11-210.pdf',
186 => 'http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf',
187 => 'https://web.archive.org/web/20170809081713/http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf',
188 => 'https://doi.org/10.1007%2F978-3-642-40763-5_51',
189 => 'https://pubmed.ncbi.nlm.nih.gov/24579167',
190 => 'https://arxiv.org/abs/1411.4555',
191 => 'https://arxiv.org/abs/1411.4952',
192 => 'https://arxiv.org/abs/1411.2539',
193 => 'http://fortune.com/ai-artificial-intelligence-deep-machine-learning/',
194 => 'https://web.archive.org/web/20180414031925/http://fortune.com/ai-artificial-intelligence-deep-machine-learning/',
195 => 'https://ui.adsabs.harvard.edu/abs/2016Natur.529..484S',
196 => 'https://doi.org/10.1038%2Fnature16961',
197 => 'https://www.worldcat.org/issn/1476-4687',
198 => 'https://pubmed.ncbi.nlm.nih.gov/26819042',
199 => 'https://api.semanticscholar.org/CorpusID:515925',
200 => 'https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks',
201 => 'https://web.archive.org/web/20201102151103/https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks',
202 => 'https://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection',
203 => 'https://web.archive.org/web/20170629172111/http://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection',
204 => 'https://openreview.net/pdf?id=SyProzZAW',
205 => 'https://web.archive.org/web/20210107183647/https://openreview.net/pdf?id=SyProzZAW',
206 => 'https://web.archive.org/web/20190331092832/https://www.technologyreview.com/s/513696/deep-learning/',
207 => 'https://www.technologyreview.com/s/513696/deep-learning/',
208 => 'http://elartu.tntu.edu.ua/handle/lib/30719',
209 => 'https://doi.org/10.1109%2F72.963769',
210 => 'https://pubmed.ncbi.nlm.nih.gov/18249962',
211 => 'https://web.archive.org/web/20200126045722/http://elartu.tntu.edu.ua/handle/lib/30719',
212 => 'https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf',
213 => 'https://arxiv.org/abs/1409.3215',
214 => 'https://ui.adsabs.harvard.edu/abs/2014arXiv1409.3215S',
215 => 'https://web.archive.org/web/20210509123145/https://papers.nips.cc/paper/2014/file/a14ac55a4f27472c5d894ec1c3c743d2-Paper.pdf',
216 => 'https://arxiv.org/abs/1602.02410',
217 => 'https://arxiv.org/abs/1512.00103',
218 => 'http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf',
219 => 'https://doi.org/10.21437%2FInterspeech.2010-343',
220 => 'https://web.archive.org/web/20170516181940/http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf',
221 => 'https://www.researchgate.net/publication/220320057',
222 => 'https://web.archive.org/web/20210509123147/https://www.researchgate.net/publication/220320057_Learning_Precise_Timing_with_LSTM_Recurrent_Networks',
223 => 'http://elartu.tntu.edu.ua/handle/lib/38369',
224 => 'https://doi.org/10.1109%2F5.726791',
225 => 'https://api.semanticscholar.org/CorpusID:14542261',
226 => 'https://doi.org/10.1109%2Ficassp.2013.6639347',
227 => 'https://api.semanticscholar.org/CorpusID:13816461',
228 => 'https://arxiv.org/abs/1212.0901',
229 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.752.9151',
230 => 'https://doi.org/10.1109%2Ficassp.2013.6639349',
231 => 'https://api.semanticscholar.org/CorpusID:12485056',
232 => 'http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf',
233 => 'https://web.archive.org/web/20170812140509/http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf',
234 => 'https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation',
235 => 'https://web.archive.org/web/20171201032606/https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation',
236 => 'https://www.researchgate.net/publication/221166159',
237 => 'https://web.archive.org/web/20210509123211/https://www.researchgate.net/publication/221166159_A_brief_introduction_to_Weightless_Neural_Systems',
238 => 'https://dl.acm.org/citation.cfm?doid=3126908.3126912',
239 => 'http://www.escholarship.org/uc/item/6ch40821',
240 => 'https://doi.org/10.1145%2F3126908.3126912',
241 => 'https://api.semanticscholar.org/CorpusID:8869270',
242 => 'https://web.archive.org/web/20200729133850/https://escholarship.org/uc/item/6ch40821',
243 => 'https://arxiv.org/abs/1702.07908',
244 => 'https://ui.adsabs.harvard.edu/abs/2017arXiv170207908V',
245 => 'https://doi.org/10.1007%2Fs11227-017-1994-x',
246 => 'https://api.semanticscholar.org/CorpusID:14135321',
247 => 'http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf',
248 => 'https://web.archive.org/web/20181118122850/http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf',
249 => 'http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/',
250 => 'https://web.archive.org/web/20160201033801/http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/',
251 => 'https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475',
252 => 'https://web.archive.org/web/20200610094310/https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475',
253 => 'https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/',
254 => 'https://web.archive.org/web/20200525144635/https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/',
255 => 'https://openai.com/blog/ai-and-compute/',
256 => 'https://web.archive.org/web/20200617200602/https://openai.com/blog/ai-and-compute/',
257 => 'https://consumer.huawei.com/en/press/news/2017/ifa2017-kirin970/',
258 => 'https://doi.org/10.1145%2F3140659.3080246',
259 => 'https://www.datanami.com/2021/11/01/cerebras-hits-the-accelerator-for-deep-learning-workloads/',
260 => 'https://venturebeat.com/2021/04/20/cerebras-systems-launches-new-ai-supercomputing-processor-with-2-6-trillion-transistors/',
261 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7116757',
262 => 'https://ui.adsabs.harvard.edu/abs/2020Natur.587...72M',
263 => 'https://doi.org/10.1038%2Fs41586-020-2861-0',
264 => 'https://pubmed.ncbi.nlm.nih.gov/33149289',
265 => 'https://arxiv.org/abs/2002.00281',
266 => 'https://doi.org/10.1038%2Fs41586-020-03070-1',
267 => 'https://pubmed.ncbi.nlm.nih.gov/33408373',
268 => 'https://api.semanticscholar.org/CorpusID:211010976',
269 => 'https://doi.org/10.13140%2FRG.2.2.15418.90567',
270 => 'https://zenodo.org/record/891433',
271 => 'https://doi.org/10.1109%2Ftaslp.2014.2339736',
272 => 'https://api.semanticscholar.org/CorpusID:206602362',
273 => 'https://web.archive.org/web/20200922180719/https://zenodo.org/record/891433',
274 => 'https://api.semanticscholar.org/CorpusID:15641618',
275 => 'http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf',
276 => 'https://doi.org/10.1186%2Fs13636-015-0068-3',
277 => 'https://api.semanticscholar.org/CorpusID:217950236',
278 => 'https://web.archive.org/web/20200924085514/http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf',
279 => 'https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/',
280 => 'https://web.archive.org/web/20170608062106/https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/',
281 => 'https://arxiv.org/abs/1412.5567',
282 => 'http://yann.lecun.com/exdb/mnist/.',
283 => 'https://web.archive.org/web/20140113175237/http://yann.lecun.com/exdb/mnist/',
284 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.226.8219',
285 => 'https://doi.org/10.1016%2Fj.neunet.2012.02.023',
286 => 'https://pubmed.ncbi.nlm.nih.gov/22386783',
287 => 'https://arxiv.org/abs/1404.3840',
288 => 'http://www.technologyreview.com/news/533936/nvidia-demos-a-car-computer-trained-with-deep-learning/',
289 => 'https://doi.org/10.3390%2Farts6020005',
290 => 'https://doi.org/10.3390%2Farts6040018',
291 => 'https://arxiv.org/abs/1402.3722',
292 => 'http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf',
293 => 'https://web.archive.org/web/20140706040227/http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf',
294 => 'http://aclweb.org/anthology/P/P13/P13-1045.pdf',
295 => 'https://web.archive.org/web/20141127005912/http://www.aclweb.org/anthology/P/P13/P13-1045.pdf',
296 => 'http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf',
297 => 'https://web.archive.org/web/20161228100300/http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf',
298 => 'https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/',
299 => 'https://web.archive.org/web/20171027050418/https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/',
300 => 'https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/',
301 => 'https://web.archive.org/web/20171027050414/https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/',
302 => 'https://doi.org/10.1109%2Ftaslp.2014.2383614',
303 => 'https://api.semanticscholar.org/CorpusID:1317136',
304 => 'https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/',
305 => 'https://web.archive.org/web/20171027050403/https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/',
306 => 'https://doi.org/10.1002%2Fdac.3259',
307 => 'https://api.semanticscholar.org/CorpusID:40745740',
308 => 'https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/',
309 => 'https://web.archive.org/web/20170313184253/https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/',
310 => 'https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/',
311 => 'https://web.archive.org/web/20170407071226/https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/',
312 => 'https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html',
313 => 'https://web.archive.org/web/20170710183732/https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html',
314 => 'https://arxiv.org/abs/1609.08144',
315 => 'https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/',
316 => 'https://web.archive.org/web/20201108101324/https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/',
317 => 'https://web.archive.org/web/20170329125916/http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf',
318 => 'http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf',
319 => 'https://doi.org/10.1038%2Fnrd4090',
320 => 'https://pubmed.ncbi.nlm.nih.gov/23903212',
321 => 'https://api.semanticscholar.org/CorpusID:20246434',
322 => 'https://doi.org/10.1016%2Fj.drudis.2014.12.014',
323 => 'https://pubmed.ncbi.nlm.nih.gov/25582842',
324 => 'https://arxiv.org/abs/1510.02855',
325 => 'https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/',
326 => 'https://web.archive.org/web/20151020040115/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/',
327 => 'http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/',
328 => 'https://web.archive.org/web/20151224104721/http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/',
329 => 'https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/',
330 => 'https://web.archive.org/web/20151225162547/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/',
331 => 'https://arxiv.org/abs/1704.01212',
332 => 'https://doi.org/10.1038%2Fs41587-019-0224-x',
333 => 'https://pubmed.ncbi.nlm.nih.gov/31477924',
334 => 'https://api.semanticscholar.org/CorpusID:201716327',
335 => 'https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/',
336 => 'https://web.archive.org/web/20200430143244/https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/',
337 => 'https://arxiv.org/abs/1504.01840',
338 => 'http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf',
339 => 'https://web.archive.org/web/20170516185259/http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf',
340 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6555124',
341 => 'https://doi.org/10.2196%2F12957',
342 => 'https://pubmed.ncbi.nlm.nih.gov/31127715',
343 => 'https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/',
344 => 'https://web.archive.org/web/20180125134534/https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/',
345 => 'http://dl.acm.org/citation.cfm?id=2649442',
346 => 'https://doi.org/10.1145%2F2649387.2649442',
347 => 'https://hdl.handle.net/11311%2F964622',
348 => 'https://api.semanticscholar.org/CorpusID:207217210',
349 => 'https://web.archive.org/web/20210509123140/https://dl.acm.org/doi/10.1145/2649387.2649442',
350 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5116102',
351 => 'https://doi.org/10.2196%2Fmhealth.6562',
352 => 'https://pubmed.ncbi.nlm.nih.gov/27815231',
353 => 'https://api.semanticscholar.org/CorpusID:3821594',
354 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5391725',
355 => 'https://doi.org/10.1093%2Fjamia%2Focw112',
356 => 'https://www.worldcat.org/issn/1067-5027',
357 => 'https://pubmed.ncbi.nlm.nih.gov/27521897',
358 => 'https://arxiv.org/abs/1702.05747',
359 => 'https://ui.adsabs.harvard.edu/abs/2017arXiv170205747L',
360 => 'https://doi.org/10.1016%2Fj.media.2017.07.005',
361 => 'https://pubmed.ncbi.nlm.nih.gov/28778026',
362 => 'https://api.semanticscholar.org/CorpusID:2088679',
363 => 'http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-326160',
364 => 'https://doi.org/10.1109%2FICCVW.2017.18',
365 => 'https://api.semanticscholar.org/CorpusID:4728736',
366 => 'https://web.archive.org/web/20210509123157/https://d1bxh8uas1mnw7.cloudfront.net/assets/embed.js',
367 => 'https://ieeexplore.ieee.org/document/9130662',
368 => 'https://doi.org/10.1109%2FACCESS.2020.3006362',
369 => 'https://www.worldcat.org/issn/2169-3536',
370 => 'https://api.semanticscholar.org/CorpusID:220733699',
371 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8997449',
372 => 'https://doi.org/10.3390%2Fcancers14071819',
373 => 'https://www.worldcat.org/issn/2072-6694',
374 => 'https://pubmed.ncbi.nlm.nih.gov/35406591',
375 => 'https://doi.org/10.1109%2FCSCITA.2017.8066548',
376 => 'https://api.semanticscholar.org/CorpusID:35350962',
377 => 'https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/',
378 => 'https://web.archive.org/web/20191011162814/https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/',
379 => 'http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf',
380 => 'https://web.archive.org/web/20180102013217/http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf',
381 => 'https://doi.org/10.1016%2Fj.knosys.2019.105048',
382 => 'https://api.semanticscholar.org/CorpusID:204092079',
383 => 'https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/',
384 => 'https://web.archive.org/web/20181116082711/https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/',
385 => 'https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php',
386 => 'https://web.archive.org/web/20180828035608/https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php',
387 => 'https://www.sciencedirect.com/science/article/pii/S0021999118307125',
388 => 'https://ui.adsabs.harvard.edu/abs/2019JCoPh.378..686R',
389 => 'https://doi.org/10.1016%2Fj.jcp.2018.10.045',
390 => 'https://www.worldcat.org/issn/0021-9991',
391 => 'https://www.osti.gov/biblio/1595805',
392 => 'https://api.semanticscholar.org/CorpusID:57379996',
393 => 'https://www.sciencedirect.com/science/article/pii/S0045782519306814',
394 => 'https://ui.adsabs.harvard.edu/abs/2020CMAME.360k2789M',
395 => 'https://doi.org/10.1016%2Fj.cma.2019.112789',
396 => 'https://www.worldcat.org/issn/0045-7825',
397 => 'https://api.semanticscholar.org/CorpusID:212755458',
398 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7219083',
399 => 'https://ui.adsabs.harvard.edu/abs/2020Sci...367.1026R',
400 => 'https://doi.org/10.1126%2Fscience.aaw4741',
401 => 'https://pubmed.ncbi.nlm.nih.gov/32001523',
402 => 'https://ieeexplore.ieee.org/document/9415140',
403 => 'https://arxiv.org/abs/2008.11625',
404 => 'https://doi.org/10.1109%2FTCI.2021.3075349',
405 => 'https://www.worldcat.org/issn/2333-9403',
406 => 'https://api.semanticscholar.org/CorpusID:235340737',
407 => 'https://ieeexplore.ieee.org/abstract/document/9144249?casa_token=NLVQbGR8h-8AAAAA:rz3cvMmiplVSRCGTDouUZkIeribqQKaMtK9t0MxlwKBMRwaTJL1onKOwkEk0qW_fFJCfd2ejOg',
408 => 'https://arxiv.org/abs/2006.14395',
409 => 'https://doi.org/10.1109%2FTUFFC.2020.3010186',
410 => 'https://www.worldcat.org/issn/1525-8955',
411 => 'https://pubmed.ncbi.nlm.nih.gov/32746211',
412 => 'https://api.semanticscholar.org/CorpusID:220055785',
413 => 'https://doi.org/10.1162%2F08997660260293319',
414 => 'https://pubmed.ncbi.nlm.nih.gov/12396572',
415 => 'https://api.semanticscholar.org/CorpusID:1119517',
416 => 'https://books.google.com/books?id=vELaRu_MrwoC',
417 => 'https://doi.org/10.1016%2F0893-6080(96)00033-0',
418 => 'https://pubmed.ncbi.nlm.nih.gov/12662587',
419 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.41.7854',
420 => 'https://doi.org/10.1017%2Fs0140525x97001581',
421 => 'https://pubmed.ncbi.nlm.nih.gov/10097006',
422 => 'https://api.semanticscholar.org/CorpusID:5818342',
423 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC51674',
424 => 'https://ui.adsabs.harvard.edu/abs/1991PNAS...88.4433M',
425 => 'https://doi.org/10.1073%2Fpnas.88.10.4433',
426 => 'https://www.worldcat.org/issn/0027-8424',
427 => 'https://pubmed.ncbi.nlm.nih.gov/1903542',
428 => 'https://doi.org/10.1162%2Fneco.1996.8.5.895',
429 => 'https://api.semanticscholar.org/CorpusID:2376781',
430 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4943066',
431 => 'https://doi.org/10.3389%2Ffncom.2016.00073',
432 => 'https://www.worldcat.org/issn/1662-5188',
433 => 'https://pubmed.ncbi.nlm.nih.gov/27468262',
434 => 'https://api.semanticscholar.org/CorpusID:9868901',
435 => 'https://doi.org/10.1038%2Fs41562-017-0186-2',
436 => 'https://www.worldcat.org/issn/2397-3374',
437 => 'https://pubmed.ncbi.nlm.nih.gov/31024135',
438 => 'https://api.semanticscholar.org/CorpusID:24504018',
439 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3207943',
440 => 'https://ui.adsabs.harvard.edu/abs/2011PLSCB...7E2211B',
441 => 'https://doi.org/10.1371%2Fjournal.pcbi.1002211',
442 => 'https://www.worldcat.org/issn/1553-7358',
443 => 'https://pubmed.ncbi.nlm.nih.gov/22096452',
444 => 'https://api.semanticscholar.org/CorpusID:7504633',
445 => 'https://doi.org/10.1016%2Fs0896-6273(00)81098-3',
446 => 'https://www.worldcat.org/issn/0896-6273',
447 => 'https://pubmed.ncbi.nlm.nih.gov/10069343',
448 => 'https://api.semanticscholar.org/CorpusID:14663106',
449 => 'https://doi.org/10.1016%2Fj.conb.2004.07.007',
450 => 'https://www.worldcat.org/issn/0959-4388',
451 => 'https://pubmed.ncbi.nlm.nih.gov/15321069',
452 => 'https://api.semanticscholar.org/CorpusID:16560320',
453 => 'https://doi.org/10.1038%2Fnn.4244',
454 => 'https://www.worldcat.org/issn/1546-1726',
455 => 'https://pubmed.ncbi.nlm.nih.gov/26906502',
456 => 'https://api.semanticscholar.org/CorpusID:16970545',
457 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5784047',
458 => 'https://doi.org/10.1098%2Frstb.2017.0043',
459 => 'https://www.worldcat.org/issn/0962-8436',
460 => 'https://pubmed.ncbi.nlm.nih.gov/29292348',
461 => 'https://api.semanticscholar.org/CorpusID:39281431',
462 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6605414',
463 => 'https://arxiv.org/abs/1411.6422',
464 => 'https://doi.org/10.1523%2Fjneurosci.5023-14.2015',
465 => 'https://pubmed.ncbi.nlm.nih.gov/26157000',
466 => 'https://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/',
467 => 'https://web.archive.org/web/20140328071226/http://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/',
468 => 'http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234',
469 => 'https://ui.adsabs.harvard.edu/abs/2016Natur.529..445G',
470 => 'https://doi.org/10.1038%2F529445a',
471 => 'https://pubmed.ncbi.nlm.nih.gov/26819021',
472 => 'https://api.semanticscholar.org/CorpusID:4460235',
473 => 'https://web.archive.org/web/20190502200837/http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234',
474 => 'https://www.worldcat.org/issn/0028-0836',
475 => 'https://web.archive.org/web/20160201140636/http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/',
476 => 'http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/',
477 => 'https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html',
478 => 'https://web.archive.org/web/20190707161547/https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html',
479 => 'https://doi.org/10.1109%2Fdevlrn.2008.4640845',
480 => 'https://api.semanticscholar.org/CorpusID:5613334',
481 => 'https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner',
482 => 'https://web.archive.org/web/20180828001727/https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner',
483 => 'https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1',
484 => 'https://web.archive.org/web/20181012035405/https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1',
485 => 'https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/',
486 => 'https://web.archive.org/web/20191104033107/https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/',
487 => 'https://www.newyorker.com/',
488 => 'https://web.archive.org/web/20091127184826/http://www.newyorker.com/',
489 => 'http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html',
490 => 'https://web.archive.org/web/20150703064823/http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html',
491 => 'https://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep',
492 => 'https://web.archive.org/web/20150619200845/http://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep',
493 => 'http://goertzel.org/DeepLearning_v1.pdf',
494 => 'https://web.archive.org/web/20150513053107/http://goertzel.org/DeepLearning_v1.pdf',
495 => 'https://arxiv.org/abs/1412.1897',
496 => 'https://arxiv.org/abs/1312.6199',
497 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.681.2190',
498 => 'https://doi.org/10.1561%2F0600000018',
499 => 'https://web.archive.org/web/20171230010335/http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/',
500 => 'http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/',
501 => 'https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425',
502 => 'https://web.archive.org/web/20191011162231/https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425',
503 => 'https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/',
504 => 'https://web.archive.org/web/20191011162230/https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/',
505 => 'https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change',
506 => 'https://web.archive.org/web/20171011233017/https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change/',
507 => 'https://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784',
508 => 'https://doi.org/10.1038%2Fnature.2017.22784',
509 => 'https://web.archive.org/web/20171010011017/http://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784',
510 => 'https://hal.science/hal-03029735',
511 => 'https://depositonce.tu-berlin.de/handle/11303/12510',
512 => 'https://doi.org/10.1177%2F1461444819885334',
513 => 'https://www.worldcat.org/issn/1461-4448',
514 => 'https://api.semanticscholar.org/CorpusID:209363848',
515 => 'https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/',
516 => 'https://www.worldcat.org/issn/1059-1028',
517 => 'https://web.archive.org/web/20190810223940/https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/',
518 => '//www.google.com/search?as_eq=wikipedia&q=%22Deep+learning%22',
519 => '//www.google.com/search?tbm=nws&q=%22Deep+learning%22+-wikipedia&tbs=ar:1',
520 => '//www.google.com/search?&q=%22Deep+learning%22&tbs=bkt:s&tbm=bks',
521 => '//www.google.com/search?tbs=bks:1&q=%22Deep+learning%22+-wikipedia',
522 => '//scholar.google.com/scholar?q=%22Deep+learning%22',
523 => 'https://www.jstor.org/action/doBasicSearch?Query=%22Deep+learning%22&acc=on&wc=on',
524 => 'http://www.deeplearningbook.org',
525 => 'https://web.archive.org/web/20160416111010/http://www.deeplearningbook.org/',
526 => 'https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt'
] |
Links in the page, before the edit (old_links) | [
0 => 'Ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf',
1 => 'ftp://ftp.idsia.ch/pub/juergen/habilitation.pdf',
2 => 'http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf',
3 => 'http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf',
4 => 'http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf',
5 => 'http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf',
6 => 'http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width',
7 => 'http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html',
8 => 'http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf',
9 => 'http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/',
10 => 'http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3',
11 => 'http://fortune.com/ai-artificial-intelligence-deep-machine-learning/',
12 => '//scholar.google.com/scholar?q=%22Deep+learning%22',
13 => '//www.google.com/search?&q=%22Deep+learning%22&tbs=bkt:s&tbm=bks',
14 => '//www.google.com/search?as_eq=wikipedia&q=%22Deep+learning%22',
15 => '//www.google.com/search?tbm=nws&q=%22Deep+learning%22+-wikipedia&tbs=ar:1',
16 => '//www.google.com/search?tbs=bks:1&q=%22Deep+learning%22+-wikipedia',
17 => 'http://yann.lecun.com/exdb/mnist/.',
18 => 'http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf',
19 => 'http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234',
20 => 'http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf',
21 => 'http://www.technologyreview.com/news/533936/nvidia-demos-a-car-computer-trained-with-deep-learning/',
22 => 'http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/',
23 => 'http://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf',
24 => 'http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf',
25 => 'http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf',
26 => 'http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf',
27 => 'http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf',
28 => 'http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf',
29 => 'http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf',
30 => 'http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf',
31 => 'http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf',
32 => 'http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html',
33 => 'http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf',
34 => 'http://gmdh.net/articles/history/polynomial.pdf',
35 => 'http://aclweb.org/anthology/P/P13/P13-1045.pdf',
36 => 'http://dl.acm.org/citation.cfm?id=1895720',
37 => 'http://dl.acm.org/citation.cfm?id=2649442',
38 => '//arxiv.org/archive/cs.CL',
39 => '//arxiv.org/archive/cs.CV',
40 => '//arxiv.org/archive/cs.LG',
41 => 'http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions',
42 => 'http://www.deeplearningbook.org',
43 => 'http://www.escholarship.org/uc/item/6ch40821',
44 => 'http://goertzel.org/DeepLearning_v1.pdf',
45 => 'http://ijcai.org/papers11/Papers/IJCAI11-210.pdf',
46 => 'http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/',
47 => 'http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf',
48 => 'http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf',
49 => 'http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-326160',
50 => 'http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/',
51 => 'http://elartu.tntu.edu.ua/handle/lib/30719',
52 => 'http://elartu.tntu.edu.ua/handle/lib/38369',
53 => 'http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf',
54 => 'http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html',
55 => 'https://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection',
56 => 'https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf',
57 => 'https://indico.cern.ch/event/510372/',
58 => 'https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/',
59 => 'https://www.datanami.com/2021/11/01/cerebras-hits-the-accelerator-for-deep-learning-workloads/',
60 => 'https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not',
61 => 'https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/',
62 => 'https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425',
63 => 'https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/',
64 => 'https://books.google.com/books?id=9CqQDwAAQBAJ&pg=PA15',
65 => 'https://books.google.com/books?id=NWOcMVA64aAC',
66 => 'https://books.google.com/books?id=NZP6AQAAQBAJ',
67 => 'https://books.google.com/books?id=Otk32Y3QkxQC&pg=PA48',
68 => 'https://books.google.com/books?id=bX4pAQAAMAAJ',
69 => 'https://books.google.com/books?id=rGFgAAAAMAAJ',
70 => 'https://books.google.com/books?id=rUBTBQAAQBAJ',
71 => 'https://books.google.com/books?id=vELaRu_MrwoC',
72 => 'https://drive.google.com/file/d/0B65v6Wo67Tk5Ml9qeW5nQ3poVTQ/view?usp=sharing',
73 => 'https://drive.google.com/file/d/0B65v6Wo67Tk5ODRzZmhSR29VeDg/view?usp=sharing',
74 => 'https://drive.google.com/file/d/0B65v6Wo67Tk5cm5DTlNGd0NPUmM/view?usp=sharing',
75 => 'https://drive.google.com/file/d/1nN_5odSG_QVae54EsQN_qSz-0ZsX6wA0/view?usp=sharing',
76 => 'https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html',
77 => 'https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf',
78 => 'https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf',
79 => 'https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner',
80 => 'https://consumer.huawei.com/en/press/news/2017/ifa2017-kirin970/',
81 => 'https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475',
82 => 'https://kaggle.com/c/MerckActivity',
83 => 'https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1',
84 => 'https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/',
85 => 'https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/',
86 => 'https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/',
87 => 'https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/',
88 => 'https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks',
89 => 'https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/',
90 => 'https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/',
91 => 'https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/',
92 => 'https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/',
93 => 'https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf',
94 => 'https://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784',
95 => 'https://www.newyorker.com/',
96 => 'https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html',
97 => 'https://openai.com/blog/ai-and-compute/',
98 => 'https://www.sciencedirect.com/science/article/pii/S0021999118307125',
99 => 'https://www.sciencedirect.com/science/article/pii/S0045782519306814',
100 => 'https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change',
101 => 'https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing',
102 => 'https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/',
103 => 'https://www.technologyreview.com/s/513696/deep-learning/',
104 => 'https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/',
105 => 'https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/',
106 => 'https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/',
107 => 'https://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep',
108 => 'https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/',
109 => 'https://venturebeat.com/2021/04/20/cerebras-systems-launches-new-ai-supercomputing-processor-with-2-6-trillion-transistors/',
110 => 'https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/',
111 => 'https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/',
112 => 'https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/',
113 => 'https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/',
114 => 'https://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/',
115 => 'https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/',
116 => 'https://depositonce.tu-berlin.de/handle/11303/12510',
117 => 'https://mediatum.ub.tum.de/doc/1289941/file.pdf',
118 => 'https://www.academia.edu/40135801',
119 => 'https://ui.adsabs.harvard.edu/abs/1990ApOpt..29.4790Z',
120 => 'https://ui.adsabs.harvard.edu/abs/1991ApOpt..30.4211Z',
121 => 'https://ui.adsabs.harvard.edu/abs/1991PNAS...88.4433M',
122 => 'https://ui.adsabs.harvard.edu/abs/1994MedPh..21..517Z',
123 => 'https://ui.adsabs.harvard.edu/abs/1994PaReL..15..807D',
124 => 'https://ui.adsabs.harvard.edu/abs/1995Sci...268.1158H',
125 => 'https://ui.adsabs.harvard.edu/abs/2004PatRe..37.1311O',
126 => 'https://ui.adsabs.harvard.edu/abs/2009ISPM...26...75B',
127 => 'https://ui.adsabs.harvard.edu/abs/2009SchpJ...4.5947H',
128 => 'https://ui.adsabs.harvard.edu/abs/2011PLSCB...7E2211B',
129 => 'https://ui.adsabs.harvard.edu/abs/2012ISPM...29...82H',
130 => 'https://ui.adsabs.harvard.edu/abs/2014arXiv1409.3215S',
131 => 'https://ui.adsabs.harvard.edu/abs/2015Natur.521..436L',
132 => 'https://ui.adsabs.harvard.edu/abs/2016Natur.529..445G',
133 => 'https://ui.adsabs.harvard.edu/abs/2016Natur.529..484S',
134 => 'https://ui.adsabs.harvard.edu/abs/2016arXiv160603813M',
135 => 'https://ui.adsabs.harvard.edu/abs/2017NatCo...8..138O',
136 => 'https://ui.adsabs.harvard.edu/abs/2017arXiv170205747L',
137 => 'https://ui.adsabs.harvard.edu/abs/2017arXiv170207908V',
138 => 'https://ui.adsabs.harvard.edu/abs/2019JCoPh.378..686R',
139 => 'https://ui.adsabs.harvard.edu/abs/2020CMAME.360k2789M',
140 => 'https://ui.adsabs.harvard.edu/abs/2020Natur.587...72M',
141 => 'https://ui.adsabs.harvard.edu/abs/2020Sci...367.1026R',
142 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.154.372',
143 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.226.8219',
144 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.41.7854',
145 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.681.2190',
146 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.701.9550',
147 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.75.6306',
148 => 'https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.752.9151',
149 => 'https://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf',
150 => 'https://hal.inria.fr/inria-00112631/document',
151 => 'https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/',
152 => 'https://pubmed.ncbi.nlm.nih.gov/10069343',
153 => 'https://pubmed.ncbi.nlm.nih.gov/10097006',
154 => 'https://pubmed.ncbi.nlm.nih.gov/12396572',
155 => 'https://pubmed.ncbi.nlm.nih.gov/12662587',
156 => 'https://pubmed.ncbi.nlm.nih.gov/15321069',
157 => 'https://pubmed.ncbi.nlm.nih.gov/16764513',
158 => 'https://pubmed.ncbi.nlm.nih.gov/17921042',
159 => 'https://pubmed.ncbi.nlm.nih.gov/18249962',
160 => 'https://pubmed.ncbi.nlm.nih.gov/1903542',
161 => 'https://pubmed.ncbi.nlm.nih.gov/20577468',
162 => 'https://pubmed.ncbi.nlm.nih.gov/20706526',
163 => 'https://pubmed.ncbi.nlm.nih.gov/20858131',
164 => 'https://pubmed.ncbi.nlm.nih.gov/22096452',
165 => 'https://pubmed.ncbi.nlm.nih.gov/22386783',
166 => 'https://pubmed.ncbi.nlm.nih.gov/23787338',
167 => 'https://pubmed.ncbi.nlm.nih.gov/23903212',
168 => 'https://pubmed.ncbi.nlm.nih.gov/24579167',
169 => 'https://pubmed.ncbi.nlm.nih.gov/25462637',
170 => 'https://pubmed.ncbi.nlm.nih.gov/25582842',
171 => 'https://pubmed.ncbi.nlm.nih.gov/26017442',
172 => 'https://pubmed.ncbi.nlm.nih.gov/26157000',
173 => 'https://pubmed.ncbi.nlm.nih.gov/26819021',
174 => 'https://pubmed.ncbi.nlm.nih.gov/26819042',
175 => 'https://pubmed.ncbi.nlm.nih.gov/26906502',
176 => 'https://pubmed.ncbi.nlm.nih.gov/27468262',
177 => 'https://pubmed.ncbi.nlm.nih.gov/27521897',
178 => 'https://pubmed.ncbi.nlm.nih.gov/27683554',
179 => 'https://pubmed.ncbi.nlm.nih.gov/27815231',
180 => 'https://pubmed.ncbi.nlm.nih.gov/28743932',
181 => 'https://pubmed.ncbi.nlm.nih.gov/28778026',
182 => 'https://pubmed.ncbi.nlm.nih.gov/29292348',
183 => 'https://pubmed.ncbi.nlm.nih.gov/31024135',
184 => 'https://pubmed.ncbi.nlm.nih.gov/31127715',
185 => 'https://pubmed.ncbi.nlm.nih.gov/31477924',
186 => 'https://pubmed.ncbi.nlm.nih.gov/32001523',
187 => 'https://pubmed.ncbi.nlm.nih.gov/32746211',
188 => 'https://pubmed.ncbi.nlm.nih.gov/33149289',
189 => 'https://pubmed.ncbi.nlm.nih.gov/33408373',
190 => 'https://pubmed.ncbi.nlm.nih.gov/35406591',
191 => 'https://pubmed.ncbi.nlm.nih.gov/7370364',
192 => 'https://pubmed.ncbi.nlm.nih.gov/7761831',
193 => 'https://pubmed.ncbi.nlm.nih.gov/8058017',
194 => 'https://pubmed.ncbi.nlm.nih.gov/9377276',
195 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3207943',
196 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4943066',
197 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5021692',
198 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5116102',
199 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC51674',
200 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5391725',
201 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5527101',
202 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5784047',
203 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6555124',
204 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6605414',
205 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7116757',
206 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7219083',
207 => 'https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8997449',
208 => 'https://tripod.nih.gov/tox21/challenge/leaderboard.jsp',
209 => 'https://www.osti.gov/biblio/1595805',
210 => 'https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks',
211 => 'https://hdl.handle.net/10338.dmlcz%2F135496',
212 => 'https://hdl.handle.net/11311%2F964622',
213 => 'https://hdl.handle.net/1721.1%2F51891',
214 => 'https://openreview.net/pdf?id=SyProzZAW',
215 => 'https://www.researchgate.net/publication/220320057',
216 => 'https://www.researchgate.net/publication/221166159',
217 => 'https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems',
218 => 'https://www.researchgate.net/publication/266030526',
219 => 'https://www.researchgate.net/publication/35657389',
220 => 'https://www.researchgate.net/publication/41229141',
221 => 'https://dl.acm.org/citation.cfm?doid=3126908.3126912',
222 => 'https://web.archive.org/web/20091127184826/http://www.newyorker.com/',
223 => 'https://web.archive.org/web/20131011071435/http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3',
224 => 'https://web.archive.org/web/20140113175237/http://yann.lecun.com/exdb/mnist/',
225 => 'https://web.archive.org/web/20140328071226/http://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/',
226 => 'https://web.archive.org/web/20140706040227/http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf',
227 => 'https://web.archive.org/web/20140929094040/http://ijcai.org/papers11/Papers/IJCAI11-210.pdf',
228 => 'https://web.archive.org/web/20141127005912/http://www.aclweb.org/anthology/P/P13/P13-1045.pdf',
229 => 'https://web.archive.org/web/20150228225709/http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html',
230 => 'https://web.archive.org/web/20150306075401/http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf',
231 => 'https://web.archive.org/web/20150513053107/http://goertzel.org/DeepLearning_v1.pdf',
232 => 'https://web.archive.org/web/20150619200845/http://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep',
233 => 'https://web.archive.org/web/20150703064823/http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html',
234 => 'https://web.archive.org/web/20150908025122/https://tripod.nih.gov/tox21/challenge/leaderboard.jsp',
235 => 'https://web.archive.org/web/20151010204407/http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf',
236 => 'https://web.archive.org/web/20151020040115/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/',
237 => 'https://web.archive.org/web/20151223164129/http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf',
238 => 'https://web.archive.org/web/20151224104721/http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/',
239 => 'https://web.archive.org/web/20151225162547/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/',
240 => 'https://web.archive.org/web/20160201033801/http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/',
241 => 'https://web.archive.org/web/20160201140636/http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/',
242 => 'https://web.archive.org/web/20160304084250/http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf',
243 => 'https://web.archive.org/web/20160309191532/http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html',
244 => 'https://web.archive.org/web/20160314152112/http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf',
245 => 'https://web.archive.org/web/20160416111010/http://www.deeplearningbook.org/',
246 => 'https://web.archive.org/web/20160419054654/https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems',
247 => 'https://web.archive.org/web/20160423021403/https://indico.cern.ch/event/510372/',
248 => 'https://web.archive.org/web/20161228100300/http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf',
249 => 'https://web.archive.org/web/20161231203934/https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not',
250 => 'https://web.archive.org/web/20170110123024/http://www.cs.toronto.edu/~kriz/imagenet_classification_with_deep_convolutional.pdf',
251 => 'https://web.archive.org/web/20170111005101/http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf',
252 => 'https://web.archive.org/web/20170313184253/https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/',
253 => 'https://web.archive.org/web/20170329125916/http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf',
254 => 'https://web.archive.org/web/20170407071226/https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/',
255 => 'https://web.archive.org/web/20170430142049/http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions',
256 => 'https://web.archive.org/web/20170516181940/http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf',
257 => 'https://web.archive.org/web/20170516185259/http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf',
258 => 'https://web.archive.org/web/20170608062106/https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/',
259 => 'https://web.archive.org/web/20170629172111/http://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection',
260 => 'https://web.archive.org/web/20170710183732/https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html',
261 => 'https://web.archive.org/web/20170721211929/http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf',
262 => 'https://web.archive.org/web/20170809081713/http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf',
263 => 'https://web.archive.org/web/20170812140509/http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf',
264 => 'https://web.archive.org/web/20170829230621/http://www.gmdh.net/articles/history/polynomial.pdf',
265 => 'https://web.archive.org/web/20170926190732/https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing',
266 => 'https://web.archive.org/web/20170926190920/https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf',
267 => 'https://web.archive.org/web/20171010011017/http://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784',
268 => 'https://web.archive.org/web/20171011233017/https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change/',
269 => 'https://web.archive.org/web/20171012044053/https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/',
270 => 'https://web.archive.org/web/20171012095148/https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/',
271 => 'https://web.archive.org/web/20171012095522/https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks/',
272 => 'https://web.archive.org/web/20171027050403/https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/',
273 => 'https://web.archive.org/web/20171027050414/https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/',
274 => 'https://web.archive.org/web/20171027050418/https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/',
275 => 'https://web.archive.org/web/20171201032606/https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation',
276 => 'https://web.archive.org/web/20171230010335/http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/',
277 => 'https://web.archive.org/web/20180102013217/http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf',
278 => 'https://web.archive.org/web/20180125134534/https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/',
279 => 'https://web.archive.org/web/20180316084821/https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/',
280 => 'https://web.archive.org/web/20180414031925/http://fortune.com/ai-artificial-intelligence-deep-machine-learning/',
281 => 'https://web.archive.org/web/20180424203806/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf',
282 => 'https://web.archive.org/web/20180522112408/http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf',
283 => 'https://web.archive.org/web/20180617065807/https://techcrunch.com/2017/05/24/alphago-beats-planets-best-human-go-player-ke-jie/amp/',
284 => 'https://web.archive.org/web/20180828001727/https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner',
285 => 'https://web.archive.org/web/20180828035608/https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php',
286 => 'https://web.archive.org/web/20181012035405/https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1',
287 => 'https://web.archive.org/web/20181116082711/https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/',
288 => 'https://web.archive.org/web/20181118122850/http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf',
289 => 'https://web.archive.org/web/20181118164457/https://mediatum.ub.tum.de/doc/1289941/file.pdf',
290 => 'https://web.archive.org/web/20190213005539/http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width',
291 => 'https://web.archive.org/web/20190331092832/https://www.technologyreview.com/s/513696/deep-learning/',
292 => 'https://web.archive.org/web/20190502200837/http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234',
293 => 'https://web.archive.org/web/20190707161547/https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html',
294 => 'https://web.archive.org/web/20190810223940/https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/',
295 => 'https://web.archive.org/web/20191011162230/https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/',
296 => 'https://web.archive.org/web/20191011162231/https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425',
297 => 'https://web.archive.org/web/20191011162814/https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/',
298 => 'https://web.archive.org/web/20191020195638/http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf',
299 => 'https://web.archive.org/web/20191104033107/https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/',
300 => 'https://web.archive.org/web/20200126045722/http://elartu.tntu.edu.ua/handle/lib/30719',
301 => 'https://web.archive.org/web/20200430143244/https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/',
302 => 'https://web.archive.org/web/20200518193413/https://hal.inria.fr/inria-00112631/document',
303 => 'https://web.archive.org/web/20200525144635/https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/',
304 => 'https://web.archive.org/web/20200610094310/https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475',
305 => 'https://web.archive.org/web/20200617200602/https://openai.com/blog/ai-and-compute/',
306 => 'https://web.archive.org/web/20200716190808/https://www.kaggle.com/c/MerckActivity',
307 => 'https://web.archive.org/web/20200729133850/https://escholarship.org/uc/item/6ch40821',
308 => 'https://web.archive.org/web/20200922180719/https://zenodo.org/record/891433',
309 => 'https://web.archive.org/web/20200924085514/http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf',
310 => 'https://web.archive.org/web/20201102151103/https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks',
311 => 'https://web.archive.org/web/20201108101324/https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/',
312 => 'https://web.archive.org/web/20201125202428/https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/',
313 => 'https://web.archive.org/web/20210107183647/https://openreview.net/pdf?id=SyProzZAW',
314 => 'https://web.archive.org/web/20210427001446/https://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf',
315 => 'https://web.archive.org/web/20210509123113/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf',
316 => 'https://web.archive.org/web/20210509123120/https://www.academia.edu/40135801/A_Survey_of_Techniques_for_Optimizing_Deep_Learning_on_GPUs',
317 => 'https://web.archive.org/web/20210509123131/https://www.researchgate.net/publication/41229141_Artificial_neural_networks_and_their_application_to_sequence_recognition',
318 => 'https://web.archive.org/web/20210509123135/https://dl.acm.org/doi/10.5555/1895550.1895720',
319 => 'https://web.archive.org/web/20210509123139/ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf',
320 => 'https://web.archive.org/web/20210509123140/https://dl.acm.org/doi/10.1145/2649387.2649442',
321 => 'https://web.archive.org/web/20210509123145/https://papers.nips.cc/paper/2014/file/a14ac55a4f27472c5d894ec1c3c743d2-Paper.pdf',
322 => 'https://web.archive.org/web/20210509123147/https://www.researchgate.net/publication/220320057_Learning_Precise_Timing_with_LSTM_Recurrent_Networks',
323 => 'https://web.archive.org/web/20210509123157/https://d1bxh8uas1mnw7.cloudfront.net/assets/embed.js',
324 => 'https://web.archive.org/web/20210509123211/https://www.researchgate.net/publication/221166159_A_brief_introduction_to_Weightless_Neural_Systems',
325 => 'https://web.archive.org/web/20210509123218/https://www.researchgate.net/publication/266030526_Acoustic_Modeling_with_Deep_Neural_Networks_Using_Raw_Time_Signal_for_LVCSR',
326 => 'https://arxiv.org/abs/1003.0358',
327 => 'https://arxiv.org/abs/1202.2745',
328 => 'https://arxiv.org/abs/1206.5533',
329 => 'https://arxiv.org/abs/1206.5538',
330 => 'https://arxiv.org/abs/1212.0901',
331 => 'https://arxiv.org/abs/1312.6199',
332 => 'https://arxiv.org/abs/1402.3722',
333 => 'https://arxiv.org/abs/1404.3840',
334 => 'https://arxiv.org/abs/1404.7828',
335 => 'https://arxiv.org/abs/1409.3215',
336 => 'https://arxiv.org/abs/1410.4281',
337 => 'https://arxiv.org/abs/1411.2539',
338 => 'https://arxiv.org/abs/1411.4555',
339 => 'https://arxiv.org/abs/1411.4952',
340 => 'https://arxiv.org/abs/1411.6422',
341 => 'https://arxiv.org/abs/1412.1897',
342 => 'https://arxiv.org/abs/1412.5567',
343 => 'https://arxiv.org/abs/1502.04156',
344 => 'https://arxiv.org/abs/1504.01840',
345 => 'https://arxiv.org/abs/1505.03654',
346 => 'https://arxiv.org/abs/1510.02855',
347 => 'https://arxiv.org/abs/1512.00103',
348 => 'https://arxiv.org/abs/1602.02410',
349 => 'https://arxiv.org/abs/1606.03813',
350 => 'https://arxiv.org/abs/1609.08144',
351 => 'https://arxiv.org/abs/1702.05747',
352 => 'https://arxiv.org/abs/1702.07908',
353 => 'https://arxiv.org/abs/1703.09039',
354 => 'https://arxiv.org/abs/1704.01212',
355 => 'https://arxiv.org/abs/2002.00281',
356 => 'https://arxiv.org/abs/2006.14395',
357 => 'https://arxiv.org/abs/2008.11625',
358 => 'https://arxiv.org/abs/2102.04029',
359 => 'https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation',
360 => 'https://doi.org/10.1002%2Fdac.3259',
361 => 'https://doi.org/10.1007%2F978-3-642-40763-5_51',
362 => 'https://doi.org/10.1007%2Fb11963',
363 => 'https://doi.org/10.1007%2Fbf00344251',
364 => 'https://doi.org/10.1007%2Fbf02551274',
365 => 'https://doi.org/10.1007%2Fs11227-017-1994-x',
366 => 'https://doi.org/10.1007%2Fs13218-012-0198-z',
367 => 'https://doi.org/10.1016%2F0167-8655(94)90009-4',
368 => 'https://doi.org/10.1016%2F0893-6080(91)90009-t',
369 => 'https://doi.org/10.1016%2F0893-6080(94)90027-2',
370 => 'https://doi.org/10.1016%2F0893-6080(96)00033-0',
371 => 'https://doi.org/10.1016%2FS0167-6393(99)00080-1',
372 => 'https://doi.org/10.1016%2Fj.acha.2015.12.005',
373 => 'https://doi.org/10.1016%2Fj.cma.2019.112789',
374 => 'https://doi.org/10.1016%2Fj.conb.2004.07.007',
375 => 'https://doi.org/10.1016%2Fj.drudis.2014.12.014',
376 => 'https://doi.org/10.1016%2Fj.jcp.2018.10.045',
377 => 'https://doi.org/10.1016%2Fj.knosys.2019.105048',
378 => 'https://doi.org/10.1016%2Fj.media.2017.07.005',
379 => 'https://doi.org/10.1016%2Fj.neunet.2012.02.023',
380 => 'https://doi.org/10.1016%2Fj.neunet.2014.09.003',
381 => 'https://doi.org/10.1016%2Fj.patcog.2004.01.013',
382 => 'https://doi.org/10.1016%2Fj.tics.2007.09.004',
383 => 'https://doi.org/10.1016%2Fs0167-6393(99)00077-1',
384 => 'https://doi.org/10.1016%2Fs0896-6273(00)81098-3',
385 => 'https://doi.org/10.1017%2Fs0140525x97001581',
386 => 'https://doi.org/10.1038%2F529445a',
387 => 'https://doi.org/10.1038%2Fnature.2017.22784',
388 => 'https://doi.org/10.1038%2Fnature14539',
389 => 'https://doi.org/10.1038%2Fnature16961',
390 => 'https://doi.org/10.1038%2Fnn.4244',
391 => 'https://doi.org/10.1038%2Fnrd4090',
392 => 'https://doi.org/10.1038%2Fs41467-017-00181-8',
393 => 'https://doi.org/10.1038%2Fs41562-017-0186-2',
394 => 'https://doi.org/10.1038%2Fs41586-020-03070-1',
395 => 'https://doi.org/10.1038%2Fs41586-020-2861-0',
396 => 'https://doi.org/10.1038%2Fs41587-019-0224-x',
397 => 'https://doi.org/10.1073%2Fpnas.88.10.4433',
398 => 'https://doi.org/10.1093%2Fjamia%2Focw112',
399 => 'https://doi.org/10.1098%2Frstb.2017.0043',
400 => 'https://doi.org/10.1109%2F29.21701',
401 => 'https://doi.org/10.1109%2F5.726791',
402 => 'https://doi.org/10.1109%2F72.963769',
403 => 'https://doi.org/10.1109%2FACCESS.2020.3006362',
404 => 'https://doi.org/10.1109%2FCSCI49370.2019.00067',
405 => 'https://doi.org/10.1109%2FCSCITA.2017.8066548',
406 => 'https://doi.org/10.1109%2FICCCI50826.2021.9402569',
407 => 'https://doi.org/10.1109%2FICCVW.2017.18',
408 => 'https://doi.org/10.1109%2FTCI.2021.3075349',
409 => 'https://doi.org/10.1109%2FTSMC.1971.4308320',
410 => 'https://doi.org/10.1109%2FTUFFC.2020.3010186',
411 => 'https://doi.org/10.1109%2Fcvpr.2012.6248110',
412 => 'https://doi.org/10.1109%2Fdevlrn.2008.4640845',
413 => 'https://doi.org/10.1109%2Ficassp.2013.6639347',
414 => 'https://doi.org/10.1109%2Ficassp.2013.6639349',
415 => 'https://doi.org/10.1109%2Fmsp.2009.932166',
416 => 'https://doi.org/10.1109%2Fmsp.2012.2205597',
417 => 'https://doi.org/10.1109%2Ftaslp.2014.2339736',
418 => 'https://doi.org/10.1109%2Ftaslp.2014.2383614',
419 => 'https://doi.org/10.1109%2Ftpami.2013.50',
420 => 'https://doi.org/10.1118%2F1.597177',
421 => 'https://doi.org/10.1126%2Fscience.7761831',
422 => 'https://doi.org/10.1126%2Fscience.aaw4741',
423 => 'https://doi.org/10.1142%2Fs0218001493000455',
424 => 'https://doi.org/10.1145%2F1553374.1553486',
425 => 'https://doi.org/10.1145%2F2649387.2649442',
426 => 'https://doi.org/10.1145%2F3126908.3126912',
427 => 'https://doi.org/10.1145%2F3140659.3080246',
428 => 'https://doi.org/10.1162%2F08997660260293319',
429 => 'https://doi.org/10.1162%2Fneco.1996.8.5.895',
430 => 'https://doi.org/10.1162%2Fneco.1997.9.8.1735',
431 => 'https://doi.org/10.1162%2Fneco.2006.18.7.1527',
432 => 'https://doi.org/10.1162%2Fneco_a_00052',
433 => 'https://doi.org/10.1177%2F1461444819885334',
434 => 'https://doi.org/10.1186%2Fs13636-015-0068-3',
435 => 'https://doi.org/10.13140%2FRG.2.2.15418.90567',
436 => 'https://doi.org/10.1364%2FAO.29.004790',
437 => 'https://doi.org/10.1364%2FAO.30.004211',
438 => 'https://doi.org/10.1371%2Fjournal.pcbi.1002211',
439 => 'https://doi.org/10.1523%2Fjneurosci.5023-14.2015',
440 => 'https://doi.org/10.1561%2F0600000018',
441 => 'https://doi.org/10.1561%2F2000000039',
442 => 'https://doi.org/10.1561%2F2200000006',
443 => 'https://doi.org/10.21437%2FInterspeech.2010-343',
444 => 'https://doi.org/10.21437%2FInterspeech.2011-169',
445 => 'https://doi.org/10.2196%2F12957',
446 => 'https://doi.org/10.2196%2Fmhealth.6562',
447 => 'https://doi.org/10.3389%2Ffncom.2016.00073',
448 => 'https://doi.org/10.3389%2Ffncom.2016.00094',
449 => 'https://doi.org/10.3390%2Farts6020005',
450 => 'https://doi.org/10.3390%2Farts6040018',
451 => 'https://doi.org/10.3390%2Fcancers14071819',
452 => 'https://doi.org/10.4249%2Fscholarpedia.5947',
453 => 'https://doi.org/10.5591%2F978-1-57735-516-8%2Fijcai11-210',
454 => 'https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php',
455 => 'https://ieeexplore.ieee.org/abstract/document/9144249?casa_token=NLVQbGR8h-8AAAAA:rz3cvMmiplVSRCGTDouUZkIeribqQKaMtK9t0MxlwKBMRwaTJL1onKOwkEk0qW_fFJCfd2ejOg',
456 => 'https://ieeexplore.ieee.org/document/9070967',
457 => 'https://ieeexplore.ieee.org/document/9130662',
458 => 'https://ieeexplore.ieee.org/document/9415140',
459 => 'https://www.jstor.org/action/doBasicSearch?Query=%22Deep+learning%22&acc=on&wc=on',
460 => 'https://api.semanticscholar.org/CorpusID:1119517',
461 => 'https://api.semanticscholar.org/CorpusID:11715509',
462 => 'https://api.semanticscholar.org/CorpusID:12149203',
463 => 'https://api.semanticscholar.org/CorpusID:12485056',
464 => 'https://api.semanticscholar.org/CorpusID:1304548',
465 => 'https://api.semanticscholar.org/CorpusID:1317136',
466 => 'https://api.semanticscholar.org/CorpusID:13816461',
467 => 'https://api.semanticscholar.org/CorpusID:14135321',
468 => 'https://api.semanticscholar.org/CorpusID:14542261',
469 => 'https://api.semanticscholar.org/CorpusID:14663106',
470 => 'https://api.semanticscholar.org/CorpusID:15066318',
471 => 'https://api.semanticscholar.org/CorpusID:15641618',
472 => 'https://api.semanticscholar.org/CorpusID:16560320',
473 => 'https://api.semanticscholar.org/CorpusID:16970545',
474 => 'https://api.semanticscholar.org/CorpusID:1915014',
475 => 'https://api.semanticscholar.org/CorpusID:1918673',
476 => 'https://api.semanticscholar.org/CorpusID:1994856',
477 => 'https://api.semanticscholar.org/CorpusID:201716327',
478 => 'https://api.semanticscholar.org/CorpusID:20246434',
479 => 'https://api.semanticscholar.org/CorpusID:204092079',
480 => 'https://api.semanticscholar.org/CorpusID:206485943',
481 => 'https://api.semanticscholar.org/CorpusID:206602362',
482 => 'https://api.semanticscholar.org/CorpusID:206775608',
483 => 'https://api.semanticscholar.org/CorpusID:207217210',
484 => 'https://api.semanticscholar.org/CorpusID:2088679',
485 => 'https://api.semanticscholar.org/CorpusID:209363848',
486 => 'https://api.semanticscholar.org/CorpusID:211010976',
487 => 'https://api.semanticscholar.org/CorpusID:212755458',
488 => 'https://api.semanticscholar.org/CorpusID:216043128',
489 => 'https://api.semanticscholar.org/CorpusID:2161592',
490 => 'https://api.semanticscholar.org/CorpusID:217950236',
491 => 'https://api.semanticscholar.org/CorpusID:220055785',
492 => 'https://api.semanticscholar.org/CorpusID:220523562',
493 => 'https://api.semanticscholar.org/CorpusID:220733699',
494 => 'https://api.semanticscholar.org/CorpusID:2309950',
495 => 'https://api.semanticscholar.org/CorpusID:231846518',
496 => 'https://api.semanticscholar.org/CorpusID:235340737',
497 => 'https://api.semanticscholar.org/CorpusID:2376781',
498 => 'https://api.semanticscholar.org/CorpusID:24504018',
499 => 'https://api.semanticscholar.org/CorpusID:3074096',
500 => 'https://api.semanticscholar.org/CorpusID:35350962',
501 => 'https://api.semanticscholar.org/CorpusID:357467',
502 => 'https://api.semanticscholar.org/CorpusID:3821594',
503 => 'https://api.semanticscholar.org/CorpusID:392458',
504 => 'https://api.semanticscholar.org/CorpusID:39281431',
505 => 'https://api.semanticscholar.org/CorpusID:393948',
506 => 'https://api.semanticscholar.org/CorpusID:3958369',
507 => 'https://api.semanticscholar.org/CorpusID:40745740',
508 => 'https://api.semanticscholar.org/CorpusID:4460235',
509 => 'https://api.semanticscholar.org/CorpusID:4728736',
510 => 'https://api.semanticscholar.org/CorpusID:515925',
511 => 'https://api.semanticscholar.org/CorpusID:5613334',
512 => 'https://api.semanticscholar.org/CorpusID:57379996',
513 => 'https://api.semanticscholar.org/CorpusID:5818342',
514 => 'https://api.semanticscholar.org/CorpusID:7504633',
515 => 'https://api.semanticscholar.org/CorpusID:8869270',
516 => 'https://api.semanticscholar.org/CorpusID:9868901',
517 => 'https://www.semanticscholar.org/paper/51a80649d16a38d41dbd20472deb3bc9b61b59a0',
518 => 'https://www.worldcat.org/issn/0021-9991',
519 => 'https://www.worldcat.org/issn/0027-8424',
520 => 'https://www.worldcat.org/issn/0028-0836',
521 => 'https://www.worldcat.org/issn/0045-7825',
522 => 'https://www.worldcat.org/issn/0096-3518',
523 => 'https://www.worldcat.org/issn/0218-0014',
524 => 'https://www.worldcat.org/issn/0896-6273',
525 => 'https://www.worldcat.org/issn/0899-7667',
526 => 'https://www.worldcat.org/issn/0959-4388',
527 => 'https://www.worldcat.org/issn/0962-8436',
528 => 'https://www.worldcat.org/issn/1059-1028',
529 => 'https://www.worldcat.org/issn/1067-5027',
530 => 'https://www.worldcat.org/issn/1364-6613',
531 => 'https://www.worldcat.org/issn/1461-4448',
532 => 'https://www.worldcat.org/issn/1476-4687',
533 => 'https://www.worldcat.org/issn/1525-8955',
534 => 'https://www.worldcat.org/issn/1546-1726',
535 => 'https://www.worldcat.org/issn/1553-7358',
536 => 'https://www.worldcat.org/issn/1610-1987',
537 => 'https://www.worldcat.org/issn/1662-5188',
538 => 'https://www.worldcat.org/issn/2072-6694',
539 => 'https://www.worldcat.org/issn/2169-3536',
540 => 'https://www.worldcat.org/issn/2333-9403',
541 => 'https://www.worldcat.org/issn/2397-3374',
542 => 'https://zenodo.org/record/891433',
543 => 'https://hal.science/hal-03029735'
] |
Parsed HTML source of the new revision (new_html) | '<div class="mw-parser-output"><p><a rel="nofollow" class="external free" href="https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt">https://github.com/android/storage-samples/blob/22784d8cbf1d990958ae554ec61afea1a9da93c1/MediaStore/app/src/main/java/com/android/samples/mediastore/MediaStoreImage.kt</a>
</p>
<div id="toc" class="toc" role="navigation" aria-labelledby="mw-toc-heading"><input type="checkbox" role="button" id="toctogglecheckbox" class="toctogglecheckbox" style="display:none" /><div class="toctitle" lang="en" dir="ltr"><h2 id="mw-toc-heading">Contents</h2><span class="toctogglespan"><label class="toctogglelabel" for="toctogglecheckbox"></label></span></div>
<ul>
<li class="toclevel-1 tocsection-1"><a href="#Definition"><span class="tocnumber">1</span> <span class="toctext">Definition</span></a></li>
<li class="toclevel-1 tocsection-2"><a href="#Overview"><span class="tocnumber">2</span> <span class="toctext">Overview</span></a></li>
<li class="toclevel-1 tocsection-3"><a href="#Interpretations"><span class="tocnumber">3</span> <span class="toctext">Interpretations</span></a></li>
<li class="toclevel-1 tocsection-4"><a href="#History"><span class="tocnumber">4</span> <span class="toctext">History</span></a>
<ul>
<li class="toclevel-2 tocsection-5"><a href="#Deep_learning_revolution"><span class="tocnumber">4.1</span> <span class="toctext">Deep learning revolution</span></a></li>
</ul>
</li>
<li class="toclevel-1 tocsection-6"><a href="#Neural_networks"><span class="tocnumber">5</span> <span class="toctext">Neural networks</span></a>
<ul>
<li class="toclevel-2 tocsection-7"><a href="#Artificial_neural_networks"><span class="tocnumber">5.1</span> <span class="toctext">Artificial neural networks</span></a></li>
<li class="toclevel-2 tocsection-8"><a href="#Deep_neural_networks"><span class="tocnumber">5.2</span> <span class="toctext">Deep neural networks</span></a>
<ul>
<li class="toclevel-3 tocsection-9"><a href="#Challenges"><span class="tocnumber">5.2.1</span> <span class="toctext">Challenges</span></a></li>
</ul>
</li>
</ul>
</li>
<li class="toclevel-1 tocsection-10"><a href="#Hardware"><span class="tocnumber">6</span> <span class="toctext">Hardware</span></a></li>
<li class="toclevel-1 tocsection-11"><a href="#Applications"><span class="tocnumber">7</span> <span class="toctext">Applications</span></a>
<ul>
<li class="toclevel-2 tocsection-12"><a href="#Automatic_speech_recognition"><span class="tocnumber">7.1</span> <span class="toctext">Automatic speech recognition</span></a></li>
<li class="toclevel-2 tocsection-13"><a href="#Image_recognition"><span class="tocnumber">7.2</span> <span class="toctext">Image recognition</span></a></li>
<li class="toclevel-2 tocsection-14"><a href="#Visual_art_processing"><span class="tocnumber">7.3</span> <span class="toctext">Visual art processing</span></a></li>
<li class="toclevel-2 tocsection-15"><a href="#Natural_language_processing"><span class="tocnumber">7.4</span> <span class="toctext">Natural language processing</span></a></li>
<li class="toclevel-2 tocsection-16"><a href="#Drug_discovery_and_toxicology"><span class="tocnumber">7.5</span> <span class="toctext">Drug discovery and toxicology</span></a></li>
<li class="toclevel-2 tocsection-17"><a href="#Customer_relationship_management"><span class="tocnumber">7.6</span> <span class="toctext">Customer relationship management</span></a></li>
<li class="toclevel-2 tocsection-18"><a href="#Recommendation_systems"><span class="tocnumber">7.7</span> <span class="toctext">Recommendation systems</span></a></li>
<li class="toclevel-2 tocsection-19"><a href="#Bioinformatics"><span class="tocnumber">7.8</span> <span class="toctext">Bioinformatics</span></a></li>
<li class="toclevel-2 tocsection-20"><a href="#Medical_image_analysis"><span class="tocnumber">7.9</span> <span class="toctext">Medical image analysis</span></a></li>
<li class="toclevel-2 tocsection-21"><a href="#Mobile_advertising"><span class="tocnumber">7.10</span> <span class="toctext">Mobile advertising</span></a></li>
<li class="toclevel-2 tocsection-22"><a href="#Image_restoration"><span class="tocnumber">7.11</span> <span class="toctext">Image restoration</span></a></li>
<li class="toclevel-2 tocsection-23"><a href="#Financial_fraud_detection"><span class="tocnumber">7.12</span> <span class="toctext">Financial fraud detection</span></a></li>
<li class="toclevel-2 tocsection-24"><a href="#Military"><span class="tocnumber">7.13</span> <span class="toctext">Military</span></a></li>
<li class="toclevel-2 tocsection-25"><a href="#Partial_differential_equations"><span class="tocnumber">7.14</span> <span class="toctext">Partial differential equations</span></a></li>
<li class="toclevel-2 tocsection-26"><a href="#Image_Reconstruction"><span class="tocnumber">7.15</span> <span class="toctext">Image Reconstruction</span></a></li>
</ul>
</li>
<li class="toclevel-1 tocsection-27"><a href="#Relation_to_human_cognitive_and_brain_development"><span class="tocnumber">8</span> <span class="toctext">Relation to human cognitive and brain development</span></a></li>
<li class="toclevel-1 tocsection-28"><a href="#Commercial_activity"><span class="tocnumber">9</span> <span class="toctext">Commercial activity</span></a></li>
<li class="toclevel-1 tocsection-29"><a href="#Criticism_and_comment"><span class="tocnumber">10</span> <span class="toctext">Criticism and comment</span></a>
<ul>
<li class="toclevel-2 tocsection-30"><a href="#Theory"><span class="tocnumber">10.1</span> <span class="toctext">Theory</span></a></li>
<li class="toclevel-2 tocsection-31"><a href="#Errors"><span class="tocnumber">10.2</span> <span class="toctext">Errors</span></a></li>
<li class="toclevel-2 tocsection-32"><a href="#Cyber_threat"><span class="tocnumber">10.3</span> <span class="toctext">Cyber threat</span></a></li>
<li class="toclevel-2 tocsection-33"><a href="#Reliance_on_human_microwork"><span class="tocnumber">10.4</span> <span class="toctext">Reliance on human microwork</span></a></li>
</ul>
</li>
<li class="toclevel-1 tocsection-34"><a href="#See_also"><span class="tocnumber">11</span> <span class="toctext">See also</span></a></li>
<li class="toclevel-1 tocsection-35"><a href="#References"><span class="tocnumber">12</span> <span class="toctext">References</span></a></li>
<li class="toclevel-1 tocsection-36"><a href="#Further_reading"><span class="tocnumber">13</span> <span class="toctext">Further reading</span></a></li>
</ul>
</div>
<h2><span class="mw-headline" id="Definition">Definition</span></h2>
<p>Deep learning is a class of <a href="/wiki/Machine_learning" title="Machine learning">machine learning</a> <a href="/wiki/Algorithm" title="Algorithm">algorithms</a> that<sup id="cite_ref-BOOK2014_1-0" class="reference"><a href="#cite_note-BOOK2014-1">[1]</a></sup><sup class="reference nowrap"><span title="Pages: 199–200">: 199–200 </span></sup> uses multiple layers to progressively extract higher-level features from the raw input. For example, in <a href="/wiki/Image_processing" class="mw-redirect" title="Image processing">image processing</a>, lower layers may identify edges, while higher layers may identify the concepts relevant to a human such as digits or letters or faces.
