-
Superconductivity of highly spin-polarized electrons in FeSe probed by $^{77}$Se NMR
Authors:
S. Molatta,
D. Opherden,
J. Wosnitza,
Z. T. Zhang,
T. Wolf,
H. v. Löhneysen,
R. Sarkar,
P. K. Biswas,
H. -J. Grafe,
H. Kühne
Abstract:
A number of recent experiments indicate that the iron-chalcogenide FeSe provides the long-sought possibility to study bulk superconductivity in the cross-over regime between the weakly coupled Bardeen--Cooper--Schrieffer (BCS) pairing and the strongly coupled Bose--Einstein condensation (BEC). We report on $^{77}$Se nuclear magnetic resonance experiments of FeSe, focused on the superconducting pha…
▽ More
A number of recent experiments indicate that the iron-chalcogenide FeSe provides the long-sought possibility to study bulk superconductivity in the cross-over regime between the weakly coupled Bardeen--Cooper--Schrieffer (BCS) pairing and the strongly coupled Bose--Einstein condensation (BEC). We report on $^{77}$Se nuclear magnetic resonance experiments of FeSe, focused on the superconducting phase for strong magnetic fields applied along the $c$ axis, where a distinct state with large spin polarization was reported. We determine this high-field state as bulk superconducting with high spatial homogeneity of the low-energy spin fluctuations. Further, we find that the static spin susceptibility becomes unusually small at temperatures approaching the superconducting state, despite the presence of pronounced spin fluctuations. Taken together, our results clearly indicate that FeSe indeed features an unusual field-induced superconducting state of a highly spin-polarized Fermi liquid in the BCS-BEC crossover regime.
△ Less
Submitted 20 October, 2020;
originally announced October 2020.
-
Freezing of molecular rotation in a paramagnetic crystal studied by $^{31}$P NMR
Authors:
D. Opherden,
F. Bärtl,
Sh. Yamamoto,
Z. T. Zhang,
S. Luther,
S. Molatta,
J. Wosnitza,
M. Baenitz,
I. Heinmaa,
R. Stern,
C. P. Landee,
H. Kühne
Abstract:
We present a detailed $^{31}$P nuclear magnetic resonance (NMR) study of the molecular rotation in the compound [Cu(pz)$_{2}$(2-HOpy)$_{2}$](PF$_{6}$)$_{2}$, where pz = C$_4$H$_4$N$_2$ and 2-HOpy = C$_5$H$_4$NHO. Here, a freezing of the PF$_6$ rotation modes is revealed by several steplike increases of the temperature-dependent second spectral moment, with accompanying broad peaks of the longitudi…
▽ More
We present a detailed $^{31}$P nuclear magnetic resonance (NMR) study of the molecular rotation in the compound [Cu(pz)$_{2}$(2-HOpy)$_{2}$](PF$_{6}$)$_{2}$, where pz = C$_4$H$_4$N$_2$ and 2-HOpy = C$_5$H$_4$NHO. Here, a freezing of the PF$_6$ rotation modes is revealed by several steplike increases of the temperature-dependent second spectral moment, with accompanying broad peaks of the longitudinal and transverse nuclear spin-relaxation rates. An analysis based on the Bloembergen-Purcell-Pound (BPP) theory quantifies the related activation energies as $E_{a}/k_{B}$ = 250 and 1400 K. Further, the anisotropy of the second spectral moment of the $^{31}$P absorption line was calculated for the rigid lattice, as well as in the presence of several sets of PF$_6$ reorientation modes, and is in excellent agreement with the experimental data. Whereas the anisotropy of the frequency shift and enhancement of nuclear spin-relaxation rates is driven by the molecular rotation with respect to the dipole fields stemming from the Cu ions, the second spectral moment is determined by the intramolecular interaction of nuclear $^{19}$F and $^{31}$P moments in the presence of the distinct rotation modes.
△ Less
Submitted 25 January, 2021; v1 submitted 13 October, 2020;
originally announced October 2020.