</p>
<h2><span class="mw-headline" id="Overview">Overview</span></h2>
<p>Most modern deep learning models are based on <a href="/wiki/Artificial_neural_network" title="Artificial neural network">artificial neural networks</a>, specifically <a href="/wiki/Convolutional_neural_network" title="Convolutional neural network">convolutional neural networks</a> (CNN)s, although they can also include <a href="/wiki/Propositional_formula" title="Propositional formula">propositional formulas</a> or latent variables organized layer-wise in deep <a href="/wiki/Generative_model" title="Generative model">generative models</a> such as the nodes in <a href="/wiki/Deep_belief_network" title="Deep belief network">deep belief networks</a> and deep <a href="/wiki/Boltzmann_machine" title="Boltzmann machine">Boltzmann machines</a>.<sup id="cite_ref-BENGIODEEP_2-0" class="reference"><a href="#cite_note-BENGIODEEP-2">[2]</a></sup>
</p><p>In deep learning, each level learns to transform its input data into a slightly more abstract and composite representation. In an image recognition application, the raw input may be a <a href="/wiki/Matrix_(mathematics)" title="Matrix (mathematics)">matrix</a> of pixels; the first representational layer may abstract the pixels and encode edges; the second layer may compose and encode arrangements of edges; the third layer may encode a nose and eyes; and the fourth layer may recognize that the image contains a face. Importantly, a deep learning process can learn which features to optimally place in which level <i>on its own</i>. This does not eliminate the need for hand-tuning; for example, varying numbers of layers and layer sizes can provide different degrees of abstraction.<sup id="cite_ref-BENGIO2012_3-0" class="reference"><a href="#cite_note-BENGIO2012-3">[3]</a></sup><sup id="cite_ref-4" class="reference"><a href="#cite_note-4">[4]</a></sup>
</p><p>The word "deep" in "deep learning" refers to the number of layers through which the data is transformed. More precisely, deep learning systems have a substantial <i>credit assignment path</i> (CAP) depth. The CAP is the chain of transformations from input to output. CAPs describe potentially causal connections between input and output. For a <a href="/wiki/Feedforward_neural_network" title="Feedforward neural network">feedforward neural network</a>, the depth of the CAPs is that of the network and is the number of hidden layers plus one (as the output layer is also parameterized). For <a href="/wiki/Recurrent_neural_network" title="Recurrent neural network">recurrent neural networks</a>, in which a signal may propagate through a layer more than once, the CAP depth is potentially unlimited.<sup id="cite_ref-SCHIDHUB_5-0" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup> No universally agreed-upon threshold of depth divides shallow learning from deep learning, but most researchers agree that deep learning involves CAP depth higher than 2. CAP of depth 2 has been shown to be a universal approximator in the sense that it can emulate any function.<sup id="cite_ref-6" class="reference"><a href="#cite_note-6">[6]</a></sup> Beyond that, more layers do not add to the function approximator ability of the network. Deep models (CAP > 2) are able to extract better features than shallow models and hence, extra layers help in learning the features effectively.
</p><p>Deep learning architectures can be constructed with a <a href="/wiki/Greedy_algorithm" title="Greedy algorithm">greedy</a> layer-by-layer method.<sup id="cite_ref-BENGIO2007_7-0" class="reference"><a href="#cite_note-BENGIO2007-7">[7]</a></sup> Deep learning helps to disentangle these abstractions and pick out which features improve performance.<sup id="cite_ref-BENGIO2012_3-1" class="reference"><a href="#cite_note-BENGIO2012-3">[3]</a></sup>
</p><p>For <a href="/wiki/Supervised_learning" title="Supervised learning">supervised learning</a> tasks, deep learning methods eliminate <a href="/wiki/Feature_engineering" title="Feature engineering">feature engineering</a>, by translating the data into compact intermediate representations akin to <a href="/wiki/Principal_Component_Analysis" class="mw-redirect" title="Principal Component Analysis">principal components</a>, and derive layered structures that remove redundancy in representation.
</p><p>Deep learning algorithms can be applied to unsupervised learning tasks. This is an important benefit because unlabeled data are more abundant than the labeled data. Examples of deep structures that can be trained in an unsupervised manner are <a href="/wiki/Deep_belief_network" title="Deep belief network">deep belief networks</a>.<sup id="cite_ref-BENGIO2012_3-2" class="reference"><a href="#cite_note-BENGIO2012-3">[3]</a></sup><sup id="cite_ref-SCHOLARDBNS_8-0" class="reference"><a href="#cite_note-SCHOLARDBNS-8">[8]</a></sup>
</p>
<h2><span class="mw-headline" id="Interpretations">Interpretations</span></h2>
<p>Deep neural networks are generally interpreted in terms of the <a href="/wiki/Universal_approximation_theorem" title="Universal approximation theorem">universal approximation theorem</a><sup id="cite_ref-cyb_9-0" class="reference"><a href="#cite_note-cyb-9">[9]</a></sup><sup id="cite_ref-horn_10-0" class="reference"><a href="#cite_note-horn-10">[10]</a></sup><sup id="cite_ref-Haykin,_Simon_1998_11-0" class="reference"><a href="#cite_note-Haykin,_Simon_1998-11">[11]</a></sup><sup id="cite_ref-Hassoun,_M._1995_p._48_12-0" class="reference"><a href="#cite_note-Hassoun,_M._1995_p._48-12">[12]</a></sup><sup id="cite_ref-ZhouLu_13-0" class="reference"><a href="#cite_note-ZhouLu-13">[13]</a></sup> or <a href="/wiki/Bayesian_inference" title="Bayesian inference">probabilistic inference</a>.<sup id="cite_ref-14" class="reference"><a href="#cite_note-14">[14]</a></sup><sup id="cite_ref-BOOK2014_1-1" class="reference"><a href="#cite_note-BOOK2014-1">[1]</a></sup><sup id="cite_ref-BENGIO2012_3-3" class="reference"><a href="#cite_note-BENGIO2012-3">[3]</a></sup><sup id="cite_ref-SCHIDHUB_5-1" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup><sup id="cite_ref-MURPHY_15-0" class="reference"><a href="#cite_note-MURPHY-15">[15]</a></sup>
</p><p>The classic universal approximation theorem concerns the capacity of <a href="/wiki/Feedforward_neural_networks" class="mw-redirect" title="Feedforward neural networks">feedforward neural networks</a> with a single hidden layer of finite size to approximate <a href="/wiki/Continuous_functions" class="mw-redirect" title="Continuous functions">continuous functions</a>.<sup id="cite_ref-cyb_9-1" class="reference"><a href="#cite_note-cyb-9">[9]</a></sup><sup id="cite_ref-horn_10-1" class="reference"><a href="#cite_note-horn-10">[10]</a></sup><sup id="cite_ref-Haykin,_Simon_1998_11-1" class="reference"><a href="#cite_note-Haykin,_Simon_1998-11">[11]</a></sup><sup id="cite_ref-Hassoun,_M._1995_p._48_12-1" class="reference"><a href="#cite_note-Hassoun,_M._1995_p._48-12">[12]</a></sup> In 1989, the first proof was published by <a href="/wiki/George_Cybenko" title="George Cybenko">George Cybenko</a> for <a href="/wiki/Sigmoid_function" title="Sigmoid function">sigmoid</a> activation functions<sup id="cite_ref-cyb_9-2" class="reference"><a href="#cite_note-cyb-9">[9]</a></sup> and was generalised to feed-forward multi-layer architectures in 1991 by Kurt Hornik.<sup id="cite_ref-horn_10-2" class="reference"><a href="#cite_note-horn-10">[10]</a></sup> Recent work also showed that universal approximation also holds for non-bounded activation functions such as the rectified linear unit.<sup id="cite_ref-sonoda17_16-0" class="reference"><a href="#cite_note-sonoda17-16">[16]</a></sup>
</p><p>The universal approximation theorem for <a href="/wiki/Deep_neural_network" class="mw-redirect" title="Deep neural network">deep neural networks</a> concerns the capacity of networks with bounded width but the depth is allowed to grow. Lu et al.<sup id="cite_ref-ZhouLu_13-1" class="reference"><a href="#cite_note-ZhouLu-13">[13]</a></sup> proved that if the width of a <a href="/wiki/Deep_neural_network" class="mw-redirect" title="Deep neural network">deep neural network</a> with <a href="/wiki/ReLU" class="mw-redirect" title="ReLU">ReLU</a> activation is strictly larger than the input dimension, then the network can approximate any <a href="/wiki/Lebesgue_integration" title="Lebesgue integration">Lebesgue integrable function</a>; If the width is smaller or equal to the input dimension, then a <a href="/wiki/Deep_neural_network" class="mw-redirect" title="Deep neural network">deep neural network</a> is not a universal approximator.
</p><p>The <a href="/wiki/Probabilistic" class="mw-redirect" title="Probabilistic">probabilistic</a> interpretation<sup id="cite_ref-MURPHY_15-1" class="reference"><a href="#cite_note-MURPHY-15">[15]</a></sup> derives from the field of <a href="/wiki/Machine_learning" title="Machine learning">machine learning</a>. It features inference,<sup id="cite_ref-BOOK2014_1-2" class="reference"><a href="#cite_note-BOOK2014-1">[1]</a></sup><sup id="cite_ref-BENGIODEEP_2-1" class="reference"><a href="#cite_note-BENGIODEEP-2">[2]</a></sup><sup id="cite_ref-BENGIO2012_3-4" class="reference"><a href="#cite_note-BENGIO2012-3">[3]</a></sup><sup id="cite_ref-SCHIDHUB_5-2" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup><sup id="cite_ref-SCHOLARDBNS_8-1" class="reference"><a href="#cite_note-SCHOLARDBNS-8">[8]</a></sup><sup id="cite_ref-MURPHY_15-2" class="reference"><a href="#cite_note-MURPHY-15">[15]</a></sup> as well as the <a href="/wiki/Optimization" class="mw-redirect" title="Optimization">optimization</a> concepts of <a href="/wiki/Training" title="Training">training</a> and <a href="/wiki/Test_(assessment)" class="mw-redirect" title="Test (assessment)">testing</a>, related to fitting and <a href="/wiki/Generalization" title="Generalization">generalization</a>, respectively. More specifically, the probabilistic interpretation considers the activation nonlinearity as a <a href="/wiki/Cumulative_distribution_function" title="Cumulative distribution function">cumulative distribution function</a>.<sup id="cite_ref-MURPHY_15-3" class="reference"><a href="#cite_note-MURPHY-15">[15]</a></sup> The probabilistic interpretation led to the introduction of <a href="/wiki/Dropout_(neural_networks)" class="mw-redirect" title="Dropout (neural networks)">dropout</a> as <a href="/wiki/Regularization_(mathematics)" title="Regularization (mathematics)">regularizer</a> in neural networks. The probabilistic interpretation was introduced by researchers including <a href="/wiki/John_Hopfield" title="John Hopfield">Hopfield</a>, <a href="/wiki/Bernard_Widrow" title="Bernard Widrow">Widrow</a> and <a href="/wiki/Kumpati_S._Narendra" title="Kumpati S. Narendra">Narendra</a> and popularized in surveys such as the one by <a href="/wiki/Christopher_Bishop" title="Christopher Bishop">Bishop</a>.<sup id="cite_ref-prml_17-0" class="reference"><a href="#cite_note-prml-17">[17]</a></sup>
</p>
<h2><span class="mw-headline" id="History">History</span></h2>
<p>Some sources point out that <a href="/wiki/Frank_Rosenblatt" title="Frank Rosenblatt">Frank Rosenblatt</a> developed and explored all of the basic ingredients of the deep learning systems of today.<sup id="cite_ref-Who_Is_the_Father_of_Deep_Learning?_18-0" class="reference"><a href="#cite_note-Who_Is_the_Father_of_Deep_Learning?-18">[18]</a></sup> He described it in his book "Principles of Neurodynamics: Perceptrons and the Theory of Brain Mechanisms", published by Cornell Aeronautical Laboratory, Inc., Cornell University in 1962.
</p><p>The first general, working learning algorithm for supervised, deep, feedforward, multilayer <a href="/wiki/Perceptron" title="Perceptron">perceptrons</a> was published by <a href="/wiki/Alexey_Ivakhnenko" title="Alexey Ivakhnenko">Alexey Ivakhnenko</a> and Lapa in 1967.<sup id="cite_ref-ivak1965_19-0" class="reference"><a href="#cite_note-ivak1965-19">[19]</a></sup> A 1971 paper described a deep network with eight layers trained by the <a href="/wiki/Group_method_of_data_handling" title="Group method of data handling">group method of data handling</a>.<sup id="cite_ref-ivak1971_20-0" class="reference"><a href="#cite_note-ivak1971-20">[20]</a></sup> Other deep learning working architectures, specifically those built for <a href="/wiki/Computer_vision" title="Computer vision">computer vision</a>, began with the <a href="/wiki/Neocognitron" title="Neocognitron">Neocognitron</a> introduced by <a href="/wiki/Kunihiko_Fukushima" title="Kunihiko Fukushima">Kunihiko Fukushima</a> in 1980.<sup id="cite_ref-FUKU1980_21-0" class="reference"><a href="#cite_note-FUKU1980-21">[21]</a></sup>
</p><p>The term <i>Deep Learning</i> was introduced to the machine learning community by <a href="/wiki/Rina_Dechter" title="Rina Dechter">Rina Dechter</a> in 1986,<sup id="cite_ref-dechter1986_22-0" class="reference"><a href="#cite_note-dechter1986-22">[22]</a></sup> and to <a href="/wiki/Artificial_Neural_Networks" class="mw-redirect" title="Artificial Neural Networks">artificial neural networks</a> by Igor Aizenberg and colleagues in 2000, in the context of <a href="/wiki/Boolean_network" title="Boolean network">Boolean</a> threshold neurons.<sup id="cite_ref-aizenberg2000_23-0" class="reference"><a href="#cite_note-aizenberg2000-23">[23]</a></sup><sup id="cite_ref-24" class="reference"><a href="#cite_note-24">[24]</a></sup>
</p><p>In 1989, <a href="/wiki/Yann_LeCun" title="Yann LeCun">Yann LeCun</a> et al. applied the standard <a href="/wiki/Backpropagation" title="Backpropagation">backpropagation</a> algorithm, which had been around as the reverse mode of <a href="/wiki/Automatic_differentiation" title="Automatic differentiation">automatic differentiation</a> since 1970,<sup id="cite_ref-lin1970_25-0" class="reference"><a href="#cite_note-lin1970-25">[25]</a></sup><sup id="cite_ref-grie2012_26-0" class="reference"><a href="#cite_note-grie2012-26">[26]</a></sup><sup id="cite_ref-WERBOS1974_27-0" class="reference"><a href="#cite_note-WERBOS1974-27">[27]</a></sup><sup id="cite_ref-werbos1982_28-0" class="reference"><a href="#cite_note-werbos1982-28">[28]</a></sup> to a deep neural network with the purpose of <a href="/wiki/Handwriting_recognition" title="Handwriting recognition">recognizing handwritten ZIP codes</a> on mail. While the algorithm worked, training required 3 days.<sup id="cite_ref-LECUN1989_29-0" class="reference"><a href="#cite_note-LECUN1989-29">[29]</a></sup>
</p><p>Independently in 1988, Wei Zhang et al. applied the backpropagation algorithm to a convolutional neural network (a simplified Neocognitron by keeping only the convolutional interconnections between the image feature layers and the last fully connected layer) for alphabets recognition and also proposed an implementation of the CNN with an optical computing system.<sup id="cite_ref-wz1988_30-0" class="reference"><a href="#cite_note-wz1988-30">[30]</a></sup><sup id="cite_ref-wz1990_31-0" class="reference"><a href="#cite_note-wz1990-31">[31]</a></sup> Subsequently, Wei Zhang, et al. modified the model by removing the last fully connected layer and applied it for medical image object segmentation in 1991<sup id="cite_ref-32" class="reference"><a href="#cite_note-32">[32]</a></sup> and breast cancer detection in mammograms in 1994.<sup id="cite_ref-33" class="reference"><a href="#cite_note-33">[33]</a></sup>
</p><p>In 1994, André de Carvalho, together with Mike Fairhurst and David Bisset, published experimental results of a multi-layer boolean neural network, also known as a weightless neural network, composed of a 3-layers self-organising feature extraction neural network module (SOFT) followed by a multi-layer classification neural network module (GSN), which were independently trained. Each layer in the feature extraction module extracted features with growing complexity regarding the previous layer.<sup id="cite_ref-34" class="reference"><a href="#cite_note-34">[34]</a></sup>
</p><p>In 1995, <a href="/wiki/Brendan_Frey" title="Brendan Frey">Brendan Frey</a> demonstrated that it was possible to train (over two days) a network containing six fully connected layers and several hundred hidden units using the <a href="/wiki/Wake-sleep_algorithm" title="Wake-sleep algorithm">wake-sleep algorithm</a>, co-developed with <a href="/wiki/Peter_Dayan" title="Peter Dayan">Peter Dayan</a> and <a href="/wiki/Geoffrey_Hinton" title="Geoffrey Hinton">Hinton</a>.<sup id="cite_ref-35" class="reference"><a href="#cite_note-35">[35]</a></sup> Many factors contribute to the slow speed, including the <a href="/wiki/Vanishing_gradient_problem" title="Vanishing gradient problem">vanishing gradient problem</a> analyzed in 1991 by <a href="/wiki/Sepp_Hochreiter" title="Sepp Hochreiter">Sepp Hochreiter</a>.<sup id="cite_ref-HOCH1991_36-0" class="reference"><a href="#cite_note-HOCH1991-36">[36]</a></sup><sup id="cite_ref-HOCH2001_37-0" class="reference"><a href="#cite_note-HOCH2001-37">[37]</a></sup>
</p><p>Since 1997, Sven Behnke extended the feed-forward hierarchical convolutional approach in the Neural Abstraction Pyramid<sup id="cite_ref-38" class="reference"><a href="#cite_note-38">[38]</a></sup> by lateral and backward connections in order to flexibly incorporate context into decisions and iteratively resolve local ambiguities.
</p><p>Simpler models that use task-specific handcrafted features such as <a href="/wiki/Gabor_filter" title="Gabor filter">Gabor filters</a> and <a href="/wiki/Support_vector_machine" title="Support vector machine">support vector machines</a> (SVMs) were a popular choice in the 1990s and 2000s, because of <a href="/wiki/Artificial_neural_network" title="Artificial neural network">artificial neural network</a>'s (ANN) computational cost and a lack of understanding of how the brain wires its biological networks.
</p><p>Both shallow and deep learning (e.g., recurrent nets) of ANNs have been explored for many years.<sup id="cite_ref-39" class="reference"><a href="#cite_note-39">[39]</a></sup><sup id="cite_ref-Robinson1992_40-0" class="reference"><a href="#cite_note-Robinson1992-40">[40]</a></sup><sup id="cite_ref-41" class="reference"><a href="#cite_note-41">[41]</a></sup> These methods never outperformed non-uniform internal-handcrafting Gaussian <a href="/wiki/Mixture_model" title="Mixture model">mixture model</a>/<a href="/wiki/Hidden_Markov_model" title="Hidden Markov model">Hidden Markov model</a> (GMM-HMM) technology based on generative models of speech trained discriminatively.<sup id="cite_ref-Baker2009_42-0" class="reference"><a href="#cite_note-Baker2009-42">[42]</a></sup> Key difficulties have been analyzed, including gradient diminishing<sup id="cite_ref-HOCH1991_36-1" class="reference"><a href="#cite_note-HOCH1991-36">[36]</a></sup> and weak temporal correlation structure in neural predictive models.<sup id="cite_ref-Bengio1991_43-0" class="reference"><a href="#cite_note-Bengio1991-43">[43]</a></sup><sup id="cite_ref-Deng1994_44-0" class="reference"><a href="#cite_note-Deng1994-44">[44]</a></sup> Additional difficulties were the lack of training data and limited computing power.
</p><p>Most <a href="/wiki/Speech_recognition" title="Speech recognition">speech recognition</a> researchers moved away from neural nets to pursue generative modeling. An exception was at <a href="/wiki/SRI_International" title="SRI International">SRI International</a> in the late 1990s. Funded by the US government's <a href="/wiki/National_Security_Agency" title="National Security Agency">NSA</a> and <a href="/wiki/DARPA" title="DARPA">DARPA</a>, SRI studied deep neural networks in speech and <a href="/wiki/Speaker_recognition" title="Speaker recognition">speaker recognition</a>. The speaker recognition team led by <a href="/wiki/Larry_Heck" title="Larry Heck">Larry Heck</a> reported significant success with deep neural networks in speech processing in the 1998 <a href="/wiki/National_Institute_of_Standards_and_Technology" title="National Institute of Standards and Technology">National Institute of Standards and Technology</a> Speaker Recognition evaluation.<sup id="cite_ref-Doddington2000_45-0" class="reference"><a href="#cite_note-Doddington2000-45">[45]</a></sup> The SRI deep neural network was then deployed in the Nuance Verifier, representing the first major industrial application of deep learning.<sup id="cite_ref-Heck2000_46-0" class="reference"><a href="#cite_note-Heck2000-46">[46]</a></sup>
</p><p>The principle of elevating "raw" features over hand-crafted optimization was first explored successfully in the architecture of deep autoencoder on the "raw" spectrogram or linear filter-bank features in the late 1990s,<sup id="cite_ref-Heck2000_46-1" class="reference"><a href="#cite_note-Heck2000-46">[46]</a></sup> showing its superiority over the Mel-Cepstral features that contain stages of fixed transformation from spectrograms. The raw features of speech, <a href="/wiki/Waveform" title="Waveform">waveforms</a>, later produced excellent larger-scale results.<sup id="cite_ref-47" class="reference"><a href="#cite_note-47">[47]</a></sup>
</p><p>Many aspects of speech recognition were taken over by a deep learning method called <a href="/wiki/Long_short-term_memory" title="Long short-term memory">long short-term memory</a> (LSTM), a recurrent neural network published by Hochreiter and <a href="/wiki/J%C3%BCrgen_Schmidhuber" title="Jürgen Schmidhuber">Schmidhuber</a> in 1997.<sup id="cite_ref-:0_48-0" class="reference"><a href="#cite_note-:0-48">[48]</a></sup> LSTM <a href="/wiki/Recurrent_neural_network" title="Recurrent neural network">RNNs</a> avoid the vanishing gradient problem and can learn "Very Deep Learning" tasks<sup id="cite_ref-SCHIDHUB_5-3" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup> that require memories of events that happened thousands of discrete time steps before, which is important for speech. In 2003, LSTM started to become competitive with traditional speech recognizers on certain tasks.<sup id="cite_ref-graves2003_49-0" class="reference"><a href="#cite_note-graves2003-49">[49]</a></sup> Later it was combined with connectionist temporal classification (CTC)<sup id="cite_ref-:1_50-0" class="reference"><a href="#cite_note-:1-50">[50]</a></sup> in stacks of LSTM RNNs.<sup id="cite_ref-fernandez2007keyword_51-0" class="reference"><a href="#cite_note-fernandez2007keyword-51">[51]</a></sup> In 2015, Google's speech recognition reportedly experienced a dramatic performance jump of 49% through CTC-trained LSTM, which they made available through <a href="/wiki/Google_Voice_Search" title="Google Voice Search">Google Voice Search</a>.<sup id="cite_ref-sak2015_52-0" class="reference"><a href="#cite_note-sak2015-52">[52]</a></sup>
</p><p>In 2006, publications by <a href="/wiki/Geoffrey_Hinton" title="Geoffrey Hinton">Geoff Hinton</a>, <a href="/wiki/Russ_Salakhutdinov" title="Russ Salakhutdinov">Ruslan Salakhutdinov</a>, Osindero and <a href="/wiki/Yee_Whye_Teh" title="Yee Whye Teh">Teh</a><sup id="cite_ref-53" class="reference"><a href="#cite_note-53">[53]</a></sup><sup id="cite_ref-hinton06_54-0" class="reference"><a href="#cite_note-hinton06-54">[54]</a></sup><sup id="cite_ref-bengio2012_55-0" class="reference"><a href="#cite_note-bengio2012-55">[55]</a></sup> showed how a many-layered <a href="/wiki/Feedforward_neural_network" title="Feedforward neural network">feedforward neural network</a> could be effectively pre-trained one layer at a time, treating each layer in turn as an unsupervised <a href="/wiki/Restricted_Boltzmann_machine" title="Restricted Boltzmann machine">restricted Boltzmann machine</a>, then fine-tuning it using supervised <a href="/wiki/Backpropagation" title="Backpropagation">backpropagation</a>.<sup id="cite_ref-HINTON2007_56-0" class="reference"><a href="#cite_note-HINTON2007-56">[56]</a></sup> The papers referred to <i>learning</i> for <i>deep belief nets.</i>
</p><p>Deep learning is part of state-of-the-art systems in various disciplines, particularly computer vision and <a href="/wiki/Automatic_speech_recognition" class="mw-redirect" title="Automatic speech recognition">automatic speech recognition</a> (ASR). Results on commonly used evaluation sets such as <a href="/wiki/TIMIT" title="TIMIT">TIMIT</a> (ASR) and <a href="/wiki/MNIST_database" title="MNIST database">MNIST</a> (<a href="/wiki/Image_classification" class="mw-redirect" title="Image classification">image classification</a>), as well as a range of large-vocabulary speech recognition tasks have steadily improved.<sup id="cite_ref-HintonDengYu2012_57-0" class="reference"><a href="#cite_note-HintonDengYu2012-57">[57]</a></sup><sup id="cite_ref-58" class="reference"><a href="#cite_note-58">[58]</a></sup> <a href="/wiki/Convolutional_neural_network" title="Convolutional neural network">Convolutional neural networks</a> (CNNs) were superseded for ASR by CTC<sup id="cite_ref-:1_50-1" class="reference"><a href="#cite_note-:1-50">[50]</a></sup> for <a href="/wiki/LSTM" class="mw-redirect" title="LSTM">LSTM</a>.<sup id="cite_ref-:0_48-1" class="reference"><a href="#cite_note-:0-48">[48]</a></sup><sup id="cite_ref-sak2015_52-1" class="reference"><a href="#cite_note-sak2015-52">[52]</a></sup><sup id="cite_ref-sak2014_59-0" class="reference"><a href="#cite_note-sak2014-59">[59]</a></sup><sup id="cite_ref-liwu2015_60-0" class="reference"><a href="#cite_note-liwu2015-60">[60]</a></sup><sup id="cite_ref-zen2015_61-0" class="reference"><a href="#cite_note-zen2015-61">[61]</a></sup> but are more successful in computer vision.
</p><p>The impact of deep learning in industry began in the early 2000s, when CNNs already processed an estimated 10% to 20% of all the checks written in the US, according to Yann LeCun.<sup id="cite_ref-lecun2016slides_62-0" class="reference"><a href="#cite_note-lecun2016slides-62">[62]</a></sup> Industrial applications of deep learning to large-scale speech recognition started around 2010.
</p><p>The 2009 NIPS Workshop on Deep Learning for Speech Recognition was motivated by the limitations of deep generative models of speech, and the possibility that given more capable hardware and large-scale data sets that deep neural nets (DNN) might become practical. It was believed that pre-training DNNs using generative models of deep belief nets (DBN) would overcome the main difficulties of neural nets. However, it was discovered that replacing pre-training with large amounts of training data for straightforward <a href="/wiki/Backpropagation" title="Backpropagation">backpropagation</a> when using DNNs with large, context-dependent output layers produced error rates dramatically lower than then-state-of-the-art Gaussian mixture model (GMM)/Hidden Markov Model (HMM) and also than more-advanced generative model-based systems.<sup id="cite_ref-HintonDengYu2012_57-1" class="reference"><a href="#cite_note-HintonDengYu2012-57">[57]</a></sup> The nature of the recognition errors produced by the two types of systems was characteristically different,<sup id="cite_ref-ReferenceICASSP2013_63-0" class="reference"><a href="#cite_note-ReferenceICASSP2013-63">[63]</a></sup> offering technical insights into how to integrate deep learning into the existing highly efficient, run-time speech decoding system deployed by all major speech recognition systems.<sup id="cite_ref-BOOK2014_1-3" class="reference"><a href="#cite_note-BOOK2014-1">[1]</a></sup><sup id="cite_ref-ReferenceA_64-0" class="reference"><a href="#cite_note-ReferenceA-64">[64]</a></sup><sup id="cite_ref-65" class="reference"><a href="#cite_note-65">[65]</a></sup> Analysis around 2009–2010, contrasting the GMM (and other generative speech models) vs. DNN models, stimulated early industrial investment in deep learning for speech recognition,<sup id="cite_ref-ReferenceICASSP2013_63-1" class="reference"><a href="#cite_note-ReferenceICASSP2013-63">[63]</a></sup> eventually leading to pervasive and dominant use in that industry. That analysis was done with comparable performance (less than 1.5% in error rate) between discriminative DNNs and generative models.<sup id="cite_ref-HintonDengYu2012_57-2" class="reference"><a href="#cite_note-HintonDengYu2012-57">[57]</a></sup><sup id="cite_ref-ReferenceICASSP2013_63-2" class="reference"><a href="#cite_note-ReferenceICASSP2013-63">[63]</a></sup><sup id="cite_ref-interspeech2014Keynote_66-0" class="reference"><a href="#cite_note-interspeech2014Keynote-66">[66]</a></sup>
</p><p>In 2010, researchers extended deep learning from TIMIT to large vocabulary speech recognition, by adopting large output layers of the DNN based on context-dependent HMM states constructed by <a href="/wiki/Decision_tree" title="Decision tree">decision trees</a>.<sup id="cite_ref-Roles2010_67-0" class="reference"><a href="#cite_note-Roles2010-67">[67]</a></sup><sup id="cite_ref-68" class="reference"><a href="#cite_note-68">[68]</a></sup><sup id="cite_ref-69" class="reference"><a href="#cite_note-69">[69]</a></sup><sup id="cite_ref-ReferenceA_64-1" class="reference"><a href="#cite_note-ReferenceA-64">[64]</a></sup>
</p><p>Advances in hardware have driven renewed interest in deep learning. In 2009, <a href="/wiki/Nvidia" title="Nvidia">Nvidia</a> was involved in what was called the “big bang” of deep learning, “as deep-learning neural networks were trained with Nvidia <a href="/wiki/Graphics_processing_unit" title="Graphics processing unit">graphics processing units</a> (GPUs).”<sup id="cite_ref-70" class="reference"><a href="#cite_note-70">[70]</a></sup> That year, <a href="/wiki/Andrew_Ng" title="Andrew Ng">Andrew Ng</a> determined that GPUs could increase the speed of deep-learning systems by about 100 times.<sup id="cite_ref-71" class="reference"><a href="#cite_note-71">[71]</a></sup> In particular, GPUs are well-suited for the matrix/vector computations involved in machine learning.<sup id="cite_ref-jung2004_72-0" class="reference"><a href="#cite_note-jung2004-72">[72]</a></sup><sup id="cite_ref-73" class="reference"><a href="#cite_note-73">[73]</a></sup><sup id="cite_ref-chellapilla2006_74-0" class="reference"><a href="#cite_note-chellapilla2006-74">[74]</a></sup> GPUs speed up training algorithms by orders of magnitude, reducing running times from weeks to days.<sup id="cite_ref-:3_75-0" class="reference"><a href="#cite_note-:3-75">[75]</a></sup><sup id="cite_ref-76" class="reference"><a href="#cite_note-76">[76]</a></sup> Further, specialized hardware and algorithm optimizations can be used for efficient processing of deep learning models.<sup id="cite_ref-sze2017_77-0" class="reference"><a href="#cite_note-sze2017-77">[77]</a></sup>
</p>
<h3><span class="mw-headline" id="Deep_learning_revolution">Deep learning revolution</span></h3>
<div class="thumb tright"><div class="thumbinner" style="width:222px;"><a href="/wiki/File:AI-ML-DL.svg" class="image"><img src="//upload.wikimedia.org/wikipedia/commons/thumb/b/bb/AI-ML-DL.svg/220px-AI-ML-DL.svg.png" decoding="async" width="220" height="243" class="thumbimage" srcset="//upload.wikimedia.org/wikipedia/commons/thumb/b/bb/AI-ML-DL.svg/330px-AI-ML-DL.svg.png 1.5x, //upload.wikimedia.org/wikipedia/commons/thumb/b/bb/AI-ML-DL.svg/440px-AI-ML-DL.svg.png 2x" data-file-width="701" data-file-height="775" /></a> <div class="thumbcaption"><div class="magnify"><a href="/wiki/File:AI-ML-DL.svg" class="internal" title="Enlarge"></a></div>How deep learning is a subset of machine learning and how machine learning is a subset of artificial intelligence (AI)</div></div></div>
<p>In 2012, a team led by George E. Dahl won the "Merck Molecular Activity Challenge" using multi-task deep neural networks to predict the <a href="/wiki/Biomolecular_target" class="mw-redirect" title="Biomolecular target">biomolecular target</a> of one drug.<sup id="cite_ref-MERCK2012_78-0" class="reference"><a href="#cite_note-MERCK2012-78">[78]</a></sup><sup id="cite_ref-:5_79-0" class="reference"><a href="#cite_note-:5-79">[79]</a></sup> In 2014, Hochreiter's group used deep learning to detect off-target and toxic effects of environmental chemicals in nutrients, household products and drugs and won the "Tox21 Data Challenge" of <a href="/wiki/NIH" class="mw-redirect" title="NIH">NIH</a>, <a href="/wiki/FDA" class="mw-redirect" title="FDA">FDA</a> and <a href="/wiki/National_Center_for_Advancing_Translational_Sciences" title="National Center for Advancing Translational Sciences">NCATS</a>.<sup id="cite_ref-TOX21_80-0" class="reference"><a href="#cite_note-TOX21-80">[80]</a></sup><sup id="cite_ref-TOX21Data_81-0" class="reference"><a href="#cite_note-TOX21Data-81">[81]</a></sup><sup id="cite_ref-:11_82-0" class="reference"><a href="#cite_note-:11-82">[82]</a></sup>
</p><p>Significant additional impacts in image or object recognition were felt from 2011 to 2012. Although CNNs trained by <a href="/wiki/Backpropagation" title="Backpropagation">backpropagation</a> had been around for decades, and GPU implementations of NNs for years, including CNNs, fast implementations of CNNs on GPUs were needed to progress on computer vision.<sup id="cite_ref-jung2004_72-1" class="reference"><a href="#cite_note-jung2004-72">[72]</a></sup><sup id="cite_ref-chellapilla2006_74-1" class="reference"><a href="#cite_note-chellapilla2006-74">[74]</a></sup><sup id="cite_ref-LECUN1989_29-1" class="reference"><a href="#cite_note-LECUN1989-29">[29]</a></sup><sup id="cite_ref-:6_83-0" class="reference"><a href="#cite_note-:6-83">[83]</a></sup><sup id="cite_ref-SCHIDHUB_5-4" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup> In 2011, this approach achieved for the first time superhuman performance in a visual pattern recognition contest. Also in 2011, it won the ICDAR Chinese handwriting contest, and in May 2012, it won the ISBI image segmentation contest.<sup id="cite_ref-:8_84-0" class="reference"><a href="#cite_note-:8-84">[84]</a></sup> Until 2011, CNNs did not play a major role at computer vision conferences, but in June 2012, a paper by Ciresan et al. at the leading conference CVPR<sup id="cite_ref-:9_85-0" class="reference"><a href="#cite_note-:9-85">[85]</a></sup> showed how max-pooling CNNs on GPU can dramatically improve many vision benchmark records. In October 2012, a similar system by Krizhevsky et al.<sup id="cite_ref-krizhevsky2012_86-0" class="reference"><a href="#cite_note-krizhevsky2012-86">[86]</a></sup> won the large-scale <a href="/wiki/ImageNet_competition" class="mw-redirect" title="ImageNet competition">ImageNet competition</a> by a significant margin over shallow machine learning methods. In November 2012, Ciresan et al.'s system also won the ICPR contest on analysis of large medical images for cancer detection, and in the following year also the MICCAI Grand Challenge on the same topic.<sup id="cite_ref-ciresan2013miccai_87-0" class="reference"><a href="#cite_note-ciresan2013miccai-87">[87]</a></sup> In 2013 and 2014, the error rate on the ImageNet task using deep learning was further reduced, following a similar trend in large-scale speech recognition.
</p><p>Image classification was then extended to the more challenging task of <a href="/wiki/Automatic_image_annotation" title="Automatic image annotation">generating descriptions</a> (captions) for images, often as a combination of CNNs and LSTMs.<sup id="cite_ref-1411.4555_88-0" class="reference"><a href="#cite_note-1411.4555-88">[88]</a></sup><sup id="cite_ref-1411.4952_89-0" class="reference"><a href="#cite_note-1411.4952-89">[89]</a></sup><sup id="cite_ref-1411.2539_90-0" class="reference"><a href="#cite_note-1411.2539-90">[90]</a></sup>
</p><p>Some researchers state that the October 2012 ImageNet victory anchored the start of a "deep learning revolution" that has transformed the AI industry.<sup id="cite_ref-91" class="reference"><a href="#cite_note-91">[91]</a></sup>
</p><p>In March 2019, <a href="/wiki/Yoshua_Bengio" title="Yoshua Bengio">Yoshua Bengio</a>, <a href="/wiki/Geoffrey_Hinton" title="Geoffrey Hinton">Geoffrey Hinton</a> and <a href="/wiki/Yann_LeCun" title="Yann LeCun">Yann LeCun</a> were awarded the <a href="/wiki/Turing_Award" title="Turing Award">Turing Award</a> for conceptual and engineering breakthroughs that have made deep neural networks a critical component of computing.
</p>
<h2><span class="mw-headline" id="Neural_networks">Neural networks</span></h2>
<h3><span class="mw-headline" id="Artificial_neural_networks">Artificial neural networks</span></h3>
<style data-mw-deduplicate="TemplateStyles:r1033289096">.mw-parser-output .hatnote{font-style:italic}.mw-parser-output div.hatnote{padding-left:1.6em;margin-bottom:0.5em}.mw-parser-output .hatnote i{font-style:normal}.mw-parser-output .hatnote+link+.hatnote{margin-top:-0.5em}</style><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Artificial_neural_network" title="Artificial neural network">Artificial neural network</a></div>
<p><b>Artificial neural networks</b> (<b>ANNs</b>) or <b><a href="/wiki/Connectionism" title="Connectionism">connectionist</a> systems</b> are computing systems inspired by the <a href="/wiki/Biological_neural_network" class="mw-redirect" title="Biological neural network">biological neural networks</a> that constitute animal brains. Such systems learn (progressively improve their ability) to do tasks by considering examples, generally without task-specific programming. For example, in image recognition, they might learn to identify images that contain cats by analyzing example images that have been manually <a href="/wiki/Labeled_data" title="Labeled data">labeled</a> as "cat" or "no cat" and using the analytic results to identify cats in other images. They have found most use in applications difficult to express with a traditional computer algorithm using <a href="/wiki/Rule-based_programming" class="mw-redirect" title="Rule-based programming">rule-based programming</a>.
</p><p>An ANN is based on a collection of connected units called <a href="/wiki/Artificial_neuron" title="Artificial neuron">artificial neurons</a>, (analogous to biological neurons in a <a href="/wiki/Brain" title="Brain">biological brain</a>). Each connection (<a href="/wiki/Synapse" title="Synapse">synapse</a>) between neurons can transmit a signal to another neuron. The receiving (postsynaptic) neuron can process the signal(s) and then signal downstream neurons connected to it. Neurons may have state, generally represented by <a href="/wiki/Real_numbers" class="mw-redirect" title="Real numbers">real numbers</a>, typically between 0 and 1. Neurons and synapses may also have a weight that varies as learning proceeds, which can increase or decrease the strength of the signal that it sends downstream.
</p><p>Typically, neurons are organized in layers. Different layers may perform different kinds of transformations on their inputs. Signals travel from the first (input), to the last (output) layer, possibly after traversing the layers multiple times.
</p><p>The original goal of the neural network approach was to solve problems in the same way that a human brain would. Over time, attention focused on matching specific mental abilities, leading to deviations from biology such as <a href="/wiki/Backpropagation" title="Backpropagation">backpropagation</a>, or passing information in the reverse direction and adjusting the network to reflect that information.
</p><p>Neural networks have been used on a variety of tasks, including computer vision, <a href="/wiki/Speech_recognition" title="Speech recognition">speech recognition</a>, <a href="/wiki/Machine_translation" title="Machine translation">machine translation</a>, <a href="/wiki/Social_network" title="Social network">social network</a> filtering, <a href="/wiki/General_game_playing" title="General game playing">playing board and video games</a> and medical diagnosis.
</p><p>As of 2017, neural networks typically have a few thousand to a few million units and millions of connections. Despite this number being several order of magnitude less than the number of neurons on a human brain, these networks can perform many tasks at a level beyond that of humans (e.g., recognizing faces, or playing "Go"<sup id="cite_ref-92" class="reference"><a href="#cite_note-92">[92]</a></sup> ).
</p>
<h3><span class="mw-headline" id="Deep_neural_networks">Deep neural networks</span></h3>
<p>A deep neural network (DNN) is an <a href="/wiki/Artificial_neural_network" title="Artificial neural network">artificial neural network</a> (ANN) with multiple layers between the input and output layers.<sup id="cite_ref-BENGIODEEP_2-2" class="reference"><a href="#cite_note-BENGIODEEP-2">[2]</a></sup><sup id="cite_ref-SCHIDHUB_5-5" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup> There are different types of neural networks but they always consist of the same components: neurons, synapses, weights, biases, and functions.<sup id="cite_ref-Nokkada_93-0" class="reference"><a href="#cite_note-Nokkada-93">[93]</a></sup> These components as a whole function similarly to a human brain, and can be trained like any other ML algorithm.<sup class="noprint Inline-Template Template-Fact" style="white-space:nowrap;">[<i><a href="/wiki/Wikipedia:Citation_needed" title="Wikipedia:Citation needed"><span title="This claim needs references to reliable sources. (November 2020)">citation needed</span></a></i>]</sup>
</p><p>For example, a DNN that is trained to recognize dog breeds will go over the given image and calculate the probability that the dog in the image is a certain breed. The user can review the results and select which probabilities the network should display (above a certain threshold, etc.) and return the proposed label. Each mathematical manipulation as such is considered a layer,<sup class="noprint Inline-Template Template-Fact" style="white-space:nowrap;">[<i><a href="/wiki/Wikipedia:Citation_needed" title="Wikipedia:Citation needed"><span title="This claim needs references to reliable sources. (March 2022)">citation needed</span></a></i>]</sup> and complex DNN have many layers, hence the name "deep" networks.
</p><p>DNNs can model complex non-linear relationships. DNN architectures generate compositional models where the object is expressed as a layered composition of <a href="/wiki/Primitive_data_type" title="Primitive data type">primitives</a>.<sup id="cite_ref-94" class="reference"><a href="#cite_note-94">[94]</a></sup> The extra layers enable composition of features from lower layers, potentially modeling complex data with fewer units than a similarly performing shallow network.<sup id="cite_ref-BENGIODEEP_2-3" class="reference"><a href="#cite_note-BENGIODEEP-2">[2]</a></sup> For instance, it was proved that sparse <a href="/wiki/Multivariate_polynomial" class="mw-redirect" title="Multivariate polynomial">multivariate polynomials</a> are exponentially easier to approximate with DNNs than with shallow networks.<sup id="cite_ref-95" class="reference"><a href="#cite_note-95">[95]</a></sup>
</p><p>Deep architectures include many variants of a few basic approaches. Each architecture has found success in specific domains. It is not always possible to compare the performance of multiple architectures, unless they have been evaluated on the same data sets.
</p><p>DNNs are typically feedforward networks in which data flows from the input layer to the output layer without looping back. At first, the DNN creates a map of virtual neurons and assigns random numerical values, or "weights", to connections between them. The weights and inputs are multiplied and return an output between 0 and 1. If the network did not accurately recognize a particular pattern, an algorithm would adjust the weights.<sup id="cite_ref-96" class="reference"><a href="#cite_note-96">[96]</a></sup> That way the algorithm can make certain parameters more influential, until it determines the correct mathematical manipulation to fully process the data.
</p><p><a href="/wiki/Recurrent_neural_networks" class="mw-redirect" title="Recurrent neural networks">Recurrent neural networks</a> (RNNs), in which data can flow in any direction, are used for applications such as <a href="/wiki/Language_model" title="Language model">language modeling</a>.<sup id="cite_ref-gers2001_97-0" class="reference"><a href="#cite_note-gers2001-97">[97]</a></sup><sup id="cite_ref-NIPS2014_98-0" class="reference"><a href="#cite_note-NIPS2014-98">[98]</a></sup><sup id="cite_ref-vinyals2016_99-0" class="reference"><a href="#cite_note-vinyals2016-99">[99]</a></sup><sup id="cite_ref-gillick2015_100-0" class="reference"><a href="#cite_note-gillick2015-100">[100]</a></sup><sup id="cite_ref-MIKO2010_101-0" class="reference"><a href="#cite_note-MIKO2010-101">[101]</a></sup> Long short-term memory is particularly effective for this use.<sup id="cite_ref-:0_48-2" class="reference"><a href="#cite_note-:0-48">[48]</a></sup><sup id="cite_ref-:10_102-0" class="reference"><a href="#cite_note-:10-102">[102]</a></sup>
</p><p><a href="/wiki/Convolutional_neural_network" title="Convolutional neural network">Convolutional deep neural networks (CNNs)</a> are used in computer vision.<sup id="cite_ref-LECUN86_103-0" class="reference"><a href="#cite_note-LECUN86-103">[103]</a></sup> CNNs also have been applied to <a href="/wiki/Acoustic_model" title="Acoustic model">acoustic modeling</a> for automatic speech recognition (ASR).<sup id="cite_ref-:2_104-0" class="reference"><a href="#cite_note-:2-104">[104]</a></sup>
</p>
<h4><span class="mw-headline" id="Challenges">Challenges</span></h4>
<p>As with ANNs, many issues can arise with naively trained DNNs. Two common issues are <a href="/wiki/Overfitting" title="Overfitting">overfitting</a> and computation time.
</p><p>DNNs are prone to overfitting because of the added layers of abstraction, which allow them to model rare dependencies in the training data. <a href="/wiki/Regularization_(mathematics)" title="Regularization (mathematics)">Regularization</a> methods such as Ivakhnenko's unit pruning<sup id="cite_ref-ivak1971_20-1" class="reference"><a href="#cite_note-ivak1971-20">[20]</a></sup> or <a href="/wiki/Weight_decay" class="mw-redirect" title="Weight decay">weight decay</a> (<span class="mwe-math-element"><span class="mwe-math-mathml-inline mwe-math-mathml-a11y" style="display: none;"><math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \ell _{2}}">
<semantics>
<mrow class="MJX-TeXAtom-ORD">
<mstyle displaystyle="true" scriptlevel="0">
<msub>
<mi>ℓ<!-- ℓ --></mi>
<mrow class="MJX-TeXAtom-ORD">
<mn>2</mn>
</mrow>
</msub>
</mstyle>
</mrow>
<annotation encoding="application/x-tex">{\displaystyle \ell _{2}}</annotation>
</semantics>
</math></span><img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/85a4571ee9be10bd3c9df2480ab3d280f99e801a" class="mwe-math-fallback-image-inline" aria-hidden="true" style="vertical-align: -0.671ex; width:2.024ex; height:2.509ex;" alt="\ell _{2}"/></span>-regularization) or <a href="/wiki/Sparse_matrix" title="Sparse matrix">sparsity</a> (<span class="mwe-math-element"><span class="mwe-math-mathml-inline mwe-math-mathml-a11y" style="display: none;"><math xmlns="http://www.w3.org/1998/Math/MathML" alttext="{\displaystyle \ell _{1}}">
<semantics>
<mrow class="MJX-TeXAtom-ORD">
<mstyle displaystyle="true" scriptlevel="0">
<msub>
<mi>ℓ<!-- ℓ --></mi>
<mrow class="MJX-TeXAtom-ORD">
<mn>1</mn>
</mrow>
</msub>
</mstyle>
</mrow>
<annotation encoding="application/x-tex">{\displaystyle \ell _{1}}</annotation>
</semantics>
</math></span><img src="https://wikimedia.org/api/rest_v1/media/math/render/svg/361ddd720474aa41cb05453e03424fb7999d3b02" class="mwe-math-fallback-image-inline" aria-hidden="true" style="vertical-align: -0.671ex; width:2.024ex; height:2.509ex;" alt="\ell _{1}"/></span>-regularization) can be applied during training to combat overfitting.<sup id="cite_ref-105" class="reference"><a href="#cite_note-105">[105]</a></sup> Alternatively dropout regularization randomly omits units from the hidden layers during training. This helps to exclude rare dependencies.<sup id="cite_ref-DAHL2013_106-0" class="reference"><a href="#cite_note-DAHL2013-106">[106]</a></sup> Finally, data can be augmented via methods such as cropping and rotating such that smaller training sets can be increased in size to reduce the chances of overfitting.<sup id="cite_ref-107" class="reference"><a href="#cite_note-107">[107]</a></sup>
</p><p>DNNs must consider many training parameters, such as the size (number of layers and number of units per layer), the <a href="/wiki/Learning_rate" title="Learning rate">learning rate</a>, and initial weights. <a href="/wiki/Hyperparameter_optimization#Grid_search" title="Hyperparameter optimization">Sweeping through the parameter space</a> for optimal parameters may not be feasible due to the cost in time and computational resources. Various tricks, such as batching (computing the gradient on several training examples at once rather than individual examples)<sup id="cite_ref-RBMTRAIN_108-0" class="reference"><a href="#cite_note-RBMTRAIN-108">[108]</a></sup> speed up computation. Large processing capabilities of many-core architectures (such as GPUs or the Intel Xeon Phi) have produced significant speedups in training, because of the suitability of such processing architectures for the matrix and vector computations.<sup id="cite_ref-109" class="reference"><a href="#cite_note-109">[109]</a></sup><sup id="cite_ref-110" class="reference"><a href="#cite_note-110">[110]</a></sup>
</p><p>Alternatively, engineers may look for other types of neural networks with more straightforward and convergent training algorithms. CMAC (<a href="/wiki/Cerebellar_model_articulation_controller" title="Cerebellar model articulation controller">cerebellar model articulation controller</a>) is one such kind of neural network. It doesn't require learning rates or randomized initial weights for CMAC. The training process can be guaranteed to converge in one step with a new batch of data, and the computational complexity of the training algorithm is linear with respect to the number of neurons involved.<sup id="cite_ref-Qin1_111-0" class="reference"><a href="#cite_note-Qin1-111">[111]</a></sup><sup id="cite_ref-Qin2_112-0" class="reference"><a href="#cite_note-Qin2-112">[112]</a></sup>
</p>
<h2><span class="mw-headline" id="Hardware">Hardware</span></h2>
<p>Since the 2010s, advances in both machine learning algorithms and <a href="/wiki/Computer_hardware" title="Computer hardware">computer hardware</a> have led to more efficient methods for training deep neural networks that contain many layers of non-linear hidden units and a very large output layer.<sup id="cite_ref-113" class="reference"><a href="#cite_note-113">[113]</a></sup> By 2019, graphic processing units (<a href="/wiki/GPU" class="mw-redirect" title="GPU">GPUs</a>), often with AI-specific enhancements, had displaced CPUs as the dominant method of training large-scale commercial cloud AI.<sup id="cite_ref-114" class="reference"><a href="#cite_note-114">[114]</a></sup> <a href="/wiki/OpenAI" title="OpenAI">OpenAI</a> estimated the hardware computation used in the largest deep learning projects from AlexNet (2012) to AlphaZero (2017), and found a 300,000-fold increase in the amount of computation required, with a doubling-time trendline of 3.4 months.<sup id="cite_ref-115" class="reference"><a href="#cite_note-115">[115]</a></sup><sup id="cite_ref-116" class="reference"><a href="#cite_note-116">[116]</a></sup>
</p><p>Special <a href="/wiki/Electronic_circuit" title="Electronic circuit">electronic circuits</a> called <a href="/wiki/Deep_learning_processor" title="Deep learning processor">deep learning processors</a> were designed to speed up deep learning algorithms. Deep learning processors include neural processing units (NPUs) in <a href="/wiki/Huawei" title="Huawei">Huawei</a> cellphones<sup id="cite_ref-117" class="reference"><a href="#cite_note-117">[117]</a></sup> and <a href="/wiki/Cloud_computing" title="Cloud computing">cloud computing</a> servers such as <a href="/wiki/Tensor_processing_unit" class="mw-redirect" title="Tensor processing unit">tensor processing units</a> (TPU) in the <a href="/wiki/Google_Cloud_Platform" title="Google Cloud Platform">Google Cloud Platform</a>.<sup id="cite_ref-118" class="reference"><a href="#cite_note-118">[118]</a></sup> <a href="/wiki/Cerebras" title="Cerebras">Cerebras Systems</a> has also built a dedicated system to handle large deep learning models, the CS-2, based on the largest processor in the industry, the second-generation Wafer Scale Engine (WSE-2).<sup id="cite_ref-119" class="reference"><a href="#cite_note-119">[119]</a></sup><sup id="cite_ref-120" class="reference"><a href="#cite_note-120">[120]</a></sup>
</p><p>Atomically thin <a href="/wiki/Semiconductors" class="mw-redirect" title="Semiconductors">semiconductors</a> are considered promising for energy-efficient deep learning hardware where the same basic device structure is used for both logic operations and data storage.
In 2020, Marega et al. published experiments with a large-area active channel material for developing logic-in-memory devices and circuits based on <a href="/wiki/Floating-gate" class="mw-redirect" title="Floating-gate">floating-gate</a> <a href="/wiki/Field-effect_transistor" title="Field-effect transistor">field-effect transistors</a> (FGFETs).<sup id="cite_ref-atomthin_121-0" class="reference"><a href="#cite_note-atomthin-121">[121]</a></sup>
</p><p>In 2021, J. Feldmann et al. proposed an integrated <a href="/wiki/Photonic" class="mw-redirect" title="Photonic">photonic</a> <a href="/wiki/Hardware_accelerator" class="mw-redirect" title="Hardware accelerator">hardware accelerator</a> for parallel convolutional processing.<sup id="cite_ref-photonic_122-0" class="reference"><a href="#cite_note-photonic-122">[122]</a></sup> The authors identify two key advantages of integrated photonics over its electronic counterparts: (1) massively parallel data transfer through <a href="/wiki/Wavelength" title="Wavelength">wavelength</a> division <a href="/wiki/Multiplexing" title="Multiplexing">multiplexing</a> in conjunction with <a href="/wiki/Frequency_comb" title="Frequency comb">frequency combs</a>, and (2) extremely high data modulation speeds.<sup id="cite_ref-photonic_122-1" class="reference"><a href="#cite_note-photonic-122">[122]</a></sup> Their system can execute trillions of multiply-accumulate operations per second, indicating the potential of <a href="/wiki/Photonic_integrated_circuit" title="Photonic integrated circuit">integrated</a> <a href="/wiki/Photonics" title="Photonics">photonics</a> in data-heavy AI applications.<sup id="cite_ref-photonic_122-2" class="reference"><a href="#cite_note-photonic-122">[122]</a></sup>
</p>
<h2><span class="mw-headline" id="Applications">Applications</span></h2>
<h3><span class="mw-headline" id="Automatic_speech_recognition">Automatic speech recognition</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Speech_recognition" title="Speech recognition">Speech recognition</a></div>
<p>Large-scale automatic speech recognition is the first and most convincing successful case of deep learning. LSTM RNNs can learn "Very Deep Learning" tasks<sup id="cite_ref-SCHIDHUB_5-6" class="reference"><a href="#cite_note-SCHIDHUB-5">[5]</a></sup> that involve multi-second intervals containing speech events separated by thousands of discrete time steps, where one time step corresponds to about 10 ms. LSTM with forget gates<sup id="cite_ref-:10_102-1" class="reference"><a href="#cite_note-:10-102">[102]</a></sup> is competitive with traditional speech recognizers on certain tasks.<sup id="cite_ref-graves2003_49-1" class="reference"><a href="#cite_note-graves2003-49">[49]</a></sup>
</p><p>The initial success in speech recognition was based on small-scale recognition tasks based on TIMIT. The data set contains 630 speakers from eight major <a href="/wiki/Dialect" title="Dialect">dialects</a> of <a href="/wiki/American_English" title="American English">American English</a>, where each speaker reads 10 sentences.<sup id="cite_ref-LDCTIMIT_123-0" class="reference"><a href="#cite_note-LDCTIMIT-123">[123]</a></sup> Its small size lets many configurations be tried. More importantly, the TIMIT task concerns phone-sequence recognition, which, unlike word-sequence recognition, allows weak phone <a href="/wiki/Bigram" title="Bigram">bigram</a> language models. This lets the strength of the acoustic modeling aspects of speech recognition be more easily analyzed. The error rates listed below, including these early results and measured as percent phone error rates (PER), have been summarized since 1991.
</p>
<table class="wikitable">
<tbody><tr>
<th>Method</th>
<th>Percent phone<br />error rate (PER) (%)
</th></tr>
<tr>
<td>Randomly Initialized RNN<sup id="cite_ref-124" class="reference"><a href="#cite_note-124">[124]</a></sup></td>
<td>26.1
</td></tr>
<tr>
<td>Bayesian Triphone GMM-HMM</td>
<td>25.6
</td></tr>
<tr>
<td>Hidden Trajectory (Generative) Model</td>
<td>24.8
</td></tr>
<tr>
<td>Monophone Randomly Initialized DNN</td>
<td>23.4
</td></tr>
<tr>
<td>Monophone DBN-DNN</td>
<td>22.4
</td></tr>
<tr>
<td>Triphone GMM-HMM with BMMI Training</td>
<td>21.7
</td></tr>
<tr>
<td>Monophone DBN-DNN on fbank</td>
<td>20.7
</td></tr>
<tr>
<td>Convolutional DNN<sup id="cite_ref-CNN-2014_125-0" class="reference"><a href="#cite_note-CNN-2014-125">[125]</a></sup></td>
<td>20.0
</td></tr>
<tr>
<td>Convolutional DNN w. Heterogeneous Pooling</td>
<td>18.7
</td></tr>
<tr>
<td>Ensemble DNN/CNN/RNN<sup id="cite_ref-EnsembleDL_126-0" class="reference"><a href="#cite_note-EnsembleDL-126">[126]</a></sup></td>
<td>18.3
</td></tr>
<tr>
<td>Bidirectional LSTM</td>
<td>17.8
</td></tr>
<tr>
<td>Hierarchical Convolutional Deep Maxout Network<sup id="cite_ref-HCDMM_127-0" class="reference"><a href="#cite_note-HCDMM-127">[127]</a></sup></td>
<td>16.5
</td></tr></tbody></table>
<p>The debut of DNNs for speaker recognition in the late 1990s and speech recognition around 2009-2011 and of LSTM around 2003–2007, accelerated progress in eight major areas:<sup id="cite_ref-BOOK2014_1-4" class="reference"><a href="#cite_note-BOOK2014-1">[1]</a></sup><sup id="cite_ref-interspeech2014Keynote_66-1" class="reference"><a href="#cite_note-interspeech2014Keynote-66">[66]</a></sup><sup id="cite_ref-ReferenceA_64-2" class="reference"><a href="#cite_note-ReferenceA-64">[64]</a></sup>
</p>
<ul><li>Scale-up/out and accelerated DNN training and decoding</li>
<li>Sequence discriminative training</li>
<li>Feature processing by deep models with solid understanding of the underlying mechanisms</li>
<li>Adaptation of DNNs and related deep models</li>
<li><a href="/wiki/Multi-task_learning" title="Multi-task learning">Multi-task</a> and <a href="/wiki/Transfer_learning" title="Transfer learning">transfer learning</a> by DNNs and related deep models</li>
<li><a href="/wiki/Convolutional_neural_network" title="Convolutional neural network">CNNs</a> and how to design them to best exploit <a href="/wiki/Domain_knowledge" title="Domain knowledge">domain knowledge</a> of speech</li>
<li><a href="/wiki/Recurrent_neural_network" title="Recurrent neural network">RNN</a> and its rich LSTM variants</li>
<li>Other types of deep models including tensor-based models and integrated deep generative/discriminative models.</li></ul>
<p>All major commercial speech recognition systems (e.g., Microsoft <a href="/wiki/Cortana_(software)" class="mw-redirect" title="Cortana (software)">Cortana</a>, <a href="/wiki/Xbox" title="Xbox">Xbox</a>, <a href="/wiki/Skype_Translator" title="Skype Translator">Skype Translator</a>, <a href="/wiki/Amazon_Alexa" title="Amazon Alexa">Amazon Alexa</a>, <a href="/wiki/Google_Now" title="Google Now">Google Now</a>, <a href="/wiki/Siri" title="Siri">Apple Siri</a>, <a href="/wiki/Baidu" title="Baidu">Baidu</a> and <a href="/wiki/IFlytek" title="IFlytek">iFlyTek</a> voice search, and a range of <a href="/wiki/Nuance_Communications" title="Nuance Communications">Nuance</a> speech products, etc.) are based on deep learning.<sup id="cite_ref-BOOK2014_1-5" class="reference"><a href="#cite_note-BOOK2014-1">[1]</a></sup><sup id="cite_ref-128" class="reference"><a href="#cite_note-128">[128]</a></sup><sup id="cite_ref-Baidu_129-0" class="reference"><a href="#cite_note-Baidu-129">[129]</a></sup>
</p>
<h3><span class="mw-headline" id="Image_recognition">Image recognition</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Computer_vision" title="Computer vision">Computer vision</a></div>
<p>A common evaluation set for image classification is the MNIST database data set. MNIST is composed of handwritten digits and includes 60,000 training examples and 10,000 test examples. As with TIMIT, its small size lets users test multiple configurations. A comprehensive list of results on this set is available.<sup id="cite_ref-YANNMNIST_130-0" class="reference"><a href="#cite_note-YANNMNIST-130">[130]</a></sup>
</p><p>Deep learning-based image recognition has become "superhuman", producing more accurate results than human contestants. This first occurred in 2011 in recognition of traffic signs, and in 2014, with recognition of human faces.<sup id="cite_ref-:7_131-0" class="reference"><a href="#cite_note-:7-131">[131]</a></sup><sup id="cite_ref-surpass1_132-0" class="reference"><a href="#cite_note-surpass1-132">[132]</a></sup>
</p><p>Deep learning-trained vehicles now interpret 360° camera views.<sup id="cite_ref-133" class="reference"><a href="#cite_note-133">[133]</a></sup> Another example is Facial Dysmorphology Novel Analysis (FDNA) used to analyze cases of human malformation connected to a large database of genetic syndromes.
</p>
<h3><span class="mw-headline" id="Visual_art_processing">Visual art processing</span></h3>
<p>Closely related to the progress that has been made in image recognition is the increasing application of deep learning techniques to various visual art tasks. DNNs have proven themselves capable, for example, of
</p>
<ul><li>identifying the style period of a given painting<sup id="cite_ref-art1_134-0" class="reference"><a href="#cite_note-art1-134">[134]</a></sup><sup id="cite_ref-art2_135-0" class="reference"><a href="#cite_note-art2-135">[135]</a></sup></li>
<li><a href="/wiki/Neural_Style_Transfer" class="mw-redirect" title="Neural Style Transfer">Neural Style Transfer</a> –  capturing the style of a given artwork and applying it in a visually pleasing manner to an arbitrary photograph or video<sup id="cite_ref-art1_134-1" class="reference"><a href="#cite_note-art1-134">[134]</a></sup><sup id="cite_ref-art2_135-1" class="reference"><a href="#cite_note-art2-135">[135]</a></sup></li>
<li>generating striking imagery based on random visual input fields.<sup id="cite_ref-art1_134-2" class="reference"><a href="#cite_note-art1-134">[134]</a></sup><sup id="cite_ref-art2_135-2" class="reference"><a href="#cite_note-art2-135">[135]</a></sup></li></ul>
<h3><span class="mw-headline" id="Natural_language_processing">Natural language processing</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Natural_language_processing" title="Natural language processing">Natural language processing</a></div>
<p>Neural networks have been used for implementing language models since the early 2000s.<sup id="cite_ref-gers2001_97-1" class="reference"><a href="#cite_note-gers2001-97">[97]</a></sup> LSTM helped to improve machine translation and language modeling.<sup id="cite_ref-NIPS2014_98-1" class="reference"><a href="#cite_note-NIPS2014-98">[98]</a></sup><sup id="cite_ref-vinyals2016_99-1" class="reference"><a href="#cite_note-vinyals2016-99">[99]</a></sup><sup id="cite_ref-gillick2015_100-1" class="reference"><a href="#cite_note-gillick2015-100">[100]</a></sup>
</p><p>Other key techniques in this field are negative sampling<sup id="cite_ref-GoldbergLevy2014_136-0" class="reference"><a href="#cite_note-GoldbergLevy2014-136">[136]</a></sup> and <a href="/wiki/Word_embedding" title="Word embedding">word embedding</a>. Word embedding, such as <i><a href="/wiki/Word2vec" title="Word2vec">word2vec</a></i>, can be thought of as a representational layer in a deep learning architecture that transforms an atomic word into a positional representation of the word relative to other words in the dataset; the position is represented as a point in a <a href="/wiki/Vector_space" title="Vector space">vector space</a>. Using word embedding as an RNN input layer allows the network to parse sentences and phrases using an effective compositional vector grammar. A compositional vector grammar can be thought of as <a href="/wiki/Probabilistic_context_free_grammar" class="mw-redirect" title="Probabilistic context free grammar">probabilistic context free grammar</a> (PCFG) implemented by an RNN.<sup id="cite_ref-SocherManning2014_137-0" class="reference"><a href="#cite_note-SocherManning2014-137">[137]</a></sup> Recursive auto-encoders built atop word embeddings can assess sentence similarity and detect paraphrasing.<sup id="cite_ref-SocherManning2014_137-1" class="reference"><a href="#cite_note-SocherManning2014-137">[137]</a></sup> Deep neural architectures provide the best results for <a href="/wiki/Statistical_parsing" title="Statistical parsing">constituency parsing</a>,<sup id="cite_ref-138" class="reference"><a href="#cite_note-138">[138]</a></sup> <a href="/wiki/Sentiment_analysis" title="Sentiment analysis">sentiment analysis</a>,<sup id="cite_ref-139" class="reference"><a href="#cite_note-139">[139]</a></sup> information retrieval,<sup id="cite_ref-140" class="reference"><a href="#cite_note-140">[140]</a></sup><sup id="cite_ref-141" class="reference"><a href="#cite_note-141">[141]</a></sup> spoken language understanding,<sup id="cite_ref-IEEE-TASL2015_142-0" class="reference"><a href="#cite_note-IEEE-TASL2015-142">[142]</a></sup> machine translation,<sup id="cite_ref-NIPS2014_98-2" class="reference"><a href="#cite_note-NIPS2014-98">[98]</a></sup><sup id="cite_ref-auto_143-0" class="reference"><a href="#cite_note-auto-143">[143]</a></sup> contextual entity linking,<sup id="cite_ref-auto_143-1" class="reference"><a href="#cite_note-auto-143">[143]</a></sup> writing style recognition,<sup id="cite_ref-BROC2017_144-0" class="reference"><a href="#cite_note-BROC2017-144">[144]</a></sup> Text classification and others.<sup id="cite_ref-145" class="reference"><a href="#cite_note-145">[145]</a></sup>
</p><p>Recent developments generalize <a href="/wiki/Word_embedding" title="Word embedding">word embedding</a> to <a href="/wiki/Sentence_embedding" title="Sentence embedding">sentence embedding</a>.
</p><p><a href="/wiki/Google_Translate" title="Google Translate">Google Translate</a> (GT) uses a large end-to-end <a href="/wiki/Long_short-term_memory" title="Long short-term memory">long short-term memory</a> (LSTM) network.<sup id="cite_ref-GT_Turovsky_2016_146-0" class="reference"><a href="#cite_note-GT_Turovsky_2016-146">[146]</a></sup><sup id="cite_ref-googleblog_GNMT_2016_147-0" class="reference"><a href="#cite_note-googleblog_GNMT_2016-147">[147]</a></sup><sup id="cite_ref-GoogleTranslate_148-0" class="reference"><a href="#cite_note-GoogleTranslate-148">[148]</a></sup><sup id="cite_ref-WiredGoogleTranslate_149-0" class="reference"><a href="#cite_note-WiredGoogleTranslate-149">[149]</a></sup> <a href="/wiki/Google_Neural_Machine_Translation" title="Google Neural Machine Translation">Google Neural Machine Translation (GNMT)</a> uses an <a href="/wiki/Example-based_machine_translation" title="Example-based machine translation">example-based machine translation</a> method in which the system "learns from millions of examples."<sup id="cite_ref-googleblog_GNMT_2016_147-1" class="reference"><a href="#cite_note-googleblog_GNMT_2016-147">[147]</a></sup> It translates "whole sentences at a time, rather than pieces. Google Translate supports over one hundred languages.<sup id="cite_ref-googleblog_GNMT_2016_147-2" class="reference"><a href="#cite_note-googleblog_GNMT_2016-147">[147]</a></sup> The network encodes the "semantics of the sentence rather than simply memorizing phrase-to-phrase translations".<sup id="cite_ref-googleblog_GNMT_2016_147-3" class="reference"><a href="#cite_note-googleblog_GNMT_2016-147">[147]</a></sup><sup id="cite_ref-Biotet_150-0" class="reference"><a href="#cite_note-Biotet-150">[150]</a></sup> GT uses English as an intermediate between most language pairs.<sup id="cite_ref-Biotet_150-1" class="reference"><a href="#cite_note-Biotet-150">[150]</a></sup>
</p>
<h3><span class="mw-headline" id="Drug_discovery_and_toxicology">Drug discovery and toxicology</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">For more information, see <a href="/wiki/Drug_discovery" title="Drug discovery">Drug discovery</a> and <a href="/wiki/Toxicology" title="Toxicology">Toxicology</a>.</div>
<p>A large percentage of candidate drugs fail to win regulatory approval. These failures are caused by insufficient efficacy (on-target effect), undesired interactions (off-target effects), or unanticipated <a href="/wiki/Toxicity" title="Toxicity">toxic effects</a>.<sup id="cite_ref-ARROWSMITH2013_151-0" class="reference"><a href="#cite_note-ARROWSMITH2013-151">[151]</a></sup><sup id="cite_ref-VERBIEST2015_152-0" class="reference"><a href="#cite_note-VERBIEST2015-152">[152]</a></sup> Research has explored use of deep learning to predict the <a href="/wiki/Biomolecular_target" class="mw-redirect" title="Biomolecular target">biomolecular targets</a>,<sup id="cite_ref-MERCK2012_78-1" class="reference"><a href="#cite_note-MERCK2012-78">[78]</a></sup><sup id="cite_ref-:5_79-1" class="reference"><a href="#cite_note-:5-79">[79]</a></sup> <a href="/wiki/Off-target" class="mw-redirect" title="Off-target">off-targets</a>, and <a href="/wiki/Toxicity" title="Toxicity">toxic effects</a> of environmental chemicals in nutrients, household products and drugs.<sup id="cite_ref-TOX21_80-1" class="reference"><a href="#cite_note-TOX21-80">[80]</a></sup><sup id="cite_ref-TOX21Data_81-1" class="reference"><a href="#cite_note-TOX21Data-81">[81]</a></sup><sup id="cite_ref-:11_82-1" class="reference"><a href="#cite_note-:11-82">[82]</a></sup>
</p><p>AtomNet is a deep learning system for structure-based <a href="/wiki/Drug_design" title="Drug design">rational drug design</a>.<sup id="cite_ref-153" class="reference"><a href="#cite_note-153">[153]</a></sup> AtomNet was used to predict novel candidate biomolecules for disease targets such as the <a href="/wiki/Ebola_virus" class="mw-redirect" title="Ebola virus">Ebola virus</a><sup id="cite_ref-154" class="reference"><a href="#cite_note-154">[154]</a></sup> and <a href="/wiki/Multiple_sclerosis" title="Multiple sclerosis">multiple sclerosis</a>.<sup id="cite_ref-155" class="reference"><a href="#cite_note-155">[155]</a></sup><sup id="cite_ref-156" class="reference"><a href="#cite_note-156">[156]</a></sup>
</p><p>In 2017 <a href="/wiki/Graph_neural_network" title="Graph neural network">graph neural networks</a> were used for the first time to predict various properties of molecules in a large toxicology data set.<sup id="cite_ref-157" class="reference"><a href="#cite_note-157">[157]</a></sup> In 2019, generative neural networks were used to produce molecules that were validated experimentally all the way into mice.<sup id="cite_ref-158" class="reference"><a href="#cite_note-158">[158]</a></sup><sup id="cite_ref-159" class="reference"><a href="#cite_note-159">[159]</a></sup>
</p>
<h3><span class="mw-headline" id="Customer_relationship_management">Customer relationship management</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Customer_relationship_management" title="Customer relationship management">Customer relationship management</a></div>
<p><a href="/wiki/Deep_reinforcement_learning" title="Deep reinforcement learning">Deep reinforcement learning</a> has been used to approximate the value of possible <a href="/wiki/Direct_marketing" title="Direct marketing">direct marketing</a> actions, defined in terms of <a href="/wiki/RFM_(customer_value)" class="mw-redirect" title="RFM (customer value)">RFM</a> variables. The estimated value function was shown to have a natural interpretation as <a href="/wiki/Customer_lifetime_value" title="Customer lifetime value">customer lifetime value</a>.<sup id="cite_ref-160" class="reference"><a href="#cite_note-160">[160]</a></sup>
</p>
<h3><span class="mw-headline" id="Recommendation_systems">Recommendation systems</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Recommender_system" title="Recommender system">Recommender system</a></div>
<p>Recommendation systems have used deep learning to extract meaningful features for a latent factor model for content-based music and journal recommendations.<sup id="cite_ref-161" class="reference"><a href="#cite_note-161">[161]</a></sup><sup id="cite_ref-162" class="reference"><a href="#cite_note-162">[162]</a></sup> Multi-view deep learning has been applied for learning user preferences from multiple domains.<sup id="cite_ref-163" class="reference"><a href="#cite_note-163">[163]</a></sup> The model uses a hybrid collaborative and content-based approach and enhances recommendations in multiple tasks.
</p>
<h3><span class="mw-headline" id="Bioinformatics">Bioinformatics</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">Main article: <a href="/wiki/Bioinformatics" title="Bioinformatics">Bioinformatics</a></div>
<p>An <a href="/wiki/Autoencoder" title="Autoencoder">autoencoder</a> ANN was used in <a href="/wiki/Bioinformatics" title="Bioinformatics">bioinformatics</a>, to predict <a href="/wiki/Gene_Ontology" title="Gene Ontology">gene ontology</a> annotations and gene-function relationships.<sup id="cite_ref-164" class="reference"><a href="#cite_note-164">[164]</a></sup>
</p><p>In medical informatics, deep learning was used to predict sleep quality based on data from wearables<sup id="cite_ref-165" class="reference"><a href="#cite_note-165">[165]</a></sup> and predictions of health complications from <a href="/wiki/Electronic_health_record" title="Electronic health record">electronic health record</a> data.<sup id="cite_ref-166" class="reference"><a href="#cite_note-166">[166]</a></sup>
</p>
<h3><span class="mw-headline" id="Medical_image_analysis">Medical image analysis</span></h3>
<p>Deep learning has been shown to produce competitive results in medical application such as cancer cell classification, lesion detection, organ segmentation and image enhancement.<sup id="cite_ref-167" class="reference"><a href="#cite_note-167">[167]</a></sup><sup id="cite_ref-168" class="reference"><a href="#cite_note-168">[168]</a></sup> Modern deep learning tools demonstrate the high accuracy of detecting various diseases and the helpfulness of their use by specialists to improve the diagnosis efficiency.<sup id="cite_ref-169" class="reference"><a href="#cite_note-169">[169]</a></sup><sup id="cite_ref-170" class="reference"><a href="#cite_note-170">[170]</a></sup>
</p>
<h3><span class="mw-headline" id="Mobile_advertising">Mobile advertising</span></h3>
<p>Finding the appropriate mobile audience for <a href="/wiki/Mobile_advertising" title="Mobile advertising">mobile advertising</a> is always challenging, since many data points must be considered and analyzed before a target segment can be created and used in ad serving by any ad server.<sup id="cite_ref-171" class="reference"><a href="#cite_note-171">[171]</a></sup> Deep learning has been used to interpret large, many-dimensioned advertising datasets. Many data points are collected during the request/serve/click internet advertising cycle. This information can form the basis of machine learning to improve ad selection.
</p>
<h3><span class="mw-headline" id="Image_restoration">Image restoration</span></h3>
<p>Deep learning has been successfully applied to <a href="/wiki/Inverse_problems" class="mw-redirect" title="Inverse problems">inverse problems</a> such as <a href="/wiki/Denoising" class="mw-redirect" title="Denoising">denoising</a>, <a href="/wiki/Super-resolution" class="mw-redirect" title="Super-resolution">super-resolution</a>, <a href="/wiki/Inpainting" title="Inpainting">inpainting</a>, and <a href="/wiki/Film_colorization" title="Film colorization">film colorization</a>.<sup id="cite_ref-172" class="reference"><a href="#cite_note-172">[172]</a></sup> These applications include learning methods such as "Shrinkage Fields for Effective Image Restoration"<sup id="cite_ref-173" class="reference"><a href="#cite_note-173">[173]</a></sup> which trains on an image dataset, and <a href="/wiki/Deep_Image_Prior" class="mw-redirect" title="Deep Image Prior">Deep Image Prior</a>, which trains on the image that needs restoration.
</p>
<h3><span class="mw-headline" id="Financial_fraud_detection">Financial fraud detection</span></h3>
<p>Deep learning is being successfully applied to financial <a href="/wiki/Fraud_detection" class="mw-redirect" title="Fraud detection">fraud detection</a>, tax evasion detection,<sup id="cite_ref-174" class="reference"><a href="#cite_note-174">[174]</a></sup> and anti-money laundering.<sup id="cite_ref-175" class="reference"><a href="#cite_note-175">[175]</a></sup>
</p>
<h3><span class="mw-headline" id="Military">Military</span></h3>
<p>The United States Department of Defense applied deep learning to train robots in new tasks through observation.<sup id="cite_ref-:12_176-0" class="reference"><a href="#cite_note-:12-176">[176]</a></sup>
</p>
<h3><span class="mw-headline" id="Partial_differential_equations">Partial differential equations</span></h3>
<p>Physics informed neural networks have been used to solve <a href="/wiki/Partial_differential_equation" title="Partial differential equation">partial differential equations</a> in both forward and inverse problems in a data driven manner.<sup id="cite_ref-177" class="reference"><a href="#cite_note-177">[177]</a></sup> One example is the reconstructing fluid flow governed by the <a href="/wiki/Navier%E2%80%93Stokes_equations" title="Navier–Stokes equations">Navier-Stokes equations</a>. Using physics informed neural networks does not require the often expensive mesh generation that conventional <a href="/wiki/Computational_fluid_dynamics" title="Computational fluid dynamics">CFD</a> methods relies on.<sup id="cite_ref-178" class="reference"><a href="#cite_note-178">[178]</a></sup><sup id="cite_ref-179" class="reference"><a href="#cite_note-179">[179]</a></sup>
</p>
<h3><span class="mw-headline" id="Image_Reconstruction">Image Reconstruction</span></h3>
<p>Image reconstruction is the reconstruction of the underlying images from the image-related measurements. Several works showed the better and superior performance of the deep learning methods compared to analytical methods for various applications, e.g., spectral imaging <sup id="cite_ref-180" class="reference"><a href="#cite_note-180">[180]</a></sup> and ultrasound imaging.<sup id="cite_ref-181" class="reference"><a href="#cite_note-181">[181]</a></sup>
</p><p><br />
<big><b>Epigenetic clock</b></big>
</p><p>For more information, see <a href="/wiki/Ageing_clock" class="mw-redirect" title="Ageing clock">Epigenetic clock</a>.
</p><p>An <b>epigenetic clock</b> is a <a href="/wiki/Biomarkers_of_aging" title="Biomarkers of aging">biochemical test</a> that can be used to measure age. Galkin et al. used deep <a href="/wiki/Neural_network" title="Neural network">neural networks</a> to train an epigenetic aging clock of unprecedented accuracy using >6,000 blood samples. The clock uses information from 1000 CpG sites and predicts people with certain conditions older than healthy controls: <a href="/wiki/Inflammatory_bowel_disease" title="Inflammatory bowel disease">IBD</a>, <a href="/wiki/Dementia" title="Dementia">frontotemporal dementia</a>, ovarian cancer, obesity. The aging clock is planned to be released for public use in 2021 by an <a href="/wiki/Insilico_Medicine" title="Insilico Medicine">Insilico Medicine</a> spinoff company Deep Longevity.
</p>
<h2><span class="mw-headline" id="Relation_to_human_cognitive_and_brain_development">Relation to human cognitive and brain development</span></h2>
<p>Deep learning is closely related to a class of theories of <a href="/wiki/Brain_development" class="mw-redirect" title="Brain development">brain development</a> (specifically, neocortical development) proposed by <a href="/wiki/Cognitive_neuroscientist" class="mw-redirect" title="Cognitive neuroscientist">cognitive neuroscientists</a> in the early 1990s.<sup id="cite_ref-UTGOFF_182-0" class="reference"><a href="#cite_note-UTGOFF-182">[182]</a></sup><sup id="cite_ref-ELMAN_183-0" class="reference"><a href="#cite_note-ELMAN-183">[183]</a></sup><sup id="cite_ref-SHRAGER_184-0" class="reference"><a href="#cite_note-SHRAGER-184">[184]</a></sup><sup id="cite_ref-QUARTZ_185-0" class="reference"><a href="#cite_note-QUARTZ-185">[185]</a></sup> These developmental theories were instantiated in computational models, making them predecessors of deep learning systems. These developmental models share the property that various proposed learning dynamics in the brain (e.g., a wave of <a href="/wiki/Nerve_growth_factor" title="Nerve growth factor">nerve growth factor</a>) support the <a href="/wiki/Self-organization" title="Self-organization">self-organization</a> somewhat analogous to the neural networks utilized in deep learning models. Like the <a href="/wiki/Neocortex" title="Neocortex">neocortex</a>, neural networks employ a hierarchy of layered filters in which each layer considers information from a prior layer (or the operating environment), and then passes its output (and possibly the original input), to other layers. This process yields a self-organizing stack of <a href="/wiki/Transducer" title="Transducer">transducers</a>, well-tuned to their operating environment. A 1995 description stated, "...the infant's brain seems to organize itself under the influence of waves of so-called trophic-factors ... different regions of the brain become connected sequentially, with one layer of tissue maturing before another and so on until the whole brain is mature."<sup id="cite_ref-BLAKESLEE_186-0" class="reference"><a href="#cite_note-BLAKESLEE-186">[186]</a></sup>
</p><p>A variety of approaches have been used to investigate the plausibility of deep learning models from a neurobiological perspective. On the one hand, several variants of the <a href="/wiki/Backpropagation" title="Backpropagation">backpropagation</a> algorithm have been proposed in order to increase its processing realism.<sup id="cite_ref-187" class="reference"><a href="#cite_note-187">[187]</a></sup><sup id="cite_ref-188" class="reference"><a href="#cite_note-188">[188]</a></sup> Other researchers have argued that unsupervised forms of deep learning, such as those based on hierarchical <a href="/wiki/Generative_model" title="Generative model">generative models</a> and <a href="/wiki/Deep_belief_network" title="Deep belief network">deep belief networks</a>, may be closer to biological reality.<sup id="cite_ref-189" class="reference"><a href="#cite_note-189">[189]</a></sup><sup id="cite_ref-190" class="reference"><a href="#cite_note-190">[190]</a></sup> In this respect, generative neural network models have been related to neurobiological evidence about sampling-based processing in the cerebral cortex.<sup id="cite_ref-191" class="reference"><a href="#cite_note-191">[191]</a></sup>
</p><p>Although a systematic comparison between the human brain organization and the neuronal encoding in deep networks has not yet been established, several analogies have been reported. For example, the computations performed by deep learning units could be similar to those of actual neurons<sup id="cite_ref-192" class="reference"><a href="#cite_note-192">[192]</a></sup> and neural populations.<sup id="cite_ref-193" class="reference"><a href="#cite_note-193">[193]</a></sup> Similarly, the representations developed by deep learning models are similar to those measured in the primate visual system<sup id="cite_ref-194" class="reference"><a href="#cite_note-194">[194]</a></sup> both at the single-unit<sup id="cite_ref-195" class="reference"><a href="#cite_note-195">[195]</a></sup> and at the population<sup id="cite_ref-196" class="reference"><a href="#cite_note-196">[196]</a></sup> levels.
</p>
<h2><span class="mw-headline" id="Commercial_activity">Commercial activity</span></h2>
<p><a href="/wiki/Facebook" title="Facebook">Facebook</a>'s AI lab performs tasks such as <a href="/wiki/Automatic_image_annotation" title="Automatic image annotation">automatically tagging uploaded pictures</a> with the names of the people in them.<sup id="cite_ref-METZ2013_197-0" class="reference"><a href="#cite_note-METZ2013-197">[197]</a></sup>
</p><p>Google's <a href="/wiki/DeepMind_Technologies" class="mw-redirect" title="DeepMind Technologies">DeepMind Technologies</a> developed a system capable of learning how to play <a href="/wiki/Atari" title="Atari">Atari</a> video games using only pixels as data input. In 2015 they demonstrated their <a href="/wiki/AlphaGo" title="AlphaGo">AlphaGo</a> system, which learned the game of <a href="/wiki/Go_(game)" title="Go (game)">Go</a> well enough to beat a professional Go player.<sup id="cite_ref-198" class="reference"><a href="#cite_note-198">[198]</a></sup><sup id="cite_ref-199" class="reference"><a href="#cite_note-199">[199]</a></sup><sup id="cite_ref-200" class="reference"><a href="#cite_note-200">[200]</a></sup> <a href="/wiki/Google_Translate" title="Google Translate">Google Translate</a> uses a neural network to translate between more than 100 languages.
</p><p>In 2017, Covariant.ai was launched, which focuses on integrating deep learning into factories.<sup id="cite_ref-201" class="reference"><a href="#cite_note-201">[201]</a></sup>
</p><p>As of 2008,<sup id="cite_ref-202" class="reference"><a href="#cite_note-202">[202]</a></sup> researchers at <a href="/wiki/University_of_Texas_at_Austin" title="University of Texas at Austin">The University of Texas at Austin</a> (UT) developed a machine learning framework called Training an Agent Manually via Evaluative Reinforcement, or TAMER, which proposed new methods for robots or computer programs to learn how to perform tasks by interacting with a human instructor.<sup id="cite_ref-:12_176-1" class="reference"><a href="#cite_note-:12-176">[176]</a></sup> First developed as TAMER, a new algorithm called Deep TAMER was later introduced in 2018 during a collaboration between <a href="/wiki/U.S._Army_Research_Laboratory" class="mw-redirect" title="U.S. Army Research Laboratory">U.S. Army Research Laboratory</a> (ARL) and UT researchers. Deep TAMER used deep learning to provide a robot the ability to learn new tasks through observation.<sup id="cite_ref-:12_176-2" class="reference"><a href="#cite_note-:12-176">[176]</a></sup> Using Deep TAMER, a robot learned a task with a human trainer, watching video streams or observing a human perform a task in-person. The robot later practiced the task with the help of some coaching from the trainer, who provided feedback such as “good job” and “bad job.”<sup id="cite_ref-203" class="reference"><a href="#cite_note-203">[203]</a></sup>
</p>
<h2><span class="mw-headline" id="Criticism_and_comment">Criticism and comment</span></h2>
<p>Deep learning has attracted both criticism and comment, in some cases from outside the field of computer science.
</p>
<h3><span class="mw-headline" id="Theory">Theory</span></h3>
<link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1033289096"/><div role="note" class="hatnote navigation-not-searchable">See also: <a href="/wiki/Explainable_artificial_intelligence" title="Explainable artificial intelligence">Explainable artificial intelligence</a></div>
<p>A main criticism concerns the lack of theory surrounding some methods.<sup id="cite_ref-204" class="reference"><a href="#cite_note-204">[204]</a></sup> Learning in the most common deep architectures is implemented using well-understood gradient descent. However, the theory surrounding other algorithms, such as contrastive divergence is less clear.<sup class="noprint Inline-Template Template-Fact" style="white-space:nowrap;">[<i><a href="/wiki/Wikipedia:Citation_needed" title="Wikipedia:Citation needed"><span title="This claim needs references to reliable sources. (July 2016)">citation needed</span></a></i>]</sup> (e.g., Does it converge? If so, how fast? What is it approximating?) Deep learning methods are often looked at as a <a href="/wiki/Black_box" title="Black box">black box</a>, with most confirmations done empirically, rather than theoretically.<sup id="cite_ref-Knight_2017_205-0" class="reference"><a href="#cite_note-Knight_2017-205">[205]</a></sup>
</p><p>
Others point out that deep learning should be looked at as a step towards realizing strong AI, not as an all-encompassing solution. Despite the power of deep learning methods, they still lack much of the functionality needed for realizing this goal entirely. Research psychologist Gary Marcus noted:</p><blockquote><p>"Realistically, deep learning is only part of the larger challenge of building intelligent machines. Such techniques lack ways of representing <a href="/wiki/Causality" title="Causality">causal relationships</a> (...) have no obvious ways of performing <a href="/wiki/Inference" title="Inference">logical inferences</a>, and they are also still a long way from integrating abstract knowledge, such as information about what objects are, what they are for, and how they are typically used. The most powerful A.I. systems, like <a href="/wiki/Watson_(computer)" class="mw-redirect" title="Watson (computer)">Watson</a> (...) use techniques like deep learning as just one element in a very complicated ensemble of techniques, ranging from the statistical technique of <a href="/wiki/Bayesian_inference" title="Bayesian inference">Bayesian inference</a> to <a href="/wiki/Deductive_reasoning" title="Deductive reasoning">deductive reasoning</a>."<sup id="cite_ref-206" class="reference"><a href="#cite_note-206">[206]</a></sup></p></blockquote>
<p>In further reference to the idea that artistic sensitivity might be inherent in relatively low levels of the cognitive hierarchy, a published series of graphic representations of the internal states of deep (20-30 layers) neural networks attempting to discern within essentially random data the images on which they were trained<sup id="cite_ref-207" class="reference"><a href="#cite_note-207">[207]</a></sup> demonstrate a visual appeal: the original research notice received well over 1,000 comments, and was the subject of what was for a time the most frequently accessed article on <i><a href="/wiki/The_Guardian" title="The Guardian">The Guardian</a>'s</i><sup id="cite_ref-208" class="reference"><a href="#cite_note-208">[208]</a></sup> website.