-
AVLnet: Learning Audio-Visual Language Representations from Instructional Videos
Authors:
Andrew Rouditchenko,
Angie Boggust,
David Harwath,
Brian Chen,
Dhiraj Joshi,
Samuel Thomas,
Kartik Audhkhasi,
Hilde Kuehne,
Rameswar Panda,
Rogerio Feris,
Brian Kingsbury,
Michael Picheny,
Antonio Torralba,
James Glass
Abstract:
Current methods for learning visually grounded language from videos often rely on text annotation, such as human generated captions or machine generated automatic speech recognition (ASR) transcripts. In this work, we introduce the Audio-Video Language Network (AVLnet), a self-supervised network that learns a shared audio-visual embedding space directly from raw video inputs. To circumvent the nee…
▽ More
Current methods for learning visually grounded language from videos often rely on text annotation, such as human generated captions or machine generated automatic speech recognition (ASR) transcripts. In this work, we introduce the Audio-Video Language Network (AVLnet), a self-supervised network that learns a shared audio-visual embedding space directly from raw video inputs. To circumvent the need for text annotation, we learn audio-visual representations from randomly segmented video clips and their raw audio waveforms. We train AVLnet on HowTo100M, a large corpus of publicly available instructional videos, and evaluate on image retrieval and video retrieval tasks, achieving state-of-the-art performance. We perform analysis of AVLnet's learned representations, showing our model utilizes speech and natural sounds to learn audio-visual concepts. Further, we propose a tri-modal model that jointly processes raw audio, video, and text captions from videos to learn a multi-modal semantic embedding space useful for text-video retrieval. Our code, data, and trained models will be released at avlnet.csail.mit.edu
△ Less
Submitted 29 June, 2021; v1 submitted 16 June, 2020;
originally announced June 2020.
-
Extremely well isolated 2D spin-$1/2$ antiferromagnetic Heisenberg layers with small exchange coupling in the molecular-based magnet CuPOF
Authors:
D. Opherden,
N. Nizar,
K. Richardson,
J. C. Monroe,
M. M. Turnbull,
M. Polson,
S. Vela,
W. J. A. Blackmore,
P. A. Goddard,
J. Singleton,
E. S. Choi,
F. Xiao,
R. C. Williams,
T. Lancaster,
F. L. Pratt,
S. J. Blundell,
Y. Skourski,
M. Uhlarz,
A. N. Ponomaryov,
S. A. Zvyagin,
J. Wosnitza,
M. Baenitz,
I. Heinmaa,
R. Stern,
H. Kühne
, et al. (1 additional authors not shown)
Abstract:
We report on a comprehensive characterization of the newly synthesized Cu$^{2+}$-based molecular magnet [Cu(pz)$_2$(2-HOpy)$_2$](PF$_6$)$_2$ (CuPOF), where pz = C$_4$H$_4$N$_2$ and 2-HOpy = C$_5$H$_4$NHO. From a comparison of theoretical modeling to results of bulk magnetometry, specific heat, $μ^+$SR, ESR, and NMR spectroscopy, this material is determined as an excellent realization of the 2D squ…
▽ More
We report on a comprehensive characterization of the newly synthesized Cu$^{2+}$-based molecular magnet [Cu(pz)$_2$(2-HOpy)$_2$](PF$_6$)$_2$ (CuPOF), where pz = C$_4$H$_4$N$_2$ and 2-HOpy = C$_5$H$_4$NHO. From a comparison of theoretical modeling to results of bulk magnetometry, specific heat, $μ^+$SR, ESR, and NMR spectroscopy, this material is determined as an excellent realization of the 2D square-lattice $S=1/2$ antiferromagnetic Heisenberg model with a moderate intraplane nearest-neighbor exchange coupling of $J/k_\mathrm{B} = 6.80(5)$ K, and an extremely small interlayer interaction of about 1 mK. At zero field, the bulk magnetometry reveals a temperature-driven crossover of spin correlations from isotropic to $XY$ type, caused by the presence of a weak intrinsic easy-plane anisotropy. A transition to long-range order, driven by the low-temperature $XY$ anisotropy under the influence of the interlayer coupling, occurs at $T_\mathrm{N} = 1.38(2)$ K, as revealed by $μ^+$SR. In applied magnetic fields, our $^1$H-NMR data reveal a strong increase of the magnetic anisotropy, manifested by a pronounced enhancement of the transition temperature to commensurate long-range order at $T_\mathrm{N} =2.8$ K and 7 T.
△ Less
Submitted 1 September, 2020; v1 submitted 25 May, 2020;
originally announced May 2020.