</p>
<h3><span class="mw-headline" id="Errors">Errors</span></h3>
<p>Some deep learning architectures display problematic behaviors,<sup id="cite_ref-goertzel_209-0" class="reference"><a href="#cite_note-goertzel-209">[209]</a></sup> such as confidently classifying unrecognizable images as belonging to a familiar category of ordinary images (2014)<sup id="cite_ref-210" class="reference"><a href="#cite_note-210">[210]</a></sup> and misclassifying minuscule perturbations of correctly classified images (2013).<sup id="cite_ref-211" class="reference"><a href="#cite_note-211">[211]</a></sup> <a href="/wiki/Ben_Goertzel" title="Ben Goertzel">Goertzel</a> hypothesized that these behaviors are due to limitations in their internal representations and that these limitations would inhibit integration into heterogeneous multi-component <a href="/wiki/Artificial_general_intelligence" title="Artificial general intelligence">artificial general intelligence</a> (AGI) architectures.<sup id="cite_ref-goertzel_209-1" class="reference"><a href="#cite_note-goertzel-209">[209]</a></sup> These issues may possibly be addressed by deep learning architectures that internally form states homologous to image-grammar<sup id="cite_ref-212" class="reference"><a href="#cite_note-212">[212]</a></sup> decompositions of observed entities and events.<sup id="cite_ref-goertzel_209-2" class="reference"><a href="#cite_note-goertzel-209">[209]</a></sup> <a href="/wiki/Grammar_induction" title="Grammar induction">Learning a grammar</a> (visual or linguistic) from training data would be equivalent to restricting the system to <a href="/wiki/Commonsense_reasoning" title="Commonsense reasoning">commonsense reasoning</a> that operates on concepts in terms of grammatical <a href="/wiki/Production_(computer_science)" title="Production (computer science)">production rules</a> and is a basic goal of both human language acquisition<sup id="cite_ref-213" class="reference"><a href="#cite_note-213">[213]</a></sup> and <a href="/wiki/Artificial_intelligence" title="Artificial intelligence">artificial intelligence</a> (AI).<sup id="cite_ref-214" class="reference"><a href="#cite_note-214">[214]</a></sup>
</p>
<h3><span class="mw-headline" id="Cyber_threat">Cyber threat</span></h3>
<p>As deep learning moves from the lab into the world, research and experience show that artificial neural networks are vulnerable to hacks and deception.<sup id="cite_ref-215" class="reference"><a href="#cite_note-215">[215]</a></sup> By identifying patterns that these systems use to function, attackers can modify inputs to ANNs in such a way that the ANN finds a match that human observers would not recognize. For example, an attacker can make subtle changes to an image such that the ANN finds a match even though the image looks to a human nothing like the search target. Such manipulation is termed an “adversarial attack.”<sup id="cite_ref-216" class="reference"><a href="#cite_note-216">[216]</a></sup>
</p><p>In 2016 researchers used one ANN to doctor images in trial and error fashion, identify another's focal points and thereby generate images that deceived it. The modified images looked no different to human eyes. Another group showed that printouts of doctored images then photographed successfully tricked an image classification system.<sup id="cite_ref-:4_217-0" class="reference"><a href="#cite_note-:4-217">[217]</a></sup> One defense is reverse image search, in which a possible fake image is submitted to a site such as <a href="/wiki/TinEye" title="TinEye">TinEye</a> that can then find other instances of it. A refinement is to search using only parts of the image, to identify images from which that piece may have been taken<b>.</b><sup id="cite_ref-218" class="reference"><a href="#cite_note-218">[218]</a></sup>
</p><p>Another group showed that certain <a href="/wiki/Psychedelic_art" title="Psychedelic art">psychedelic</a> spectacles could fool a <a href="/wiki/Facial_recognition_system" title="Facial recognition system">facial recognition system</a> into thinking ordinary people were celebrities, potentially allowing one person to impersonate another. In 2017 researchers added stickers to <a href="/wiki/Stop_sign" title="Stop sign">stop signs</a> and caused an ANN to misclassify them.<sup id="cite_ref-:4_217-1" class="reference"><a href="#cite_note-:4-217">[217]</a></sup>
</p><p>ANNs can however be further trained to detect attempts at deception, potentially leading attackers and defenders into an arms race similar to the kind that already defines the <a href="/wiki/Malware" title="Malware">malware</a> defense industry. ANNs have been trained to defeat ANN-based anti-malware software by repeatedly attacking a defense with malware that was continually altered by a <a href="/wiki/Genetic_algorithm" title="Genetic algorithm">genetic algorithm</a> until it tricked the anti-malware while retaining its ability to damage the target.<sup id="cite_ref-:4_217-2" class="reference"><a href="#cite_note-:4-217">[217]</a></sup>
</p><p>In 2016, another group demonstrated that certain sounds could make the <a href="/wiki/Google_Now" title="Google Now">Google Now</a> voice command system open a particular web address, and hypothesized that this could "serve as a stepping stone for further attacks (e.g., opening a web page hosting drive-by malware)."<sup id="cite_ref-:4_217-3" class="reference"><a href="#cite_note-:4-217">[217]</a></sup>
</p><p>In “data poisoning,” false data is continually smuggled into a machine learning system's training set to prevent it from achieving mastery.<sup id="cite_ref-:4_217-4" class="reference"><a href="#cite_note-:4-217">[217]</a></sup>
</p>
<h3><span class="mw-headline" id="Reliance_on_human_microwork">Reliance on human <a href="/wiki/Microwork" title="Microwork">microwork</a></span></h3>
<style data-mw-deduplicate="TemplateStyles:r1097763485">.mw-parser-output .ambox{border:1px solid #a2a9b1;border-left:10px solid #36c;background-color:#fbfbfb;box-sizing:border-box}.mw-parser-output .ambox+link+.ambox,.mw-parser-output .ambox+link+style+.ambox,.mw-parser-output .ambox+link+link+.ambox,.mw-parser-output .ambox+.mw-empty-elt+link+.ambox,.mw-parser-output .ambox+.mw-empty-elt+link+style+.ambox,.mw-parser-output .ambox+.mw-empty-elt+link+link+.ambox{margin-top:-1px}html body.mediawiki .mw-parser-output .ambox.mbox-small-left{margin:4px 1em 4px 0;overflow:hidden;width:238px;border-collapse:collapse;font-size:88%;line-height:1.25em}.mw-parser-output .ambox-speedy{border-left:10px solid #b32424;background-color:#fee7e6}.mw-parser-output .ambox-delete{border-left:10px solid #b32424}.mw-parser-output .ambox-content{border-left:10px solid #f28500}.mw-parser-output .ambox-style{border-left:10px solid #fc3}.mw-parser-output .ambox-move{border-left:10px solid #9932cc}.mw-parser-output .ambox-protection{border-left:10px solid #a2a9b1}.mw-parser-output .ambox .mbox-text{border:none;padding:0.25em 0.5em;width:100%}.mw-parser-output .ambox .mbox-image{border:none;padding:2px 0 2px 0.5em;text-align:center}.mw-parser-output .ambox .mbox-imageright{border:none;padding:2px 0.5em 2px 0;text-align:center}.mw-parser-output .ambox .mbox-empty-cell{border:none;padding:0;width:1px}.mw-parser-output .ambox .mbox-image-div{width:52px}html.client-js body.skin-minerva .mw-parser-output .mbox-text-span{margin-left:23px!important}@media(min-width:720px){.mw-parser-output .ambox{margin:0 10%}}</style><table class="box-More_citations_needed plainlinks metadata ambox ambox-content ambox-Refimprove" role="presentation"><tbody><tr><td class="mbox-image"><div class="mbox-image-div"><a href="/wiki/File:Question_book-new.svg" class="image"><img alt="" src="//upload.wikimedia.org/wikipedia/en/thumb/9/99/Question_book-new.svg/50px-Question_book-new.svg.png" decoding="async" width="50" height="39" srcset="//upload.wikimedia.org/wikipedia/en/thumb/9/99/Question_book-new.svg/75px-Question_book-new.svg.png 1.5x, //upload.wikimedia.org/wikipedia/en/thumb/9/99/Question_book-new.svg/100px-Question_book-new.svg.png 2x" data-file-width="512" data-file-height="399" /></a></div></td><td class="mbox-text"><div class="mbox-text-span">This section <b>needs additional citations for <a href="/wiki/Wikipedia:Verifiability" title="Wikipedia:Verifiability">verification</a></b>.<span class="hide-when-compact"> Please help <a class="external text" href="https://en.wikipedia.org/w/index.php?title=Deep_learning&action=edit">improve this article</a> by <a href="/wiki/Help:Referencing_for_beginners" title="Help:Referencing for beginners">adding citations to reliable sources</a>. Unsourced material may be challenged and removed.<br /><small><span class="plainlinks"><i>Find sources:</i> <a rel="nofollow" class="external text" href="//www.google.com/search?as_eq=wikipedia&q=%22Deep+learning%22">"Deep learning"</a> – <a rel="nofollow" class="external text" href="//www.google.com/search?tbm=nws&q=%22Deep+learning%22+-wikipedia&tbs=ar:1">news</a> <b>·</b> <a rel="nofollow" class="external text" href="//www.google.com/search?&q=%22Deep+learning%22&tbs=bkt:s&tbm=bks">newspapers</a> <b>·</b> <a rel="nofollow" class="external text" href="//www.google.com/search?tbs=bks:1&q=%22Deep+learning%22+-wikipedia">books</a> <b>·</b> <a rel="nofollow" class="external text" href="//scholar.google.com/scholar?q=%22Deep+learning%22">scholar</a> <b>·</b> <a rel="nofollow" class="external text" href="https://www.jstor.org/action/doBasicSearch?Query=%22Deep+learning%22&acc=on&wc=on">JSTOR</a></span></small></span> <span class="date-container"><i>(<span class="date">April 2021</span>)</i></span><span class="hide-when-compact"><i> (<small><a href="/wiki/Help:Maintenance_template_removal" title="Help:Maintenance template removal">Learn how and when to remove this template message</a></small>)</i></span></div></td></tr></tbody></table>
<p>Most Deep Learning systems rely on training and verification data that is generated and/or annotated by humans.<sup id="cite_ref-219" class="reference"><a href="#cite_note-219">[219]</a></sup> It has been argued in <a href="/wiki/Media_studies" title="Media studies">media philosophy</a> that not only low-paid <a href="/wiki/Clickworkers" title="Clickworkers">clickwork</a> (e.g. on <a href="/wiki/Amazon_Mechanical_Turk" title="Amazon Mechanical Turk">Amazon Mechanical Turk</a>) is regularly deployed for this purpose, but also implicit forms of human <a href="/wiki/Microwork" title="Microwork">microwork</a> that are often not recognized as such.<sup id="cite_ref-:13_220-0" class="reference"><a href="#cite_note-:13-220">[220]</a></sup> The philosopher Rainer Mühlhoff distinguishes five types of "machinic capture" of human microwork to generate training data: (1) <a href="/wiki/Gamification" title="Gamification">gamification</a> (the embedding of annotation or computation tasks in the flow of a game), (2) "trapping and tracking" (e.g. <a href="/wiki/CAPTCHA" title="CAPTCHA">CAPTCHAs</a> for image recognition or click-tracking on Google <a href="/wiki/Search_engine_results_page" title="Search engine results page">search results pages</a>), (3) exploitation of social motivations (e.g. <a href="/wiki/Tag_(Facebook)" class="mw-redirect" title="Tag (Facebook)">tagging faces</a> on <a href="/wiki/Facebook" title="Facebook">Facebook</a> to obtain labeled facial images), (4) <a href="/wiki/Information_mining" class="mw-redirect" title="Information mining">information mining</a> (e.g. by leveraging <a href="/wiki/Quantified_self" title="Quantified self">quantified-self</a> devices such as <a href="/wiki/Activity_tracker" title="Activity tracker">activity trackers</a>) and (5) <a href="/wiki/Clickworkers" title="Clickworkers">clickwork</a>.<sup id="cite_ref-:13_220-1" class="reference"><a href="#cite_note-:13-220">[220]</a></sup>
</p><p>Mühlhoff argues that in most commercial end-user applications of Deep Learning such as <a href="/wiki/DeepFace" title="DeepFace">Facebook's face recognition system</a>, the need for training data does not stop once an ANN is trained. Rather, there is a continued demand for human-generated verification data to constantly calibrate and update the ANN. For this purpose Facebook introduced the feature that once a user is automatically recognized in an image, they receive a notification. They can choose whether of not they like to be publicly labeled on the image, or tell Facebook that it is not them in the picture.<sup id="cite_ref-221" class="reference"><a href="#cite_note-221">[221]</a></sup> This user interface is a mechanism to generate "a constant stream of verification data"<sup id="cite_ref-:13_220-2" class="reference"><a href="#cite_note-:13-220">[220]</a></sup> to further train the network in real-time. As Mühlhoff argues, involvement of human users to generate training and verification data is so typical for most commercial end-user applications of Deep Learning that such systems may be referred to as "human-aided artificial intelligence".<sup id="cite_ref-:13_220-3" class="reference"><a href="#cite_note-:13-220">[220]</a></sup>
</p>
<h2><span class="mw-headline" id="See_also">See also</span></h2>
<ul><li><a href="/wiki/Applications_of_artificial_intelligence" title="Applications of artificial intelligence">Applications of artificial intelligence</a></li>
<li><a href="/wiki/Comparison_of_deep_learning_software" title="Comparison of deep learning software">Comparison of deep learning software</a></li>
<li><a href="/wiki/Compressed_sensing" title="Compressed sensing">Compressed sensing</a></li>
<li><a href="/wiki/Differentiable_programming" title="Differentiable programming">Differentiable programming</a></li>
<li><a href="/wiki/Echo_state_network" title="Echo state network">Echo state network</a></li>
<li><a href="/wiki/List_of_artificial_intelligence_projects" title="List of artificial intelligence projects">List of artificial intelligence projects</a></li>
<li><a href="/wiki/Liquid_state_machine" title="Liquid state machine">Liquid state machine</a></li>
<li><a href="/wiki/List_of_datasets_for_machine-learning_research" title="List of datasets for machine-learning research">List of datasets for machine-learning research</a></li>
<li><a href="/wiki/Reservoir_computing" title="Reservoir computing">Reservoir computing</a></li>
<li><a href="/wiki/Scale_space#Deep_learning_and_scale_space" title="Scale space">Scale space and deep learning</a></li>
<li><a href="/wiki/Sparse_coding" class="mw-redirect" title="Sparse coding">Sparse coding</a></li></ul>
<h2><span class="mw-headline" id="References">References</span></h2>
<style data-mw-deduplicate="TemplateStyles:r1011085734">.mw-parser-output .reflist{font-size:90%;margin-bottom:0.5em;list-style-type:decimal}.mw-parser-output .reflist .references{font-size:100%;margin-bottom:0;list-style-type:inherit}.mw-parser-output .reflist-columns-2{column-width:30em}.mw-parser-output .reflist-columns-3{column-width:25em}.mw-parser-output .reflist-columns{margin-top:0.3em}.mw-parser-output .reflist-columns ol{margin-top:0}.mw-parser-output .reflist-columns li{page-break-inside:avoid;break-inside:avoid-column}.mw-parser-output .reflist-upper-alpha{list-style-type:upper-alpha}.mw-parser-output .reflist-upper-roman{list-style-type:upper-roman}.mw-parser-output .reflist-lower-alpha{list-style-type:lower-alpha}.mw-parser-output .reflist-lower-greek{list-style-type:lower-greek}.mw-parser-output .reflist-lower-roman{list-style-type:lower-roman}</style><div class="reflist reflist-columns references-column-width" style="column-width: 30em;">
<ol class="references">
<li id="cite_note-BOOK2014-1"><span class="mw-cite-backlink">^ <a href="#cite_ref-BOOK2014_1-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-BOOK2014_1-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-BOOK2014_1-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-BOOK2014_1-3"><sup><i><b>d</b></i></sup></a> <a href="#cite_ref-BOOK2014_1-4"><sup><i><b>e</b></i></sup></a> <a href="#cite_ref-BOOK2014_1-5"><sup><i><b>f</b></i></sup></a></span> <span class="reference-text"><style data-mw-deduplicate="TemplateStyles:r1133582631">.mw-parser-output cite.citation{font-style:inherit;word-wrap:break-word}.mw-parser-output .citation q{quotes:"\"""\"""'""'"}.mw-parser-output .citation:target{background-color:rgba(0,127,255,0.133)}.mw-parser-output .id-lock-free a,.mw-parser-output .citation .cs1-lock-free a{background:url("//upload.wikimedia.org/wikipedia/commons/6/65/Lock-green.svg")right 0.1em center/9px no-repeat}.mw-parser-output .id-lock-limited a,.mw-parser-output .id-lock-registration a,.mw-parser-output .citation .cs1-lock-limited a,.mw-parser-output .citation .cs1-lock-registration a{background:url("//upload.wikimedia.org/wikipedia/commons/d/d6/Lock-gray-alt-2.svg")right 0.1em center/9px no-repeat}.mw-parser-output .id-lock-subscription a,.mw-parser-output .citation .cs1-lock-subscription a{background:url("//upload.wikimedia.org/wikipedia/commons/a/aa/Lock-red-alt-2.svg")right 0.1em center/9px no-repeat}.mw-parser-output .cs1-ws-icon a{background:url("//upload.wikimedia.org/wikipedia/commons/4/4c/Wikisource-logo.svg")right 0.1em center/12px no-repeat}.mw-parser-output .cs1-code{color:inherit;background:inherit;border:none;padding:inherit}.mw-parser-output .cs1-hidden-error{display:none;color:#d33}.mw-parser-output .cs1-visible-error{color:#d33}.mw-parser-output .cs1-maint{display:none;color:#3a3;margin-left:0.3em}.mw-parser-output .cs1-format{font-size:95%}.mw-parser-output .cs1-kern-left{padding-left:0.2em}.mw-parser-output .cs1-kern-right{padding-right:0.2em}.mw-parser-output .citation .mw-selflink{font-weight:inherit}</style><cite id="CITEREFDengYu2014" class="citation journal cs1">Deng, L.; Yu, D. (2014). <a rel="nofollow" class="external text" href="http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf">"Deep Learning: Methods and Applications"</a> <span class="cs1-format">(PDF)</span>. <i>Foundations and Trends in Signal Processing</i>. <b>7</b> (3–4): 1–199. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1561%2F2000000039">10.1561/2000000039</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160314152112/http://research.microsoft.com/pubs/209355/DeepLearning-NowPublishing-Vol7-SIG-039.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2016-03-14<span class="reference-accessdate">. Retrieved <span class="nowrap">2014-10-18</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Foundations+and+Trends+in+Signal+Processing&rft.atitle=Deep+Learning%3A+Methods+and+Applications&rft.volume=7&rft.issue=3%E2%80%934&rft.pages=1-199&rft.date=2014&rft_id=info%3Adoi%2F10.1561%2F2000000039&rft.aulast=Deng&rft.aufirst=L.&rft.au=Yu%2C+D.&rft_id=http%3A%2F%2Fresearch.microsoft.com%2Fpubs%2F209355%2FDeepLearning-NowPublishing-Vol7-SIG-039.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-BENGIODEEP-2"><span class="mw-cite-backlink">^ <a href="#cite_ref-BENGIODEEP_2-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-BENGIODEEP_2-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-BENGIODEEP_2-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-BENGIODEEP_2-3"><sup><i><b>d</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBengio2009" class="citation journal cs1">Bengio, Yoshua (2009). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160304084250/http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20(2009).pdf">"Learning Deep Architectures for AI"</a> <span class="cs1-format">(PDF)</span>. <i>Foundations and Trends in Machine Learning</i>. <b>2</b> (1): 1–127. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.701.9550">10.1.1.701.9550</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1561%2F2200000006">10.1561/2200000006</a>. Archived from <a rel="nofollow" class="external text" href="http://sanghv.com/download/soft/machine%20learning,%20artificial%20intelligence,%20mathematics%20ebooks/ML/learning%20deep%20architectures%20for%20AI%20%282009%29.pdf">the original</a> <span class="cs1-format">(PDF)</span> on 4 March 2016<span class="reference-accessdate">. Retrieved <span class="nowrap">3 September</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Foundations+and+Trends+in+Machine+Learning&rft.atitle=Learning+Deep+Architectures+for+AI&rft.volume=2&rft.issue=1&rft.pages=1-127&rft.date=2009&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.701.9550%23id-name%3DCiteSeerX&rft_id=info%3Adoi%2F10.1561%2F2200000006&rft.aulast=Bengio&rft.aufirst=Yoshua&rft_id=http%3A%2F%2Fsanghv.com%2Fdownload%2Fsoft%2Fmachine%2520learning%2C%2520artificial%2520intelligence%2C%2520mathematics%2520ebooks%2FML%2Flearning%2520deep%2520architectures%2520for%2520AI%2520%25282009%2529.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-BENGIO2012-3"><span class="mw-cite-backlink">^ <a href="#cite_ref-BENGIO2012_3-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-BENGIO2012_3-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-BENGIO2012_3-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-BENGIO2012_3-3"><sup><i><b>d</b></i></sup></a> <a href="#cite_ref-BENGIO2012_3-4"><sup><i><b>e</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBengioCourvilleVincent2013" class="citation journal cs1">Bengio, Y.; Courville, A.; Vincent, P. (2013). "Representation Learning: A Review and New Perspectives". <i>IEEE Transactions on Pattern Analysis and Machine Intelligence</i>. <b>35</b> (8): 1798–1828. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1206.5538">1206.5538</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Ftpami.2013.50">10.1109/tpami.2013.50</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/23787338">23787338</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:393948">393948</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Pattern+Analysis+and+Machine+Intelligence&rft.atitle=Representation+Learning%3A+A+Review+and+New+Perspectives&rft.volume=35&rft.issue=8&rft.pages=1798-1828&rft.date=2013&rft_id=info%3Aarxiv%2F1206.5538&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A393948%23id-name%3DS2CID&rft_id=info%3Apmid%2F23787338&rft_id=info%3Adoi%2F10.1109%2Ftpami.2013.50&rft.aulast=Bengio&rft.aufirst=Y.&rft.au=Courville%2C+A.&rft.au=Vincent%2C+P.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-4"><span class="mw-cite-backlink"><b><a href="#cite_ref-4">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFLeCunBengioHinton2015" class="citation journal cs1">LeCun, Yann; Bengio, Yoshua; Hinton, Geoffrey (28 May 2015). "Deep learning". <i>Nature</i>. <b>521</b> (7553): 436–444. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2015Natur.521..436L">2015Natur.521..436L</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fnature14539">10.1038/nature14539</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/26017442">26017442</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:3074096">3074096</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=Deep+learning&rft.volume=521&rft.issue=7553&rft.pages=436-444&rft.date=2015-05-28&rft_id=info%3Adoi%2F10.1038%2Fnature14539&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A3074096%23id-name%3DS2CID&rft_id=info%3Apmid%2F26017442&rft_id=info%3Abibcode%2F2015Natur.521..436L&rft.aulast=LeCun&rft.aufirst=Yann&rft.au=Bengio%2C+Yoshua&rft.au=Hinton%2C+Geoffrey&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-SCHIDHUB-5"><span class="mw-cite-backlink">^ <a href="#cite_ref-SCHIDHUB_5-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-SCHIDHUB_5-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-SCHIDHUB_5-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-SCHIDHUB_5-3"><sup><i><b>d</b></i></sup></a> <a href="#cite_ref-SCHIDHUB_5-4"><sup><i><b>e</b></i></sup></a> <a href="#cite_ref-SCHIDHUB_5-5"><sup><i><b>f</b></i></sup></a> <a href="#cite_ref-SCHIDHUB_5-6"><sup><i><b>g</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSchmidhuber2015" class="citation journal cs1">Schmidhuber, J. (2015). "Deep Learning in Neural Networks: An Overview". <i>Neural Networks</i>. <b>61</b>: 85–117. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1404.7828">1404.7828</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.neunet.2014.09.003">10.1016/j.neunet.2014.09.003</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/25462637">25462637</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:11715509">11715509</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Networks&rft.atitle=Deep+Learning+in+Neural+Networks%3A+An+Overview&rft.volume=61&rft.pages=85-117&rft.date=2015&rft_id=info%3Aarxiv%2F1404.7828&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A11715509%23id-name%3DS2CID&rft_id=info%3Apmid%2F25462637&rft_id=info%3Adoi%2F10.1016%2Fj.neunet.2014.09.003&rft.aulast=Schmidhuber&rft.aufirst=J.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-6"><span class="mw-cite-backlink"><b><a href="#cite_ref-6">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFShigeki2019" class="citation book cs1">Shigeki, Sugiyama (12 April 2019). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=9CqQDwAAQBAJ&pg=PA15"><i>Human Behavior and Another Kind in Consciousness: Emerging Research and Opportunities: Emerging Research and Opportunities</i></a>. IGI Global. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-5225-8218-2" title="Special:BookSources/978-1-5225-8218-2"><bdi>978-1-5225-8218-2</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Human+Behavior+and+Another+Kind+in+Consciousness%3A+Emerging+Research+and+Opportunities%3A+Emerging+Research+and+Opportunities&rft.pub=IGI+Global&rft.date=2019-04-12&rft.isbn=978-1-5225-8218-2&rft.aulast=Shigeki&rft.aufirst=Sugiyama&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3D9CqQDwAAQBAJ%26pg%3DPA15&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-BENGIO2007-7"><span class="mw-cite-backlink"><b><a href="#cite_ref-BENGIO2007_7-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBengioLamblinPopoviciLarochelle2007" class="citation conference cs1">Bengio, Yoshua; Lamblin, Pascal; Popovici, Dan; Larochelle, Hugo (2007). <a rel="nofollow" class="external text" href="http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf"><i>Greedy layer-wise training of deep networks</i></a> <span class="cs1-format">(PDF)</span>. Advances in neural information processing systems. pp. 153–160. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20191020195638/http://papers.nips.cc/paper/3048-greedy-layer-wise-training-of-deep-networks.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2019-10-20<span class="reference-accessdate">. Retrieved <span class="nowrap">2019-10-06</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=conference&rft.btitle=Greedy+layer-wise+training+of+deep+networks&rft.pages=153-160&rft.date=2007&rft.aulast=Bengio&rft.aufirst=Yoshua&rft.au=Lamblin%2C+Pascal&rft.au=Popovici%2C+Dan&rft.au=Larochelle%2C+Hugo&rft_id=http%3A%2F%2Fpapers.nips.cc%2Fpaper%2F3048-greedy-layer-wise-training-of-deep-networks.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-SCHOLARDBNS-8"><span class="mw-cite-backlink">^ <a href="#cite_ref-SCHOLARDBNS_8-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-SCHOLARDBNS_8-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHinton2009" class="citation journal cs1">Hinton, G.E. (2009). <a rel="nofollow" class="external text" href="https://doi.org/10.4249%2Fscholarpedia.5947">"Deep belief networks"</a>. <i>Scholarpedia</i>. <b>4</b> (5): 5947. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2009SchpJ...4.5947H">2009SchpJ...4.5947H</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.4249%2Fscholarpedia.5947">10.4249/scholarpedia.5947</a></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Scholarpedia&rft.atitle=Deep+belief+networks&rft.volume=4&rft.issue=5&rft.pages=5947&rft.date=2009&rft_id=info%3Adoi%2F10.4249%2Fscholarpedia.5947&rft_id=info%3Abibcode%2F2009SchpJ...4.5947H&rft.aulast=Hinton&rft.aufirst=G.E.&rft_id=https%3A%2F%2Fdoi.org%2F10.4249%252Fscholarpedia.5947&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-cyb-9"><span class="mw-cite-backlink">^ <a href="#cite_ref-cyb_9-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-cyb_9-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-cyb_9-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCybenko1989" class="citation journal cs1">Cybenko (1989). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20151010204407/http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf">"Approximations by superpositions of sigmoidal functions"</a> <span class="cs1-format">(PDF)</span>. <i><a href="/wiki/Mathematics_of_Control,_Signals,_and_Systems" title="Mathematics of Control, Signals, and Systems">Mathematics of Control, Signals, and Systems</a></i>. <b>2</b> (4): 303–314. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2Fbf02551274">10.1007/bf02551274</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:3958369">3958369</a>. Archived from <a rel="nofollow" class="external text" href="http://deeplearning.cs.cmu.edu/pdfs/Cybenko.pdf">the original</a> <span class="cs1-format">(PDF)</span> on 10 October 2015.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Mathematics+of+Control%2C+Signals%2C+and+Systems&rft.atitle=Approximations+by+superpositions+of+sigmoidal+functions&rft.volume=2&rft.issue=4&rft.pages=303-314&rft.date=1989&rft_id=info%3Adoi%2F10.1007%2Fbf02551274&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A3958369%23id-name%3DS2CID&rft.au=Cybenko&rft_id=http%3A%2F%2Fdeeplearning.cs.cmu.edu%2Fpdfs%2FCybenko.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-horn-10"><span class="mw-cite-backlink">^ <a href="#cite_ref-horn_10-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-horn_10-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-horn_10-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHornik1991" class="citation journal cs1">Hornik, Kurt (1991). "Approximation Capabilities of Multilayer Feedforward Networks". <i>Neural Networks</i>. <b>4</b> (2): 251–257. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2F0893-6080%2891%2990009-t">10.1016/0893-6080(91)90009-t</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Networks&rft.atitle=Approximation+Capabilities+of+Multilayer+Feedforward+Networks&rft.volume=4&rft.issue=2&rft.pages=251-257&rft.date=1991&rft_id=info%3Adoi%2F10.1016%2F0893-6080%2891%2990009-t&rft.aulast=Hornik&rft.aufirst=Kurt&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Haykin,_Simon_1998-11"><span class="mw-cite-backlink">^ <a href="#cite_ref-Haykin,_Simon_1998_11-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-Haykin,_Simon_1998_11-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHaykin1999" class="citation book cs1">Haykin, Simon S. (1999). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=bX4pAQAAMAAJ"><i>Neural Networks: A Comprehensive Foundation</i></a>. Prentice Hall. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-13-273350-2" title="Special:BookSources/978-0-13-273350-2"><bdi>978-0-13-273350-2</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Neural+Networks%3A+A+Comprehensive+Foundation&rft.pub=Prentice+Hall&rft.date=1999&rft.isbn=978-0-13-273350-2&rft.aulast=Haykin&rft.aufirst=Simon+S.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DbX4pAQAAMAAJ&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Hassoun,_M._1995_p._48-12"><span class="mw-cite-backlink">^ <a href="#cite_ref-Hassoun,_M._1995_p._48_12-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-Hassoun,_M._1995_p._48_12-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHassoun1995" class="citation book cs1">Hassoun, Mohamad H. (1995). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=Otk32Y3QkxQC&pg=PA48"><i>Fundamentals of Artificial Neural Networks</i></a>. MIT Press. p. 48. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-262-08239-6" title="Special:BookSources/978-0-262-08239-6"><bdi>978-0-262-08239-6</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Fundamentals+of+Artificial+Neural+Networks&rft.pages=48&rft.pub=MIT+Press&rft.date=1995&rft.isbn=978-0-262-08239-6&rft.aulast=Hassoun&rft.aufirst=Mohamad+H.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DOtk32Y3QkxQC%26pg%3DPA48&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-ZhouLu-13"><span class="mw-cite-backlink">^ <a href="#cite_ref-ZhouLu_13-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-ZhouLu_13-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text">Lu, Z., Pu, H., Wang, F., Hu, Z., & Wang, L. (2017). <a rel="nofollow" class="external text" href="http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width">The Expressive Power of Neural Networks: A View from the Width</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20190213005539/http://papers.nips.cc/paper/7203-the-expressive-power-of-neural-networks-a-view-from-the-width">Archived</a> 2019-02-13 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a>. Neural Information Processing Systems, 6231-6239.</span>
</li>
<li id="cite_note-14"><span class="mw-cite-backlink"><b><a href="#cite_ref-14">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFOrhanMa2017" class="citation journal cs1">Orhan, A. E.; Ma, W. J. (2017). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5527101">"Efficient probabilistic inference in generic neural networks trained with non-probabilistic feedback"</a>. <i>Nature Communications</i>. <b>8</b> (1): 138. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2017NatCo...8..138O">2017NatCo...8..138O</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fs41467-017-00181-8">10.1038/s41467-017-00181-8</a></span>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5527101">5527101</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/28743932">28743932</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature+Communications&rft.atitle=Efficient+probabilistic+inference+in+generic+neural+networks+trained+with+non-probabilistic+feedback&rft.volume=8&rft.issue=1&rft.pages=138&rft.date=2017&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5527101%23id-name%3DPMC&rft_id=info%3Apmid%2F28743932&rft_id=info%3Adoi%2F10.1038%2Fs41467-017-00181-8&rft_id=info%3Abibcode%2F2017NatCo...8..138O&rft.aulast=Orhan&rft.aufirst=A.+E.&rft.au=Ma%2C+W.+J.&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5527101&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-MURPHY-15"><span class="mw-cite-backlink">^ <a href="#cite_ref-MURPHY_15-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-MURPHY_15-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-MURPHY_15-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-MURPHY_15-3"><sup><i><b>d</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMurphy2012" class="citation book cs1">Murphy, Kevin P. (24 August 2012). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=NZP6AQAAQBAJ"><i>Machine Learning: A Probabilistic Perspective</i></a>. MIT Press. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-262-01802-9" title="Special:BookSources/978-0-262-01802-9"><bdi>978-0-262-01802-9</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Machine+Learning%3A+A+Probabilistic+Perspective&rft.pub=MIT+Press&rft.date=2012-08-24&rft.isbn=978-0-262-01802-9&rft.aulast=Murphy&rft.aufirst=Kevin+P.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DNZP6AQAAQBAJ&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-sonoda17-16"><span class="mw-cite-backlink"><b><a href="#cite_ref-sonoda17_16-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSonodaMurata2017" class="citation journal cs1">Sonoda, Sho; Murata, Noboru (2017). "Neural network with unbounded activation functions is universal approximator". <i>Applied and Computational Harmonic Analysis</i>. <b>43</b> (2): 233–268. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1505.03654">1505.03654</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.acha.2015.12.005">10.1016/j.acha.2015.12.005</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:12149203">12149203</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Applied+and+Computational+Harmonic+Analysis&rft.atitle=Neural+network+with+unbounded+activation+functions+is+universal+approximator&rft.volume=43&rft.issue=2&rft.pages=233-268&rft.date=2017&rft_id=info%3Aarxiv%2F1505.03654&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A12149203%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1016%2Fj.acha.2015.12.005&rft.aulast=Sonoda&rft.aufirst=Sho&rft.au=Murata%2C+Noboru&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-prml-17"><span class="mw-cite-backlink"><b><a href="#cite_ref-prml_17-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBishop,_Christopher_M.2006" class="citation book cs1">Bishop, Christopher M. (2006). <a rel="nofollow" class="external text" href="http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf"><i>Pattern Recognition and Machine Learning</i></a> <span class="cs1-format">(PDF)</span>. Springer. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-387-31073-2" title="Special:BookSources/978-0-387-31073-2"><bdi>978-0-387-31073-2</bdi></a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170111005101/http://users.isr.ist.utl.pt/~wurmd/Livros/school/Bishop%20-%20Pattern%20Recognition%20And%20Machine%20Learning%20-%20Springer%20%202006.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-01-11<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-08-06</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Pattern+Recognition+and+Machine+Learning&rft.pub=Springer&rft.date=2006&rft.isbn=978-0-387-31073-2&rft.au=Bishop%2C+Christopher+M.&rft_id=http%3A%2F%2Fusers.isr.ist.utl.pt%2F~wurmd%2FLivros%2Fschool%2FBishop%2520-%2520Pattern%2520Recognition%2520And%2520Machine%2520Learning%2520-%2520Springer%2520%25202006.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Who_Is_the_Father_of_Deep_Learning?-18"><span class="mw-cite-backlink"><b><a href="#cite_ref-Who_Is_the_Father_of_Deep_Learning?_18-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTappert2019" class="citation book cs1">Tappert, Charles C. (2019). <a rel="nofollow" class="external text" href="https://ieeexplore.ieee.org/document/9070967">"Who Is the Father of Deep Learning?"</a>. <i>2019 International Conference on Computational Science and Computational Intelligence (CSCI)</i>. IEEE. pp. 343–348. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FCSCI49370.2019.00067">10.1109/CSCI49370.2019.00067</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-7281-5584-5" title="Special:BookSources/978-1-7281-5584-5"><bdi>978-1-7281-5584-5</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:216043128">216043128</a><span class="reference-accessdate">. Retrieved <span class="nowrap">31 May</span> 2021</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Who+Is+the+Father+of+Deep+Learning%3F&rft.btitle=2019+International+Conference+on+Computational+Science+and+Computational+Intelligence+%28CSCI%29&rft.pages=343-348&rft.pub=IEEE&rft.date=2019&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A216043128%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2FCSCI49370.2019.00067&rft.isbn=978-1-7281-5584-5&rft.aulast=Tappert&rft.aufirst=Charles+C.&rft_id=https%3A%2F%2Fieeexplore.ieee.org%2Fdocument%2F9070967&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-ivak1965-19"><span class="mw-cite-backlink"><b><a href="#cite_ref-ivak1965_19-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFIvakhnenkoLapa1967" class="citation book cs1">Ivakhnenko, A. G.; Lapa, V. G. (1967). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=rGFgAAAAMAAJ"><i>Cybernetics and Forecasting Techniques</i></a>. American Elsevier Publishing Co. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-444-00020-0" title="Special:BookSources/978-0-444-00020-0"><bdi>978-0-444-00020-0</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Cybernetics+and+Forecasting+Techniques&rft.pub=American+Elsevier+Publishing+Co.&rft.date=1967&rft.isbn=978-0-444-00020-0&rft.aulast=Ivakhnenko&rft.aufirst=A.+G.&rft.au=Lapa%2C+V.+G.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DrGFgAAAAMAAJ&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-ivak1971-20"><span class="mw-cite-backlink">^ <a href="#cite_ref-ivak1971_20-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-ivak1971_20-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFIvakhnenko1971" class="citation journal cs1 cs1-prop-long-vol">Ivakhnenko, Alexey (1971). <a rel="nofollow" class="external text" href="http://gmdh.net/articles/history/polynomial.pdf">"Polynomial theory of complex systems"</a> <span class="cs1-format">(PDF)</span>. <i>IEEE Transactions on Systems, Man and Cybernetics</i>. SMC-1 (4): 364–378. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FTSMC.1971.4308320">10.1109/TSMC.1971.4308320</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170829230621/http://www.gmdh.net/articles/history/polynomial.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-08-29<span class="reference-accessdate">. Retrieved <span class="nowrap">2019-11-05</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Systems%2C+Man+and+Cybernetics&rft.atitle=Polynomial+theory+of+complex+systems&rft.volume=SMC-1&rft.issue=4&rft.pages=364-378&rft.date=1971&rft_id=info%3Adoi%2F10.1109%2FTSMC.1971.4308320&rft.aulast=Ivakhnenko&rft.aufirst=Alexey&rft_id=http%3A%2F%2Fgmdh.net%2Farticles%2Fhistory%2Fpolynomial.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-FUKU1980-21"><span class="mw-cite-backlink"><b><a href="#cite_ref-FUKU1980_21-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFFukushima1980" class="citation journal cs1">Fukushima, K. (1980). "Neocognitron: A self-organizing neural network model for a mechanism of pattern recognition unaffected by shift in position". <i>Biol. Cybern</i>. <b>36</b> (4): 193–202. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2Fbf00344251">10.1007/bf00344251</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/7370364">7370364</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:206775608">206775608</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Biol.+Cybern.&rft.atitle=Neocognitron%3A+A+self-organizing+neural+network+model+for+a+mechanism+of+pattern+recognition+unaffected+by+shift+in+position&rft.volume=36&rft.issue=4&rft.pages=193-202&rft.date=1980&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A206775608%23id-name%3DS2CID&rft_id=info%3Apmid%2F7370364&rft_id=info%3Adoi%2F10.1007%2Fbf00344251&rft.aulast=Fukushima&rft.aufirst=K.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-dechter1986-22"><span class="mw-cite-backlink"><b><a href="#cite_ref-dechter1986_22-0">^</a></b></span> <span class="reference-text"><a href="/wiki/Rina_Dechter" title="Rina Dechter">Rina Dechter</a> (1986). Learning while searching in constraint-satisfaction problems. University of California, Computer Science Department, Cognitive Systems Laboratory.<a rel="nofollow" class="external text" href="https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems">Online</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160419054654/https://www.researchgate.net/publication/221605378_Learning_While_Searching_in_Constraint-Satisfaction-Problems">Archived</a> 2016-04-19 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a></span>
</li>
<li id="cite_note-aizenberg2000-23"><span class="mw-cite-backlink"><b><a href="#cite_ref-aizenberg2000_23-0">^</a></b></span> <span class="reference-text">Igor Aizenberg, Naum N. Aizenberg, Joos P.L. Vandewalle (2000). Multi-Valued and Universal Binary Neurons: Theory, Learning and Applications. Springer Science & Business Media.</span>
</li>
<li id="cite_note-24"><span class="mw-cite-backlink"><b><a href="#cite_ref-24">^</a></b></span> <span class="reference-text">Co-evolving recurrent neurons learn deep memory POMDPs. Proc. GECCO, Washington, D. C., pp. 1795-1802, ACM Press, New York, NY, USA, 2005.</span>
</li>
<li id="cite_note-lin1970-25"><span class="mw-cite-backlink"><b><a href="#cite_ref-lin1970_25-0">^</a></b></span> <span class="reference-text"><a href="/wiki/Seppo_Linnainmaa" title="Seppo Linnainmaa">Seppo Linnainmaa</a> (1970). The representation of the cumulative rounding error of an algorithm as a Taylor expansion of the local rounding errors. Master's Thesis (in Finnish), Univ. Helsinki, 6-7.</span>
</li>
<li id="cite_note-grie2012-26"><span class="mw-cite-backlink"><b><a href="#cite_ref-grie2012_26-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGriewank2012" class="citation journal cs1">Griewank, Andreas (2012). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170721211929/http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf">"Who Invented the Reverse Mode of Differentiation?"</a> <span class="cs1-format">(PDF)</span>. <i>Documenta Mathematica</i> (Extra Volume ISMP): 389–400. Archived from <a rel="nofollow" class="external text" href="http://www.math.uiuc.edu/documenta/vol-ismp/52_griewank-andreas-b.pdf">the original</a> <span class="cs1-format">(PDF)</span> on 21 July 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">11 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Documenta+Mathematica&rft.atitle=Who+Invented+the+Reverse+Mode+of+Differentiation%3F&rft.issue=Extra+Volume+ISMP&rft.pages=389-400&rft.date=2012&rft.aulast=Griewank&rft.aufirst=Andreas&rft_id=http%3A%2F%2Fwww.math.uiuc.edu%2Fdocumenta%2Fvol-ismp%2F52_griewank-andreas-b.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-WERBOS1974-27"><span class="mw-cite-backlink"><b><a href="#cite_ref-WERBOS1974_27-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFWerbos1974" class="citation journal cs1">Werbos, P. (1974). <a rel="nofollow" class="external text" href="https://www.researchgate.net/publication/35657389">"Beyond Regression: New Tools for Prediction and Analysis in the Behavioral Sciences"</a>. <i>Harvard University</i><span class="reference-accessdate">. Retrieved <span class="nowrap">12 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Harvard+University&rft.atitle=Beyond+Regression%3A+New+Tools+for+Prediction+and+Analysis+in+the+Behavioral+Sciences&rft.date=1974&rft.aulast=Werbos&rft.aufirst=P.&rft_id=https%3A%2F%2Fwww.researchgate.net%2Fpublication%2F35657389&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-werbos1982-28"><span class="mw-cite-backlink"><b><a href="#cite_ref-werbos1982_28-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFWerbos1982" class="citation book cs1">Werbos, Paul (1982). <a rel="nofollow" class="external text" href="ftp://ftp.idsia.ch/pub/juergen/habilitation.pdf">"Applications of advances in nonlinear sensitivity analysis"</a> <span class="cs1-format">(PDF)</span>. <i>System modeling and optimization</i>. Springer. pp. 762–770.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Applications+of+advances+in+nonlinear+sensitivity+analysis&rft.btitle=System+modeling+and+optimization&rft.pages=762-770&rft.pub=Springer&rft.date=1982&rft.aulast=Werbos&rft.aufirst=Paul&rft_id=ftp%3A%2F%2Fftp.idsia.ch%2Fpub%2Fjuergen%2Fhabilitation.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-LECUN1989-29"><span class="mw-cite-backlink">^ <a href="#cite_ref-LECUN1989_29-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-LECUN1989_29-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text">LeCun <i>et al.</i>, "Backpropagation Applied to Handwritten Zip Code Recognition," <i>Neural Computation</i>, 1, pp. 541–551, 1989.</span>
</li>
<li id="cite_note-wz1988-30"><span class="mw-cite-backlink"><b><a href="#cite_ref-wz1988_30-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZhang1988" class="citation journal cs1">Zhang, Wei (1988). <a rel="nofollow" class="external text" href="https://drive.google.com/file/d/1nN_5odSG_QVae54EsQN_qSz-0ZsX6wA0/view?usp=sharing">"Shift-invariant pattern recognition neural network and its optical architecture"</a>. <i>Proceedings of Annual Conference of the Japan Society of Applied Physics</i>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+of+Annual+Conference+of+the+Japan+Society+of+Applied+Physics&rft.atitle=Shift-invariant+pattern+recognition+neural+network+and+its+optical+architecture&rft.date=1988&rft.aulast=Zhang&rft.aufirst=Wei&rft_id=https%3A%2F%2Fdrive.google.com%2Ffile%2Fd%2F1nN_5odSG_QVae54EsQN_qSz-0ZsX6wA0%2Fview%3Fusp%3Dsharing&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-wz1990-31"><span class="mw-cite-backlink"><b><a href="#cite_ref-wz1990_31-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZhang1990" class="citation journal cs1">Zhang, Wei (1990). <a rel="nofollow" class="external text" href="https://drive.google.com/file/d/0B65v6Wo67Tk5ODRzZmhSR29VeDg/view?usp=sharing">"Parallel distributed processing model with local space-invariant interconnections and its optical architecture"</a>. <i>Applied Optics</i>. <b>29</b> (32): 4790–7. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/1990ApOpt..29.4790Z">1990ApOpt..29.4790Z</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1364%2FAO.29.004790">10.1364/AO.29.004790</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/20577468">20577468</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Applied+Optics&rft.atitle=Parallel+distributed+processing+model+with+local+space-invariant+interconnections+and+its+optical+architecture&rft.volume=29&rft.issue=32&rft.pages=4790-7&rft.date=1990&rft_id=info%3Apmid%2F20577468&rft_id=info%3Adoi%2F10.1364%2FAO.29.004790&rft_id=info%3Abibcode%2F1990ApOpt..29.4790Z&rft.aulast=Zhang&rft.aufirst=Wei&rft_id=https%3A%2F%2Fdrive.google.com%2Ffile%2Fd%2F0B65v6Wo67Tk5ODRzZmhSR29VeDg%2Fview%3Fusp%3Dsharing&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-32"><span class="mw-cite-backlink"><b><a href="#cite_ref-32">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZhang1991" class="citation journal cs1">Zhang, Wei (1991). <a rel="nofollow" class="external text" href="https://drive.google.com/file/d/0B65v6Wo67Tk5cm5DTlNGd0NPUmM/view?usp=sharing">"Image processing of human corneal endothelium based on a learning network"</a>. <i>Applied Optics</i>. <b>30</b> (29): 4211–7. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/1991ApOpt..30.4211Z">1991ApOpt..30.4211Z</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1364%2FAO.30.004211">10.1364/AO.30.004211</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/20706526">20706526</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Applied+Optics&rft.atitle=Image+processing+of+human+corneal+endothelium+based+on+a+learning+network&rft.volume=30&rft.issue=29&rft.pages=4211-7&rft.date=1991&rft_id=info%3Apmid%2F20706526&rft_id=info%3Adoi%2F10.1364%2FAO.30.004211&rft_id=info%3Abibcode%2F1991ApOpt..30.4211Z&rft.aulast=Zhang&rft.aufirst=Wei&rft_id=https%3A%2F%2Fdrive.google.com%2Ffile%2Fd%2F0B65v6Wo67Tk5cm5DTlNGd0NPUmM%2Fview%3Fusp%3Dsharing&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-33"><span class="mw-cite-backlink"><b><a href="#cite_ref-33">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZhang1994" class="citation journal cs1">Zhang, Wei (1994). <a rel="nofollow" class="external text" href="https://drive.google.com/file/d/0B65v6Wo67Tk5Ml9qeW5nQ3poVTQ/view?usp=sharing">"Computerized detection of clustered microcalcifications in digital mammograms using a shift-invariant artificial neural network"</a>. <i>Medical Physics</i>. <b>21</b> (4): 517–24. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/1994MedPh..21..517Z">1994MedPh..21..517Z</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1118%2F1.597177">10.1118/1.597177</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/8058017">8058017</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Medical+Physics&rft.atitle=Computerized+detection+of+clustered+microcalcifications+in+digital+mammograms+using+a+shift-invariant+artificial+neural+network&rft.volume=21&rft.issue=4&rft.pages=517-24&rft.date=1994&rft_id=info%3Apmid%2F8058017&rft_id=info%3Adoi%2F10.1118%2F1.597177&rft_id=info%3Abibcode%2F1994MedPh..21..517Z&rft.aulast=Zhang&rft.aufirst=Wei&rft_id=https%3A%2F%2Fdrive.google.com%2Ffile%2Fd%2F0B65v6Wo67Tk5Ml9qeW5nQ3poVTQ%2Fview%3Fusp%3Dsharing&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-34"><span class="mw-cite-backlink"><b><a href="#cite_ref-34">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFde_CarvalhoFairhurstBisset1994" class="citation journal cs1">de Carvalho, Andre C. L. F.; Fairhurst, Mike C.; Bisset, David (8 August 1994). "An integrated Boolean neural network for pattern classification". <i>Pattern Recognition Letters</i>. <b>15</b> (8): 807–813. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/1994PaReL..15..807D">1994PaReL..15..807D</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2F0167-8655%2894%2990009-4">10.1016/0167-8655(94)90009-4</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Pattern+Recognition+Letters&rft.atitle=An+integrated+Boolean+neural+network+for+pattern+classification&rft.volume=15&rft.issue=8&rft.pages=807-813&rft.date=1994-08-08&rft_id=info%3Adoi%2F10.1016%2F0167-8655%2894%2990009-4&rft_id=info%3Abibcode%2F1994PaReL..15..807D&rft.aulast=de+Carvalho&rft.aufirst=Andre+C.+L.+F.&rft.au=Fairhurst%2C+Mike+C.&rft.au=Bisset%2C+David&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-35"><span class="mw-cite-backlink"><b><a href="#cite_ref-35">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHintonDayanFreyNeal1995" class="citation journal cs1">Hinton, Geoffrey E.; Dayan, Peter; Frey, Brendan J.; Neal, Radford (26 May 1995). "The wake-sleep algorithm for unsupervised neural networks". <i>Science</i>. <b>268</b> (5214): 1158–1161. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/1995Sci...268.1158H">1995Sci...268.1158H</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1126%2Fscience.7761831">10.1126/science.7761831</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/7761831">7761831</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Science&rft.atitle=The+wake-sleep+algorithm+for+unsupervised+neural+networks&rft.volume=268&rft.issue=5214&rft.pages=1158-1161&rft.date=1995-05-26&rft_id=info%3Apmid%2F7761831&rft_id=info%3Adoi%2F10.1126%2Fscience.7761831&rft_id=info%3Abibcode%2F1995Sci...268.1158H&rft.aulast=Hinton&rft.aufirst=Geoffrey+E.&rft.au=Dayan%2C+Peter&rft.au=Frey%2C+Brendan+J.&rft.au=Neal%2C+Radford&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-HOCH1991-36"><span class="mw-cite-backlink">^ <a href="#cite_ref-HOCH1991_36-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-HOCH1991_36-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text">S. Hochreiter., "<a rel="nofollow" class="external text" href="http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf">Untersuchungen zu dynamischen neuronalen Netzen</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150306075401/http://people.idsia.ch/~juergen/SeppHochreiter1991ThesisAdvisorSchmidhuber.pdf">Archived</a> 2015-03-06 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a>," <i>Diploma thesis. Institut f. Informatik, Technische Univ. Munich. Advisor: J. Schmidhuber</i>, 1991.</span>
</li>
<li id="cite_note-HOCH2001-37"><span class="mw-cite-backlink"><b><a href="#cite_ref-HOCH2001_37-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHochreiter2001" class="citation book cs1">Hochreiter, S.; et al. (15 January 2001). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=NWOcMVA64aAC">"Gradient flow in recurrent nets: the difficulty of learning long-term dependencies"</a>. In Kolen, John F.; Kremer, Stefan C. (eds.). <i>A Field Guide to Dynamical Recurrent Networks</i>. John Wiley & Sons. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-7803-5369-5" title="Special:BookSources/978-0-7803-5369-5"><bdi>978-0-7803-5369-5</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Gradient+flow+in+recurrent+nets%3A+the+difficulty+of+learning+long-term+dependencies&rft.btitle=A+Field+Guide+to+Dynamical+Recurrent+Networks&rft.pub=John+Wiley+%26+Sons&rft.date=2001-01-15&rft.isbn=978-0-7803-5369-5&rft.aulast=Hochreiter&rft.aufirst=S.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DNWOcMVA64aAC&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-38"><span class="mw-cite-backlink"><b><a href="#cite_ref-38">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBehnke2003" class="citation book cs1">Behnke, Sven (2003). <i>Hierarchical Neural Networks for Image Interpretation</i>. Lecture Notes in Computer Science. Vol. 2766. Springer. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2Fb11963">10.1007/b11963</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/3-540-40722-7" title="Special:BookSources/3-540-40722-7"><bdi>3-540-40722-7</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:1304548">1304548</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Hierarchical+Neural+Networks+for+Image+Interpretation&rft.series=Lecture+Notes+in+Computer+Science&rft.pub=Springer&rft.date=2003&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A1304548%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1007%2Fb11963&rft.isbn=3-540-40722-7&rft.aulast=Behnke&rft.aufirst=Sven&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-39"><span class="mw-cite-backlink"><b><a href="#cite_ref-39">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMorganBourlardRenalsCohen1993" class="citation journal cs1">Morgan, Nelson; Bourlard, Hervé; Renals, Steve; Cohen, Michael; Franco, Horacio (1 August 1993). "Hybrid neural network/hidden markov model systems for continuous speech recognition". <i>International Journal of Pattern Recognition and Artificial Intelligence</i>. <b>07</b> (4): 899–916. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1142%2Fs0218001493000455">10.1142/s0218001493000455</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0218-0014">0218-0014</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=International+Journal+of+Pattern+Recognition+and+Artificial+Intelligence&rft.atitle=Hybrid+neural+network%2Fhidden+markov+model+systems+for+continuous+speech+recognition&rft.volume=07&rft.issue=4&rft.pages=899-916&rft.date=1993-08-01&rft_id=info%3Adoi%2F10.1142%2Fs0218001493000455&rft.issn=0218-0014&rft.aulast=Morgan&rft.aufirst=Nelson&rft.au=Bourlard%2C+Herv%C3%A9&rft.au=Renals%2C+Steve&rft.au=Cohen%2C+Michael&rft.au=Franco%2C+Horacio&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Robinson1992-40"><span class="mw-cite-backlink"><b><a href="#cite_ref-Robinson1992_40-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRobinson1992" class="citation journal cs1"><a href="/wiki/Tony_Robinson_(speech_recognition)" title="Tony Robinson (speech recognition)">Robinson, T.</a> (1992). <a rel="nofollow" class="external text" href="http://dl.acm.org/citation.cfm?id=1895720">"A real-time recurrent error propagation network word recognition system"</a>. <i>ICASSP</i>. Icassp'92: 617–620. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/9780780305328" title="Special:BookSources/9780780305328"><bdi>9780780305328</bdi></a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123135/https://dl.acm.org/doi/10.5555/1895550.1895720">Archived</a> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-12</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=ICASSP&rft.atitle=A+real-time+recurrent+error+propagation+network+word+recognition+system&rft.pages=617-620&rft.date=1992&rft.isbn=9780780305328&rft.aulast=Robinson&rft.aufirst=T.&rft_id=http%3A%2F%2Fdl.acm.org%2Fcitation.cfm%3Fid%3D1895720&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-41"><span class="mw-cite-backlink"><b><a href="#cite_ref-41">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFWaibelHanazawaHintonShikano1989" class="citation journal cs1">Waibel, A.; Hanazawa, T.; Hinton, G.; Shikano, K.; Lang, K. J. (March 1989). <a rel="nofollow" class="external text" href="http://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf">"Phoneme recognition using time-delay neural networks"</a> <span class="cs1-format">(PDF)</span>. <i>IEEE Transactions on Acoustics, Speech, and Signal Processing</i>. <b>37</b> (3): 328–339. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2F29.21701">10.1109/29.21701</a>. <a href="/wiki/Hdl_(identifier)" class="mw-redirect" title="Hdl (identifier)">hdl</a>:<a rel="nofollow" class="external text" href="https://hdl.handle.net/10338.dmlcz%2F135496">10338.dmlcz/135496</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0096-3518">0096-3518</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210427001446/https://dml.cz/bitstream/handle/10338.dmlcz/135496/Kybernetika_38-2002-6_2.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2021-04-27<span class="reference-accessdate">. Retrieved <span class="nowrap">2019-09-24</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Acoustics%2C+Speech%2C+and+Signal+Processing&rft.atitle=Phoneme+recognition+using+time-delay+neural+networks&rft.volume=37&rft.issue=3&rft.pages=328-339&rft.date=1989-03&rft_id=info%3Ahdl%2F10338.dmlcz%2F135496&rft.issn=0096-3518&rft_id=info%3Adoi%2F10.1109%2F29.21701&rft.aulast=Waibel&rft.aufirst=A.&rft.au=Hanazawa%2C+T.&rft.au=Hinton%2C+G.&rft.au=Shikano%2C+K.&rft.au=Lang%2C+K.+J.&rft_id=http%3A%2F%2Fdml.cz%2Fbitstream%2Fhandle%2F10338.dmlcz%2F135496%2FKybernetika_38-2002-6_2.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Baker2009-42"><span class="mw-cite-backlink"><b><a href="#cite_ref-Baker2009_42-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBakerDengGlassKhudanpur2009" class="citation journal cs1">Baker, J.; Deng, Li; Glass, Jim; Khudanpur, S.; Lee, C.-H.; Morgan, N.; O'Shaughnessy, D. (2009). "Research Developments and Directions in Speech Recognition and Understanding, Part 1". <i>IEEE Signal Processing Magazine</i>. <b>26</b> (3): 75–80. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2009ISPM...26...75B">2009ISPM...26...75B</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Fmsp.2009.932166">10.1109/msp.2009.932166</a>. <a href="/wiki/Hdl_(identifier)" class="mw-redirect" title="Hdl (identifier)">hdl</a>:<a rel="nofollow" class="external text" href="https://hdl.handle.net/1721.1%2F51891">1721.1/51891</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:357467">357467</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Signal+Processing+Magazine&rft.atitle=Research+Developments+and+Directions+in+Speech+Recognition+and+Understanding%2C+Part+1&rft.volume=26&rft.issue=3&rft.pages=75-80&rft.date=2009&rft_id=info%3Ahdl%2F1721.1%2F51891&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A357467%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2Fmsp.2009.932166&rft_id=info%3Abibcode%2F2009ISPM...26...75B&rft.aulast=Baker&rft.aufirst=J.&rft.au=Deng%2C+Li&rft.au=Glass%2C+Jim&rft.au=Khudanpur%2C+S.&rft.au=Lee%2C+C.-H.&rft.au=Morgan%2C+N.&rft.au=O%27Shaughnessy%2C+D.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Bengio1991-43"><span class="mw-cite-backlink"><b><a href="#cite_ref-Bengio1991_43-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBengio1991" class="citation web cs1">Bengio, Y. (1991). <a rel="nofollow" class="external text" href="https://www.researchgate.net/publication/41229141">"Artificial Neural Networks and their Application to Speech/Sequence Recognition"</a>. McGill University Ph.D. thesis. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123131/https://www.researchgate.net/publication/41229141_Artificial_neural_networks_and_their_application_to_sequence_recognition">Archived</a> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-12</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Artificial+Neural+Networks+and+their+Application+to+Speech%2FSequence+Recognition&rft.pub=McGill+University+Ph.D.+thesis&rft.date=1991&rft.aulast=Bengio&rft.aufirst=Y.&rft_id=https%3A%2F%2Fwww.researchgate.net%2Fpublication%2F41229141&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Deng1994-44"><span class="mw-cite-backlink"><b><a href="#cite_ref-Deng1994_44-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDengHassaneinElmasry1994" class="citation journal cs1">Deng, L.; Hassanein, K.; Elmasry, M. (1994). "Analysis of correlation structure for a neural predictive model with applications to speech recognition". <i>Neural Networks</i>. <b>7</b> (2): 331–339. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2F0893-6080%2894%2990027-2">10.1016/0893-6080(94)90027-2</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Networks&rft.atitle=Analysis+of+correlation+structure+for+a+neural+predictive+model+with+applications+to+speech+recognition&rft.volume=7&rft.issue=2&rft.pages=331-339&rft.date=1994&rft_id=info%3Adoi%2F10.1016%2F0893-6080%2894%2990027-2&rft.aulast=Deng&rft.aufirst=L.&rft.au=Hassanein%2C+K.&rft.au=Elmasry%2C+M.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Doddington2000-45"><span class="mw-cite-backlink"><b><a href="#cite_ref-Doddington2000_45-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDoddingtonPrzybockiMartinReynolds2000" class="citation journal cs1">Doddington, G.; Przybocki, M.; Martin, A.; Reynolds, D. (2000). "The NIST speaker recognition evaluation ± Overview, methodology, systems, results, perspective". <i>Speech Communication</i>. <b>31</b> (2): 225–254. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2FS0167-6393%2899%2900080-1">10.1016/S0167-6393(99)00080-1</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Speech+Communication&rft.atitle=The+NIST+speaker+recognition+evaluation+%C2%B1+Overview%2C+methodology%2C+systems%2C+results%2C+perspective&rft.volume=31&rft.issue=2&rft.pages=225-254&rft.date=2000&rft_id=info%3Adoi%2F10.1016%2FS0167-6393%2899%2900080-1&rft.aulast=Doddington&rft.aufirst=G.&rft.au=Przybocki%2C+M.&rft.au=Martin%2C+A.&rft.au=Reynolds%2C+D.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Heck2000-46"><span class="mw-cite-backlink">^ <a href="#cite_ref-Heck2000_46-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-Heck2000_46-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHeckKonigSonmezWeintraub2000" class="citation journal cs1">Heck, L.; Konig, Y.; Sonmez, M.; Weintraub, M. (2000). "Robustness to Telephone Handset Distortion in Speaker Recognition by Discriminative Feature Design". <i>Speech Communication</i>. <b>31</b> (2): 181–192. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fs0167-6393%2899%2900077-1">10.1016/s0167-6393(99)00077-1</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Speech+Communication&rft.atitle=Robustness+to+Telephone+Handset+Distortion+in+Speaker+Recognition+by+Discriminative+Feature+Design&rft.volume=31&rft.issue=2&rft.pages=181-192&rft.date=2000&rft_id=info%3Adoi%2F10.1016%2Fs0167-6393%2899%2900077-1&rft.aulast=Heck&rft.aufirst=L.&rft.au=Konig%2C+Y.&rft.au=Sonmez%2C+M.&rft.au=Weintraub%2C+M.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-47"><span class="mw-cite-backlink"><b><a href="#cite_ref-47">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.researchgate.net/publication/266030526">"Acoustic Modeling with Deep Neural Networks Using Raw Time Signal for LVCSR (PDF Download Available)"</a>. <i>ResearchGate</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123218/https://www.researchgate.net/publication/266030526_Acoustic_Modeling_with_Deep_Neural_Networks_Using_Raw_Time_Signal_for_LVCSR">Archived</a> from the original on 9 May 2021<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=ResearchGate&rft.atitle=Acoustic+Modeling+with+Deep+Neural+Networks+Using+Raw+Time+Signal+for+LVCSR+%28PDF+Download+Available%29&rft_id=https%3A%2F%2Fwww.researchgate.net%2Fpublication%2F266030526&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:0-48"><span class="mw-cite-backlink">^ <a href="#cite_ref-:0_48-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:0_48-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-:0_48-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHochreiterSchmidhuber1997" class="citation journal cs1">Hochreiter, Sepp; Schmidhuber, Jürgen (1 November 1997). "Long Short-Term Memory". <i>Neural Computation</i>. <b>9</b> (8): 1735–1780. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1162%2Fneco.1997.9.8.1735">10.1162/neco.1997.9.8.1735</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0899-7667">0899-7667</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/9377276">9377276</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:1915014">1915014</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Computation&rft.atitle=Long+Short-Term+Memory&rft.volume=9&rft.issue=8&rft.pages=1735-1780&rft.date=1997-11-01&rft.issn=0899-7667&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A1915014%23id-name%3DS2CID&rft_id=info%3Apmid%2F9377276&rft_id=info%3Adoi%2F10.1162%2Fneco.1997.9.8.1735&rft.aulast=Hochreiter&rft.aufirst=Sepp&rft.au=Schmidhuber%2C+J%C3%BCrgen&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-graves2003-49"><span class="mw-cite-backlink">^ <a href="#cite_ref-graves2003_49-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-graves2003_49-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGravesEckBeringerSchmidhuber2003" class="citation web cs1">Graves, Alex; Eck, Douglas; Beringer, Nicole; Schmidhuber, Jürgen (2003). <a rel="nofollow" class="external text" href="Ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf">"Biologically Plausible Speech Recognition with LSTM Neural Nets"</a> <span class="cs1-format">(PDF)</span>. <i>1st Intl. Workshop on Biologically Inspired Approaches to Advanced Information Technology, Bio-ADIT 2004, Lausanne, Switzerland</i>. pp. 175–184. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123139/ftp://ftp.idsia.ch/pub/juergen/bioadit2004.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2016-04-09</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=1st+Intl.+Workshop+on+Biologically+Inspired+Approaches+to+Advanced+Information+Technology%2C+Bio-ADIT+2004%2C+Lausanne%2C+Switzerland&rft.atitle=Biologically+Plausible+Speech+Recognition+with+LSTM+Neural+Nets&rft.pages=175-184&rft.date=2003&rft.aulast=Graves&rft.aufirst=Alex&rft.au=Eck%2C+Douglas&rft.au=Beringer%2C+Nicole&rft.au=Schmidhuber%2C+J%C3%BCrgen&rft_id=Ftp%3A%2F%2Fftp.idsia.ch%2Fpub%2Fjuergen%2Fbioadit2004.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:1-50"><span class="mw-cite-backlink">^ <a href="#cite_ref-:1_50-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:1_50-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGravesFernándezGomez2006" class="citation journal cs1">Graves, Alex; Fernández, Santiago; Gomez, Faustino (2006). "Connectionist temporal classification: Labelling unsegmented sequence data with recurrent neural networks". <i>Proceedings of the International Conference on Machine Learning, ICML 2006</i>: 369–376. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.75.6306">10.1.1.75.6306</a></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+of+the+International+Conference+on+Machine+Learning%2C+ICML+2006&rft.atitle=Connectionist+temporal+classification%3A+Labelling+unsegmented+sequence+data+with+recurrent+neural+networks&rft.pages=369-376&rft.date=2006&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.75.6306%23id-name%3DCiteSeerX&rft.aulast=Graves&rft.aufirst=Alex&rft.au=Fern%C3%A1ndez%2C+Santiago&rft.au=Gomez%2C+Faustino&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-fernandez2007keyword-51"><span class="mw-cite-backlink"><b><a href="#cite_ref-fernandez2007keyword_51-0">^</a></b></span> <span class="reference-text">Santiago Fernandez, Alex Graves, and Jürgen Schmidhuber (2007). <a rel="nofollow" class="external text" href="https://mediatum.ub.tum.de/doc/1289941/file.pdf">An application of recurrent neural networks to discriminative keyword spotting</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20181118164457/https://mediatum.ub.tum.de/doc/1289941/file.pdf">Archived</a> 2018-11-18 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a>. Proceedings of ICANN (2), pp. 220–229.</span>
</li>
<li id="cite_note-sak2015-52"><span class="mw-cite-backlink">^ <a href="#cite_ref-sak2015_52-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-sak2015_52-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSakSeniorRaoBeaufays2015" class="citation web cs1">Sak, Haşim; Senior, Andrew; Rao, Kanishka; Beaufays, Françoise; Schalkwyk, Johan (September 2015). <a rel="nofollow" class="external text" href="http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html">"Google voice search: faster and more accurate"</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160309191532/http://googleresearch.blogspot.ch/2015/09/google-voice-search-faster-and-more.html">Archived</a> from the original on 2016-03-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2016-04-09</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Google+voice+search%3A+faster+and+more+accurate&rft.date=2015-09&rft.aulast=Sak&rft.aufirst=Ha%C5%9Fim&rft.au=Senior%2C+Andrew&rft.au=Rao%2C+Kanishka&rft.au=Beaufays%2C+Fran%C3%A7oise&rft.au=Schalkwyk%2C+Johan&rft_id=http%3A%2F%2Fgoogleresearch.blogspot.ch%2F2015%2F09%2Fgoogle-voice-search-faster-and-more.html&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-53"><span class="mw-cite-backlink"><b><a href="#cite_ref-53">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHinton2007" class="citation journal cs1">Hinton, Geoffrey E. (1 October 2007). <a rel="nofollow" class="external text" href="http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3">"Learning multiple layers of representation"</a>. <i>Trends in Cognitive Sciences</i>. <b>11</b> (10): 428–434. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.tics.2007.09.004">10.1016/j.tics.2007.09.004</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1364-6613">1364-6613</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/17921042">17921042</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:15066318">15066318</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20131011071435/http://www.cell.com/trends/cognitive-sciences/abstract/S1364-6613(07)00217-3">Archived</a> from the original on 11 October 2013<span class="reference-accessdate">. Retrieved <span class="nowrap">12 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Trends+in+Cognitive+Sciences&rft.atitle=Learning+multiple+layers+of+representation&rft.volume=11&rft.issue=10&rft.pages=428-434&rft.date=2007-10-01&rft.issn=1364-6613&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A15066318%23id-name%3DS2CID&rft_id=info%3Apmid%2F17921042&rft_id=info%3Adoi%2F10.1016%2Fj.tics.2007.09.004&rft.aulast=Hinton&rft.aufirst=Geoffrey+E.&rft_id=http%3A%2F%2Fwww.cell.com%2Ftrends%2Fcognitive-sciences%2Fabstract%2FS1364-6613%2807%2900217-3&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-hinton06-54"><span class="mw-cite-backlink"><b><a href="#cite_ref-hinton06_54-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHintonOsinderoTeh2006" class="citation journal cs1"><a href="/wiki/Geoff_Hinton" class="mw-redirect" title="Geoff Hinton">Hinton, G. E.</a>; Osindero, S.; Teh, Y. W. (2006). <a rel="nofollow" class="external text" href="http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf">"A Fast Learning Algorithm for Deep Belief Nets"</a> <span class="cs1-format">(PDF)</span>. <i><a href="/wiki/Neural_Computation_(journal)" title="Neural Computation (journal)">Neural Computation</a></i>. <b>18</b> (7): 1527–1554. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1162%2Fneco.2006.18.7.1527">10.1162/neco.2006.18.7.1527</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/16764513">16764513</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:2309950">2309950</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20151223164129/http://www.cs.toronto.edu/~hinton/absps/fastnc.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2015-12-23<span class="reference-accessdate">. Retrieved <span class="nowrap">2011-07-20</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Computation&rft.atitle=A+Fast+Learning+Algorithm+for+Deep+Belief+Nets&rft.volume=18&rft.issue=7&rft.pages=1527-1554&rft.date=2006&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A2309950%23id-name%3DS2CID&rft_id=info%3Apmid%2F16764513&rft_id=info%3Adoi%2F10.1162%2Fneco.2006.18.7.1527&rft.aulast=Hinton&rft.aufirst=G.+E.&rft.au=Osindero%2C+S.&rft.au=Teh%2C+Y.+W.&rft_id=http%3A%2F%2Fwww.cs.toronto.edu%2F~hinton%2Fabsps%2Ffastnc.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-bengio2012-55"><span class="mw-cite-backlink"><b><a href="#cite_ref-bengio2012_55-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBengio2012" class="citation arxiv cs1"><a href="/wiki/Yoshua_Bengio" title="Yoshua Bengio">Bengio, Yoshua</a> (2012). "Practical recommendations for gradient-based training of deep architectures". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1206.5533">1206.5533</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.LG">cs.LG</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Practical+recommendations+for+gradient-based+training+of+deep+architectures&rft.date=2012&rft_id=info%3Aarxiv%2F1206.5533&rft.aulast=Bengio&rft.aufirst=Yoshua&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-HINTON2007-56"><span class="mw-cite-backlink"><b><a href="#cite_ref-HINTON2007_56-0">^</a></b></span> <span class="reference-text">G. E. Hinton., "<a rel="nofollow" class="external text" href="http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf">Learning multiple layers of representation</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180522112408/http://www.csri.utoronto.ca/~hinton/absps/ticsdraft.pdf">Archived</a> 2018-05-22 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a>," <i>Trends in Cognitive Sciences</i>, 11, pp. 428–434, 2007.</span>
</li>
<li id="cite_note-HintonDengYu2012-57"><span class="mw-cite-backlink">^ <a href="#cite_ref-HintonDengYu2012_57-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-HintonDengYu2012_57-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-HintonDengYu2012_57-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHintonDengYuDahl2012" class="citation journal cs1">Hinton, G.; Deng, L.; Yu, D.; Dahl, G.; Mohamed, A.; Jaitly, N.; Senior, A.; Vanhoucke, V.; Nguyen, P.; Sainath, T.; Kingsbury, B. (2012). "Deep Neural Networks for Acoustic Modeling in Speech Recognition: The Shared Views of Four Research Groups". <i>IEEE Signal Processing Magazine</i>. <b>29</b> (6): 82–97. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2012ISPM...29...82H">2012ISPM...29...82H</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Fmsp.2012.2205597">10.1109/msp.2012.2205597</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:206485943">206485943</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Signal+Processing+Magazine&rft.atitle=Deep+Neural+Networks+for+Acoustic+Modeling+in+Speech+Recognition%3A+The+Shared+Views+of+Four+Research+Groups&rft.volume=29&rft.issue=6&rft.pages=82-97&rft.date=2012&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A206485943%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2Fmsp.2012.2205597&rft_id=info%3Abibcode%2F2012ISPM...29...82H&rft.aulast=Hinton&rft.aufirst=G.&rft.au=Deng%2C+L.&rft.au=Yu%2C+D.&rft.au=Dahl%2C+G.&rft.au=Mohamed%2C+A.&rft.au=Jaitly%2C+N.&rft.au=Senior%2C+A.&rft.au=Vanhoucke%2C+V.&rft.au=Nguyen%2C+P.&rft.au=Sainath%2C+T.&rft.au=Kingsbury%2C+B.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-58"><span class="mw-cite-backlink"><b><a href="#cite_ref-58">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSinghSahaSahidullah2021" class="citation book cs1">Singh, Premjeet; Saha, Goutam; Sahidullah, Md (2021). "Non-linear frequency warping using constant-Q transformation for speech emotion recognition". <i>2021 International Conference on Computer Communication and Informatics (ICCCI)</i>. pp. 1–4. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/2102.04029">2102.04029</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FICCCI50826.2021.9402569">10.1109/ICCCI50826.2021.9402569</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-7281-5875-4" title="Special:BookSources/978-1-7281-5875-4"><bdi>978-1-7281-5875-4</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:231846518">231846518</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Non-linear+frequency+warping+using+constant-Q+transformation+for+speech+emotion+recognition&rft.btitle=2021+International+Conference+on+Computer+Communication+and+Informatics+%28ICCCI%29&rft.pages=1-4&rft.date=2021&rft_id=info%3Aarxiv%2F2102.04029&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A231846518%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2FICCCI50826.2021.9402569&rft.isbn=978-1-7281-5875-4&rft.aulast=Singh&rft.aufirst=Premjeet&rft.au=Saha%2C+Goutam&rft.au=Sahidullah%2C+Md&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-sak2014-59"><span class="mw-cite-backlink"><b><a href="#cite_ref-sak2014_59-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSakSeniorBeaufays2014" class="citation web cs1">Sak, Hasim; Senior, Andrew; Beaufays, Francoise (2014). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180424203806/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf">"Long Short-Term Memory recurrent neural network architectures for large scale acoustic modeling"</a> <span class="cs1-format">(PDF)</span>. Archived from <a rel="nofollow" class="external text" href="https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43905.pdf">the original</a> <span class="cs1-format">(PDF)</span> on 24 April 2018.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Long+Short-Term+Memory+recurrent+neural+network+architectures+for+large+scale+acoustic+modeling&rft.date=2014&rft.aulast=Sak&rft.aufirst=Hasim&rft.au=Senior%2C+Andrew&rft.au=Beaufays%2C+Francoise&rft_id=https%3A%2F%2Fstatic.googleusercontent.com%2Fmedia%2Fresearch.google.com%2Fen%2F%2Fpubs%2Farchive%2F43905.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-liwu2015-60"><span class="mw-cite-backlink"><b><a href="#cite_ref-liwu2015_60-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFLiWu2014" class="citation arxiv cs1">Li, Xiangang; Wu, Xihong (2014). "Constructing Long Short-Term Memory based Deep Recurrent Neural Networks for Large Vocabulary Speech Recognition". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1410.4281">1410.4281</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CL">cs.CL</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Constructing+Long+Short-Term+Memory+based+Deep+Recurrent+Neural+Networks+for+Large+Vocabulary+Speech+Recognition&rft.date=2014&rft_id=info%3Aarxiv%2F1410.4281&rft.aulast=Li&rft.aufirst=Xiangang&rft.au=Wu%2C+Xihong&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-zen2015-61"><span class="mw-cite-backlink"><b><a href="#cite_ref-zen2015_61-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZenSak2015" class="citation web cs1">Zen, Heiga; Sak, Hasim (2015). <a rel="nofollow" class="external text" href="https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf">"Unidirectional Long Short-Term Memory Recurrent Neural Network with Recurrent Output Layer for Low-Latency Speech Synthesis"</a> <span class="cs1-format">(PDF)</span>. <i>Google.com</i>. ICASSP. pp. 4470–4474. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123113/https://static.googleusercontent.com/media/research.google.com/en//pubs/archive/43266.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Google.com&rft.atitle=Unidirectional+Long+Short-Term+Memory+Recurrent+Neural+Network+with+Recurrent+Output+Layer+for+Low-Latency+Speech+Synthesis&rft.pages=4470-4474&rft.date=2015&rft.aulast=Zen&rft.aufirst=Heiga&rft.au=Sak%2C+Hasim&rft_id=https%3A%2F%2Fstatic.googleusercontent.com%2Fmedia%2Fresearch.google.com%2Fen%2F%2Fpubs%2Farchive%2F43266.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-lecun2016slides-62"><span class="mw-cite-backlink"><b><a href="#cite_ref-lecun2016slides_62-0">^</a></b></span> <span class="reference-text"><a href="/wiki/Yann_LeCun" title="Yann LeCun">Yann LeCun</a> (2016). Slides on Deep Learning <a rel="nofollow" class="external text" href="https://indico.cern.ch/event/510372/">Online</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160423021403/https://indico.cern.ch/event/510372/">Archived</a> 2016-04-23 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a></span>
</li>
<li id="cite_note-ReferenceICASSP2013-63"><span class="mw-cite-backlink">^ <a href="#cite_ref-ReferenceICASSP2013_63-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-ReferenceICASSP2013_63-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-ReferenceICASSP2013_63-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDengHintonKingsbury2013" class="citation journal cs1">Deng, L.; Hinton, G.; Kingsbury, B. (2013). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf">"New types of deep neural network learning for speech recognition and related applications: An overview (ICASSP)"</a> <span class="cs1-format">(PDF)</span>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170926190920/https://www.microsoft.com/en-us/research/wp-content/uploads/2016/02/ICASSP-2013-DengHintonKingsbury-revised.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-09-26<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-12</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=New+types+of+deep+neural+network+learning+for+speech+recognition+and+related+applications%3A+An+overview+%28ICASSP%29&rft.date=2013&rft.aulast=Deng&rft.aufirst=L.&rft.au=Hinton%2C+G.&rft.au=Kingsbury%2C+B.&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fwp-content%2Fuploads%2F2016%2F02%2FICASSP-2013-DengHintonKingsbury-revised.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span> <span class="cs1-hidden-error citation-comment"><code class="cs1-code">{{<a href="/wiki/Template:Cite_journal" title="Template:Cite journal">cite journal</a>}}</code>: </span><span class="cs1-hidden-error citation-comment">Cite journal requires <code class="cs1-code">|journal=</code> (<a href="/wiki/Help:CS1_errors#missing_periodical" title="Help:CS1 errors">help</a>)</span></span>
</li>
<li id="cite_note-ReferenceA-64"><span class="mw-cite-backlink">^ <a href="#cite_ref-ReferenceA_64-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-ReferenceA_64-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-ReferenceA_64-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFYuDeng2014" class="citation book cs1">Yu, D.; Deng, L. (2014). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=rUBTBQAAQBAJ"><i>Automatic Speech Recognition: A Deep Learning Approach (Publisher: Springer)</i></a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-4471-5779-3" title="Special:BookSources/978-1-4471-5779-3"><bdi>978-1-4471-5779-3</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Automatic+Speech+Recognition%3A+A+Deep+Learning+Approach+%28Publisher%3A+Springer%29&rft.date=2014&rft.isbn=978-1-4471-5779-3&rft.aulast=Yu&rft.aufirst=D.&rft.au=Deng%2C+L.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DrUBTBQAAQBAJ&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-65"><span class="mw-cite-backlink"><b><a href="#cite_ref-65">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/">"Deng receives prestigious IEEE Technical Achievement Award - Microsoft Research"</a>. <i>Microsoft Research</i>. 3 December 2015. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180316084821/https://www.microsoft.com/en-us/research/blog/deng-receives-prestigious-ieee-technical-achievement-award/">Archived</a> from the original on 16 March 2018<span class="reference-accessdate">. Retrieved <span class="nowrap">16 March</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Microsoft+Research&rft.atitle=Deng+receives+prestigious+IEEE+Technical+Achievement+Award+-+Microsoft+Research&rft.date=2015-12-03&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fblog%2Fdeng-receives-prestigious-ieee-technical-achievement-award%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-interspeech2014Keynote-66"><span class="mw-cite-backlink">^ <a href="#cite_ref-interspeech2014Keynote_66-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-interspeech2014Keynote_66-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFLi2014" class="citation web cs1">Li, Deng (September 2014). <a rel="nofollow" class="external text" href="https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing">"Keynote talk: 'Achievements and Challenges of Deep Learning - From Speech Analysis and Recognition To Language and Multimodal Processing'<span class="cs1-kern-right"></span>"</a>. <i>Interspeech</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170926190732/https://www.superlectures.com/interspeech2014/downloadFile?id=6&type=slides&filename=achievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing">Archived</a> from the original on 2017-09-26<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-12</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Interspeech&rft.atitle=Keynote+talk%3A+%27Achievements+and+Challenges+of+Deep+Learning+-+From+Speech+Analysis+and+Recognition+To+Language+and+Multimodal+Processing%27&rft.date=2014-09&rft.aulast=Li&rft.aufirst=Deng&rft_id=https%3A%2F%2Fwww.superlectures.com%2Finterspeech2014%2FdownloadFile%3Fid%3D6%26type%3Dslides%26filename%3Dachievements-and-challenges-of-deep-learning-from-speech-analysis-and-recognition-to-language-and-multimodal-processing&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Roles2010-67"><span class="mw-cite-backlink"><b><a href="#cite_ref-Roles2010_67-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFYuDeng2010" class="citation journal cs1">Yu, D.; Deng, L. (2010). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/">"Roles of Pre-Training and Fine-Tuning in Context-Dependent DBN-HMMs for Real-World Speech Recognition"</a>. <i>NIPS Workshop on Deep Learning and Unsupervised Feature Learning</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171012095148/https://www.microsoft.com/en-us/research/publication/roles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition/">Archived</a> from the original on 2017-10-12<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-14</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=NIPS+Workshop+on+Deep+Learning+and+Unsupervised+Feature+Learning&rft.atitle=Roles+of+Pre-Training+and+Fine-Tuning+in+Context-Dependent+DBN-HMMs+for+Real-World+Speech+Recognition&rft.date=2010&rft.aulast=Yu&rft.aufirst=D.&rft.au=Deng%2C+L.&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Froles-of-pre-training-and-fine-tuning-in-context-dependent-dbn-hmms-for-real-world-speech-recognition%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-68"><span class="mw-cite-backlink"><b><a href="#cite_ref-68">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSeideLiYu2011" class="citation journal cs1">Seide, F.; Li, G.; Yu, D. (2011). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks">"Conversational speech transcription using context-dependent deep neural networks"</a>. <i>Interspeech</i>: 437–440. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.21437%2FInterspeech.2011-169">10.21437/Interspeech.2011-169</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171012095522/https://www.microsoft.com/en-us/research/publication/conversational-speech-transcription-using-context-dependent-deep-neural-networks/">Archived</a> from the original on 2017-10-12<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-14</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Interspeech&rft.atitle=Conversational+speech+transcription+using+context-dependent+deep+neural+networks&rft.pages=437-440&rft.date=2011&rft_id=info%3Adoi%2F10.21437%2FInterspeech.2011-169&rft.aulast=Seide&rft.aufirst=F.&rft.au=Li%2C+G.&rft.au=Yu%2C+D.&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Fconversational-speech-transcription-using-context-dependent-deep-neural-networks&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-69"><span class="mw-cite-backlink"><b><a href="#cite_ref-69">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDengLiHuangYao2013" class="citation journal cs1">Deng, Li; Li, Jinyu; Huang, Jui-Ting; Yao, Kaisheng; Yu, Dong; Seide, Frank; Seltzer, Mike; Zweig, Geoff; He, Xiaodong (1 May 2013). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/">"Recent Advances in Deep Learning for Speech Research at Microsoft"</a>. <i>Microsoft Research</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171012044053/https://www.microsoft.com/en-us/research/publication/recent-advances-in-deep-learning-for-speech-research-at-microsoft/">Archived</a> from the original on 12 October 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Microsoft+Research&rft.atitle=Recent+Advances+in+Deep+Learning+for+Speech+Research+at+Microsoft&rft.date=2013-05-01&rft.aulast=Deng&rft.aufirst=Li&rft.au=Li%2C+Jinyu&rft.au=Huang%2C+Jui-Ting&rft.au=Yao%2C+Kaisheng&rft.au=Yu%2C+Dong&rft.au=Seide%2C+Frank&rft.au=Seltzer%2C+Mike&rft.au=Zweig%2C+Geoff&rft.au=He%2C+Xiaodong&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Frecent-advances-in-deep-learning-for-speech-research-at-microsoft%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-70"><span class="mw-cite-backlink"><b><a href="#cite_ref-70">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/">"Nvidia CEO bets big on deep learning and VR"</a>. <a href="/wiki/Venture_Beat" class="mw-redirect" title="Venture Beat">Venture Beat</a>. 5 April 2016. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20201125202428/https://venturebeat.com/2016/04/05/nvidia-ceo-bets-big-on-deep-learning-and-vr/">Archived</a> from the original on 25 November 2020<span class="reference-accessdate">. Retrieved <span class="nowrap">21 April</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Nvidia+CEO+bets+big+on+deep+learning+and+VR&rft.pub=Venture+Beat&rft.date=2016-04-05&rft_id=https%3A%2F%2Fventurebeat.com%2F2016%2F04%2F05%2Fnvidia-ceo-bets-big-on-deep-learning-and-vr%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-71"><span class="mw-cite-backlink"><b><a href="#cite_ref-71">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation news cs1"><a rel="nofollow" class="external text" href="https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not">"From not working to neural networking"</a>. <i><a href="/wiki/The_Economist" title="The Economist">The Economist</a></i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20161231203934/https://www.economist.com/news/special-report/21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not">Archived</a> from the original on 2016-12-31<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-08-26</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=The+Economist&rft.atitle=From+not+working+to+neural+networking&rft_id=https%3A%2F%2Fwww.economist.com%2Fnews%2Fspecial-report%2F21700756-artificial-intelligence-boom-based-old-idea-modern-twist-not&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-jung2004-72"><span class="mw-cite-backlink">^ <a href="#cite_ref-jung2004_72-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-jung2004_72-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFOhJung2004" class="citation journal cs1">Oh, K.-S.; Jung, K. (2004). "GPU implementation of neural networks". <i>Pattern Recognition</i>. <b>37</b> (6): 1311–1314. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2004PatRe..37.1311O">2004PatRe..37.1311O</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.patcog.2004.01.013">10.1016/j.patcog.2004.01.013</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Pattern+Recognition&rft.atitle=GPU+implementation+of+neural+networks&rft.volume=37&rft.issue=6&rft.pages=1311-1314&rft.date=2004&rft_id=info%3Adoi%2F10.1016%2Fj.patcog.2004.01.013&rft_id=info%3Abibcode%2F2004PatRe..37.1311O&rft.aulast=Oh&rft.aufirst=K.-S.&rft.au=Jung%2C+K.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-73"><span class="mw-cite-backlink"><b><a href="#cite_ref-73">^</a></b></span> <span class="reference-text">"<a rel="nofollow" class="external text" href="https://www.academia.edu/40135801">A Survey of Techniques for Optimizing Deep Learning on GPUs</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123120/https://www.academia.edu/40135801/A_Survey_of_Techniques_for_Optimizing_Deep_Learning_on_GPUs">Archived</a> 2021-05-09 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a>", S. Mittal and S. Vaishay, Journal of Systems Architecture, 2019</span>
</li>
<li id="cite_note-chellapilla2006-74"><span class="mw-cite-backlink">^ <a href="#cite_ref-chellapilla2006_74-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-chellapilla2006_74-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFChellapillaPuriSimard2006" class="citation cs2">Chellapilla, Kumar; Puri, Sidd; Simard, Patrice (2006), <a rel="nofollow" class="external text" href="https://hal.inria.fr/inria-00112631/document"><i>High performance convolutional neural networks for document processing</i></a>, <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200518193413/https://hal.inria.fr/inria-00112631/document">archived</a> from the original on 2020-05-18<span class="reference-accessdate">, retrieved <span class="nowrap">2021-02-14</span></span></cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=High+performance+convolutional+neural+networks+for+document+processing&rft.date=2006&rft.aulast=Chellapilla&rft.aufirst=Kumar&rft.au=Puri%2C+Sidd&rft.au=Simard%2C+Patrice&rft_id=https%3A%2F%2Fhal.inria.fr%2Finria-00112631%2Fdocument&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:3-75"><span class="mw-cite-backlink"><b><a href="#cite_ref-:3_75-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCireşanMeierGambardellaSchmidhuber2010" class="citation journal cs1">Cireşan, Dan Claudiu; Meier, Ueli; Gambardella, Luca Maria; Schmidhuber, Jürgen (21 September 2010). "Deep, Big, Simple Neural Nets for Handwritten Digit Recognition". <i>Neural Computation</i>. <b>22</b> (12): 3207–3220. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1003.0358">1003.0358</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1162%2Fneco_a_00052">10.1162/neco_a_00052</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0899-7667">0899-7667</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/20858131">20858131</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:1918673">1918673</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Computation&rft.atitle=Deep%2C+Big%2C+Simple+Neural+Nets+for+Handwritten+Digit+Recognition&rft.volume=22&rft.issue=12&rft.pages=3207-3220&rft.date=2010-09-21&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A1918673%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1162%2Fneco_a_00052&rft_id=info%3Aarxiv%2F1003.0358&rft.issn=0899-7667&rft_id=info%3Apmid%2F20858131&rft.aulast=Cire%C5%9Fan&rft.aufirst=Dan+Claudiu&rft.au=Meier%2C+Ueli&rft.au=Gambardella%2C+Luca+Maria&rft.au=Schmidhuber%2C+J%C3%BCrgen&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-76"><span class="mw-cite-backlink"><b><a href="#cite_ref-76">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRainaMadhavanNg2009" class="citation journal cs1">Raina, Rajat; Madhavan, Anand; Ng, Andrew Y. (2009). "Large-scale Deep Unsupervised Learning Using Graphics Processors". <i>Proceedings of the 26th Annual International Conference on Machine Learning</i>. ICML '09. New York, NY, USA: ACM: 873–880. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.154.372">10.1.1.154.372</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F1553374.1553486">10.1145/1553374.1553486</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/9781605585161" title="Special:BookSources/9781605585161"><bdi>9781605585161</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:392458">392458</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+of+the+26th+Annual+International+Conference+on+Machine+Learning&rft.atitle=Large-scale+Deep+Unsupervised+Learning+Using+Graphics+Processors&rft.pages=873-880&rft.date=2009&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.154.372%23id-name%3DCiteSeerX&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A392458%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1145%2F1553374.1553486&rft.isbn=9781605585161&rft.aulast=Raina&rft.aufirst=Rajat&rft.au=Madhavan%2C+Anand&rft.au=Ng%2C+Andrew+Y.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-sze2017-77"><span class="mw-cite-backlink"><b><a href="#cite_ref-sze2017_77-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSzeChenYangEmer2017" class="citation arxiv cs1"><a href="/wiki/Vivienne_Sze" title="Vivienne Sze">Sze, Vivienne</a>; Chen, Yu-Hsin; Yang, Tien-Ju; Emer, Joel (2017). "Efficient Processing of Deep Neural Networks: A Tutorial and Survey". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1703.09039">1703.09039</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CV">cs.CV</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Efficient+Processing+of+Deep+Neural+Networks%3A+A+Tutorial+and+Survey&rft.date=2017&rft_id=info%3Aarxiv%2F1703.09039&rft.aulast=Sze&rft.aufirst=Vivienne&rft.au=Chen%2C+Yu-Hsin&rft.au=Yang%2C+Tien-Ju&rft.au=Emer%2C+Joel&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-MERCK2012-78"><span class="mw-cite-backlink">^ <a href="#cite_ref-MERCK2012_78-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-MERCK2012_78-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://kaggle.com/c/MerckActivity">"Merck Molecular Activity Challenge"</a>. <i>kaggle.com</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200716190808/https://www.kaggle.com/c/MerckActivity">Archived</a> from the original on 2020-07-16<span class="reference-accessdate">. Retrieved <span class="nowrap">2020-07-16</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=kaggle.com&rft.atitle=Merck+Molecular+Activity+Challenge&rft_id=https%3A%2F%2Fkaggle.com%2Fc%2FMerckActivity&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:5-79"><span class="mw-cite-backlink">^ <a href="#cite_ref-:5_79-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:5_79-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions">"Multi-task Neural Networks for QSAR Predictions | Data Science Association"</a>. <i>www.datascienceassn.org</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170430142049/http://www.datascienceassn.org/content/multi-task-neural-networks-qsar-predictions">Archived</a> from the original on 30 April 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=www.datascienceassn.org&rft.atitle=Multi-task+Neural+Networks+for+QSAR+Predictions+%7C+Data+Science+Association&rft_id=http%3A%2F%2Fwww.datascienceassn.org%2Fcontent%2Fmulti-task-neural-networks-qsar-predictions&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-TOX21-80"><span class="mw-cite-backlink">^ <a href="#cite_ref-TOX21_80-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-TOX21_80-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text">"Toxicology in the 21st century Data Challenge"</span>
</li>
<li id="cite_note-TOX21Data-81"><span class="mw-cite-backlink">^ <a href="#cite_ref-TOX21Data_81-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-TOX21Data_81-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://tripod.nih.gov/tox21/challenge/leaderboard.jsp">"NCATS Announces Tox21 Data Challenge Winners"</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150908025122/https://tripod.nih.gov/tox21/challenge/leaderboard.jsp">Archived</a> from the original on 2015-09-08<span class="reference-accessdate">. Retrieved <span class="nowrap">2015-03-05</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=NCATS+Announces+Tox21+Data+Challenge+Winners&rft_id=https%3A%2F%2Ftripod.nih.gov%2Ftox21%2Fchallenge%2Fleaderboard.jsp&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:11-82"><span class="mw-cite-backlink">^ <a href="#cite_ref-:11_82-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:11_82-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://web.archive.org/web/20150228225709/http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html">"NCATS Announces Tox21 Data Challenge Winners"</a>. Archived from <a rel="nofollow" class="external text" href="http://www.ncats.nih.gov/news-and-events/features/tox21-challenge-winners.html">the original</a> on 28 February 2015<span class="reference-accessdate">. Retrieved <span class="nowrap">5 March</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=NCATS+Announces+Tox21+Data+Challenge+Winners&rft_id=http%3A%2F%2Fwww.ncats.nih.gov%2Fnews-and-events%2Ffeatures%2Ftox21-challenge-winners.html&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:6-83"><span class="mw-cite-backlink"><b><a href="#cite_ref-:6_83-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCiresanMeierMasciGambardella2011" class="citation journal cs1">Ciresan, D. C.; Meier, U.; Masci, J.; Gambardella, L. M.; Schmidhuber, J. (2011). <a rel="nofollow" class="external text" href="http://ijcai.org/papers11/Papers/IJCAI11-210.pdf">"Flexible, High Performance Convolutional Neural Networks for Image Classification"</a> <span class="cs1-format">(PDF)</span>. <i>International Joint Conference on Artificial Intelligence</i>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.5591%2F978-1-57735-516-8%2Fijcai11-210">10.5591/978-1-57735-516-8/ijcai11-210</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20140929094040/http://ijcai.org/papers11/Papers/IJCAI11-210.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2014-09-29<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=International+Joint+Conference+on+Artificial+Intelligence&rft.atitle=Flexible%2C+High+Performance+Convolutional+Neural+Networks+for+Image+Classification&rft.date=2011&rft_id=info%3Adoi%2F10.5591%2F978-1-57735-516-8%2Fijcai11-210&rft.aulast=Ciresan&rft.aufirst=D.+C.&rft.au=Meier%2C+U.&rft.au=Masci%2C+J.&rft.au=Gambardella%2C+L.+M.&rft.au=Schmidhuber%2C+J.&rft_id=http%3A%2F%2Fijcai.org%2Fpapers11%2FPapers%2FIJCAI11-210.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:8-84"><span class="mw-cite-backlink"><b><a href="#cite_ref-:8_84-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCiresanGiustiGambardellaSchmidhuber2012" class="citation book cs1">Ciresan, Dan; Giusti, Alessandro; Gambardella, Luca M.; Schmidhuber, Juergen (2012). Pereira, F.; Burges, C. J. C.; Bottou, L.; Weinberger, K. Q. (eds.). <a rel="nofollow" class="external text" href="http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf"><i>Advances in Neural Information Processing Systems 25</i></a> <span class="cs1-format">(PDF)</span>. Curran Associates, Inc. pp. 2843–2851. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170809081713/http://papers.nips.cc/paper/4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-08-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Advances+in+Neural+Information+Processing+Systems+25&rft.pages=2843-2851&rft.pub=Curran+Associates%2C+Inc.&rft.date=2012&rft.aulast=Ciresan&rft.aufirst=Dan&rft.au=Giusti%2C+Alessandro&rft.au=Gambardella%2C+Luca+M.&rft.au=Schmidhuber%2C+Juergen&rft_id=http%3A%2F%2Fpapers.nips.cc%2Fpaper%2F4741-deep-neural-networks-segment-neuronal-membranes-in-electron-microscopy-images.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:9-85"><span class="mw-cite-backlink"><b><a href="#cite_ref-:9_85-0">^</a></b></span> <span class="error mw-ext-cite-error" lang="en" dir="ltr">Cite error: The named reference <code>:9</code> was invoked but never defined (see the <a href="/wiki/Help:Cite_errors/Cite_error_references_no_text" title="Help:Cite errors/Cite error references no text">help page</a>).</span></li>
<li id="cite_note-krizhevsky2012-86"><span class="mw-cite-backlink"><b><a href="#cite_ref-krizhevsky2012_86-0">^</a></b></span> <span class="error mw-ext-cite-error" lang="en" dir="ltr">Cite error: The named reference <code>krizhevsky2012</code> was invoked but never defined (see the <a href="/wiki/Help:Cite_errors/Cite_error_references_no_text" title="Help:Cite errors/Cite error references no text">help page</a>).</span></li>
<li id="cite_note-ciresan2013miccai-87"><span class="mw-cite-backlink"><b><a href="#cite_ref-ciresan2013miccai_87-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCiresanGiustiGambardellaSchmidhuber2013" class="citation journal cs1">Ciresan, D.; Giusti, A.; Gambardella, L.M.; Schmidhuber, J. (2013). "Mitosis Detection in Breast Cancer Histology Images using Deep Neural Networks". <i>Proceedings MICCAI</i>. Lecture Notes in Computer Science. <b>7908</b> (Pt 2): 411–418. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2F978-3-642-40763-5_51">10.1007/978-3-642-40763-5_51</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-3-642-38708-1" title="Special:BookSources/978-3-642-38708-1"><bdi>978-3-642-38708-1</bdi></a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/24579167">24579167</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+MICCAI&rft.atitle=Mitosis+Detection+in+Breast+Cancer+Histology+Images+using+Deep+Neural+Networks&rft.volume=7908&rft.issue=Pt+2&rft.pages=411-418&rft.date=2013&rft_id=info%3Apmid%2F24579167&rft_id=info%3Adoi%2F10.1007%2F978-3-642-40763-5_51&rft.isbn=978-3-642-38708-1&rft.aulast=Ciresan&rft.aufirst=D.&rft.au=Giusti%2C+A.&rft.au=Gambardella%2C+L.M.&rft.au=Schmidhuber%2C+J.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-1411.4555-88"><span class="mw-cite-backlink"><b><a href="#cite_ref-1411.4555_88-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFVinyalsToshevBengioErhan2014" class="citation arxiv cs1">Vinyals, Oriol; Toshev, Alexander; Bengio, Samy; Erhan, Dumitru (2014). "Show and Tell: A Neural Image Caption Generator". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1411.4555">1411.4555</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CV">cs.CV</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Show+and+Tell%3A+A+Neural+Image+Caption+Generator&rft.date=2014&rft_id=info%3Aarxiv%2F1411.4555&rft.aulast=Vinyals&rft.aufirst=Oriol&rft.au=Toshev%2C+Alexander&rft.au=Bengio%2C+Samy&rft.au=Erhan%2C+Dumitru&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span>.</span>
</li>
<li id="cite_note-1411.4952-89"><span class="mw-cite-backlink"><b><a href="#cite_ref-1411.4952_89-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFFangGuptaIandolaSrivastava2014" class="citation arxiv cs1">Fang, Hao; Gupta, Saurabh; Iandola, Forrest; Srivastava, Rupesh; Deng, Li; Dollár, Piotr; Gao, Jianfeng; He, Xiaodong; Mitchell, Margaret; Platt, John C; Lawrence Zitnick, C; Zweig, Geoffrey (2014). "From Captions to Visual Concepts and Back". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1411.4952">1411.4952</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CV">cs.CV</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=From+Captions+to+Visual+Concepts+and+Back&rft.date=2014&rft_id=info%3Aarxiv%2F1411.4952&rft.aulast=Fang&rft.aufirst=Hao&rft.au=Gupta%2C+Saurabh&rft.au=Iandola%2C+Forrest&rft.au=Srivastava%2C+Rupesh&rft.au=Deng%2C+Li&rft.au=Doll%C3%A1r%2C+Piotr&rft.au=Gao%2C+Jianfeng&rft.au=He%2C+Xiaodong&rft.au=Mitchell%2C+Margaret&rft.au=Platt%2C+John+C&rft.au=Lawrence+Zitnick%2C+C&rft.au=Zweig%2C+Geoffrey&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span>.</span>
</li>
<li id="cite_note-1411.2539-90"><span class="mw-cite-backlink"><b><a href="#cite_ref-1411.2539_90-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFKirosSalakhutdinovZemel2014" class="citation arxiv cs1">Kiros, Ryan; Salakhutdinov, Ruslan; Zemel, Richard S (2014). "Unifying Visual-Semantic Embeddings with Multimodal Neural Language Models". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1411.2539">1411.2539</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.LG">cs.LG</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Unifying+Visual-Semantic+Embeddings+with+Multimodal+Neural+Language+Models&rft.date=2014&rft_id=info%3Aarxiv%2F1411.2539&rft.aulast=Kiros&rft.aufirst=Ryan&rft.au=Salakhutdinov%2C+Ruslan&rft.au=Zemel%2C+Richard+S&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span>.</span>
</li>