-
Long-range magnetic order in the ${\tilde S}=1/2$ triangular lattice antiferromagnet KCeS$_2$
Authors:
G. Bastien,
B. Rubrecht,
E. Haeussler,
P. Schlender,
Z. Zangeneh,
S. Avdoshenko,
R. Sarkar,
A. Alfonsov,
S. Luther,
Y. A. Onykiienko,
H. C. Walker,
H. Kühne,
V. Grinenko,
Z. Guguchia,
V. Kataev,
H. -H. Klauss,
L. Hozoi,
J. van den Brink,
D. S. Inosov,
B. Büchner,
A. U. B. Wolter,
T. Doert
Abstract:
Recently, several putative quantum spin liquid (QSL) states were discovered in ${\tilde S} = 1/2$ rare-earth based triangular-lattice antiferromagnets (TLAF) with the delafossite structure. A way to clarify the origin of the QSL state in these systems is to identify ways to tune them from the putative QSL state towards long-range magnetic order. Here, we introduce the Ce-based TLAF KCeS$_2$ and sh…
▽ More
Recently, several putative quantum spin liquid (QSL) states were discovered in ${\tilde S} = 1/2$ rare-earth based triangular-lattice antiferromagnets (TLAF) with the delafossite structure. A way to clarify the origin of the QSL state in these systems is to identify ways to tune them from the putative QSL state towards long-range magnetic order. Here, we introduce the Ce-based TLAF KCeS$_2$ and show via low-temperature specific heat and $μ$SR investigations that it yields magnetic order below $T_{\mathrm N} = 0.38$ K despite the same delafossite structure. We identify a well separated ${\tilde S} = 1/2$ ground state for KCeS$_2$ from inelastic neutron scattering and embedded-cluster quantum chemical calculations. Magnetization and electron spin resonance measurements on single crystals indicate a strong easy-plane $g$~factor anisotropy, in agreement with the ab initio calculations. Finally, our specific-heat studies reveal an in-plane anisotropy of the magnetic field-temperature phase diagram which may indicate anisotropic magnetic interactions in KCeS$_2$.
△ Less
Submitted 22 July, 2020; v1 submitted 13 May, 2020;
originally announced May 2020.
-
Joint Visual-Temporal Embedding for Unsupervised Learning of Actions in Untrimmed Sequences
Authors:
Rosaura G. VidalMata,
Walter J. Scheirer,
Anna Kukleva,
David Cox,
Hilde Kuehne
Abstract:
Understanding the structure of complex activities in untrimmed videos is a challenging task in the area of action recognition. One problem here is that this task usually requires a large amount of hand-annotated minute- or even hour-long video data, but annotating such data is very time consuming and can not easily be automated or scaled. To address this problem, this paper proposes an approach fo…
▽ More
Understanding the structure of complex activities in untrimmed videos is a challenging task in the area of action recognition. One problem here is that this task usually requires a large amount of hand-annotated minute- or even hour-long video data, but annotating such data is very time consuming and can not easily be automated or scaled. To address this problem, this paper proposes an approach for the unsupervised learning of actions in untrimmed video sequences based on a joint visual-temporal embedding space. To this end, we combine a visual embedding based on a predictive U-Net architecture with a temporal continuous function. The resulting representation space allows detecting relevant action clusters based on their visual as well as their temporal appearance. The proposed method is evaluated on three standard benchmark datasets, Breakfast Actions, INRIA YouTube Instructional Videos, and 50 Salads. We show that the proposed approach is able to provide a meaningful visual and temporal embedding out of the visual cues present in contiguous video frames and is suitable for the task of unsupervised temporal segmentation of actions.
△ Less
Submitted 30 September, 2020; v1 submitted 29 January, 2020;
originally announced January 2020.
-
More Is Less: Learning Efficient Video Representations by Big-Little Network and Depthwise Temporal Aggregation
Authors:
Quanfu Fan,
Chun-Fu Chen,
Hilde Kuehne,
Marco Pistoia,
David Cox
Abstract:
Current state-of-the-art models for video action recognition are mostly based on expensive 3D ConvNets. This results in a need for large GPU clusters to train and evaluate such architectures. To address this problem, we present a lightweight and memory-friendly architecture for action recognition that performs on par with or better than current architectures by using only a fraction of resources.…
▽ More
Current state-of-the-art models for video action recognition are mostly based on expensive 3D ConvNets. This results in a need for large GPU clusters to train and evaluate such architectures. To address this problem, we present a lightweight and memory-friendly architecture for action recognition that performs on par with or better than current architectures by using only a fraction of resources. The proposed architecture is based on a combination of a deep subnet operating on low-resolution frames with a compact subnet operating on high-resolution frames, allowing for high efficiency and accuracy at the same time. We demonstrate that our approach achieves a reduction by $3\sim4$ times in FLOPs and $\sim2$ times in memory usage compared to the baseline. This enables training deeper models with more input frames under the same computational budget. To further obviate the need for large-scale 3D convolutions, a temporal aggregation module is proposed to model temporal dependencies in a video at very small additional computational costs. Our models achieve strong performance on several action recognition benchmarks including Kinetics, Something-Something and Moments-in-time. The code and models are available at https://github.com/IBM/bLVNet-TAM.
△ Less
Submitted 2 December, 2019;
originally announced December 2019.