<li id="cite_note-91"><span class="mw-cite-backlink"><b><a href="#cite_ref-91">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation news cs1"><a rel="nofollow" class="external text" href="http://fortune.com/ai-artificial-intelligence-deep-machine-learning/">"Why Deep Learning Is Suddenly Changing Your Life"</a>. <i>Fortune</i>. 2016. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180414031925/http://fortune.com/ai-artificial-intelligence-deep-machine-learning/">Archived</a> from the original on 14 April 2018<span class="reference-accessdate">. Retrieved <span class="nowrap">13 April</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Fortune&rft.atitle=Why+Deep+Learning+Is+Suddenly+Changing+Your+Life&rft.date=2016&rft_id=http%3A%2F%2Ffortune.com%2Fai-artificial-intelligence-deep-machine-learning%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-92"><span class="mw-cite-backlink"><b><a href="#cite_ref-92">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSilverHuangMaddisonGuez2016" class="citation journal cs1">Silver, David; Huang, Aja; Maddison, Chris J.; Guez, Arthur; Sifre, Laurent; Driessche, George van den; Schrittwieser, Julian; Antonoglou, Ioannis; Panneershelvam, Veda (January 2016). "Mastering the game of Go with deep neural networks and tree search". <i>Nature</i>. <b>529</b> (7587): 484–489. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2016Natur.529..484S">2016Natur.529..484S</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fnature16961">10.1038/nature16961</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1476-4687">1476-4687</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/26819042">26819042</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:515925">515925</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=Mastering+the+game+of+Go+with+deep+neural+networks+and+tree+search&rft.volume=529&rft.issue=7587&rft.pages=484-489&rft.date=2016-01&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A515925%23id-name%3DS2CID&rft_id=info%3Abibcode%2F2016Natur.529..484S&rft.issn=1476-4687&rft_id=info%3Adoi%2F10.1038%2Fnature16961&rft_id=info%3Apmid%2F26819042&rft.aulast=Silver&rft.aufirst=David&rft.au=Huang%2C+Aja&rft.au=Maddison%2C+Chris+J.&rft.au=Guez%2C+Arthur&rft.au=Sifre%2C+Laurent&rft.au=Driessche%2C+George+van+den&rft.au=Schrittwieser%2C+Julian&rft.au=Antonoglou%2C+Ioannis&rft.au=Panneershelvam%2C+Veda&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Nokkada-93"><span class="mw-cite-backlink"><b><a href="#cite_ref-Nokkada_93-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation cs2"><a rel="nofollow" class="external text" href="https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks"><i>A Guide to Deep Learning and Neural Networks</i></a>, <a rel="nofollow" class="external text" href="https://web.archive.org/web/20201102151103/https://serokell.io/blog/deep-learning-and-neural-network-guide#components-of-neural-networks">archived</a> from the original on 2020-11-02<span class="reference-accessdate">, retrieved <span class="nowrap">2020-11-16</span></span></cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=A+Guide+to+Deep+Learning+and+Neural+Networks&rft_id=https%3A%2F%2Fserokell.io%2Fblog%2Fdeep-learning-and-neural-network-guide%23components-of-neural-networks&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-94"><span class="mw-cite-backlink"><b><a href="#cite_ref-94">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSzegedyToshevErhan2013" class="citation journal cs1">Szegedy, Christian; Toshev, Alexander; Erhan, Dumitru (2013). <a rel="nofollow" class="external text" href="https://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection">"Deep neural networks for object detection"</a>. <i>Advances in Neural Information Processing Systems</i>: 2553–2561. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170629172111/http://papers.nips.cc/paper/5207-deep-neural-networks-for-object-detection">Archived</a> from the original on 2017-06-29<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Advances+in+Neural+Information+Processing+Systems&rft.atitle=Deep+neural+networks+for+object+detection&rft.pages=2553-2561&rft.date=2013&rft.aulast=Szegedy&rft.aufirst=Christian&rft.au=Toshev%2C+Alexander&rft.au=Erhan%2C+Dumitru&rft_id=https%3A%2F%2Fpapers.nips.cc%2Fpaper%2F5207-deep-neural-networks-for-object-detection&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-95"><span class="mw-cite-backlink"><b><a href="#cite_ref-95">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRolnickTegmark2018" class="citation conference cs1">Rolnick, David; Tegmark, Max (2018). <a rel="nofollow" class="external text" href="https://openreview.net/pdf?id=SyProzZAW">"The power of deeper networks for expressing natural functions"</a>. <i>International Conference on Learning Representations</i>. ICLR 2018. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210107183647/https://openreview.net/pdf?id=SyProzZAW">Archived</a> from the original on 2021-01-07<span class="reference-accessdate">. Retrieved <span class="nowrap">2021-01-05</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=conference&rft.atitle=The+power+of+deeper+networks+for+expressing+natural+functions&rft.btitle=International+Conference+on+Learning+Representations&rft.date=2018&rft.aulast=Rolnick&rft.aufirst=David&rft.au=Tegmark%2C+Max&rft_id=https%3A%2F%2Fopenreview.net%2Fpdf%3Fid%3DSyProzZAW&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-96"><span class="mw-cite-backlink"><b><a href="#cite_ref-96">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHof" class="citation news cs1">Hof, Robert D. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20190331092832/https://www.technologyreview.com/s/513696/deep-learning/">"Is Artificial Intelligence Finally Coming into Its Own?"</a>. <i>MIT Technology Review</i>. Archived from <a rel="nofollow" class="external text" href="https://www.technologyreview.com/s/513696/deep-learning/">the original</a> on 31 March 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">10 July</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=MIT+Technology+Review&rft.atitle=Is+Artificial+Intelligence+Finally+Coming+into+Its+Own%3F&rft.aulast=Hof&rft.aufirst=Robert+D.&rft_id=https%3A%2F%2Fwww.technologyreview.com%2Fs%2F513696%2Fdeep-learning%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-gers2001-97"><span class="mw-cite-backlink">^ <a href="#cite_ref-gers2001_97-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-gers2001_97-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGersSchmidhuber2001" class="citation journal cs1">Gers, Felix A.; Schmidhuber, Jürgen (2001). <a rel="nofollow" class="external text" href="http://elartu.tntu.edu.ua/handle/lib/30719">"LSTM Recurrent Networks Learn Simple Context Free and Context Sensitive Languages"</a>. <i>IEEE Transactions on Neural Networks</i>. <b>12</b> (6): 1333–1340. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2F72.963769">10.1109/72.963769</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/18249962">18249962</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200126045722/http://elartu.tntu.edu.ua/handle/lib/30719">Archived</a> from the original on 2020-01-26<span class="reference-accessdate">. Retrieved <span class="nowrap">2020-02-25</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Neural+Networks&rft.atitle=LSTM+Recurrent+Networks+Learn+Simple+Context+Free+and+Context+Sensitive+Languages&rft.volume=12&rft.issue=6&rft.pages=1333-1340&rft.date=2001&rft_id=info%3Adoi%2F10.1109%2F72.963769&rft_id=info%3Apmid%2F18249962&rft.aulast=Gers&rft.aufirst=Felix+A.&rft.au=Schmidhuber%2C+J%C3%BCrgen&rft_id=http%3A%2F%2Felartu.tntu.edu.ua%2Fhandle%2Flib%2F30719&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-NIPS2014-98"><span class="mw-cite-backlink">^ <a href="#cite_ref-NIPS2014_98-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-NIPS2014_98-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-NIPS2014_98-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSutskeverVinyalsLe2014" class="citation journal cs1">Sutskever, L.; Vinyals, O.; Le, Q. (2014). <a rel="nofollow" class="external text" href="https://papers.nips.cc/paper/5346-sequence-to-sequence-learning-with-neural-networks.pdf">"Sequence to Sequence Learning with Neural Networks"</a> <span class="cs1-format">(PDF)</span>. <i>Proc. NIPS</i>. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1409.3215">1409.3215</a></span>. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2014arXiv1409.3215S">2014arXiv1409.3215S</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123145/https://papers.nips.cc/paper/2014/file/a14ac55a4f27472c5d894ec1c3c743d2-Paper.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proc.+NIPS&rft.atitle=Sequence+to+Sequence+Learning+with+Neural+Networks&rft.date=2014&rft_id=info%3Aarxiv%2F1409.3215&rft_id=info%3Abibcode%2F2014arXiv1409.3215S&rft.aulast=Sutskever&rft.aufirst=L.&rft.au=Vinyals%2C+O.&rft.au=Le%2C+Q.&rft_id=https%3A%2F%2Fpapers.nips.cc%2Fpaper%2F5346-sequence-to-sequence-learning-with-neural-networks.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-vinyals2016-99"><span class="mw-cite-backlink">^ <a href="#cite_ref-vinyals2016_99-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-vinyals2016_99-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFJozefowiczVinyalsSchusterShazeer2016" class="citation arxiv cs1">Jozefowicz, Rafal; Vinyals, Oriol; Schuster, Mike; Shazeer, Noam; Wu, Yonghui (2016). "Exploring the Limits of Language Modeling". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1602.02410">1602.02410</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CL">cs.CL</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Exploring+the+Limits+of+Language+Modeling&rft.date=2016&rft_id=info%3Aarxiv%2F1602.02410&rft.aulast=Jozefowicz&rft.aufirst=Rafal&rft.au=Vinyals%2C+Oriol&rft.au=Schuster%2C+Mike&rft.au=Shazeer%2C+Noam&rft.au=Wu%2C+Yonghui&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-gillick2015-100"><span class="mw-cite-backlink">^ <a href="#cite_ref-gillick2015_100-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-gillick2015_100-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGillickBrunkVinyalsSubramanya2015" class="citation arxiv cs1">Gillick, Dan; Brunk, Cliff; Vinyals, Oriol; Subramanya, Amarnag (2015). "Multilingual Language Processing from Bytes". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1512.00103">1512.00103</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CL">cs.CL</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Multilingual+Language+Processing+from+Bytes&rft.date=2015&rft_id=info%3Aarxiv%2F1512.00103&rft.aulast=Gillick&rft.aufirst=Dan&rft.au=Brunk%2C+Cliff&rft.au=Vinyals%2C+Oriol&rft.au=Subramanya%2C+Amarnag&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-MIKO2010-101"><span class="mw-cite-backlink"><b><a href="#cite_ref-MIKO2010_101-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMikolov2010" class="citation journal cs1">Mikolov, T.; et al. (2010). <a rel="nofollow" class="external text" href="http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf">"Recurrent neural network based language model"</a> <span class="cs1-format">(PDF)</span>. <i>Interspeech</i>: 1045–1048. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.21437%2FInterspeech.2010-343">10.21437/Interspeech.2010-343</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170516181940/http://www.fit.vutbr.cz/research/groups/speech/servite/2010/rnnlm_mikolov.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-05-16<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Interspeech&rft.atitle=Recurrent+neural+network+based+language+model&rft.pages=1045-1048&rft.date=2010&rft_id=info%3Adoi%2F10.21437%2FInterspeech.2010-343&rft.aulast=Mikolov&rft.aufirst=T.&rft_id=http%3A%2F%2Fwww.fit.vutbr.cz%2Fresearch%2Fgroups%2Fspeech%2Fservite%2F2010%2Frnnlm_mikolov.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:10-102"><span class="mw-cite-backlink">^ <a href="#cite_ref-:10_102-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:10_102-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.researchgate.net/publication/220320057">"Learning Precise Timing with LSTM Recurrent Networks (PDF Download Available)"</a>. <i>ResearchGate</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123147/https://www.researchgate.net/publication/220320057_Learning_Precise_Timing_with_LSTM_Recurrent_Networks">Archived</a> from the original on 9 May 2021<span class="reference-accessdate">. Retrieved <span class="nowrap">13 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=ResearchGate&rft.atitle=Learning+Precise+Timing+with+LSTM+Recurrent+Networks+%28PDF+Download+Available%29&rft_id=https%3A%2F%2Fwww.researchgate.net%2Fpublication%2F220320057&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-LECUN86-103"><span class="mw-cite-backlink"><b><a href="#cite_ref-LECUN86_103-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFLeCun1998" class="citation journal cs1">LeCun, Y.; et al. (1998). <a rel="nofollow" class="external text" href="http://elartu.tntu.edu.ua/handle/lib/38369">"Gradient-based learning applied to document recognition"</a>. <i>Proceedings of the IEEE</i>. <b>86</b> (11): 2278–2324. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2F5.726791">10.1109/5.726791</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:14542261">14542261</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+of+the+IEEE&rft.atitle=Gradient-based+learning+applied+to+document+recognition&rft.volume=86&rft.issue=11&rft.pages=2278-2324&rft.date=1998&rft_id=info%3Adoi%2F10.1109%2F5.726791&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A14542261%23id-name%3DS2CID&rft.aulast=LeCun&rft.aufirst=Y.&rft_id=http%3A%2F%2Felartu.tntu.edu.ua%2Fhandle%2Flib%2F38369&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:2-104"><span class="mw-cite-backlink"><b><a href="#cite_ref-:2_104-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSainathMohamedKingsburyRamabhadran2013" class="citation book cs1">Sainath, Tara N.; Mohamed, Abdel-Rahman; Kingsbury, Brian; Ramabhadran, Bhuvana (2013). "Deep convolutional neural networks for LVCSR". <i>2013 IEEE International Conference on Acoustics, Speech and Signal Processing</i>. pp. 8614–8618. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Ficassp.2013.6639347">10.1109/icassp.2013.6639347</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-4799-0356-6" title="Special:BookSources/978-1-4799-0356-6"><bdi>978-1-4799-0356-6</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:13816461">13816461</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Deep+convolutional+neural+networks+for+LVCSR&rft.btitle=2013+IEEE+International+Conference+on+Acoustics%2C+Speech+and+Signal+Processing&rft.pages=8614-8618&rft.date=2013&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A13816461%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2Ficassp.2013.6639347&rft.isbn=978-1-4799-0356-6&rft.aulast=Sainath&rft.aufirst=Tara+N.&rft.au=Mohamed%2C+Abdel-Rahman&rft.au=Kingsbury%2C+Brian&rft.au=Ramabhadran%2C+Bhuvana&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-105"><span class="mw-cite-backlink"><b><a href="#cite_ref-105">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBengioBoulanger-LewandowskiPascanu2013" class="citation book cs1">Bengio, Yoshua; Boulanger-Lewandowski, Nicolas; Pascanu, Razvan (2013). "Advances in optimizing recurrent networks". <i>2013 IEEE International Conference on Acoustics, Speech and Signal Processing</i>. pp. 8624–8628. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1212.0901">1212.0901</a></span>. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.752.9151">10.1.1.752.9151</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Ficassp.2013.6639349">10.1109/icassp.2013.6639349</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-4799-0356-6" title="Special:BookSources/978-1-4799-0356-6"><bdi>978-1-4799-0356-6</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:12485056">12485056</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Advances+in+optimizing+recurrent+networks&rft.btitle=2013+IEEE+International+Conference+on+Acoustics%2C+Speech+and+Signal+Processing&rft.pages=8624-8628&rft.date=2013&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A12485056%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2Ficassp.2013.6639349&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.752.9151%23id-name%3DCiteSeerX&rft_id=info%3Aarxiv%2F1212.0901&rft.isbn=978-1-4799-0356-6&rft.aulast=Bengio&rft.aufirst=Yoshua&rft.au=Boulanger-Lewandowski%2C+Nicolas&rft.au=Pascanu%2C+Razvan&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-DAHL2013-106"><span class="mw-cite-backlink"><b><a href="#cite_ref-DAHL2013_106-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDahl2013" class="citation journal cs1">Dahl, G.; et al. (2013). <a rel="nofollow" class="external text" href="http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf">"Improving DNNs for LVCSR using rectified linear units and dropout"</a> <span class="cs1-format">(PDF)</span>. <i>ICASSP</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170812140509/http://www.cs.toronto.edu/~gdahl/papers/reluDropoutBN_icassp2013.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-08-12<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=ICASSP&rft.atitle=Improving+DNNs+for+LVCSR+using+rectified+linear+units+and+dropout&rft.date=2013&rft.aulast=Dahl&rft.aufirst=G.&rft_id=http%3A%2F%2Fwww.cs.toronto.edu%2F~gdahl%2Fpapers%2FreluDropoutBN_icassp2013.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-107"><span class="mw-cite-backlink"><b><a href="#cite_ref-107">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation">"Data Augmentation - deeplearning.ai | Coursera"</a>. <i>Coursera</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171201032606/https://www.coursera.org/learn/convolutional-neural-networks/lecture/AYzbX/data-augmentation">Archived</a> from the original on 1 December 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">30 November</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Coursera&rft.atitle=Data+Augmentation+-+deeplearning.ai+%7C+Coursera&rft_id=https%3A%2F%2Fwww.coursera.org%2Flearn%2Fconvolutional-neural-networks%2Flecture%2FAYzbX%2Fdata-augmentation&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-RBMTRAIN-108"><span class="mw-cite-backlink"><b><a href="#cite_ref-RBMTRAIN_108-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHinton2010" class="citation journal cs1">Hinton, G. E. (2010). <a rel="nofollow" class="external text" href="https://www.researchgate.net/publication/221166159">"A Practical Guide to Training Restricted Boltzmann Machines"</a>. <i>Tech. Rep. UTML TR 2010-003</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123211/https://www.researchgate.net/publication/221166159_A_brief_introduction_to_Weightless_Neural_Systems">Archived</a> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-13</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Tech.+Rep.+UTML+TR+2010-003&rft.atitle=A+Practical+Guide+to+Training+Restricted+Boltzmann+Machines&rft.date=2010&rft.aulast=Hinton&rft.aufirst=G.+E.&rft_id=https%3A%2F%2Fwww.researchgate.net%2Fpublication%2F221166159&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-109"><span class="mw-cite-backlink"><b><a href="#cite_ref-109">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFYouBuluçDemmel2017" class="citation book cs1">You, Yang; Buluç, Aydın; Demmel, James (November 2017). <a rel="nofollow" class="external text" href="https://dl.acm.org/citation.cfm?doid=3126908.3126912">"Scaling deep learning on GPU and knights landing clusters"</a>. <a rel="nofollow" class="external text" href="http://www.escholarship.org/uc/item/6ch40821"><i>Proceedings of the International Conference for High Performance Computing, Networking, Storage and Analysis on - SC '17</i></a>. SC '17, ACM. pp. 1–12. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F3126908.3126912">10.1145/3126908.3126912</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/9781450351140" title="Special:BookSources/9781450351140"><bdi>9781450351140</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:8869270">8869270</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200729133850/https://escholarship.org/uc/item/6ch40821">Archived</a> from the original on 29 July 2020<span class="reference-accessdate">. Retrieved <span class="nowrap">5 March</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Scaling+deep+learning+on+GPU+and+knights+landing+clusters&rft.btitle=Proceedings+of+the+International+Conference+for+High+Performance+Computing%2C+Networking%2C+Storage+and+Analysis+on+-+SC+%2717&rft.pages=1-12&rft.pub=SC+%2717%2C+ACM&rft.date=2017-11&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A8869270%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1145%2F3126908.3126912&rft.isbn=9781450351140&rft.aulast=You&rft.aufirst=Yang&rft.au=Bulu%C3%A7%2C+Ayd%C4%B1n&rft.au=Demmel%2C+James&rft_id=https%3A%2F%2Fdl.acm.org%2Fcitation.cfm%3Fdoid%3D3126908.3126912&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-110"><span class="mw-cite-backlink"><b><a href="#cite_ref-110">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFViebkeMemetiPllanaAbraham2019" class="citation journal cs1">Viebke, André; Memeti, Suejb; Pllana, Sabri; Abraham, Ajith (2019). "CHAOS: a parallelization scheme for training convolutional neural networks on Intel Xeon Phi". <i>The Journal of Supercomputing</i>. <b>75</b>: 197–227. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1702.07908">1702.07908</a></span>. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2017arXiv170207908V">2017arXiv170207908V</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1007%2Fs11227-017-1994-x">10.1007/s11227-017-1994-x</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:14135321">14135321</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=The+Journal+of+Supercomputing&rft.atitle=CHAOS%3A+a+parallelization+scheme+for+training+convolutional+neural+networks+on+Intel+Xeon+Phi&rft.volume=75&rft.pages=197-227&rft.date=2019&rft_id=info%3Aarxiv%2F1702.07908&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A14135321%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1007%2Fs11227-017-1994-x&rft_id=info%3Abibcode%2F2017arXiv170207908V&rft.aulast=Viebke&rft.aufirst=Andr%C3%A9&rft.au=Memeti%2C+Suejb&rft.au=Pllana%2C+Sabri&rft.au=Abraham%2C+Ajith&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Qin1-111"><span class="mw-cite-backlink"><b><a href="#cite_ref-Qin1_111-0">^</a></b></span> <span class="reference-text">Ting Qin, et al. "A learning algorithm of CMAC based on RLS." Neural Processing Letters 19.1 (2004): 49-61.</span>
</li>
<li id="cite_note-Qin2-112"><span class="mw-cite-backlink"><b><a href="#cite_ref-Qin2_112-0">^</a></b></span> <span class="reference-text">Ting Qin, et al. "<a rel="nofollow" class="external text" href="http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf">Continuous CMAC-QRLS and its systolic array</a> <a rel="nofollow" class="external text" href="https://web.archive.org/web/20181118122850/http://www-control.eng.cam.ac.uk/Homepage/papers/cued_control_997.pdf">Archived</a> 2018-11-18 at the <a href="/wiki/Wayback_Machine" title="Wayback Machine">Wayback Machine</a>." Neural Processing Letters 22.1 (2005): 1-16.</span>
</li>
<li id="cite_note-113"><span class="mw-cite-backlink"><b><a href="#cite_ref-113">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFResearch2015" class="citation web cs1">Research, AI (23 October 2015). <a rel="nofollow" class="external text" href="http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/">"Deep Neural Networks for Acoustic Modeling in Speech Recognition"</a>. <i>airesearch.com</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160201033801/http://airesearch.com/ai-research-papers/deep-neural-networks-for-acoustic-modeling-in-speech-recognition/">Archived</a> from the original on 1 February 2016<span class="reference-accessdate">. Retrieved <span class="nowrap">23 October</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=airesearch.com&rft.atitle=Deep+Neural+Networks+for+Acoustic+Modeling+in+Speech+Recognition&rft.date=2015-10-23&rft.aulast=Research&rft.aufirst=AI&rft_id=http%3A%2F%2Fairesearch.com%2Fai-research-papers%2Fdeep-neural-networks-for-acoustic-modeling-in-speech-recognition%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-114"><span class="mw-cite-backlink"><b><a href="#cite_ref-114">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation news cs1"><a rel="nofollow" class="external text" href="https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475">"GPUs Continue to Dominate the AI Accelerator Market for Now"</a>. <i>InformationWeek</i>. December 2019. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200610094310/https://www.informationweek.com/big-data/ai-machine-learning/gpus-continue-to-dominate-the-ai-accelerator-market-for-now/a/d-id/1336475">Archived</a> from the original on 10 June 2020<span class="reference-accessdate">. Retrieved <span class="nowrap">11 June</span> 2020</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=InformationWeek&rft.atitle=GPUs+Continue+to+Dominate+the+AI+Accelerator+Market+for+Now&rft.date=2019-12&rft_id=https%3A%2F%2Fwww.informationweek.com%2Fbig-data%2Fai-machine-learning%2Fgpus-continue-to-dominate-the-ai-accelerator-market-for-now%2Fa%2Fd-id%2F1336475&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-115"><span class="mw-cite-backlink"><b><a href="#cite_ref-115">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRay2019" class="citation news cs1">Ray, Tiernan (2019). <a rel="nofollow" class="external text" href="https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/">"AI is changing the entire nature of computation"</a>. <i>ZDNet</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200525144635/https://www.zdnet.com/article/ai-is-changing-the-entire-nature-of-compute/">Archived</a> from the original on 25 May 2020<span class="reference-accessdate">. Retrieved <span class="nowrap">11 June</span> 2020</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=ZDNet&rft.atitle=AI+is+changing+the+entire+nature+of+computation&rft.date=2019&rft.aulast=Ray&rft.aufirst=Tiernan&rft_id=https%3A%2F%2Fwww.zdnet.com%2Farticle%2Fai-is-changing-the-entire-nature-of-compute%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-116"><span class="mw-cite-backlink"><b><a href="#cite_ref-116">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://openai.com/blog/ai-and-compute/">"AI and Compute"</a>. <i>OpenAI</i>. 16 May 2018. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200617200602/https://openai.com/blog/ai-and-compute/">Archived</a> from the original on 17 June 2020<span class="reference-accessdate">. Retrieved <span class="nowrap">11 June</span> 2020</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=OpenAI&rft.atitle=AI+and+Compute&rft.date=2018-05-16&rft_id=https%3A%2F%2Fopenai.com%2Fblog%2Fai-and-compute%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-117"><span class="mw-cite-backlink"><b><a href="#cite_ref-117">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://consumer.huawei.com/en/press/news/2017/ifa2017-kirin970/">"HUAWEI Reveals the Future of Mobile AI at IFA 2017 | HUAWEI Latest News | HUAWEI Global"</a>. <i>consumer.huawei.com</i>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=consumer.huawei.com&rft.atitle=HUAWEI+Reveals+the+Future+of+Mobile+AI+at+IFA+2017+%26%23124%3B+HUAWEI+Latest+News+%26%23124%3B+HUAWEI+Global&rft_id=https%3A%2F%2Fconsumer.huawei.com%2Fen%2Fpress%2Fnews%2F2017%2Fifa2017-kirin970%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-118"><span class="mw-cite-backlink"><b><a href="#cite_ref-118">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFPYoungCliffPatilNishantPattersonDavid2017" class="citation journal cs1">P, JouppiNorman; YoungCliff; PatilNishant; PattersonDavid; AgrawalGaurav; BajwaRaminder; BatesSarah; BhatiaSuresh; BodenNan; BorchersAl; BoyleRick (2017-06-24). <a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F3140659.3080246">"In-Datacenter Performance Analysis of a Tensor Processing Unit"</a>. <i>ACM SIGARCH Computer Architecture News</i>. <b>45</b> (2): 1–12. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F3140659.3080246">10.1145/3140659.3080246</a></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=ACM+SIGARCH+Computer+Architecture+News&rft.atitle=In-Datacenter+Performance+Analysis+of+a+Tensor+Processing+Unit&rft.volume=45&rft.issue=2&rft.pages=1-12&rft.date=2017-06-24&rft_id=info%3Adoi%2F10.1145%2F3140659.3080246&rft.aulast=P&rft.aufirst=JouppiNorman&rft.au=YoungCliff&rft.au=PatilNishant&rft.au=PattersonDavid&rft.au=AgrawalGaurav&rft.au=BajwaRaminder&rft.au=BatesSarah&rft.au=BhatiaSuresh&rft.au=BodenNan&rft.au=BorchersAl&rft.au=BoyleRick&rft_id=https%3A%2F%2Fdoi.org%2F10.1145%252F3140659.3080246&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-119"><span class="mw-cite-backlink"><b><a href="#cite_ref-119">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFWoodie2021" class="citation web cs1">Woodie, Alex (2021-11-01). <a rel="nofollow" class="external text" href="https://www.datanami.com/2021/11/01/cerebras-hits-the-accelerator-for-deep-learning-workloads/">"Cerebras Hits the Accelerator for Deep Learning Workloads"</a>. <i>Datanami</i><span class="reference-accessdate">. Retrieved <span class="nowrap">2022-08-03</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Datanami&rft.atitle=Cerebras+Hits+the+Accelerator+for+Deep+Learning+Workloads&rft.date=2021-11-01&rft.aulast=Woodie&rft.aufirst=Alex&rft_id=https%3A%2F%2Fwww.datanami.com%2F2021%2F11%2F01%2Fcerebras-hits-the-accelerator-for-deep-learning-workloads%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-120"><span class="mw-cite-backlink"><b><a href="#cite_ref-120">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://venturebeat.com/2021/04/20/cerebras-systems-launches-new-ai-supercomputing-processor-with-2-6-trillion-transistors/">"Cerebras launches new AI supercomputing processor with 2.6 trillion transistors"</a>. <i>VentureBeat</i>. 2021-04-20<span class="reference-accessdate">. Retrieved <span class="nowrap">2022-08-03</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=VentureBeat&rft.atitle=Cerebras+launches+new+AI+supercomputing+processor+with+2.6+trillion+transistors&rft.date=2021-04-20&rft_id=https%3A%2F%2Fventurebeat.com%2F2021%2F04%2F20%2Fcerebras-systems-launches-new-ai-supercomputing-processor-with-2-6-trillion-transistors%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-atomthin-121"><span class="mw-cite-backlink"><b><a href="#cite_ref-atomthin_121-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMaregaZhaoAvsarWang2020" class="citation journal cs1">Marega, Guilherme Migliato; Zhao, Yanfei; Avsar, Ahmet; Wang, Zhenyu; Tripati, Mukesh; Radenovic, Aleksandra; Kis, Anras (2020). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7116757">"Logic-in-memory based on an atomically thin semiconductor"</a>. <i>Nature</i>. <b>587</b> (2): 72–77. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2020Natur.587...72M">2020Natur.587...72M</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fs41586-020-2861-0">10.1038/s41586-020-2861-0</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7116757">7116757</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/33149289">33149289</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=Logic-in-memory+based+on+an+atomically+thin+semiconductor&rft.volume=587&rft.issue=2&rft.pages=72-77&rft.date=2020&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC7116757%23id-name%3DPMC&rft_id=info%3Apmid%2F33149289&rft_id=info%3Adoi%2F10.1038%2Fs41586-020-2861-0&rft_id=info%3Abibcode%2F2020Natur.587...72M&rft.aulast=Marega&rft.aufirst=Guilherme+Migliato&rft.au=Zhao%2C+Yanfei&rft.au=Avsar%2C+Ahmet&rft.au=Wang%2C+Zhenyu&rft.au=Tripati%2C+Mukesh&rft.au=Radenovic%2C+Aleksandra&rft.au=Kis%2C+Anras&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC7116757&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-photonic-122"><span class="mw-cite-backlink">^ <a href="#cite_ref-photonic_122-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-photonic_122-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-photonic_122-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFFeldmannYoungbloodKarpovGehring2021" class="citation journal cs1">Feldmann, J.; Youngblood, N.; Karpov, M.; et al. (2021). "Parallel convolutional processing using an integrated photonic tensor". <i>Nature</i>. <b>589</b> (2): 52–58. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/2002.00281">2002.00281</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fs41586-020-03070-1">10.1038/s41586-020-03070-1</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/33408373">33408373</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:211010976">211010976</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=Parallel+convolutional+processing+using+an+integrated+photonic+tensor&rft.volume=589&rft.issue=2&rft.pages=52-58&rft.date=2021&rft_id=info%3Aarxiv%2F2002.00281&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A211010976%23id-name%3DS2CID&rft_id=info%3Apmid%2F33408373&rft_id=info%3Adoi%2F10.1038%2Fs41586-020-03070-1&rft.aulast=Feldmann&rft.aufirst=J.&rft.au=Youngblood%2C+N.&rft.au=Karpov%2C+M.&rft.au=Gehring%2C+H.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-LDCTIMIT-123"><span class="mw-cite-backlink"><b><a href="#cite_ref-LDCTIMIT_123-0">^</a></b></span> <span class="reference-text"><i>TIMIT Acoustic-Phonetic Continuous Speech Corpus</i> Linguistic Data Consortium, Philadelphia.</span>
</li>
<li id="cite_note-124"><span class="mw-cite-backlink"><b><a href="#cite_ref-124">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRobinson1991" class="citation journal cs1 cs1-prop-long-vol"><a href="/wiki/Tony_Robinson_(speech_recognition)" title="Tony Robinson (speech recognition)">Robinson, Tony</a> (30 September 1991). "Several Improvements to a Recurrent Error Propagation Network Phone Recognition System". <i>Cambridge University Engineering Department Technical Report</i>. CUED/F-INFENG/TR82. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.13140%2FRG.2.2.15418.90567">10.13140/RG.2.2.15418.90567</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Cambridge+University+Engineering+Department+Technical+Report&rft.atitle=Several+Improvements+to+a+Recurrent+Error+Propagation+Network+Phone+Recognition+System&rft.volume=CUED%2FF-INFENG%2FTR82&rft.date=1991-09-30&rft_id=info%3Adoi%2F10.13140%2FRG.2.2.15418.90567&rft.aulast=Robinson&rft.aufirst=Tony&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-CNN-2014-125"><span class="mw-cite-backlink"><b><a href="#cite_ref-CNN-2014_125-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFAbdel-Hamid2014" class="citation journal cs1">Abdel-Hamid, O.; et al. (2014). <a rel="nofollow" class="external text" href="https://zenodo.org/record/891433">"Convolutional Neural Networks for Speech Recognition"</a>. <i>IEEE/ACM Transactions on Audio, Speech, and Language Processing</i>. <b>22</b> (10): 1533–1545. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Ftaslp.2014.2339736">10.1109/taslp.2014.2339736</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:206602362">206602362</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200922180719/https://zenodo.org/record/891433">Archived</a> from the original on 2020-09-22<span class="reference-accessdate">. Retrieved <span class="nowrap">2018-04-20</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE%2FACM+Transactions+on+Audio%2C+Speech%2C+and+Language+Processing&rft.atitle=Convolutional+Neural+Networks+for+Speech+Recognition&rft.volume=22&rft.issue=10&rft.pages=1533-1545&rft.date=2014&rft_id=info%3Adoi%2F10.1109%2Ftaslp.2014.2339736&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A206602362%23id-name%3DS2CID&rft.aulast=Abdel-Hamid&rft.aufirst=O.&rft_id=https%3A%2F%2Fzenodo.org%2Frecord%2F891433&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-EnsembleDL-126"><span class="mw-cite-backlink"><b><a href="#cite_ref-EnsembleDL_126-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDengPlatt2014" class="citation journal cs1">Deng, L.; Platt, J. (2014). "Ensemble Deep Learning for Speech Recognition". <i>Proc. Interspeech</i>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:15641618">15641618</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proc.+Interspeech&rft.atitle=Ensemble+Deep+Learning+for+Speech+Recognition&rft.date=2014&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A15641618%23id-name%3DS2CID&rft.aulast=Deng&rft.aufirst=L.&rft.au=Platt%2C+J.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-HCDMM-127"><span class="mw-cite-backlink"><b><a href="#cite_ref-HCDMM_127-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTóth2015" class="citation journal cs1">Tóth, Laszló (2015). <a rel="nofollow" class="external text" href="http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf">"Phone Recognition with Hierarchical Convolutional Deep Maxout Networks"</a> <span class="cs1-format">(PDF)</span>. <i>EURASIP Journal on Audio, Speech, and Music Processing</i>. <b>2015</b>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1186%2Fs13636-015-0068-3">10.1186/s13636-015-0068-3</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:217950236">217950236</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200924085514/http://publicatio.bibl.u-szeged.hu/5976/1/EURASIP2015.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2020-09-24<span class="reference-accessdate">. Retrieved <span class="nowrap">2019-04-01</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=EURASIP+Journal+on+Audio%2C+Speech%2C+and+Music+Processing&rft.atitle=Phone+Recognition+with+Hierarchical+Convolutional+Deep+Maxout+Networks&rft.volume=2015&rft.date=2015&rft_id=info%3Adoi%2F10.1186%2Fs13636-015-0068-3&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A217950236%23id-name%3DS2CID&rft.aulast=T%C3%B3th&rft.aufirst=Laszl%C3%B3&rft_id=http%3A%2F%2Fpublicatio.bibl.u-szeged.hu%2F5976%2F1%2FEURASIP2015.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-128"><span class="mw-cite-backlink"><b><a href="#cite_ref-128">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMcMillan2014" class="citation magazine cs1">McMillan, Robert (17 December 2014). <a rel="nofollow" class="external text" href="https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/">"How Skype Used AI to Build Its Amazing New Language Translator | WIRED"</a>. <i>Wired</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170608062106/https://www.wired.com/2014/12/skype-used-ai-build-amazing-new-language-translator/">Archived</a> from the original on 8 June 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Wired&rft.atitle=How+Skype+Used+AI+to+Build+Its+Amazing+New+Language+Translator+%7C+WIRED&rft.date=2014-12-17&rft.aulast=McMillan&rft.aufirst=Robert&rft_id=https%3A%2F%2Fwww.wired.com%2F2014%2F12%2Fskype-used-ai-build-amazing-new-language-translator%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Baidu-129"><span class="mw-cite-backlink"><b><a href="#cite_ref-Baidu_129-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHannunCaseCasperCatanzaro2014" class="citation arxiv cs1">Hannun, Awni; Case, Carl; Casper, Jared; Catanzaro, Bryan; Diamos, Greg; Elsen, Erich; Prenger, Ryan; Satheesh, Sanjeev; Sengupta, Shubho; Coates, Adam; Ng, Andrew Y (2014). "Deep Speech: Scaling up end-to-end speech recognition". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1412.5567">1412.5567</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CL">cs.CL</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Deep+Speech%3A+Scaling+up+end-to-end+speech+recognition&rft.date=2014&rft_id=info%3Aarxiv%2F1412.5567&rft.aulast=Hannun&rft.aufirst=Awni&rft.au=Case%2C+Carl&rft.au=Casper%2C+Jared&rft.au=Catanzaro%2C+Bryan&rft.au=Diamos%2C+Greg&rft.au=Elsen%2C+Erich&rft.au=Prenger%2C+Ryan&rft.au=Satheesh%2C+Sanjeev&rft.au=Sengupta%2C+Shubho&rft.au=Coates%2C+Adam&rft.au=Ng%2C+Andrew+Y&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-YANNMNIST-130"><span class="mw-cite-backlink"><b><a href="#cite_ref-YANNMNIST_130-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://yann.lecun.com/exdb/mnist/.">"MNIST handwritten digit database, Yann LeCun, Corinna Cortes and Chris Burges"</a>. <i>yann.lecun.com</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20140113175237/http://yann.lecun.com/exdb/mnist/">Archived</a> from the original on 2014-01-13<span class="reference-accessdate">. Retrieved <span class="nowrap">2014-01-28</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=yann.lecun.com&rft.atitle=MNIST+handwritten+digit+database%2C+Yann+LeCun%2C+Corinna+Cortes+and+Chris+Burges&rft_id=http%3A%2F%2Fyann.lecun.com%2Fexdb%2Fmnist%2F.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:7-131"><span class="mw-cite-backlink"><b><a href="#cite_ref-:7_131-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCireşanMeierMasciSchmidhuber2012" class="citation journal cs1">Cireşan, Dan; Meier, Ueli; Masci, Jonathan; Schmidhuber, Jürgen (August 2012). "Multi-column deep neural network for traffic sign classification". <i>Neural Networks</i>. Selected Papers from IJCNN 2011. <b>32</b>: 333–338. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.226.8219">10.1.1.226.8219</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.neunet.2012.02.023">10.1016/j.neunet.2012.02.023</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/22386783">22386783</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Networks&rft.atitle=Multi-column+deep+neural+network+for+traffic+sign+classification&rft.volume=32&rft.pages=333-338&rft.date=2012-08&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.226.8219%23id-name%3DCiteSeerX&rft_id=info%3Apmid%2F22386783&rft_id=info%3Adoi%2F10.1016%2Fj.neunet.2012.02.023&rft.aulast=Cire%C5%9Fan&rft.aufirst=Dan&rft.au=Meier%2C+Ueli&rft.au=Masci%2C+Jonathan&rft.au=Schmidhuber%2C+J%C3%BCrgen&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-surpass1-132"><span class="mw-cite-backlink"><b><a href="#cite_ref-surpass1_132-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFChaochao_LuXiaoou_Tang2014" class="citation arxiv cs1">Chaochao Lu; Xiaoou Tang (2014). "Surpassing Human Level Face Recognition". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1404.3840">1404.3840</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CV">cs.CV</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Surpassing+Human+Level+Face+Recognition&rft.date=2014&rft_id=info%3Aarxiv%2F1404.3840&rft.au=Chaochao+Lu&rft.au=Xiaoou+Tang&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-133"><span class="mw-cite-backlink"><b><a href="#cite_ref-133">^</a></b></span> <span class="reference-text"><a rel="nofollow" class="external text" href="http://www.technologyreview.com/news/533936/nvidia-demos-a-car-computer-trained-with-deep-learning/">Nvidia Demos a Car Computer Trained with "Deep Learning"</a> (6 January 2015), David Talbot, <i><a href="/wiki/MIT_Technology_Review" title="MIT Technology Review">MIT Technology Review</a></i></span>
</li>
<li id="cite_note-art1-134"><span class="mw-cite-backlink">^ <a href="#cite_ref-art1_134-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-art1_134-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-art1_134-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFG._W._SmithFrederic_Fol_Leymarie2017" class="citation journal cs1">G. W. Smith; Frederic Fol Leymarie (10 April 2017). <a rel="nofollow" class="external text" href="https://doi.org/10.3390%2Farts6020005">"The Machine as Artist: An Introduction"</a>. <i>Arts</i>. <b>6</b> (4): 5. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.3390%2Farts6020005">10.3390/arts6020005</a></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Arts&rft.atitle=The+Machine+as+Artist%3A+An+Introduction&rft.volume=6&rft.issue=4&rft.pages=5&rft.date=2017-04-10&rft_id=info%3Adoi%2F10.3390%2Farts6020005&rft.au=G.+W.+Smith&rft.au=Frederic+Fol+Leymarie&rft_id=https%3A%2F%2Fdoi.org%2F10.3390%252Farts6020005&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-art2-135"><span class="mw-cite-backlink">^ <a href="#cite_ref-art2_135-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-art2_135-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-art2_135-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBlaise_Agüera_y_Arcas2017" class="citation journal cs1">Blaise Agüera y Arcas (29 September 2017). <a rel="nofollow" class="external text" href="https://doi.org/10.3390%2Farts6040018">"Art in the Age of Machine Intelligence"</a>. <i>Arts</i>. <b>6</b> (4): 18. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.3390%2Farts6040018">10.3390/arts6040018</a></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Arts&rft.atitle=Art+in+the+Age+of+Machine+Intelligence&rft.volume=6&rft.issue=4&rft.pages=18&rft.date=2017-09-29&rft_id=info%3Adoi%2F10.3390%2Farts6040018&rft.au=Blaise+Ag%C3%BCera+y+Arcas&rft_id=https%3A%2F%2Fdoi.org%2F10.3390%252Farts6040018&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-GoldbergLevy2014-136"><span class="mw-cite-backlink"><b><a href="#cite_ref-GoldbergLevy2014_136-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGoldbergLevy2014" class="citation arxiv cs1">Goldberg, Yoav; Levy, Omar (2014). "word2vec Explained: Deriving Mikolov et al.'s Negative-Sampling Word-Embedding Method". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1402.3722">1402.3722</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CL">cs.CL</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=word2vec+Explained%3A+Deriving+Mikolov+et+al.%27s+Negative-Sampling+Word-Embedding+Method&rft.date=2014&rft_id=info%3Aarxiv%2F1402.3722&rft.aulast=Goldberg&rft.aufirst=Yoav&rft.au=Levy%2C+Omar&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-SocherManning2014-137"><span class="mw-cite-backlink">^ <a href="#cite_ref-SocherManning2014_137-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-SocherManning2014_137-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSocherManning" class="citation web cs1">Socher, Richard; Manning, Christopher. <a rel="nofollow" class="external text" href="http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf">"Deep Learning for NLP"</a> <span class="cs1-format">(PDF)</span>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20140706040227/http://nlp.stanford.edu/courses/NAACL2013/NAACL2013-Socher-Manning-DeepLearning.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 6 July 2014<span class="reference-accessdate">. Retrieved <span class="nowrap">26 October</span> 2014</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Deep+Learning+for+NLP&rft.aulast=Socher&rft.aufirst=Richard&rft.au=Manning%2C+Christopher&rft_id=http%3A%2F%2Fnlp.stanford.edu%2Fcourses%2FNAACL2013%2FNAACL2013-Socher-Manning-DeepLearning.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-138"><span class="mw-cite-backlink"><b><a href="#cite_ref-138">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSocherBauerManningNg2013" class="citation journal cs1">Socher, Richard; Bauer, John; Manning, Christopher; Ng, Andrew (2013). <a rel="nofollow" class="external text" href="http://aclweb.org/anthology/P/P13/P13-1045.pdf">"Parsing With Compositional Vector Grammars"</a> <span class="cs1-format">(PDF)</span>. <i>Proceedings of the ACL 2013 Conference</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20141127005912/http://www.aclweb.org/anthology/P/P13/P13-1045.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2014-11-27<span class="reference-accessdate">. Retrieved <span class="nowrap">2014-09-03</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+of+the+ACL+2013+Conference&rft.atitle=Parsing+With+Compositional+Vector+Grammars&rft.date=2013&rft.aulast=Socher&rft.aufirst=Richard&rft.au=Bauer%2C+John&rft.au=Manning%2C+Christopher&rft.au=Ng%2C+Andrew&rft_id=http%3A%2F%2Faclweb.org%2Fanthology%2FP%2FP13%2FP13-1045.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-139"><span class="mw-cite-backlink"><b><a href="#cite_ref-139">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSocher2013" class="citation journal cs1">Socher, Richard (2013). <a rel="nofollow" class="external text" href="http://nlp.stanford.edu/~socherr/EMNLP2013_RNTN.pdf">"Recursive Deep Models for Semantic Compositionality Over a Sentiment Treebank"</a> <span class="cs1-format">(PDF)</span>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20161228100300/http://nlp.stanford.edu/%7Esocherr/EMNLP2013_RNTN.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2016-12-28<span class="reference-accessdate">. Retrieved <span class="nowrap">2014-09-03</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.atitle=Recursive+Deep+Models+for+Semantic+Compositionality+Over+a+Sentiment+Treebank&rft.date=2013&rft.aulast=Socher&rft.aufirst=Richard&rft_id=http%3A%2F%2Fnlp.stanford.edu%2F~socherr%2FEMNLP2013_RNTN.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span> <span class="cs1-hidden-error citation-comment"><code class="cs1-code">{{<a href="/wiki/Template:Cite_journal" title="Template:Cite journal">cite journal</a>}}</code>: </span><span class="cs1-hidden-error citation-comment">Cite journal requires <code class="cs1-code">|journal=</code> (<a href="/wiki/Help:CS1_errors#missing_periodical" title="Help:CS1 errors">help</a>)</span></span>
</li>
<li id="cite_note-140"><span class="mw-cite-backlink"><b><a href="#cite_ref-140">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFShenHeGaoDeng2014" class="citation journal cs1">Shen, Yelong; He, Xiaodong; Gao, Jianfeng; Deng, Li; Mesnil, Gregoire (1 November 2014). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/">"A Latent Semantic Model with Convolutional-Pooling Structure for Information Retrieval"</a>. <i>Microsoft Research</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171027050418/https://www.microsoft.com/en-us/research/publication/a-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval/">Archived</a> from the original on 27 October 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Microsoft+Research&rft.atitle=A+Latent+Semantic+Model+with+Convolutional-Pooling+Structure+for+Information+Retrieval&rft.date=2014-11-01&rft.aulast=Shen&rft.aufirst=Yelong&rft.au=He%2C+Xiaodong&rft.au=Gao%2C+Jianfeng&rft.au=Deng%2C+Li&rft.au=Mesnil%2C+Gregoire&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Fa-latent-semantic-model-with-convolutional-pooling-structure-for-information-retrieval%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-141"><span class="mw-cite-backlink"><b><a href="#cite_ref-141">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFHuangHeGaoDeng2013" class="citation journal cs1">Huang, Po-Sen; He, Xiaodong; Gao, Jianfeng; Deng, Li; Acero, Alex; Heck, Larry (1 October 2013). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/">"Learning Deep Structured Semantic Models for Web Search using Clickthrough Data"</a>. <i>Microsoft Research</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171027050414/https://www.microsoft.com/en-us/research/publication/learning-deep-structured-semantic-models-for-web-search-using-clickthrough-data/">Archived</a> from the original on 27 October 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Microsoft+Research&rft.atitle=Learning+Deep+Structured+Semantic+Models+for+Web+Search+using+Clickthrough+Data&rft.date=2013-10-01&rft.aulast=Huang&rft.aufirst=Po-Sen&rft.au=He%2C+Xiaodong&rft.au=Gao%2C+Jianfeng&rft.au=Deng%2C+Li&rft.au=Acero%2C+Alex&rft.au=Heck%2C+Larry&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Flearning-deep-structured-semantic-models-for-web-search-using-clickthrough-data%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-IEEE-TASL2015-142"><span class="mw-cite-backlink"><b><a href="#cite_ref-IEEE-TASL2015_142-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMesnilDauphinYaoBengio2015" class="citation journal cs1">Mesnil, G.; Dauphin, Y.; Yao, K.; Bengio, Y.; Deng, L.; Hakkani-Tur, D.; He, X.; Heck, L.; Tur, G.; Yu, D.; Zweig, G. (2015). "Using recurrent neural networks for slot filling in spoken language understanding". <i>IEEE Transactions on Audio, Speech, and Language Processing</i>. <b>23</b> (3): 530–539. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Ftaslp.2014.2383614">10.1109/taslp.2014.2383614</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:1317136">1317136</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Audio%2C+Speech%2C+and+Language+Processing&rft.atitle=Using+recurrent+neural+networks+for+slot+filling+in+spoken+language+understanding&rft.volume=23&rft.issue=3&rft.pages=530-539&rft.date=2015&rft_id=info%3Adoi%2F10.1109%2Ftaslp.2014.2383614&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A1317136%23id-name%3DS2CID&rft.aulast=Mesnil&rft.aufirst=G.&rft.au=Dauphin%2C+Y.&rft.au=Yao%2C+K.&rft.au=Bengio%2C+Y.&rft.au=Deng%2C+L.&rft.au=Hakkani-Tur%2C+D.&rft.au=He%2C+X.&rft.au=Heck%2C+L.&rft.au=Tur%2C+G.&rft.au=Yu%2C+D.&rft.au=Zweig%2C+G.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-auto-143"><span class="mw-cite-backlink">^ <a href="#cite_ref-auto_143-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-auto_143-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGaoHeYihDeng2014" class="citation journal cs1">Gao, Jianfeng; He, Xiaodong; Yih, Scott Wen-tau; Deng, Li (1 June 2014). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/">"Learning Continuous Phrase Representations for Translation Modeling"</a>. <i>Microsoft Research</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171027050403/https://www.microsoft.com/en-us/research/publication/learning-continuous-phrase-representations-for-translation-modeling/">Archived</a> from the original on 27 October 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Microsoft+Research&rft.atitle=Learning+Continuous+Phrase+Representations+for+Translation+Modeling&rft.date=2014-06-01&rft.aulast=Gao&rft.aufirst=Jianfeng&rft.au=He%2C+Xiaodong&rft.au=Yih%2C+Scott+Wen-tau&rft.au=Deng%2C+Li&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Flearning-continuous-phrase-representations-for-translation-modeling%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-BROC2017-144"><span class="mw-cite-backlink"><b><a href="#cite_ref-BROC2017_144-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBrocardoTraoreWoungangObaidat2017" class="citation journal cs1">Brocardo, Marcelo Luiz; Traore, Issa; Woungang, Isaac; Obaidat, Mohammad S. (2017). "Authorship verification using deep belief network systems". <i>International Journal of Communication Systems</i>. <b>30</b> (12): e3259. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1002%2Fdac.3259">10.1002/dac.3259</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:40745740">40745740</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=International+Journal+of+Communication+Systems&rft.atitle=Authorship+verification+using+deep+belief+network+systems&rft.volume=30&rft.issue=12&rft.pages=e3259&rft.date=2017&rft_id=info%3Adoi%2F10.1002%2Fdac.3259&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A40745740%23id-name%3DS2CID&rft.aulast=Brocardo&rft.aufirst=Marcelo+Luiz&rft.au=Traore%2C+Issa&rft.au=Woungang%2C+Isaac&rft.au=Obaidat%2C+Mohammad+S.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-145"><span class="mw-cite-backlink"><b><a href="#cite_ref-145">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation news cs1"><a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/">"Deep Learning for Natural Language Processing: Theory and Practice (CIKM2014 Tutorial) - Microsoft Research"</a>. <i>Microsoft Research</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170313184253/https://www.microsoft.com/en-us/research/project/deep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial/">Archived</a> from the original on 13 March 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Microsoft+Research&rft.atitle=Deep+Learning+for+Natural+Language+Processing%3A+Theory+and+Practice+%28CIKM2014+Tutorial%29+-+Microsoft+Research&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fproject%2Fdeep-learning-for-natural-language-processing-theory-and-practice-cikm2014-tutorial%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-GT_Turovsky_2016-146"><span class="mw-cite-backlink"><b><a href="#cite_ref-GT_Turovsky_2016_146-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTurovsky2016" class="citation web cs1">Turovsky, Barak (15 November 2016). <a rel="nofollow" class="external text" href="https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/">"Found in translation: More accurate, fluent sentences in Google Translate"</a>. <i>The Keyword Google Blog</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170407071226/https://blog.google/products/translate/found-translation-more-accurate-fluent-sentences-google-translate/">Archived</a> from the original on 7 April 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">23 March</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=The+Keyword+Google+Blog&rft.atitle=Found+in+translation%3A+More+accurate%2C+fluent+sentences+in+Google+Translate&rft.date=2016-11-15&rft.aulast=Turovsky&rft.aufirst=Barak&rft_id=https%3A%2F%2Fblog.google%2Fproducts%2Ftranslate%2Ffound-translation-more-accurate-fluent-sentences-google-translate%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-googleblog_GNMT_2016-147"><span class="mw-cite-backlink">^ <a href="#cite_ref-googleblog_GNMT_2016_147-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-googleblog_GNMT_2016_147-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-googleblog_GNMT_2016_147-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-googleblog_GNMT_2016_147-3"><sup><i><b>d</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSchusterJohnsonThorat2016" class="citation web cs1">Schuster, Mike; Johnson, Melvin; Thorat, Nikhil (22 November 2016). <a rel="nofollow" class="external text" href="https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html">"Zero-Shot Translation with Google's Multilingual Neural Machine Translation System"</a>. <i>Google Research Blog</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170710183732/https://research.googleblog.com/2016/11/zero-shot-translation-with-googles.html">Archived</a> from the original on 10 July 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">23 March</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Google+Research+Blog&rft.atitle=Zero-Shot+Translation+with+Google%27s+Multilingual+Neural+Machine+Translation+System&rft.date=2016-11-22&rft.aulast=Schuster&rft.aufirst=Mike&rft.au=Johnson%2C+Melvin&rft.au=Thorat%2C+Nikhil&rft_id=https%3A%2F%2Fresearch.googleblog.com%2F2016%2F11%2Fzero-shot-translation-with-googles.html&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-GoogleTranslate-148"><span class="mw-cite-backlink"><b><a href="#cite_ref-GoogleTranslate_148-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFWuSchusterChenLe2016" class="citation arxiv cs1">Wu, Yonghui; Schuster, Mike; Chen, Zhifeng; Le, Quoc V; Norouzi, Mohammad; Macherey, Wolfgang; Krikun, Maxim; Cao, Yuan; Gao, Qin; Macherey, Klaus; Klingner, Jeff; Shah, Apurva; Johnson, Melvin; Liu, Xiaobing; Kaiser, Łukasz; Gouws, Stephan; Kato, Yoshikiyo; Kudo, Taku; Kazawa, Hideto; Stevens, Keith; Kurian, George; Patil, Nishant; Wang, Wei; Young, Cliff; Smith, Jason; Riesa, Jason; Rudnick, Alex; Vinyals, Oriol; Corrado, Greg; et al. (2016). "Google's Neural Machine Translation System: Bridging the Gap between Human and Machine Translation". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1609.08144">1609.08144</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CL">cs.CL</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Google%27s+Neural+Machine+Translation+System%3A+Bridging+the+Gap+between+Human+and+Machine+Translation&rft.date=2016&rft_id=info%3Aarxiv%2F1609.08144&rft.aulast=Wu&rft.aufirst=Yonghui&rft.au=Schuster%2C+Mike&rft.au=Chen%2C+Zhifeng&rft.au=Le%2C+Quoc+V&rft.au=Norouzi%2C+Mohammad&rft.au=Macherey%2C+Wolfgang&rft.au=Krikun%2C+Maxim&rft.au=Cao%2C+Yuan&rft.au=Gao%2C+Qin&rft.au=Macherey%2C+Klaus&rft.au=Klingner%2C+Jeff&rft.au=Shah%2C+Apurva&rft.au=Johnson%2C+Melvin&rft.au=Liu%2C+Xiaobing&rft.au=Kaiser%2C+%C5%81ukasz&rft.au=Gouws%2C+Stephan&rft.au=Kato%2C+Yoshikiyo&rft.au=Kudo%2C+Taku&rft.au=Kazawa%2C+Hideto&rft.au=Stevens%2C+Keith&rft.au=Kurian%2C+George&rft.au=Patil%2C+Nishant&rft.au=Wang%2C+Wei&rft.au=Young%2C+Cliff&rft.au=Smith%2C+Jason&rft.au=Riesa%2C+Jason&rft.au=Rudnick%2C+Alex&rft.au=Vinyals%2C+Oriol&rft.au=Corrado%2C+Greg&rft.au=Hughes%2C+Macduff&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-WiredGoogleTranslate-149"><span class="mw-cite-backlink"><b><a href="#cite_ref-WiredGoogleTranslate_149-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMetz2016" class="citation magazine cs1">Metz, Cade (27 September 2016). <a rel="nofollow" class="external text" href="https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/">"An Infusion of AI Makes Google Translate More Powerful Than Ever"</a>. <i><a href="/wiki/Wired_(magazine)" title="Wired (magazine)">Wired</a></i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20201108101324/https://www.wired.com/2016/09/google-claims-ai-breakthrough-machine-translation/">Archived</a> from the original on 8 November 2020<span class="reference-accessdate">. Retrieved <span class="nowrap">12 October</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Wired&rft.atitle=An+Infusion+of+AI+Makes+Google+Translate+More+Powerful+Than+Ever&rft.date=2016-09-27&rft.aulast=Metz&rft.aufirst=Cade&rft_id=https%3A%2F%2Fwww.wired.com%2F2016%2F09%2Fgoogle-claims-ai-breakthrough-machine-translation%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Biotet-150"><span class="mw-cite-backlink">^ <a href="#cite_ref-Biotet_150-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-Biotet_150-1"><sup><i><b>b</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBoitetBlanchonSeligmanBellynck2010" class="citation web cs1">Boitet, Christian; Blanchon, Hervé; Seligman, Mark; Bellynck, Valérie (2010). <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170329125916/http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf">"MT on and for the Web"</a> <span class="cs1-format">(PDF)</span>. Archived from <a rel="nofollow" class="external text" href="http://www-clips.imag.fr/geta/herve.blanchon/Pdfs/NLP-KE-10.pdf">the original</a> <span class="cs1-format">(PDF)</span> on 29 March 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">1 December</span> 2016</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=MT+on+and+for+the+Web&rft.date=2010&rft.aulast=Boitet&rft.aufirst=Christian&rft.au=Blanchon%2C+Herv%C3%A9&rft.au=Seligman%2C+Mark&rft.au=Bellynck%2C+Val%C3%A9rie&rft_id=http%3A%2F%2Fwww-clips.imag.fr%2Fgeta%2Fherve.blanchon%2FPdfs%2FNLP-KE-10.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-ARROWSMITH2013-151"><span class="mw-cite-backlink"><b><a href="#cite_ref-ARROWSMITH2013_151-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFArrowsmithMiller2013" class="citation journal cs1">Arrowsmith, J; Miller, P (2013). "Trial watch: Phase II and phase III attrition rates 2011-2012". <i>Nature Reviews Drug Discovery</i>. <b>12</b> (8): 569. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fnrd4090">10.1038/nrd4090</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/23903212">23903212</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:20246434">20246434</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature+Reviews+Drug+Discovery&rft.atitle=Trial+watch%3A+Phase+II+and+phase+III+attrition+rates+2011-2012&rft.volume=12&rft.issue=8&rft.pages=569&rft.date=2013&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A20246434%23id-name%3DS2CID&rft_id=info%3Apmid%2F23903212&rft_id=info%3Adoi%2F10.1038%2Fnrd4090&rft.aulast=Arrowsmith&rft.aufirst=J&rft.au=Miller%2C+P&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-VERBIEST2015-152"><span class="mw-cite-backlink"><b><a href="#cite_ref-VERBIEST2015_152-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFVerbistKlambauerVervoortTalloen2015" class="citation journal cs1">Verbist, B; Klambauer, G; Vervoort, L; Talloen, W; The Qstar, Consortium; Shkedy, Z; Thas, O; Bender, A; Göhlmann, H. W.; Hochreiter, S (2015). <a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.drudis.2014.12.014">"Using transcriptomics to guide lead optimization in drug discovery projects: Lessons learned from the QSTAR project"</a>. <i>Drug Discovery Today</i>. <b>20</b> (5): 505–513. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.drudis.2014.12.014">10.1016/j.drudis.2014.12.014</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/25582842">25582842</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Drug+Discovery+Today&rft.atitle=Using+transcriptomics+to+guide+lead+optimization+in+drug+discovery+projects%3A+Lessons+learned+from+the+QSTAR+project&rft.volume=20&rft.issue=5&rft.pages=505-513&rft.date=2015&rft_id=info%3Adoi%2F10.1016%2Fj.drudis.2014.12.014&rft_id=info%3Apmid%2F25582842&rft.aulast=Verbist&rft.aufirst=B&rft.au=Klambauer%2C+G&rft.au=Vervoort%2C+L&rft.au=Talloen%2C+W&rft.au=The+Qstar%2C+Consortium&rft.au=Shkedy%2C+Z&rft.au=Thas%2C+O&rft.au=Bender%2C+A&rft.au=G%C3%B6hlmann%2C+H.+W.&rft.au=Hochreiter%2C+S&rft_id=https%3A%2F%2Fdoi.org%2F10.1016%252Fj.drudis.2014.12.014&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-153"><span class="mw-cite-backlink"><b><a href="#cite_ref-153">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFWallachDzambaHeifets2015" class="citation arxiv cs1">Wallach, Izhar; Dzamba, Michael; Heifets, Abraham (9 October 2015). "AtomNet: A Deep Convolutional Neural Network for Bioactivity Prediction in Structure-based Drug Discovery". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1510.02855">1510.02855</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.LG">cs.LG</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=AtomNet%3A+A+Deep+Convolutional+Neural+Network+for+Bioactivity+Prediction+in+Structure-based+Drug+Discovery&rft.date=2015-10-09&rft_id=info%3Aarxiv%2F1510.02855&rft.aulast=Wallach&rft.aufirst=Izhar&rft.au=Dzamba%2C+Michael&rft.au=Heifets%2C+Abraham&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-154"><span class="mw-cite-backlink"><b><a href="#cite_ref-154">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation news cs1"><a rel="nofollow" class="external text" href="https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/">"Toronto startup has a faster way to discover effective medicines"</a>. <i>The Globe and Mail</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20151020040115/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/">Archived</a> from the original on 20 October 2015<span class="reference-accessdate">. Retrieved <span class="nowrap">9 November</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=The+Globe+and+Mail&rft.atitle=Toronto+startup+has+a+faster+way+to+discover+effective+medicines&rft_id=https%3A%2F%2Fwww.theglobeandmail.com%2Freport-on-business%2Fsmall-business%2Fstarting-out%2Ftoronto-startup-has-a-faster-way-to-discover-effective-medicines%2Farticle25660419%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-155"><span class="mw-cite-backlink"><b><a href="#cite_ref-155">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/">"Startup Harnesses Supercomputers to Seek Cures"</a>. <i>KQED Future of You</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20151224104721/http://ww2.kqed.org/futureofyou/2015/05/27/startup-harnesses-supercomputers-to-seek-cures/">Archived</a> from the original on 24 December 2015<span class="reference-accessdate">. Retrieved <span class="nowrap">9 November</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=KQED+Future+of+You&rft.atitle=Startup+Harnesses+Supercomputers+to+Seek+Cures&rft_id=http%3A%2F%2Fww2.kqed.org%2Ffutureofyou%2F2015%2F05%2F27%2Fstartup-harnesses-supercomputers-to-seek-cures%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-156"><span class="mw-cite-backlink"><b><a href="#cite_ref-156">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5D%20and%20multiple%20sclerosis%20%5B/">"Toronto startup has a faster way to discover effective medicines"</a>. <i><a href="/wiki/The_Globe_and_Mail" title="The Globe and Mail">The Globe and Mail</a></i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20151225162547/http://www.theglobeandmail.com/report-on-business/small-business/starting-out/toronto-startup-has-a-faster-way-to-discover-effective-medicines/article25660419/%5d%20and%20multiple%20sclerosis%20%5b/">Archived</a> from the original on 2015-12-25<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-08-26</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=The+Globe+and+Mail&rft.atitle=Toronto+startup+has+a+faster+way+to+discover+effective+medicines&rft_id=https%3A%2F%2Fwww.theglobeandmail.com%2Freport-on-business%2Fsmall-business%2Fstarting-out%2Ftoronto-startup-has-a-faster-way-to-discover-effective-medicines%2Farticle25660419%2F%255D%2520and%2520multiple%2520sclerosis%2520%255B%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-157"><span class="mw-cite-backlink"><b><a href="#cite_ref-157">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGilmerSchoenholzRileyVinyals2017" class="citation arxiv cs1">Gilmer, Justin; Schoenholz, Samuel S.; Riley, Patrick F.; Vinyals, Oriol; Dahl, George E. (2017-06-12). "Neural Message Passing for Quantum Chemistry". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1704.01212">1704.01212</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.LG">cs.LG</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Neural+Message+Passing+for+Quantum+Chemistry&rft.date=2017-06-12&rft_id=info%3Aarxiv%2F1704.01212&rft.aulast=Gilmer&rft.aufirst=Justin&rft.au=Schoenholz%2C+Samuel+S.&rft.au=Riley%2C+Patrick+F.&rft.au=Vinyals%2C+Oriol&rft.au=Dahl%2C+George+E.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-158"><span class="mw-cite-backlink"><b><a href="#cite_ref-158">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZhavoronkov2019" class="citation journal cs1">Zhavoronkov, Alex (2019). "Deep learning enables rapid identification of potent DDR1 kinase inhibitors". <i>Nature Biotechnology</i>. <b>37</b> (9): 1038–1040. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fs41587-019-0224-x">10.1038/s41587-019-0224-x</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/31477924">31477924</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:201716327">201716327</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature+Biotechnology&rft.atitle=Deep+learning+enables+rapid+identification+of+potent+DDR1+kinase+inhibitors&rft.volume=37&rft.issue=9&rft.pages=1038-1040&rft.date=2019&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A201716327%23id-name%3DS2CID&rft_id=info%3Apmid%2F31477924&rft_id=info%3Adoi%2F10.1038%2Fs41587-019-0224-x&rft.aulast=Zhavoronkov&rft.aufirst=Alex&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-159"><span class="mw-cite-backlink"><b><a href="#cite_ref-159">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGregory" class="citation magazine cs1">Gregory, Barber. <a rel="nofollow" class="external text" href="https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/">"A Molecule Designed By AI Exhibits 'Druglike' Qualities"</a>. <i>Wired</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20200430143244/https://www.wired.com/story/molecule-designed-ai-exhibits-druglike-qualities/">Archived</a> from the original on 2020-04-30<span class="reference-accessdate">. Retrieved <span class="nowrap">2019-09-05</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Wired&rft.atitle=A+Molecule+Designed+By+AI+Exhibits+%27Druglike%27+Qualities&rft.aulast=Gregory&rft.aufirst=Barber&rft_id=https%3A%2F%2Fwww.wired.com%2Fstory%2Fmolecule-designed-ai-exhibits-druglike-qualities%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-160"><span class="mw-cite-backlink"><b><a href="#cite_ref-160">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTkachenko2015" class="citation arxiv cs1">Tkachenko, Yegor (8 April 2015). "Autonomous CRM Control via CLV Approximation with Deep Reinforcement Learning in Discrete and Continuous Action Space". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1504.01840">1504.01840</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.LG">cs.LG</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Autonomous+CRM+Control+via+CLV+Approximation+with+Deep+Reinforcement+Learning+in+Discrete+and+Continuous+Action+Space&rft.date=2015-04-08&rft_id=info%3Aarxiv%2F1504.01840&rft.aulast=Tkachenko&rft.aufirst=Yegor&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-161"><span class="mw-cite-backlink"><b><a href="#cite_ref-161">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFvan_den_OordDielemanSchrauwen2013" class="citation book cs1">van den Oord, Aaron; Dieleman, Sander; Schrauwen, Benjamin (2013). Burges, C. J. C.; Bottou, L.; Welling, M.; Ghahramani, Z.; Weinberger, K. Q. (eds.). <a rel="nofollow" class="external text" href="http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf"><i>Advances in Neural Information Processing Systems 26</i></a> <span class="cs1-format">(PDF)</span>. Curran Associates, Inc. pp. 2643–2651. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20170516185259/http://papers.nips.cc/paper/5004-deep-content-based-music-recommendation.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2017-05-16<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-14</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Advances+in+Neural+Information+Processing+Systems+26&rft.pages=2643-2651&rft.pub=Curran+Associates%2C+Inc.&rft.date=2013&rft.aulast=van+den+Oord&rft.aufirst=Aaron&rft.au=Dieleman%2C+Sander&rft.au=Schrauwen%2C+Benjamin&rft_id=http%3A%2F%2Fpapers.nips.cc%2Fpaper%2F5004-deep-content-based-music-recommendation.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-162"><span class="mw-cite-backlink"><b><a href="#cite_ref-162">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFFengZhangRenShang2019" class="citation journal cs1">Feng, X.Y.; Zhang, H.; Ren, Y.J.; Shang, P.H.; Zhu, Y.; Liang, Y.C.; Guan, R.C.; Xu, D. (2019). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6555124">"The Deep Learning–Based Recommender System "Pubmender" for Choosing a Biomedical Publication Venue: Development and Validation Study"</a>. <i><a href="/wiki/Journal_of_Medical_Internet_Research" title="Journal of Medical Internet Research">Journal of Medical Internet Research</a></i>. <b>21</b> (5): e12957. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.2196%2F12957">10.2196/12957</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6555124">6555124</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/31127715">31127715</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Journal+of+Medical+Internet+Research&rft.atitle=The+Deep+Learning%E2%80%93Based+Recommender+System+%22Pubmender%22+for+Choosing+a+Biomedical+Publication+Venue%3A+Development+and+Validation+Study&rft.volume=21&rft.issue=5&rft.pages=e12957&rft.date=2019&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC6555124%23id-name%3DPMC&rft_id=info%3Apmid%2F31127715&rft_id=info%3Adoi%2F10.2196%2F12957&rft.aulast=Feng&rft.aufirst=X.Y.&rft.au=Zhang%2C+H.&rft.au=Ren%2C+Y.J.&rft.au=Shang%2C+P.H.&rft.au=Zhu%2C+Y.&rft.au=Liang%2C+Y.C.&rft.au=Guan%2C+R.C.&rft.au=Xu%2C+D.&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC6555124&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-163"><span class="mw-cite-backlink"><b><a href="#cite_ref-163">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFElkahkySongHe2015" class="citation journal cs1">Elkahky, Ali Mamdouh; Song, Yang; He, Xiaodong (1 May 2015). <a rel="nofollow" class="external text" href="https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/">"A Multi-View Deep Learning Approach for Cross Domain User Modeling in Recommendation Systems"</a>. <i>Microsoft Research</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180125134534/https://www.microsoft.com/en-us/research/publication/a-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems/">Archived</a> from the original on 25 January 2018<span class="reference-accessdate">. Retrieved <span class="nowrap">14 June</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Microsoft+Research&rft.atitle=A+Multi-View+Deep+Learning+Approach+for+Cross+Domain+User+Modeling+in+Recommendation+Systems&rft.date=2015-05-01&rft.aulast=Elkahky&rft.aufirst=Ali+Mamdouh&rft.au=Song%2C+Yang&rft.au=He%2C+Xiaodong&rft_id=https%3A%2F%2Fwww.microsoft.com%2Fen-us%2Fresearch%2Fpublication%2Fa-multi-view-deep-learning-approach-for-cross-domain-user-modeling-in-recommendation-systems%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-164"><span class="mw-cite-backlink"><b><a href="#cite_ref-164">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFChiccoSadowskiBaldi2014" class="citation book cs1">Chicco, Davide; Sadowski, Peter; Baldi, Pierre (1 January 2014). <a rel="nofollow" class="external text" href="http://dl.acm.org/citation.cfm?id=2649442"><i>Deep Autoencoder Neural Networks for Gene Ontology Annotation Predictions</i></a>. <i>Proceedings of the 5th ACM Conference on Bioinformatics, Computational Biology, and Health Informatics - BCB '14</i>. ACM. pp. 533–540. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1145%2F2649387.2649442">10.1145/2649387.2649442</a>. <a href="/wiki/Hdl_(identifier)" class="mw-redirect" title="Hdl (identifier)">hdl</a>:<a rel="nofollow" class="external text" href="https://hdl.handle.net/11311%2F964622">11311/964622</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/9781450328944" title="Special:BookSources/9781450328944"><bdi>9781450328944</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:207217210">207217210</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123140/https://dl.acm.org/doi/10.1145/2649387.2649442">Archived</a> from the original on 9 May 2021<span class="reference-accessdate">. Retrieved <span class="nowrap">23 November</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Deep+Autoencoder+Neural+Networks+for+Gene+Ontology+Annotation+Predictions&rft.pages=533-540&rft.pub=ACM&rft.date=2014-01-01&rft_id=info%3Ahdl%2F11311%2F964622&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A207217210%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1145%2F2649387.2649442&rft.isbn=9781450328944&rft.aulast=Chicco&rft.aufirst=Davide&rft.au=Sadowski%2C+Peter&rft.au=Baldi%2C+Pierre&rft_id=http%3A%2F%2Fdl.acm.org%2Fcitation.cfm%3Fid%3D2649442&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-165"><span class="mw-cite-backlink"><b><a href="#cite_ref-165">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSathyanarayana2016" class="citation journal cs1">Sathyanarayana, Aarti (1 January 2016). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5116102">"Sleep Quality Prediction From Wearable Data Using Deep Learning"</a>. <i>JMIR mHealth and uHealth</i>. <b>4</b> (4): e125. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.2196%2Fmhealth.6562">10.2196/mhealth.6562</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5116102">5116102</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/27815231">27815231</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:3821594">3821594</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=JMIR+mHealth+and+uHealth&rft.atitle=Sleep+Quality+Prediction+From+Wearable+Data+Using+Deep+Learning&rft.volume=4&rft.issue=4&rft.pages=e125&rft.date=2016-01-01&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5116102%23id-name%3DPMC&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A3821594%23id-name%3DS2CID&rft_id=info%3Apmid%2F27815231&rft_id=info%3Adoi%2F10.2196%2Fmhealth.6562&rft.aulast=Sathyanarayana&rft.aufirst=Aarti&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5116102&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-166"><span class="mw-cite-backlink"><b><a href="#cite_ref-166">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFChoiSchuetzStewartSun2016" class="citation journal cs1">Choi, Edward; Schuetz, Andy; Stewart, Walter F.; Sun, Jimeng (13 August 2016). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5391725">"Using recurrent neural network models for early detection of heart failure onset"</a>. <i>Journal of the American Medical Informatics Association</i>. <b>24</b> (2): 361–370. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1093%2Fjamia%2Focw112">10.1093/jamia/ocw112</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1067-5027">1067-5027</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5391725">5391725</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/27521897">27521897</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Journal+of+the+American+Medical+Informatics+Association&rft.atitle=Using+recurrent+neural+network+models+for+early+detection+of+heart+failure+onset&rft.volume=24&rft.issue=2&rft.pages=361-370&rft.date=2016-08-13&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5391725%23id-name%3DPMC&rft.issn=1067-5027&rft_id=info%3Apmid%2F27521897&rft_id=info%3Adoi%2F10.1093%2Fjamia%2Focw112&rft.aulast=Choi&rft.aufirst=Edward&rft.au=Schuetz%2C+Andy&rft.au=Stewart%2C+Walter+F.&rft.au=Sun%2C+Jimeng&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5391725&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-167"><span class="mw-cite-backlink"><b><a href="#cite_ref-167">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFLitjensKooiBejnordiSetio2017" class="citation journal cs1">Litjens, Geert; Kooi, Thijs; Bejnordi, Babak Ehteshami; Setio, Arnaud Arindra Adiyoso; Ciompi, Francesco; Ghafoorian, Mohsen; van der Laak, Jeroen A.W.M.; van Ginneken, Bram; Sánchez, Clara I. (December 2017). "A survey on deep learning in medical image analysis". <i>Medical Image Analysis</i>. <b>42</b>: 60–88. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1702.05747">1702.05747</a></span>. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2017arXiv170205747L">2017arXiv170205747L</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.media.2017.07.005">10.1016/j.media.2017.07.005</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/28778026">28778026</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:2088679">2088679</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Medical+Image+Analysis&rft.atitle=A+survey+on+deep+learning+in+medical+image+analysis&rft.volume=42&rft.pages=60-88&rft.date=2017-12&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A2088679%23id-name%3DS2CID&rft_id=info%3Abibcode%2F2017arXiv170205747L&rft_id=info%3Aarxiv%2F1702.05747&rft_id=info%3Apmid%2F28778026&rft_id=info%3Adoi%2F10.1016%2Fj.media.2017.07.005&rft.aulast=Litjens&rft.aufirst=Geert&rft.au=Kooi%2C+Thijs&rft.au=Bejnordi%2C+Babak+Ehteshami&rft.au=Setio%2C+Arnaud+Arindra+Adiyoso&rft.au=Ciompi%2C+Francesco&rft.au=Ghafoorian%2C+Mohsen&rft.au=van+der+Laak%2C+Jeroen+A.W.M.&rft.au=van+Ginneken%2C+Bram&rft.au=S%C3%A1nchez%2C+Clara+I.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-168"><span class="mw-cite-backlink"><b><a href="#cite_ref-168">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFForslidWieslanderBengtssonWahlby2017" class="citation book cs1">Forslid, Gustav; Wieslander, Hakan; Bengtsson, Ewert; Wahlby, Carolina; Hirsch, Jan-Michael; Stark, Christina Runow; Sadanandan, Sajith Kecheril (2017). <a rel="nofollow" class="external text" href="http://urn.kb.se/resolve?urn=urn:nbn:se:uu:diva-326160">"Deep Convolutional Neural Networks for Detecting Cellular Changes Due to Malignancy"</a>. <i>2017 IEEE International Conference on Computer Vision Workshops (ICCVW)</i>. pp. 82–89. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FICCVW.2017.18">10.1109/ICCVW.2017.18</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/9781538610343" title="Special:BookSources/9781538610343"><bdi>9781538610343</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:4728736">4728736</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20210509123157/https://d1bxh8uas1mnw7.cloudfront.net/assets/embed.js">Archived</a> from the original on 2021-05-09<span class="reference-accessdate">. Retrieved <span class="nowrap">2019-11-12</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Deep+Convolutional+Neural+Networks+for+Detecting+Cellular+Changes+Due+to+Malignancy&rft.btitle=2017+IEEE+International+Conference+on+Computer+Vision+Workshops+%28ICCVW%29&rft.pages=82-89&rft.date=2017&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A4728736%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2FICCVW.2017.18&rft.isbn=9781538610343&rft.aulast=Forslid&rft.aufirst=Gustav&rft.au=Wieslander%2C+Hakan&rft.au=Bengtsson%2C+Ewert&rft.au=Wahlby%2C+Carolina&rft.au=Hirsch%2C+Jan-Michael&rft.au=Stark%2C+Christina+Runow&rft.au=Sadanandan%2C+Sajith+Kecheril&rft_id=http%3A%2F%2Furn.kb.se%2Fresolve%3Furn%3Durn%3Anbn%3Ase%3Auu%3Adiva-326160&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-169"><span class="mw-cite-backlink"><b><a href="#cite_ref-169">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDongZhouWangPeng2020" class="citation journal cs1">Dong, Xin; Zhou, Yizhao; Wang, Lantian; Peng, Jingfeng; Lou, Yanbo; Fan, Yiqun (2020). <a rel="nofollow" class="external text" href="https://ieeexplore.ieee.org/document/9130662">"Liver Cancer Detection Using Hybridized Fully Convolutional Neural Network Based on Deep Learning Framework"</a>. <i>IEEE Access</i>. <b>8</b>: 129889–129898. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FACCESS.2020.3006362">10.1109/ACCESS.2020.3006362</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/2169-3536">2169-3536</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:220733699">220733699</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Access&rft.atitle=Liver+Cancer+Detection+Using+Hybridized+Fully+Convolutional+Neural+Network+Based+on+Deep+Learning+Framework&rft.volume=8&rft.pages=129889-129898&rft.date=2020&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A220733699%23id-name%3DS2CID&rft.issn=2169-3536&rft_id=info%3Adoi%2F10.1109%2FACCESS.2020.3006362&rft.aulast=Dong&rft.aufirst=Xin&rft.au=Zhou%2C+Yizhao&rft.au=Wang%2C+Lantian&rft.au=Peng%2C+Jingfeng&rft.au=Lou%2C+Yanbo&rft.au=Fan%2C+Yiqun&rft_id=https%3A%2F%2Fieeexplore.ieee.org%2Fdocument%2F9130662&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-170"><span class="mw-cite-backlink"><b><a href="#cite_ref-170">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFLyakhovLyakhovaNagornov2022" class="citation journal cs1">Lyakhov, Pavel Alekseevich; Lyakhova, Ulyana Alekseevna; Nagornov, Nikolay Nikolaevich (2022-04-03). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8997449">"System for the Recognizing of Pigmented Skin Lesions with Fusion and Analysis of Heterogeneous Data Based on a Multimodal Neural Network"</a>. <i>Cancers</i>. <b>14</b> (7): 1819. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.3390%2Fcancers14071819">10.3390/cancers14071819</a></span>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/2072-6694">2072-6694</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC8997449">8997449</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/35406591">35406591</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Cancers&rft.atitle=System+for+the+Recognizing+of+Pigmented+Skin+Lesions+with+Fusion+and+Analysis+of+Heterogeneous+Data+Based+on+a+Multimodal+Neural+Network&rft.volume=14&rft.issue=7&rft.pages=1819&rft.date=2022-04-03&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC8997449%23id-name%3DPMC&rft.issn=2072-6694&rft_id=info%3Apmid%2F35406591&rft_id=info%3Adoi%2F10.3390%2Fcancers14071819&rft.aulast=Lyakhov&rft.aufirst=Pavel+Alekseevich&rft.au=Lyakhova%2C+Ulyana+Alekseevna&rft.au=Nagornov%2C+Nikolay+Nikolaevich&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC8997449&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-171"><span class="mw-cite-backlink"><b><a href="#cite_ref-171">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFDeMaityGoelShitole2017" class="citation book cs1">De, Shaunak; Maity, Abhishek; Goel, Vritti; Shitole, Sanjay; Bhattacharya, Avik (2017). "Predicting the popularity of instagram posts for a lifestyle magazine using deep learning". <i>2017 2nd International Conference on Communication Systems, Computing and IT Applications (CSCITA)</i>. pp. 174–177. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FCSCITA.2017.8066548">10.1109/CSCITA.2017.8066548</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-5090-4381-1" title="Special:BookSources/978-1-5090-4381-1"><bdi>978-1-5090-4381-1</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:35350962">35350962</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=bookitem&rft.atitle=Predicting+the+popularity+of+instagram+posts+for+a+lifestyle+magazine+using+deep+learning&rft.btitle=2017+2nd+International+Conference+on+Communication+Systems%2C+Computing+and+IT+Applications+%28CSCITA%29&rft.pages=174-177&rft.date=2017&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A35350962%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2FCSCITA.2017.8066548&rft.isbn=978-1-5090-4381-1&rft.aulast=De&rft.aufirst=Shaunak&rft.au=Maity%2C+Abhishek&rft.au=Goel%2C+Vritti&rft.au=Shitole%2C+Sanjay&rft.au=Bhattacharya%2C+Avik&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-172"><span class="mw-cite-backlink"><b><a href="#cite_ref-172">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/">"Colorizing and Restoring Old Images with Deep Learning"</a>. <i>FloydHub Blog</i>. 13 November 2018. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20191011162814/https://blog.floydhub.com/colorizing-and-restoring-old-images-with-deep-learning/">Archived</a> from the original on 11 October 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">11 October</span> 2019</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=FloydHub+Blog&rft.atitle=Colorizing+and+Restoring+Old+Images+with+Deep+Learning&rft.date=2018-11-13&rft_id=https%3A%2F%2Fblog.floydhub.com%2Fcolorizing-and-restoring-old-images-with-deep-learning%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-173"><span class="mw-cite-backlink"><b><a href="#cite_ref-173">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSchmidtRoth" class="citation conference cs1">Schmidt, Uwe; Roth, Stefan. <a rel="nofollow" class="external text" href="http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf"><i>Shrinkage Fields for Effective Image Restoration</i></a> <span class="cs1-format">(PDF)</span>. Computer Vision and Pattern Recognition (CVPR), 2014 IEEE Conference on. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180102013217/http://research.uweschmidt.org/pubs/cvpr14schmidt.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2018-01-02<span class="reference-accessdate">. Retrieved <span class="nowrap">2018-01-01</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=conference&rft.btitle=Shrinkage+Fields+for+Effective+Image+Restoration&rft.aulast=Schmidt&rft.aufirst=Uwe&rft.au=Roth%2C+Stefan&rft_id=http%3A%2F%2Fresearch.uweschmidt.org%2Fpubs%2Fcvpr14schmidt.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-174"><span class="mw-cite-backlink"><b><a href="#cite_ref-174">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFKleanthousChatzis2020" class="citation journal cs1">Kleanthous, Christos; Chatzis, Sotirios (2020). "Gated Mixture Variational Autoencoders for Value Added Tax audit case selection". <i>Knowledge-Based Systems</i>. <b>188</b>: 105048. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.knosys.2019.105048">10.1016/j.knosys.2019.105048</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:204092079">204092079</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Knowledge-Based+Systems&rft.atitle=Gated+Mixture+Variational+Autoencoders+for+Value+Added+Tax+audit+case+selection&rft.volume=188&rft.pages=105048&rft.date=2020&rft_id=info%3Adoi%2F10.1016%2Fj.knosys.2019.105048&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A204092079%23id-name%3DS2CID&rft.aulast=Kleanthous&rft.aufirst=Christos&rft.au=Chatzis%2C+Sotirios&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-175"><span class="mw-cite-backlink"><b><a href="#cite_ref-175">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCzech2018" class="citation journal cs1">Czech, Tomasz (28 June 2018). <a rel="nofollow" class="external text" href="https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/">"Deep learning: the next frontier for money laundering detection"</a>. <i>Global Banking and Finance Review</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20181116082711/https://www.globalbankingandfinance.com/deep-learning-the-next-frontier-for-money-laundering-detection/">Archived</a> from the original on 2018-11-16<span class="reference-accessdate">. Retrieved <span class="nowrap">2018-07-15</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Global+Banking+and+Finance+Review&rft.atitle=Deep+learning%3A+the+next+frontier+for+money+laundering+detection&rft.date=2018-06-28&rft.aulast=Czech&rft.aufirst=Tomasz&rft_id=https%3A%2F%2Fwww.globalbankingandfinance.com%2Fdeep-learning-the-next-frontier-for-money-laundering-detection%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:12-176"><span class="mw-cite-backlink">^ <a href="#cite_ref-:12_176-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:12_176-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-:12_176-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php">"Army researchers develop new algorithms to train robots"</a>. <i>EurekAlert!</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180828035608/https://www.eurekalert.org/pub_releases/2018-02/uarl-ard020218.php">Archived</a> from the original on 28 August 2018<span class="reference-accessdate">. Retrieved <span class="nowrap">29 August</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=EurekAlert%21&rft.atitle=Army+researchers+develop+new+algorithms+to+train+robots&rft_id=https%3A%2F%2Fwww.eurekalert.org%2Fpub_releases%2F2018-02%2Fuarl-ard020218.php&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-177"><span class="mw-cite-backlink"><b><a href="#cite_ref-177">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRaissiPerdikarisKarniadakis2019" class="citation journal cs1">Raissi, M.; Perdikaris, P.; Karniadakis, G. E. (2019-02-01). <a rel="nofollow" class="external text" href="https://www.sciencedirect.com/science/article/pii/S0021999118307125">"Physics-informed neural networks: A deep learning framework for solving forward and inverse problems involving nonlinear partial differential equations"</a>. <i>Journal of Computational Physics</i>. <b>378</b>: 686–707. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2019JCoPh.378..686R">2019JCoPh.378..686R</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.jcp.2018.10.045">10.1016/j.jcp.2018.10.045</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0021-9991">0021-9991</a>. <a href="/wiki/OSTI_(identifier)" class="mw-redirect" title="OSTI (identifier)">OSTI</a> <a rel="nofollow" class="external text" href="https://www.osti.gov/biblio/1595805">1595805</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:57379996">57379996</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Journal+of+Computational+Physics&rft.atitle=Physics-informed+neural+networks%3A+A+deep+learning+framework+for+solving+forward+and+inverse+problems+involving+nonlinear+partial+differential+equations&rft.volume=378&rft.pages=686-707&rft.date=2019-02-01&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A57379996%23id-name%3DS2CID&rft_id=https%3A%2F%2Fwww.osti.gov%2Fbiblio%2F1595805%23id-name%3DOSTI&rft_id=info%3Abibcode%2F2019JCoPh.378..686R&rft.issn=0021-9991&rft_id=info%3Adoi%2F10.1016%2Fj.jcp.2018.10.045&rft.aulast=Raissi&rft.aufirst=M.&rft.au=Perdikaris%2C+P.&rft.au=Karniadakis%2C+G.+E.&rft_id=https%3A%2F%2Fwww.sciencedirect.com%2Fscience%2Farticle%2Fpii%2FS0021999118307125&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-178"><span class="mw-cite-backlink"><b><a href="#cite_ref-178">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMaoJagtapKarniadakis2020" class="citation journal cs1">Mao, Zhiping; Jagtap, Ameya D.; Karniadakis, George Em (2020-03-01). <a rel="nofollow" class="external text" href="https://www.sciencedirect.com/science/article/pii/S0045782519306814">"Physics-informed neural networks for high-speed flows"</a>. <i>Computer Methods in Applied Mechanics and Engineering</i>. <b>360</b>: 112789. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2020CMAME.360k2789M">2020CMAME.360k2789M</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.cma.2019.112789">10.1016/j.cma.2019.112789</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0045-7825">0045-7825</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:212755458">212755458</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Computer+Methods+in+Applied+Mechanics+and+Engineering&rft.atitle=Physics-informed+neural+networks+for+high-speed+flows&rft.volume=360&rft.pages=112789&rft.date=2020-03-01&rft_id=info%3Adoi%2F10.1016%2Fj.cma.2019.112789&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A212755458%23id-name%3DS2CID&rft.issn=0045-7825&rft_id=info%3Abibcode%2F2020CMAME.360k2789M&rft.aulast=Mao&rft.aufirst=Zhiping&rft.au=Jagtap%2C+Ameya+D.&rft.au=Karniadakis%2C+George+Em&rft_id=https%3A%2F%2Fwww.sciencedirect.com%2Fscience%2Farticle%2Fpii%2FS0045782519306814&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-179"><span class="mw-cite-backlink"><b><a href="#cite_ref-179">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFRaissiYazdaniKarniadakis2020" class="citation journal cs1">Raissi, Maziar; Yazdani, Alireza; Karniadakis, George Em (2020-02-28). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7219083">"Hidden fluid mechanics: Learning velocity and pressure fields from flow visualizations"</a>. <i>Science</i>. <b>367</b> (6481): 1026–1030. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2020Sci...367.1026R">2020Sci...367.1026R</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1126%2Fscience.aaw4741">10.1126/science.aaw4741</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC7219083">7219083</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/32001523">32001523</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Science&rft.atitle=Hidden+fluid+mechanics%3A+Learning+velocity+and+pressure+fields+from+flow+visualizations&rft.volume=367&rft.issue=6481&rft.pages=1026-1030&rft.date=2020-02-28&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC7219083%23id-name%3DPMC&rft_id=info%3Apmid%2F32001523&rft_id=info%3Adoi%2F10.1126%2Fscience.aaw4741&rft_id=info%3Abibcode%2F2020Sci...367.1026R&rft.aulast=Raissi&rft.aufirst=Maziar&rft.au=Yazdani%2C+Alireza&rft.au=Karniadakis%2C+George+Em&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC7219083&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-180"><span class="mw-cite-backlink"><b><a href="#cite_ref-180">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFOktemKarBezekKamalabadi2021" class="citation journal cs1">Oktem, Figen S.; Kar, Oğuzhan Fatih; Bezek, Can Deniz; Kamalabadi, Farzad (2021). <a rel="nofollow" class="external text" href="https://ieeexplore.ieee.org/document/9415140">"High-Resolution Multi-Spectral Imaging With Diffractive Lenses and Learned Reconstruction"</a>. <i>IEEE Transactions on Computational Imaging</i>. <b>7</b>: 489–504. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/2008.11625">2008.11625</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FTCI.2021.3075349">10.1109/TCI.2021.3075349</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/2333-9403">2333-9403</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:235340737">235340737</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Computational+Imaging&rft.atitle=High-Resolution+Multi-Spectral+Imaging+With+Diffractive+Lenses+and+Learned+Reconstruction&rft.volume=7&rft.pages=489-504&rft.date=2021&rft_id=info%3Aarxiv%2F2008.11625&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A235340737%23id-name%3DS2CID&rft.issn=2333-9403&rft_id=info%3Adoi%2F10.1109%2FTCI.2021.3075349&rft.aulast=Oktem&rft.aufirst=Figen+S.&rft.au=Kar%2C+O%C4%9Fuzhan+Fatih&rft.au=Bezek%2C+Can+Deniz&rft.au=Kamalabadi%2C+Farzad&rft_id=https%3A%2F%2Fieeexplore.ieee.org%2Fdocument%2F9415140&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-181"><span class="mw-cite-backlink"><b><a href="#cite_ref-181">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBernhardtVishnevskiyRauGoksel2020" class="citation journal cs1">Bernhardt, Melanie; Vishnevskiy, Valery; Rau, Richard; Goksel, Orcun (December 2020). <a rel="nofollow" class="external text" href="https://ieeexplore.ieee.org/abstract/document/9144249?casa_token=NLVQbGR8h-8AAAAA:rz3cvMmiplVSRCGTDouUZkIeribqQKaMtK9t0MxlwKBMRwaTJL1onKOwkEk0qW_fFJCfd2ejOg">"Training Variational Networks With Multidomain Simulations: Speed-of-Sound Image Reconstruction"</a>. <i>IEEE Transactions on Ultrasonics, Ferroelectrics, and Frequency Control</i>. <b>67</b> (12): 2584–2594. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/2006.14395">2006.14395</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2FTUFFC.2020.3010186">10.1109/TUFFC.2020.3010186</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1525-8955">1525-8955</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/32746211">32746211</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:220055785">220055785</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=IEEE+Transactions+on+Ultrasonics%2C+Ferroelectrics%2C+and+Frequency+Control&rft.atitle=Training+Variational+Networks+With+Multidomain+Simulations%3A+Speed-of-Sound+Image+Reconstruction&rft.volume=67&rft.issue=12&rft.pages=2584-2594&rft.date=2020-12&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A220055785%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2FTUFFC.2020.3010186&rft_id=info%3Aarxiv%2F2006.14395&rft.issn=1525-8955&rft_id=info%3Apmid%2F32746211&rft.aulast=Bernhardt&rft.aufirst=Melanie&rft.au=Vishnevskiy%2C+Valery&rft.au=Rau%2C+Richard&rft.au=Goksel%2C+Orcun&rft_id=https%3A%2F%2Fieeexplore.ieee.org%2Fabstract%2Fdocument%2F9144249%3Fcasa_token%3DNLVQbGR8h-8AAAAA%3Arz3cvMmiplVSRCGTDouUZkIeribqQKaMtK9t0MxlwKBMRwaTJL1onKOwkEk0qW_fFJCfd2ejOg&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-UTGOFF-182"><span class="mw-cite-backlink"><b><a href="#cite_ref-UTGOFF_182-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFUtgoffStracuzzi2002" class="citation journal cs1">Utgoff, P. E.; Stracuzzi, D. J. (2002). "Many-layered learning". <i>Neural Computation</i>. <b>14</b> (10): 2497–2529. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1162%2F08997660260293319">10.1162/08997660260293319</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/12396572">12396572</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:1119517">1119517</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Computation&rft.atitle=Many-layered+learning&rft.volume=14&rft.issue=10&rft.pages=2497-2529&rft.date=2002&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A1119517%23id-name%3DS2CID&rft_id=info%3Apmid%2F12396572&rft_id=info%3Adoi%2F10.1162%2F08997660260293319&rft.aulast=Utgoff&rft.aufirst=P.+E.&rft.au=Stracuzzi%2C+D.+J.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-ELMAN-183"><span class="mw-cite-backlink"><b><a href="#cite_ref-ELMAN_183-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFElman1998" class="citation book cs1">Elman, Jeffrey L. (1998). <a rel="nofollow" class="external text" href="https://books.google.com/books?id=vELaRu_MrwoC"><i>Rethinking Innateness: A Connectionist Perspective on Development</i></a>. MIT Press. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-262-55030-7" title="Special:BookSources/978-0-262-55030-7"><bdi>978-0-262-55030-7</bdi></a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Rethinking+Innateness%3A+A+Connectionist+Perspective+on+Development&rft.pub=MIT+Press&rft.date=1998&rft.isbn=978-0-262-55030-7&rft.aulast=Elman&rft.aufirst=Jeffrey+L.&rft_id=https%3A%2F%2Fbooks.google.com%2Fbooks%3Fid%3DvELaRu_MrwoC&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-SHRAGER-184"><span class="mw-cite-backlink"><b><a href="#cite_ref-SHRAGER_184-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFShragerJohnson1996" class="citation journal cs1">Shrager, J.; Johnson, MH (1996). "Dynamic plasticity influences the emergence of function in a simple cortical array". <i>Neural Networks</i>. <b>9</b> (7): 1119–1129. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2F0893-6080%2896%2900033-0">10.1016/0893-6080(96)00033-0</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/12662587">12662587</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Networks&rft.atitle=Dynamic+plasticity+influences+the+emergence+of+function+in+a+simple+cortical+array&rft.volume=9&rft.issue=7&rft.pages=1119-1129&rft.date=1996&rft_id=info%3Adoi%2F10.1016%2F0893-6080%2896%2900033-0&rft_id=info%3Apmid%2F12662587&rft.aulast=Shrager&rft.aufirst=J.&rft.au=Johnson%2C+MH&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-QUARTZ-185"><span class="mw-cite-backlink"><b><a href="#cite_ref-QUARTZ_185-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFQuartzSejnowski1997" class="citation journal cs1">Quartz, SR; Sejnowski, TJ (1997). "The neural basis of cognitive development: A constructivist manifesto". <i>Behavioral and Brain Sciences</i>. <b>20</b> (4): 537–556. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.41.7854">10.1.1.41.7854</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1017%2Fs0140525x97001581">10.1017/s0140525x97001581</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/10097006">10097006</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:5818342">5818342</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Behavioral+and+Brain+Sciences&rft.atitle=The+neural+basis+of+cognitive+development%3A+A+constructivist+manifesto&rft.volume=20&rft.issue=4&rft.pages=537-556&rft.date=1997&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.41.7854%23id-name%3DCiteSeerX&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A5818342%23id-name%3DS2CID&rft_id=info%3Apmid%2F10097006&rft_id=info%3Adoi%2F10.1017%2Fs0140525x97001581&rft.aulast=Quartz&rft.aufirst=SR&rft.au=Sejnowski%2C+TJ&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-BLAKESLEE-186"><span class="mw-cite-backlink"><b><a href="#cite_ref-BLAKESLEE_186-0">^</a></b></span> <span class="reference-text">S. Blakeslee., "In brain's early growth, timetable may be critical," <i>The New York Times, Science Section</i>, pp. B5–B6, 1995.</span>
</li>
<li id="cite_note-187"><span class="mw-cite-backlink"><b><a href="#cite_ref-187">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMazzoniAndersenJordan1991" class="citation journal cs1">Mazzoni, P.; Andersen, R. A.; Jordan, M. I. (15 May 1991). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC51674">"A more biologically plausible learning rule for neural networks"</a>. <i>Proceedings of the National Academy of Sciences</i>. <b>88</b> (10): 4433–4437. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/1991PNAS...88.4433M">1991PNAS...88.4433M</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.1073%2Fpnas.88.10.4433">10.1073/pnas.88.10.4433</a></span>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0027-8424">0027-8424</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC51674">51674</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/1903542">1903542</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Proceedings+of+the+National+Academy+of+Sciences&rft.atitle=A+more+biologically+plausible+learning+rule+for+neural+networks.&rft.volume=88&rft.issue=10&rft.pages=4433-4437&rft.date=1991-05-15&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC51674%23id-name%3DPMC&rft_id=info%3Abibcode%2F1991PNAS...88.4433M&rft_id=info%3Apmid%2F1903542&rft_id=info%3Adoi%2F10.1073%2Fpnas.88.10.4433&rft.issn=0027-8424&rft.aulast=Mazzoni&rft.aufirst=P.&rft.au=Andersen%2C+R.+A.&rft.au=Jordan%2C+M.+I.&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC51674&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-188"><span class="mw-cite-backlink"><b><a href="#cite_ref-188">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFO'Reilly1996" class="citation journal cs1">O'Reilly, Randall C. (1 July 1996). "Biologically Plausible Error-Driven Learning Using Local Activation Differences: The Generalized Recirculation Algorithm". <i>Neural Computation</i>. <b>8</b> (5): 895–938. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1162%2Fneco.1996.8.5.895">10.1162/neco.1996.8.5.895</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0899-7667">0899-7667</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:2376781">2376781</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neural+Computation&rft.atitle=Biologically+Plausible+Error-Driven+Learning+Using+Local+Activation+Differences%3A+The+Generalized+Recirculation+Algorithm&rft.volume=8&rft.issue=5&rft.pages=895-938&rft.date=1996-07-01&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A2376781%23id-name%3DS2CID&rft.issn=0899-7667&rft_id=info%3Adoi%2F10.1162%2Fneco.1996.8.5.895&rft.aulast=O%27Reilly&rft.aufirst=Randall+C.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-189"><span class="mw-cite-backlink"><b><a href="#cite_ref-189">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTestolinZorzi2016" class="citation journal cs1">Testolin, Alberto; Zorzi, Marco (2016). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4943066">"Probabilistic Models and Generative Neural Networks: Towards an Unified Framework for Modeling Normal and Impaired Neurocognitive Functions"</a>. <i>Frontiers in Computational Neuroscience</i>. <b>10</b>: 73. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://doi.org/10.3389%2Ffncom.2016.00073">10.3389/fncom.2016.00073</a></span>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1662-5188">1662-5188</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC4943066">4943066</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/27468262">27468262</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:9868901">9868901</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Frontiers+in+Computational+Neuroscience&rft.atitle=Probabilistic+Models+and+Generative+Neural+Networks%3A+Towards+an+Unified+Framework+for+Modeling+Normal+and+Impaired+Neurocognitive+Functions&rft.volume=10&rft.pages=73&rft.date=2016&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC4943066%23id-name%3DPMC&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A9868901%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.3389%2Ffncom.2016.00073&rft.issn=1662-5188&rft_id=info%3Apmid%2F27468262&rft.aulast=Testolin&rft.aufirst=Alberto&rft.au=Zorzi%2C+Marco&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC4943066&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-190"><span class="mw-cite-backlink"><b><a href="#cite_ref-190">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTestolinStoianovZorzi2017" class="citation journal cs1">Testolin, Alberto; Stoianov, Ivilin; Zorzi, Marco (September 2017). "Letter perception emerges from unsupervised deep learning and recycling of natural image features". <i>Nature Human Behaviour</i>. <b>1</b> (9): 657–664. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fs41562-017-0186-2">10.1038/s41562-017-0186-2</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/2397-3374">2397-3374</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/31024135">31024135</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:24504018">24504018</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature+Human+Behaviour&rft.atitle=Letter+perception+emerges+from+unsupervised+deep+learning+and+recycling+of+natural+image+features&rft.volume=1&rft.issue=9&rft.pages=657-664&rft.date=2017-09&rft.issn=2397-3374&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A24504018%23id-name%3DS2CID&rft_id=info%3Apmid%2F31024135&rft_id=info%3Adoi%2F10.1038%2Fs41562-017-0186-2&rft.aulast=Testolin&rft.aufirst=Alberto&rft.au=Stoianov%2C+Ivilin&rft.au=Zorzi%2C+Marco&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-191"><span class="mw-cite-backlink"><b><a href="#cite_ref-191">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBuesingBillNesslerMaass2011" class="citation journal cs1">Buesing, Lars; Bill, Johannes; Nessler, Bernhard; Maass, Wolfgang (3 November 2011). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3207943">"Neural Dynamics as Sampling: A Model for Stochastic Computation in Recurrent Networks of Spiking Neurons"</a>. <i>PLOS Computational Biology</i>. <b>7</b> (11): e1002211. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2011PLSCB...7E2211B">2011PLSCB...7E2211B</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1371%2Fjournal.pcbi.1002211">10.1371/journal.pcbi.1002211</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1553-7358">1553-7358</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC3207943">3207943</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/22096452">22096452</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:7504633">7504633</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=PLOS+Computational+Biology&rft.atitle=Neural+Dynamics+as+Sampling%3A+A+Model+for+Stochastic+Computation+in+Recurrent+Networks+of+Spiking+Neurons&rft.volume=7&rft.issue=11&rft.pages=e1002211&rft.date=2011-11-03&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC3207943%23id-name%3DPMC&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A7504633%23id-name%3DS2CID&rft_id=info%3Abibcode%2F2011PLSCB...7E2211B&rft.issn=1553-7358&rft_id=info%3Adoi%2F10.1371%2Fjournal.pcbi.1002211&rft_id=info%3Apmid%2F22096452&rft.aulast=Buesing&rft.aufirst=Lars&rft.au=Bill%2C+Johannes&rft.au=Nessler%2C+Bernhard&rft.au=Maass%2C+Wolfgang&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC3207943&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-192"><span class="mw-cite-backlink"><b><a href="#cite_ref-192">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFCashYuste1999" class="citation journal cs1">Cash, S.; Yuste, R. (February 1999). "Linear summation of excitatory inputs by CA1 pyramidal neurons". <i>Neuron</i>. <b>22</b> (2): 383–394. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fs0896-6273%2800%2981098-3">10.1016/s0896-6273(00)81098-3</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0896-6273">0896-6273</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/10069343">10069343</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:14663106">14663106</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Neuron&rft.atitle=Linear+summation+of+excitatory+inputs+by+CA1+pyramidal+neurons&rft.volume=22&rft.issue=2&rft.pages=383-394&rft.date=1999-02&rft.issn=0896-6273&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A14663106%23id-name%3DS2CID&rft_id=info%3Apmid%2F10069343&rft_id=info%3Adoi%2F10.1016%2Fs0896-6273%2800%2981098-3&rft.aulast=Cash&rft.aufirst=S.&rft.au=Yuste%2C+R.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-193"><span class="mw-cite-backlink"><b><a href="#cite_ref-193">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFOlshausenField2004" class="citation journal cs1">Olshausen, B; Field, D (1 August 2004). "Sparse coding of sensory inputs". <i>Current Opinion in Neurobiology</i>. <b>14</b> (4): 481–487. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1016%2Fj.conb.2004.07.007">10.1016/j.conb.2004.07.007</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0959-4388">0959-4388</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/15321069">15321069</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:16560320">16560320</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Current+Opinion+in+Neurobiology&rft.atitle=Sparse+coding+of+sensory+inputs&rft.volume=14&rft.issue=4&rft.pages=481-487&rft.date=2004-08-01&rft.issn=0959-4388&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A16560320%23id-name%3DS2CID&rft_id=info%3Apmid%2F15321069&rft_id=info%3Adoi%2F10.1016%2Fj.conb.2004.07.007&rft.aulast=Olshausen&rft.aufirst=B&rft.au=Field%2C+D&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-194"><span class="mw-cite-backlink"><b><a href="#cite_ref-194">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFYaminsDiCarlo2016" class="citation journal cs1">Yamins, Daniel L K; DiCarlo, James J (March 2016). "Using goal-driven deep learning models to understand sensory cortex". <i>Nature Neuroscience</i>. <b>19</b> (3): 356–365. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fnn.4244">10.1038/nn.4244</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1546-1726">1546-1726</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/26906502">26906502</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:16970545">16970545</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature+Neuroscience&rft.atitle=Using+goal-driven+deep+learning+models+to+understand+sensory+cortex&rft.volume=19&rft.issue=3&rft.pages=356-365&rft.date=2016-03&rft.issn=1546-1726&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A16970545%23id-name%3DS2CID&rft_id=info%3Apmid%2F26906502&rft_id=info%3Adoi%2F10.1038%2Fnn.4244&rft.aulast=Yamins&rft.aufirst=Daniel+L+K&rft.au=DiCarlo%2C+James+J&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-195"><span class="mw-cite-backlink"><b><a href="#cite_ref-195">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZorziTestolin2018" class="citation journal cs1">Zorzi, Marco; Testolin, Alberto (19 February 2018). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5784047">"An emergentist perspective on the origin of number sense"</a>. <i>Phil. Trans. R. Soc. B</i>. <b>373</b> (1740): 20170043. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1098%2Frstb.2017.0043">10.1098/rstb.2017.0043</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0962-8436">0962-8436</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC5784047">5784047</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/29292348">29292348</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:39281431">39281431</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Phil.+Trans.+R.+Soc.+B&rft.atitle=An+emergentist+perspective+on+the+origin+of+number+sense&rft.volume=373&rft.issue=1740&rft.pages=20170043&rft.date=2018-02-19&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5784047%23id-name%3DPMC&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A39281431%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1098%2Frstb.2017.0043&rft.issn=0962-8436&rft_id=info%3Apmid%2F29292348&rft.aulast=Zorzi&rft.aufirst=Marco&rft.au=Testolin%2C+Alberto&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC5784047&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-196"><span class="mw-cite-backlink"><b><a href="#cite_ref-196">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGüçlüvan_Gerven2015" class="citation journal cs1">Güçlü, Umut; van Gerven, Marcel A. J. (8 July 2015). <a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6605414">"Deep Neural Networks Reveal a Gradient in the Complexity of Neural Representations across the Ventral Stream"</a>. <i>Journal of Neuroscience</i>. <b>35</b> (27): 10005–10014. <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1411.6422">1411.6422</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1523%2Fjneurosci.5023-14.2015">10.1523/jneurosci.5023-14.2015</a>. <a href="/wiki/PMC_(identifier)" class="mw-redirect" title="PMC (identifier)">PMC</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://www.ncbi.nlm.nih.gov/pmc/articles/PMC6605414">6605414</a></span>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/26157000">26157000</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Journal+of+Neuroscience&rft.atitle=Deep+Neural+Networks+Reveal+a+Gradient+in+the+Complexity+of+Neural+Representations+across+the+Ventral+Stream&rft.volume=35&rft.issue=27&rft.pages=10005-10014&rft.date=2015-07-08&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC6605414%23id-name%3DPMC&rft_id=info%3Apmid%2F26157000&rft_id=info%3Aarxiv%2F1411.6422&rft_id=info%3Adoi%2F10.1523%2Fjneurosci.5023-14.2015&rft.aulast=G%C3%BC%C3%A7l%C3%BC&rft.aufirst=Umut&rft.au=van+Gerven%2C+Marcel+A.+J.&rft_id=https%3A%2F%2Fwww.ncbi.nlm.nih.gov%2Fpmc%2Farticles%2FPMC6605414&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-METZ2013-197"><span class="mw-cite-backlink"><b><a href="#cite_ref-METZ2013_197-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMetz2013" class="citation magazine cs1">Metz, C. (12 December 2013). <a rel="nofollow" class="external text" href="https://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/">"Facebook's 'Deep Learning' Guru Reveals the Future of AI"</a>. <i>Wired</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20140328071226/http://www.wired.com/wiredenterprise/2013/12/facebook-yann-lecun-qa/">Archived</a> from the original on 28 March 2014<span class="reference-accessdate">. Retrieved <span class="nowrap">26 August</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Wired&rft.atitle=Facebook%27s+%27Deep+Learning%27+Guru+Reveals+the+Future+of+AI&rft.date=2013-12-12&rft.aulast=Metz&rft.aufirst=C.&rft_id=https%3A%2F%2Fwww.wired.com%2Fwiredenterprise%2F2013%2F12%2Ffacebook-yann-lecun-qa%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-198"><span class="mw-cite-backlink"><b><a href="#cite_ref-198">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGibney2016" class="citation journal cs1">Gibney, Elizabeth (2016). <a rel="nofollow" class="external text" href="http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234">"Google AI algorithm masters ancient game of Go"</a>. <i>Nature</i>. <b>529</b> (7587): 445–446. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2016Natur.529..445G">2016Natur.529..445G</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2F529445a">10.1038/529445a</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/26819021">26819021</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:4460235">4460235</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20190502200837/http://www.nature.com/news/google-ai-algorithm-masters-ancient-game-of-go-1.19234">Archived</a> from the original on 2 May 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">30 January</span> 2016</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=Google+AI+algorithm+masters+ancient+game+of+Go&rft.volume=529&rft.issue=7587&rft.pages=445-446&rft.date=2016&rft_id=info%3Adoi%2F10.1038%2F529445a&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A4460235%23id-name%3DS2CID&rft_id=info%3Apmid%2F26819021&rft_id=info%3Abibcode%2F2016Natur.529..445G&rft.aulast=Gibney&rft.aufirst=Elizabeth&rft_id=http%3A%2F%2Fwww.nature.com%2Fnews%2Fgoogle-ai-algorithm-masters-ancient-game-of-go-1.19234&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-199"><span class="mw-cite-backlink"><b><a href="#cite_ref-199">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSilverHuangMaddisonGuez2016" class="citation journal cs1"><a href="/wiki/David_Silver_(programmer)" class="mw-redirect" title="David Silver (programmer)">Silver, David</a>; <a href="/wiki/Aja_Huang" title="Aja Huang">Huang, Aja</a>; Maddison, Chris J.; Guez, Arthur; Sifre, Laurent; Driessche, George van den; Schrittwieser, Julian; Antonoglou, Ioannis; Panneershelvam, Veda; Lanctot, Marc; Dieleman, Sander; Grewe, Dominik; Nham, John; Kalchbrenner, Nal; <a href="/wiki/Ilya_Sutskever" title="Ilya Sutskever">Sutskever, Ilya</a>; Lillicrap, Timothy; Leach, Madeleine; Kavukcuoglu, Koray; Graepel, Thore; <a href="/wiki/Demis_Hassabis" title="Demis Hassabis">Hassabis, Demis</a> (28 January 2016). "Mastering the game of Go with deep neural networks and tree search". <i><a href="/wiki/Nature_(journal)" title="Nature (journal)">Nature</a></i>. <b>529</b> (7587): 484–489. <a href="/wiki/Bibcode_(identifier)" class="mw-redirect" title="Bibcode (identifier)">Bibcode</a>:<a rel="nofollow" class="external text" href="https://ui.adsabs.harvard.edu/abs/2016Natur.529..484S">2016Natur.529..484S</a>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fnature16961">10.1038/nature16961</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/0028-0836">0028-0836</a>. <a href="/wiki/PMID_(identifier)" class="mw-redirect" title="PMID (identifier)">PMID</a> <a rel="nofollow" class="external text" href="https://pubmed.ncbi.nlm.nih.gov/26819042">26819042</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:515925">515925</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=Mastering+the+game+of+Go+with+deep+neural+networks+and+tree+search&rft.volume=529&rft.issue=7587&rft.pages=484-489&rft.date=2016-01-28&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A515925%23id-name%3DS2CID&rft_id=info%3Abibcode%2F2016Natur.529..484S&rft.issn=0028-0836&rft_id=info%3Adoi%2F10.1038%2Fnature16961&rft_id=info%3Apmid%2F26819042&rft.aulast=Silver&rft.aufirst=David&rft.au=Huang%2C+Aja&rft.au=Maddison%2C+Chris+J.&rft.au=Guez%2C+Arthur&rft.au=Sifre%2C+Laurent&rft.au=Driessche%2C+George+van+den&rft.au=Schrittwieser%2C+Julian&rft.au=Antonoglou%2C+Ioannis&rft.au=Panneershelvam%2C+Veda&rft.au=Lanctot%2C+Marc&rft.au=Dieleman%2C+Sander&rft.au=Grewe%2C+Dominik&rft.au=Nham%2C+John&rft.au=Kalchbrenner%2C+Nal&rft.au=Sutskever%2C+Ilya&rft.au=Lillicrap%2C+Timothy&rft.au=Leach%2C+Madeleine&rft.au=Kavukcuoglu%2C+Koray&rft.au=Graepel%2C+Thore&rft.au=Hassabis%2C+Demis&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span><span style="position:relative; top: -2px;"><a href="/wiki/Paywall" title="closed access publication – behind paywall"><img alt="closed access" src="//upload.wikimedia.org/wikipedia/commons/thumb/0/0e/Closed_Access_logo_transparent.svg/9px-Closed_Access_logo_transparent.svg.png" decoding="async" width="9" height="14" srcset="//upload.wikimedia.org/wikipedia/commons/thumb/0/0e/Closed_Access_logo_transparent.svg/14px-Closed_Access_logo_transparent.svg.png 1.5x, //upload.wikimedia.org/wikipedia/commons/thumb/0/0e/Closed_Access_logo_transparent.svg/18px-Closed_Access_logo_transparent.svg.png 2x" data-file-width="640" data-file-height="1000" /></a></span></span>
</li>
<li id="cite_note-200"><span class="mw-cite-backlink"><b><a href="#cite_ref-200">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://web.archive.org/web/20160201140636/http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/">"A Google DeepMind Algorithm Uses Deep Learning and More to Master the Game of Go | MIT Technology Review"</a>. <i>MIT Technology Review</i>. Archived from <a rel="nofollow" class="external text" href="http://www.technologyreview.com/news/546066/googles-ai-masters-the-game-of-go-a-decade-earlier-than-expected/">the original</a> on 1 February 2016<span class="reference-accessdate">. Retrieved <span class="nowrap">30 January</span> 2016</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=MIT+Technology+Review&rft.atitle=A+Google+DeepMind+Algorithm+Uses+Deep+Learning+and+More+to+Master+the+Game+of+Go+%7C+MIT+Technology+Review&rft_id=http%3A%2F%2Fwww.technologyreview.com%2Fnews%2F546066%2Fgoogles-ai-masters-the-game-of-go-a-decade-earlier-than-expected%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-201"><span class="mw-cite-backlink"><b><a href="#cite_ref-201">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMetz2017" class="citation news cs1">Metz, Cade (6 November 2017). <a rel="nofollow" class="external text" href="https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html">"A.I. Researchers Leave Elon Musk Lab to Begin Robotics Start-Up"</a>. <i>The New York Times</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20190707161547/https://www.nytimes.com/2017/11/06/technology/artificial-intelligence-start-up.html">Archived</a> from the original on 7 July 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">5 July</span> 2019</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=The+New+York+Times&rft.atitle=A.I.+Researchers+Leave+Elon+Musk+Lab+to+Begin+Robotics+Start-Up&rft.date=2017-11-06&rft.aulast=Metz&rft.aufirst=Cade&rft_id=https%3A%2F%2Fwww.nytimes.com%2F2017%2F11%2F06%2Ftechnology%2Fartificial-intelligence-start-up.html&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-202"><span class="mw-cite-backlink"><b><a href="#cite_ref-202">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFBradley_Knox,_W.Stone,_Peter2008" class="citation journal cs1">Bradley Knox, W.; Stone, Peter (2008). "TAMER: Training an Agent Manually via Evaluative Reinforcement". <i>2008 7th IEEE International Conference on Development and Learning</i>: 292–297. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1109%2Fdevlrn.2008.4640845">10.1109/devlrn.2008.4640845</a>. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-1-4244-2661-4" title="Special:BookSources/978-1-4244-2661-4"><bdi>978-1-4244-2661-4</bdi></a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:5613334">5613334</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=2008+7th+IEEE+International+Conference+on+Development+and+Learning&rft.atitle=TAMER%3A+Training+an+Agent+Manually+via+Evaluative+Reinforcement&rft.pages=292-297&rft.date=2008&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A5613334%23id-name%3DS2CID&rft_id=info%3Adoi%2F10.1109%2Fdevlrn.2008.4640845&rft.isbn=978-1-4244-2661-4&rft.au=Bradley+Knox%2C+W.&rft.au=Stone%2C+Peter&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-203"><span class="mw-cite-backlink"><b><a href="#cite_ref-203">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner">"Talk to the Algorithms: AI Becomes a Faster Learner"</a>. <i>governmentciomedia.com</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20180828001727/https://governmentciomedia.com/talk-algorithms-ai-becomes-faster-learner">Archived</a> from the original on 28 August 2018<span class="reference-accessdate">. Retrieved <span class="nowrap">29 August</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=governmentciomedia.com&rft.atitle=Talk+to+the+Algorithms%3A+AI+Becomes+a+Faster+Learner&rft_id=https%3A%2F%2Fgovernmentciomedia.com%2Ftalk-algorithms-ai-becomes-faster-learner&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-204"><span class="mw-cite-backlink"><b><a href="#cite_ref-204">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMarcus2018" class="citation web cs1">Marcus, Gary (14 January 2018). <a rel="nofollow" class="external text" href="https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1">"In defense of skepticism about deep learning"</a>. <i>Gary Marcus</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20181012035405/https://medium.com/@GaryMarcus/in-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1">Archived</a> from the original on 12 October 2018<span class="reference-accessdate">. Retrieved <span class="nowrap">11 October</span> 2018</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Gary+Marcus&rft.atitle=In+defense+of+skepticism+about+deep+learning&rft.date=2018-01-14&rft.aulast=Marcus&rft.aufirst=Gary&rft_id=https%3A%2F%2Fmedium.com%2F%40GaryMarcus%2Fin-defense-of-skepticism-about-deep-learning-6e8bfd5ae0f1&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-Knight_2017-205"><span class="mw-cite-backlink"><b><a href="#cite_ref-Knight_2017_205-0">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFKnight2017" class="citation web cs1">Knight, Will (14 March 2017). <a rel="nofollow" class="external text" href="https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/">"DARPA is funding projects that will try to open up AI's black boxes"</a>. <i>MIT Technology Review</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20191104033107/https://www.technologyreview.com/s/603795/the-us-military-wants-its-autonomous-machines-to-explain-themselves/">Archived</a> from the original on 4 November 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">2 November</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=MIT+Technology+Review&rft.atitle=DARPA+is+funding+projects+that+will+try+to+open+up+AI%27s+black+boxes&rft.date=2017-03-14&rft.aulast=Knight&rft.aufirst=Will&rft_id=https%3A%2F%2Fwww.technologyreview.com%2Fs%2F603795%2Fthe-us-military-wants-its-autonomous-machines-to-explain-themselves%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-206"><span class="mw-cite-backlink"><b><a href="#cite_ref-206">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMarcus2012" class="citation magazine cs1">Marcus, Gary (November 25, 2012). <a rel="nofollow" class="external text" href="https://www.newyorker.com/">"Is "Deep Learning" a Revolution in Artificial Intelligence?"</a>. <i>The New Yorker</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20091127184826/http://www.newyorker.com/">Archived</a> from the original on 2009-11-27<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-06-14</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=The+New+Yorker&rft.atitle=Is+%22Deep+Learning%22+a+Revolution+in+Artificial+Intelligence%3F&rft.date=2012-11-25&rft.aulast=Marcus&rft.aufirst=Gary&rft_id=https%3A%2F%2Fwww.newyorker.com%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-207"><span class="mw-cite-backlink"><b><a href="#cite_ref-207">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFAlexander_MordvintsevChristopher_OlahMike_Tyka2015" class="citation web cs1">Alexander Mordvintsev; Christopher Olah; Mike Tyka (17 June 2015). <a rel="nofollow" class="external text" href="http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html">"Inceptionism: Going Deeper into Neural Networks"</a>. Google Research Blog. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150703064823/http://googleresearch.blogspot.co.uk/2015/06/inceptionism-going-deeper-into-neural.html">Archived</a> from the original on 3 July 2015<span class="reference-accessdate">. Retrieved <span class="nowrap">20 June</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Inceptionism%3A+Going+Deeper+into+Neural+Networks&rft.pub=Google+Research+Blog&rft.date=2015-06-17&rft.au=Alexander+Mordvintsev&rft.au=Christopher+Olah&rft.au=Mike+Tyka&rft_id=http%3A%2F%2Fgoogleresearch.blogspot.co.uk%2F2015%2F06%2Finceptionism-going-deeper-into-neural.html&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-208"><span class="mw-cite-backlink"><b><a href="#cite_ref-208">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFAlex_Hern2015" class="citation news cs1">Alex Hern (18 June 2015). <a rel="nofollow" class="external text" href="https://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep">"Yes, androids do dream of electric sheep"</a>. <i>The Guardian</i>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150619200845/http://www.theguardian.com/technology/2015/jun/18/google-image-recognition-neural-network-androids-dream-electric-sheep">Archived</a> from the original on 19 June 2015<span class="reference-accessdate">. Retrieved <span class="nowrap">20 June</span> 2015</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=The+Guardian&rft.atitle=Yes%2C+androids+do+dream+of+electric+sheep&rft.date=2015-06-18&rft.au=Alex+Hern&rft_id=https%3A%2F%2Fwww.theguardian.com%2Ftechnology%2F2015%2Fjun%2F18%2Fgoogle-image-recognition-neural-network-androids-dream-electric-sheep&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-goertzel-209"><span class="mw-cite-backlink">^ <a href="#cite_ref-goertzel_209-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-goertzel_209-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-goertzel_209-2"><sup><i><b>c</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGoertzel2015" class="citation web cs1">Goertzel, Ben (2015). <a rel="nofollow" class="external text" href="http://goertzel.org/DeepLearning_v1.pdf">"Are there Deep Reasons Underlying the Pathologies of Today's Deep Learning Algorithms?"</a> <span class="cs1-format">(PDF)</span>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20150513053107/http://goertzel.org/DeepLearning_v1.pdf">Archived</a> <span class="cs1-format">(PDF)</span> from the original on 2015-05-13<span class="reference-accessdate">. Retrieved <span class="nowrap">2015-05-10</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Are+there+Deep+Reasons+Underlying+the+Pathologies+of+Today%27s+Deep+Learning+Algorithms%3F&rft.date=2015&rft.aulast=Goertzel&rft.aufirst=Ben&rft_id=http%3A%2F%2Fgoertzel.org%2FDeepLearning_v1.pdf&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-210"><span class="mw-cite-backlink"><b><a href="#cite_ref-210">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFNguyenYosinskiClune2014" class="citation arxiv cs1">Nguyen, Anh; Yosinski, Jason; Clune, Jeff (2014). "Deep Neural Networks are Easily Fooled: High Confidence Predictions for Unrecognizable Images". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1412.1897">1412.1897</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CV">cs.CV</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Deep+Neural+Networks+are+Easily+Fooled%3A+High+Confidence+Predictions+for+Unrecognizable+Images&rft.date=2014&rft_id=info%3Aarxiv%2F1412.1897&rft.aulast=Nguyen&rft.aufirst=Anh&rft.au=Yosinski%2C+Jason&rft.au=Clune%2C+Jeff&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-211"><span class="mw-cite-backlink"><b><a href="#cite_ref-211">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFSzegedyZarembaSutskeverBruna2013" class="citation arxiv cs1">Szegedy, Christian; Zaremba, Wojciech; Sutskever, Ilya; Bruna, Joan; Erhan, Dumitru; Goodfellow, Ian; Fergus, Rob (2013). "Intriguing properties of neural networks". <a href="/wiki/ArXiv_(identifier)" class="mw-redirect" title="ArXiv (identifier)">arXiv</a>:<span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://arxiv.org/abs/1312.6199">1312.6199</a></span> [<a rel="nofollow" class="external text" href="//arxiv.org/archive/cs.CV">cs.CV</a>].</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=preprint&rft.jtitle=arXiv&rft.atitle=Intriguing+properties+of+neural+networks&rft.date=2013&rft_id=info%3Aarxiv%2F1312.6199&rft.aulast=Szegedy&rft.aufirst=Christian&rft.au=Zaremba%2C+Wojciech&rft.au=Sutskever%2C+Ilya&rft.au=Bruna%2C+Joan&rft.au=Erhan%2C+Dumitru&rft.au=Goodfellow%2C+Ian&rft.au=Fergus%2C+Rob&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-212"><span class="mw-cite-backlink"><b><a href="#cite_ref-212">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFZhuMumford2006" class="citation journal cs1">Zhu, S.C.; Mumford, D. (2006). "A stochastic grammar of images". <i>Found. Trends Comput. Graph. Vis</i>. <b>2</b> (4): 259–362. <a href="/wiki/CiteSeerX_(identifier)" class="mw-redirect" title="CiteSeerX (identifier)">CiteSeerX</a> <span class="cs1-lock-free" title="Freely accessible"><a rel="nofollow" class="external text" href="https://citeseerx.ist.psu.edu/viewdoc/summary?doi=10.1.1.681.2190">10.1.1.681.2190</a></span>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1561%2F0600000018">10.1561/0600000018</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Found.+Trends+Comput.+Graph.+Vis.&rft.atitle=A+stochastic+grammar+of+images&rft.volume=2&rft.issue=4&rft.pages=259-362&rft.date=2006&rft_id=https%3A%2F%2Fciteseerx.ist.psu.edu%2Fviewdoc%2Fsummary%3Fdoi%3D10.1.1.681.2190%23id-name%3DCiteSeerX&rft_id=info%3Adoi%2F10.1561%2F0600000018&rft.aulast=Zhu&rft.aufirst=S.C.&rft.au=Mumford%2C+D.&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-213"><span class="mw-cite-backlink"><b><a href="#cite_ref-213">^</a></b></span> <span class="reference-text">Miller, G. A., and N. Chomsky. "Pattern conception." Paper for Conference on pattern detection, University of Michigan. 1957.</span>
</li>
<li id="cite_note-214"><span class="mw-cite-backlink"><b><a href="#cite_ref-214">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFEisner" class="citation web cs1">Eisner, Jason. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171230010335/http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/">"Deep Learning of Recursive Structure: Grammar Induction"</a>. Archived from <a rel="nofollow" class="external text" href="http://techtalks.tv/talks/deep-learning-of-recursive-structure-grammar-induction/58089/">the original</a> on 2017-12-30<span class="reference-accessdate">. Retrieved <span class="nowrap">2015-05-10</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=unknown&rft.btitle=Deep+Learning+of+Recursive+Structure%3A+Grammar+Induction&rft.aulast=Eisner&rft.aufirst=Jason&rft_id=http%3A%2F%2Ftechtalks.tv%2Ftalks%2Fdeep-learning-of-recursive-structure-grammar-induction%2F58089%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-215"><span class="mw-cite-backlink"><b><a href="#cite_ref-215">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425">"Hackers Have Already Started to Weaponize Artificial Intelligence"</a>. <i>Gizmodo</i>. 11 September 2017. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20191011162231/https://gizmodo.com/hackers-have-already-started-to-weaponize-artificial-in-1797688425">Archived</a> from the original on 11 October 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">11 October</span> 2019</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=Gizmodo&rft.atitle=Hackers+Have+Already+Started+to+Weaponize+Artificial+Intelligence&rft.date=2017-09-11&rft_id=https%3A%2F%2Fgizmodo.com%2Fhackers-have-already-started-to-weaponize-artificial-in-1797688425&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-216"><span class="mw-cite-backlink"><b><a href="#cite_ref-216">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation web cs1"><a rel="nofollow" class="external text" href="https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/">"How hackers can force AI to make dumb mistakes"</a>. <i>The Daily Dot</i>. 18 June 2018. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20191011162230/https://www.dailydot.com/debug/adversarial-attacks-ai-mistakes/">Archived</a> from the original on 11 October 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">11 October</span> 2019</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=unknown&rft.jtitle=The+Daily+Dot&rft.atitle=How+hackers+can+force+AI+to+make+dumb+mistakes&rft.date=2018-06-18&rft_id=https%3A%2F%2Fwww.dailydot.com%2Fdebug%2Fadversarial-attacks-ai-mistakes%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:4-217"><span class="mw-cite-backlink">^ <a href="#cite_ref-:4_217-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:4_217-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-:4_217-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-:4_217-3"><sup><i><b>d</b></i></sup></a> <a href="#cite_ref-:4_217-4"><sup><i><b>e</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation news cs1"><a rel="nofollow" class="external text" href="https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change">"AI Is Easy to Fool—Why That Needs to Change"</a>. <i>Singularity Hub</i>. 10 October 2017. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171011233017/https://singularityhub.com/2017/10/10/ai-is-easy-to-fool-why-that-needs-to-change/">Archived</a> from the original on 11 October 2017<span class="reference-accessdate">. Retrieved <span class="nowrap">11 October</span> 2017</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Singularity+Hub&rft.atitle=AI+Is+Easy+to+Fool%E2%80%94Why+That+Needs+to+Change&rft.date=2017-10-10&rft_id=https%3A%2F%2Fsingularityhub.com%2F2017%2F10%2F10%2Fai-is-easy-to-fool-why-that-needs-to-change&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-218"><span class="mw-cite-backlink"><b><a href="#cite_ref-218">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGibney2017" class="citation journal cs1">Gibney, Elizabeth (2017). <a rel="nofollow" class="external text" href="https://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784">"The scientist who spots fake videos"</a>. <i>Nature</i>. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1038%2Fnature.2017.22784">10.1038/nature.2017.22784</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20171010011017/http://www.nature.com/news/the-scientist-who-spots-fake-videos-1.22784">Archived</a> from the original on 2017-10-10<span class="reference-accessdate">. Retrieved <span class="nowrap">2017-10-11</span></span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Nature&rft.atitle=The+scientist+who+spots+fake+videos&rft.date=2017&rft_id=info%3Adoi%2F10.1038%2Fnature.2017.22784&rft.aulast=Gibney&rft.aufirst=Elizabeth&rft_id=https%3A%2F%2Fwww.nature.com%2Fnews%2Fthe-scientist-who-spots-fake-videos-1.22784&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-219"><span class="mw-cite-backlink"><b><a href="#cite_ref-219">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFTubaro2020" class="citation journal cs1">Tubaro, Paola (2020). <a rel="nofollow" class="external text" href="https://hal.science/hal-03029735">"Whose intelligence is artificial intelligence?"</a>. <i>Global Dialogue</i>: 38.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Global+Dialogue&rft.atitle=Whose+intelligence+is+artificial+intelligence%3F&rft.pages=38&rft.date=2020&rft.aulast=Tubaro&rft.aufirst=Paola&rft_id=https%3A%2F%2Fhal.science%2Fhal-03029735&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-:13-220"><span class="mw-cite-backlink">^ <a href="#cite_ref-:13_220-0"><sup><i><b>a</b></i></sup></a> <a href="#cite_ref-:13_220-1"><sup><i><b>b</b></i></sup></a> <a href="#cite_ref-:13_220-2"><sup><i><b>c</b></i></sup></a> <a href="#cite_ref-:13_220-3"><sup><i><b>d</b></i></sup></a></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFMühlhoff2019" class="citation journal cs1">Mühlhoff, Rainer (6 November 2019). <a rel="nofollow" class="external text" href="https://depositonce.tu-berlin.de/handle/11303/12510">"Human-aided artificial intelligence: Or, how to run large computations in human brains? Toward a media sociology of machine learning"</a>. <i>New Media & Society</i>. <b>22</b> (10): 1868–1884. <a href="/wiki/Doi_(identifier)" class="mw-redirect" title="Doi (identifier)">doi</a>:<a rel="nofollow" class="external text" href="https://doi.org/10.1177%2F1461444819885334">10.1177/1461444819885334</a>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1461-4448">1461-4448</a>. <a href="/wiki/S2CID_(identifier)" class="mw-redirect" title="S2CID (identifier)">S2CID</a> <a rel="nofollow" class="external text" href="https://api.semanticscholar.org/CorpusID:209363848">209363848</a>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=New+Media+%26+Society&rft.atitle=Human-aided+artificial+intelligence%3A+Or%2C+how+to+run+large+computations+in+human+brains%3F+Toward+a+media+sociology+of+machine+learning&rft.volume=22&rft.issue=10&rft.pages=1868-1884&rft.date=2019-11-06&rft_id=https%3A%2F%2Fapi.semanticscholar.org%2FCorpusID%3A209363848%23id-name%3DS2CID&rft.issn=1461-4448&rft_id=info%3Adoi%2F10.1177%2F1461444819885334&rft.aulast=M%C3%BChlhoff&rft.aufirst=Rainer&rft_id=https%3A%2F%2Fdepositonce.tu-berlin.de%2Fhandle%2F11303%2F12510&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
<li id="cite_note-221"><span class="mw-cite-backlink"><b><a href="#cite_ref-221">^</a></b></span> <span class="reference-text"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite class="citation magazine cs1"><a rel="nofollow" class="external text" href="https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/">"Facebook Can Now Find Your Face, Even When It's Not Tagged"</a>. <i>Wired</i>. <a href="/wiki/ISSN_(identifier)" class="mw-redirect" title="ISSN (identifier)">ISSN</a> <a rel="nofollow" class="external text" href="https://www.worldcat.org/issn/1059-1028">1059-1028</a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20190810223940/https://www.wired.com/story/facebook-will-find-your-face-even-when-its-not-tagged/">Archived</a> from the original on 10 August 2019<span class="reference-accessdate">. Retrieved <span class="nowrap">22 November</span> 2019</span>.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Ajournal&rft.genre=article&rft.jtitle=Wired&rft.atitle=Facebook+Can+Now+Find+Your+Face%2C+Even+When+It%27s+Not+Tagged&rft.issn=1059-1028&rft_id=https%3A%2F%2Fwww.wired.com%2Fstory%2Ffacebook-will-find-your-face-even-when-its-not-tagged%2F&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span></span>
</li>
</ol></div>
<h2><span class="mw-headline" id="Further_reading">Further reading</span></h2>
<style data-mw-deduplicate="TemplateStyles:r1054258005">.mw-parser-output .refbegin{font-size:90%;margin-bottom:0.5em}.mw-parser-output .refbegin-hanging-indents>ul{margin-left:0}.mw-parser-output .refbegin-hanging-indents>ul>li{margin-left:0;padding-left:3.2em;text-indent:-3.2em}.mw-parser-output .refbegin-hanging-indents ul,.mw-parser-output .refbegin-hanging-indents ul li{list-style:none}@media(max-width:720px){.mw-parser-output .refbegin-hanging-indents>ul>li{padding-left:1.6em;text-indent:-1.6em}}.mw-parser-output .refbegin-columns{margin-top:0.3em}.mw-parser-output .refbegin-columns ul{margin-top:0}.mw-parser-output .refbegin-columns li{page-break-inside:avoid;break-inside:avoid-column}</style><div class="refbegin" style="">
<ul><li><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1133582631"/><cite id="CITEREFGoodfellowBengioCourville2016" class="citation book cs1"><a href="/wiki/Ian_Goodfellow" title="Ian Goodfellow">Goodfellow, Ian</a>; <a href="/wiki/Yoshua_Bengio" title="Yoshua Bengio">Bengio, Yoshua</a>; Courville, Aaron (2016). <a rel="nofollow" class="external text" href="http://www.deeplearningbook.org"><i>Deep Learning</i></a>. MIT Press. <a href="/wiki/ISBN_(identifier)" class="mw-redirect" title="ISBN (identifier)">ISBN</a> <a href="/wiki/Special:BookSources/978-0-26203561-3" title="Special:BookSources/978-0-26203561-3"><bdi>978-0-26203561-3</bdi></a>. <a rel="nofollow" class="external text" href="https://web.archive.org/web/20160416111010/http://www.deeplearningbook.org/">Archived</a> from the original on 2016-04-16<span class="reference-accessdate">. Retrieved <span class="nowrap">2021-05-09</span></span>, introductory textbook.</cite><span title="ctx_ver=Z39.88-2004&rft_val_fmt=info%3Aofi%2Ffmt%3Akev%3Amtx%3Abook&rft.genre=book&rft.btitle=Deep+Learning&rft.pub=MIT+Press&rft.date=2016&rft.isbn=978-0-26203561-3&rft.aulast=Goodfellow&rft.aufirst=Ian&rft.au=Bengio%2C+Yoshua&rft.au=Courville%2C+Aaron&rft_id=http%3A%2F%2Fwww.deeplearningbook.org&rfr_id=info%3Asid%2Fen.wikipedia.org%3ADeep+learning" class="Z3988"></span><span class="cs1-maint citation-comment"><code class="cs1-code">{{<a href="/wiki/Template:Cite_book" title="Template:Cite book">cite book</a>}}</code>: CS1 maint: postscript (<a href="/wiki/Category:CS1_maint:_postscript" title="Category:CS1 maint: postscript">link</a>)</span></li></ul>
</div>
<div class="navbox-styles"><style data-mw-deduplicate="TemplateStyles:r1129693374">.mw-parser-output .hlist dl,.mw-parser-output .hlist ol,.mw-parser-output .hlist ul{margin:0;padding:0}.mw-parser-output .hlist dd,.mw-parser-output .hlist dt,.mw-parser-output .hlist li{margin:0;display:inline}.mw-parser-output .hlist.inline,.mw-parser-output .hlist.inline dl,.mw-parser-output .hlist.inline ol,.mw-parser-output .hlist.inline ul,.mw-parser-output .hlist dl dl,.mw-parser-output .hlist dl ol,.mw-parser-output .hlist dl ul,.mw-parser-output .hlist ol dl,.mw-parser-output .hlist ol ol,.mw-parser-output .hlist ol ul,.mw-parser-output .hlist ul dl,.mw-parser-output .hlist ul ol,.mw-parser-output .hlist ul ul{display:inline}.mw-parser-output .hlist .mw-empty-li{display:none}.mw-parser-output .hlist dt::after{content:": "}.mw-parser-output .hlist dd::after,.mw-parser-output .hlist li::after{content:" · ";font-weight:bold}.mw-parser-output .hlist dd:last-child::after,.mw-parser-output .hlist dt:last-child::after,.mw-parser-output .hlist li:last-child::after{content:none}.mw-parser-output .hlist dd dd:first-child::before,.mw-parser-output .hlist dd dt:first-child::before,.mw-parser-output .hlist dd li:first-child::before,.mw-parser-output .hlist dt dd:first-child::before,.mw-parser-output .hlist dt dt:first-child::before,.mw-parser-output .hlist dt li:first-child::before,.mw-parser-output .hlist li dd:first-child::before,.mw-parser-output .hlist li dt:first-child::before,.mw-parser-output .hlist li li:first-child::before{content:" (";font-weight:normal}.mw-parser-output .hlist dd dd:last-child::after,.mw-parser-output .hlist dd dt:last-child::after,.mw-parser-output .hlist dd li:last-child::after,.mw-parser-output .hlist dt dd:last-child::after,.mw-parser-output .hlist dt dt:last-child::after,.mw-parser-output .hlist dt li:last-child::after,.mw-parser-output .hlist li dd:last-child::after,.mw-parser-output .hlist li dt:last-child::after,.mw-parser-output .hlist li li:last-child::after{content:")";font-weight:normal}.mw-parser-output .hlist ol{counter-reset:listitem}.mw-parser-output .hlist ol>li{counter-increment:listitem}.mw-parser-output .hlist ol>li::before{content:" "counter(listitem)"\a0 "}.mw-parser-output .hlist dd ol>li:first-child::before,.mw-parser-output .hlist dt ol>li:first-child::before,.mw-parser-output .hlist li ol>li:first-child::before{content:" ("counter(listitem)"\a0 "}</style><style data-mw-deduplicate="TemplateStyles:r1061467846">.mw-parser-output .navbox{box-sizing:border-box;border:1px solid #a2a9b1;width:100%;clear:both;font-size:88%;text-align:center;padding:1px;margin:1em auto 0}.mw-parser-output .navbox .navbox{margin-top:0}.mw-parser-output .navbox+.navbox,.mw-parser-output .navbox+.navbox-styles+.navbox{margin-top:-1px}.mw-parser-output .navbox-inner,.mw-parser-output .navbox-subgroup{width:100%}.mw-parser-output .navbox-group,.mw-parser-output .navbox-title,.mw-parser-output .navbox-abovebelow{padding:0.25em 1em;line-height:1.5em;text-align:center}.mw-parser-output .navbox-group{white-space:nowrap;text-align:right}.mw-parser-output .navbox,.mw-parser-output .navbox-subgroup{background-color:#fdfdfd}.mw-parser-output .navbox-list{line-height:1.5em;border-color:#fdfdfd}.mw-parser-output .navbox-list-with-group{text-align:left;border-left-width:2px;border-left-style:solid}.mw-parser-output tr+tr>.navbox-abovebelow,.mw-parser-output tr+tr>.navbox-group,.mw-parser-output tr+tr>.navbox-image,.mw-parser-output tr+tr>.navbox-list{border-top:2px solid #fdfdfd}.mw-parser-output .navbox-title{background-color:#ccf}.mw-parser-output .navbox-abovebelow,.mw-parser-output .navbox-group,.mw-parser-output .navbox-subgroup .navbox-title{background-color:#ddf}.mw-parser-output .navbox-subgroup .navbox-group,.mw-parser-output .navbox-subgroup .navbox-abovebelow{background-color:#e6e6ff}.mw-parser-output .navbox-even{background-color:#f7f7f7}.mw-parser-output .navbox-odd{background-color:transparent}.mw-parser-output .navbox .hlist td dl,.mw-parser-output .navbox .hlist td ol,.mw-parser-output .navbox .hlist td ul,.mw-parser-output .navbox td.hlist dl,.mw-parser-output .navbox td.hlist ol,.mw-parser-output .navbox td.hlist ul{padding:0.125em 0}.mw-parser-output .navbox .navbar{display:block;font-size:100%}.mw-parser-output .navbox-title .navbar{float:left;text-align:left;margin-right:0.5em}</style></div><div role="navigation" class="navbox" aria-labelledby="Differentiable_computing" style="padding:3px"><table class="nowraplinks hlist mw-collapsible autocollapse navbox-inner" style="border-spacing:0;background:transparent;color:inherit"><tbody><tr><th scope="col" class="navbox-title" colspan="2"><link rel="mw-deduplicated-inline-style" href="mw-data:TemplateStyles:r1129693374"/><style data-mw-deduplicate="TemplateStyles:r1063604349">.mw-parser-output .navbar{display:inline;font-size:88%;font-weight:normal}.mw-parser-output .navbar-collapse{float:left;text-align:left}.mw-parser-output .navbar-boxtext{word-spacing:0}.mw-parser-output .navbar ul{display:inline-block;white-space:nowrap;line-height:inherit}.mw-parser-output .navbar-brackets::before{margin-right:-0.125em;content:"[ "}.mw-parser-output .navbar-brackets::after{margin-left:-0.125em;content:" ]"}.mw-parser-output .navbar li{word-spacing:-0.125em}.mw-parser-output .navbar a>span,.mw-parser-output .navbar a>abbr{text-decoration:inherit}.mw-parser-output .navbar-mini abbr{font-variant:small-caps;border-bottom:none;text-decoration:none;cursor:inherit}.mw-parser-output .navbar-ct-full{font-size:114%;margin:0 7em}.mw-parser-output .navbar-ct-mini{font-size:114%;margin:0 4em}</style><div class="navbar plainlinks hlist navbar-mini"><ul><li class="nv-view"><a href="/wiki/Template:Differentiable_computing" title="Template:Differentiable computing"><abbr title="View this template" style=";;background:none transparent;border:none;box-shadow:none;padding:0;">v</abbr></a></li><li class="nv-talk"><a href="/wiki/Template_talk:Differentiable_computing" title="Template talk:Differentiable computing"><abbr title="Discuss this template" style=";;background:none transparent;border:none;box-shadow:none;padding:0;">t</abbr></a></li><li class="nv-edit"><a class="external text" href="https://en.wikipedia.org/w/index.php?title=Template:Differentiable_computing&action=edit"><abbr title="Edit this template" style=";;background:none transparent;border:none;box-shadow:none;padding:0;">e</abbr></a></li></ul></div><div id="Differentiable_computing" style="font-size:114%;margin:0 4em">Differentiable computing</div></th></tr><tr><th scope="row" class="navbox-group" style="width:1%"><a href="/wiki/Differentiable_function" title="Differentiable function">General</a></th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><b><a href="/wiki/Differentiable_programming" title="Differentiable programming">Differentiable programming</a></b></li>
<li><a href="/wiki/Information_geometry" title="Information geometry">Information geometry</a></li>
<li><a href="/wiki/Statistical_manifold" title="Statistical manifold">Statistical manifold</a><br /></li></ul>
<ul><li><a href="/wiki/Automatic_differentiation" title="Automatic differentiation">Automatic differentiation</a></li>
<li><a href="/wiki/Neuromorphic_engineering" title="Neuromorphic engineering">Neuromorphic engineering</a></li>
<li><a href="/wiki/Cable_theory" title="Cable theory">Cable theory</a></li>
<li><a href="/wiki/Pattern_recognition" title="Pattern recognition">Pattern recognition</a></li>
<li><a href="/wiki/Tensor_calculus" title="Tensor calculus">Tensor calculus</a></li>
<li><a href="/wiki/Computational_learning_theory" title="Computational learning theory">Computational learning theory</a></li>
<li><a href="/wiki/Inductive_bias" title="Inductive bias">Inductive bias</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Concepts</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Gradient_descent" title="Gradient descent">Gradient descent</a>
<ul><li><a href="/wiki/Stochastic_gradient_descent" title="Stochastic gradient descent">SGD</a></li></ul></li>
<li><a href="/wiki/Cluster_analysis" title="Cluster analysis">Clustering</a></li>
<li><a href="/wiki/Regression_analysis" title="Regression analysis">Regression</a>
<ul><li><a href="/wiki/Overfitting" title="Overfitting">Overfitting</a></li></ul></li>
<li><a href="/wiki/Hallucination_(artificial_intelligence)" title="Hallucination (artificial intelligence)">Hallucination</a></li>
<li><a href="/wiki/Adversarial_machine_learning" title="Adversarial machine learning">Adversary</a></li>
<li><a href="/wiki/Algorithmic_bias" title="Algorithmic bias">Algorithmic bias</a></li>
<li><a href="/wiki/Attention_(machine_learning)" title="Attention (machine learning)">Attention</a></li>
<li><a href="/wiki/Convolution" title="Convolution">Convolution</a></li>
<li><a href="/wiki/Loss_functions_for_classification" title="Loss functions for classification">Loss functions</a></li>
<li><a href="/wiki/Backpropagation" title="Backpropagation">Backpropagation</a></li>
<li><a href="/wiki/Batch_normalization" title="Batch normalization">Normalization</a></li>
<li><a href="/wiki/Activation_function" title="Activation function">Activation</a>
<ul><li><a href="/wiki/Softmax_function" title="Softmax function">Softmax</a></li>
<li><a href="/wiki/Sigmoid_function" title="Sigmoid function">Sigmoid</a></li>
<li><a href="/wiki/Rectifier_(neural_networks)" title="Rectifier (neural networks)">Rectifier</a></li></ul></li>
<li><a href="/wiki/Regularization_(mathematics)" title="Regularization (mathematics)">Regularization</a></li>
<li><a href="/wiki/Training,_validation,_and_test_sets" class="mw-redirect" title="Training, validation, and test sets">Datasets</a>
<ul><li><a href="/wiki/Data_augmentation" title="Data augmentation">Augmentation</a></li></ul></li>
<li><a href="/wiki/Diffusion_process" title="Diffusion process">Diffusion</a></li>
<li><a href="/wiki/Autoregressive_model" title="Autoregressive model">Autoregression</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Programming languages</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Python_(programming_language)" title="Python (programming language)">Python</a></li>
<li><a href="/wiki/Julia_(programming_language)" title="Julia (programming language)">Julia</a></li>
<li><a href="/wiki/Swift_(programming_language)" title="Swift (programming language)">Swift</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Application</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Machine_learning" title="Machine learning">Machine learning</a></li>
<li><a href="/wiki/Artificial_neural_network" title="Artificial neural network">Artificial neural network</a>
<ul><li><a class="mw-selflink selflink">Deep learning</a></li></ul></li>
<li><a href="/wiki/Computational_science" title="Computational science">Scientific computing</a></li>
<li><a href="/wiki/Artificial_intelligence" title="Artificial intelligence">Artificial Intelligence</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Hardware</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Graphcore" title="Graphcore">IPU</a></li>
<li><a href="/wiki/Tensor_Processing_Unit" title="Tensor Processing Unit">TPU</a></li>
<li><a href="/wiki/Vision_processing_unit" title="Vision processing unit">VPU</a></li>
<li><a href="/wiki/Memristor" title="Memristor">Memristor</a></li>
<li><a href="/wiki/SpiNNaker" title="SpiNNaker">SpiNNaker</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Software library</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/TensorFlow" title="TensorFlow">TensorFlow</a></li>
<li><a href="/wiki/PyTorch" title="PyTorch">PyTorch</a></li>
<li><a href="/wiki/Keras" title="Keras">Keras</a></li>
<li><a href="/wiki/Theano_(software)" title="Theano (software)">Theano</a></li>
<li><a href="/wiki/Google_JAX" title="Google JAX">JAX</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Implementation</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em"></div><table class="nowraplinks navbox-subgroup" style="border-spacing:0"><tbody><tr><th scope="row" class="navbox-group" style="width:1%">Audio–visual</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/AlexNet" title="AlexNet">AlexNet</a></li>
<li><a href="/wiki/WaveNet" title="WaveNet">WaveNet</a></li>
<li><a href="/wiki/Human_image_synthesis" title="Human image synthesis">Human image synthesis</a></li>
<li><a href="/wiki/Handwriting_recognition" title="Handwriting recognition">HWR</a></li>
<li><a href="/wiki/Optical_character_recognition" title="Optical character recognition">OCR</a></li>
<li><a href="/wiki/Deep_learning_speech_synthesis" title="Deep learning speech synthesis">Speech synthesis</a></li>
<li><a href="/wiki/15.ai" title="15.ai">15.ai</a></li>
<li><a href="/wiki/Speech_recognition" title="Speech recognition">Speech recognition</a></li>
<li><a href="/wiki/Facial_recognition_system" title="Facial recognition system">Facial recognition</a></li>
<li><a href="/wiki/AlphaFold" title="AlphaFold">AlphaFold</a></li>
<li><a href="/wiki/DALL-E" title="DALL-E">DALL-E</a></li>
<li><a href="/wiki/Midjourney" title="Midjourney">Midjourney</a></li>
<li><a href="/wiki/Stable_Diffusion" title="Stable Diffusion">Stable Diffusion</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Verbal</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Word2vec" title="Word2vec">Word2vec</a></li>
<li><a href="/wiki/Seq2seq" title="Seq2seq">Seq2seq</a></li>
<li><a href="/wiki/Transformer_(machine_learning_model)" title="Transformer (machine learning model)">Transformer</a></li>
<li><a href="/wiki/BERT_(language_model)" title="BERT (language model)">BERT</a></li>
<li><a href="/wiki/LaMDA" title="LaMDA">LaMDA</a></li>
<li><a href="/wiki/Neural_machine_translation" title="Neural machine translation">NMT</a></li>
<li><a href="/wiki/Project_Debater" title="Project Debater">Project Debater</a></li>
<li><a href="/wiki/IBM_Watson" title="IBM Watson">IBM Watson</a></li>
<li><a href="/wiki/GPT-2" title="GPT-2">GPT-2</a></li>
<li><a href="/wiki/GPT-3" title="GPT-3">GPT-3</a></li>
<li><i><a href="/wiki/GPT-4" title="GPT-4">GPT-4</a> (unreleased)</i></li>
<li><a href="/wiki/GPT-J" title="GPT-J">GPT-J</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Decisional</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/AlphaGo" title="AlphaGo">AlphaGo</a></li>
<li><a href="/wiki/AlphaZero" title="AlphaZero">AlphaZero</a></li>
<li><a href="/wiki/Q-learning" title="Q-learning">Q-learning</a></li>
<li><a href="/wiki/State%E2%80%93action%E2%80%93reward%E2%80%93state%E2%80%93action" title="State–action–reward–state–action">SARSA</a></li>
<li><a href="/wiki/OpenAI_Five" title="OpenAI Five">OpenAI Five</a></li>
<li><a href="/wiki/Self-driving_car" title="Self-driving car">Self-driving car</a></li>
<li><a href="/wiki/MuZero" title="MuZero">MuZero</a></li>
<li><a href="/wiki/Action_selection" title="Action selection">Action selection</a></li>
<li><a href="/wiki/Robot_control" title="Robot control">Robot control</a></li></ul>
</div></td></tr></tbody></table><div></div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">People</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Yoshua_Bengio" title="Yoshua Bengio">Yoshua Bengio</a></li>
<li><a href="/wiki/Alex_Graves_(computer_scientist)" title="Alex Graves (computer scientist)">Alex Graves</a></li>
<li><a href="/wiki/Ian_Goodfellow" title="Ian Goodfellow">Ian Goodfellow</a></li>
<li><a href="/wiki/Demis_Hassabis" title="Demis Hassabis">Demis Hassabis</a></li>
<li><a href="/wiki/Geoffrey_Hinton" title="Geoffrey Hinton">Geoffrey Hinton</a></li>
<li><a href="/wiki/Yann_LeCun" title="Yann LeCun">Yann LeCun</a></li>
<li><a href="/wiki/Fei-Fei_Li" title="Fei-Fei Li">Fei-Fei Li</a></li>
<li><a href="/wiki/Andrew_Ng" title="Andrew Ng">Andrew Ng</a></li>
<li><a href="/wiki/J%C3%BCrgen_Schmidhuber" title="Jürgen Schmidhuber">Jürgen Schmidhuber</a></li>
<li><a href="/wiki/David_Silver_(computer_scientist)" title="David Silver (computer scientist)">David Silver</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Organizations</th><td class="navbox-list-with-group navbox-list navbox-odd" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/DeepMind" title="DeepMind">DeepMind</a></li>
<li><a href="/wiki/OpenAI" title="OpenAI">OpenAI</a></li>
<li><a href="/wiki/MIT_Computer_Science_and_Artificial_Intelligence_Laboratory" title="MIT Computer Science and Artificial Intelligence Laboratory">MIT CSAIL</a></li>
<li><a href="/wiki/Mila_(research_institute)" title="Mila (research institute)">Mila</a></li>
<li><a href="/wiki/Google_Brain" title="Google Brain">Google Brain</a></li>
<li><a href="/wiki/Meta_AI" title="Meta AI">Meta AI</a></li>
<li><a href="/wiki/Anthropic" title="Anthropic">Anthropic</a></li></ul>
</div></td></tr><tr><th scope="row" class="navbox-group" style="width:1%">Architectures</th><td class="navbox-list-with-group navbox-list navbox-even" style="width:100%;padding:0"><div style="padding:0 0.25em">
<ul><li><a href="/wiki/Neural_Turing_machine" title="Neural Turing machine">Neural Turing machine</a></li>
<li><a href="/wiki/Differentiable_neural_computer" title="Differentiable neural computer">Differentiable neural computer</a></li>
<li><a href="/wiki/Transformer_(machine_learning_model)" title="Transformer (machine learning model)">Transformer</a></li>
<li><a href="/wiki/Recurrent_neural_network" title="Recurrent neural network">Recurrent neural network (RNN)</a></li>
<li><a href="/wiki/Long_short-term_memory" title="Long short-term memory">Long short-term memory (LSTM)</a></li>
<li><a href="/wiki/Gated_recurrent_unit" title="Gated recurrent unit">Gated recurrent unit (GRU)</a></li>
<li><a href="/wiki/Echo_state_network" title="Echo state network">Echo state network</a></li>
<li><a href="/wiki/Multilayer_perceptron" title="Multilayer perceptron">Multilayer perceptron (MLP)</a></li>
<li><a href="/wiki/Convolutional_neural_network" title="Convolutional neural network">Convolutional neural network</a></li>
<li><a href="/wiki/Residual_network" class="mw-redirect" title="Residual network">Residual network</a></li>
<li><a href="/wiki/Autoencoder" title="Autoencoder">Autoencoder</a></li>
<li><a href="/wiki/Variational_autoencoder" title="Variational autoencoder">Variational autoencoder (VAE)</a></li>
<li><a href="/wiki/Generative_adversarial_network" title="Generative adversarial network">Generative adversarial network (GAN)</a></li>
<li><a href="/wiki/Graph_neural_network" title="Graph neural network">Graph neural network</a></li></ul>
</div></td></tr><tr><td class="navbox-abovebelow" colspan="2"><div>
<ul><li><a href="/wiki/File:Symbol_portal_class.svg" class="image" title="Portal"><img alt="" src="//upload.wikimedia.org/wikipedia/en/thumb/e/e2/Symbol_portal_class.svg/16px-Symbol_portal_class.svg.png" decoding="async" width="16" height="16" class="noviewer" srcset="//upload.wikimedia.org/wikipedia/en/thumb/e/e2/Symbol_portal_class.svg/23px-Symbol_portal_class.svg.png 1.5x, //upload.wikimedia.org/wikipedia/en/thumb/e/e2/Symbol_portal_class.svg/31px-Symbol_portal_class.svg.png 2x" data-file-width="180" data-file-height="185" /></a> Portals
<ul><li><a href="/wiki/Portal:Computer_programming" title="Portal:Computer programming">Computer programming</a></li>
<li><a href="/wiki/Portal:Technology" title="Portal:Technology">Technology</a></li></ul></li>
<li><img alt="" src="//upload.wikimedia.org/wikipedia/en/thumb/9/96/Symbol_category_class.svg/16px-Symbol_category_class.svg.png" decoding="async" title="Category" width="16" height="16" class="noviewer" srcset="//upload.wikimedia.org/wikipedia/en/thumb/9/96/Symbol_category_class.svg/23px-Symbol_category_class.svg.png 1.5x, //upload.wikimedia.org/wikipedia/en/thumb/9/96/Symbol_category_class.svg/31px-Symbol_category_class.svg.png 2x" data-file-width="180" data-file-height="185" /> Category
<ul><li><a href="/wiki/Category:Artificial_neural_networks" title="Category:Artificial neural networks">Artificial neural networks</a></li>
<li><a href="/wiki/Category:Machine_learning" title="Category:Machine learning">Machine learning</a></li></ul></li></ul>
</div></td></tr></tbody></table></div></div>' |
Whether or not the change was made through a Tor exit node (tor_exit_node) | false |
Unix timestamp of change (timestamp) | '1677556143' |