-
Anisotropic field-induced ordering in the triangular-lattice quantum spin liquid NaYbSe$_2$
Authors:
K. M. Ranjith,
S. Luther,
T. Reimann,
B. Schmidt,
Ph. Schlender,
J. Sichelschmidt,
H. Yasuoka,
A. M. Strydom,
Y. Skourski,
J. Wosnitza,
H. Kühne,
Th. Doert,
M. Baenitz
Abstract:
High-quality single crystals of NaYbSe$_2$, which resembles a perfect triangular-lattice antiferromagnet without the intrinsic disorder, are investigated by magnetization and specific heat, as well as the local probe techniques nuclear magnetic resonance (NMR) and electron spin resonance (ESR). The low-field measurements confirm the absence of any spin freezing or long-range magnetic order down to…
▽ More
High-quality single crystals of NaYbSe$_2$, which resembles a perfect triangular-lattice antiferromagnet without the intrinsic disorder, are investigated by magnetization and specific heat, as well as the local probe techniques nuclear magnetic resonance (NMR) and electron spin resonance (ESR). The low-field measurements confirm the absence of any spin freezing or long-range magnetic order down to 50~mK, which suggests a quantum spin liquid ground (QSL) state with gapless excitations. The instability of the QSL state is observed upon applying magnetic fields. For the $H\bot c$ direction, a field-induced magnetic phase transition is observed above 2~T from the $C_{\rm p}(T)$ data, agreeing with a clear $\frac{M_s}{3}$ plateau of $M(H)$, which is associated with an up-up-down (uud) spin arrangement. For the $H\|c$ direction, a field-induced transition could be evidenced at a much higher field range (9 - 21~T). The $^{23}$Na NMR measurements provide microscopic evidence for field-induced ordering for both directions. A reentrant behavior of $T_{\rm N}$, originating from the thermal and quantum spin fluctuations, is observed for both directions. The anisotropic exchange interactions $J_{\perp}\simeq$ 4.7~K and $J_z\simeq$2.33~K are extracted from the modified bond-dependent XXZ model for the spin-$\frac{1}{2}$ triangular-lattice antiferromagnet. The absence of magnetic long-range order at zero fields is assigned to the effect of strong bond-frustration, arising from the complex spin-orbit entangled $4f$ ground state. Finally, we derive the highly anisotropic magnetic phase diagram, which is discussed in comparison with the existing theoretical models for spin-$\frac{1}{2}$ triangular-lattice antiferromagnets.
△ Less
Submitted 28 November, 2019;
originally announced November 2019.
-
A Hybrid RNN-HMM Approach for Weakly Supervised Temporal Action Segmentation
Authors:
Hilde Kuehne,
Alexander Richard,
Juergen Gall
Abstract:
Action recognition has become a rapidly developing research field within the last decade. But with the increasing demand for large scale data, the need of hand annotated data for the training becomes more and more impractical. One way to avoid frame-based human annotation is the use of action order information to learn the respective action classes. In this context, we propose a hierarchical appro…
▽ More
Action recognition has become a rapidly developing research field within the last decade. But with the increasing demand for large scale data, the need of hand annotated data for the training becomes more and more impractical. One way to avoid frame-based human annotation is the use of action order information to learn the respective action classes. In this context, we propose a hierarchical approach to address the problem of weakly supervised learning of human actions from ordered action labels by structuring recognition in a coarse-to-fine manner. Given a set of videos and an ordered list of the occurring actions, the task is to infer start and end frames of the related action classes within the video and to train the respective action classifiers without any need for hand labeled frame boundaries. We address this problem by combining a framewise RNN model with a coarse probabilistic inference. This combination allows for the temporal alignment of long sequences and thus, for an iterative training of both elements. While this system alone already generates good results, we show that the performance can be further improved by approximating the number of subactions to the characteristics of the different action classes as well as by the introduction of a regularizing length prior. The proposed system is evaluated on two benchmark datasets, the Breakfast and the Hollywood extended dataset, showing a competitive performance on various weak learning tasks such as temporal action segmentation and action alignment.
△ Less
Submitted 3 June, 2019;
originally announced June 2019.
-
Mining YouTube - A dataset for learning fine-grained action concepts from webly supervised video data
Authors:
Hilde Kuehne,
Ahsan Iqbal,
Alexander Richard,
Juergen Gall
Abstract:
Action recognition is so far mainly focusing on the problem of classification of hand selected preclipped actions and reaching impressive results in this field. But with the performance even ceiling on current datasets, it also appears that the next steps in the field will have to go beyond this fully supervised classification. One way to overcome those problems is to move towards less restricted…
▽ More
Action recognition is so far mainly focusing on the problem of classification of hand selected preclipped actions and reaching impressive results in this field. But with the performance even ceiling on current datasets, it also appears that the next steps in the field will have to go beyond this fully supervised classification. One way to overcome those problems is to move towards less restricted scenarios. In this context we present a large-scale real-world dataset designed to evaluate learning techniques for human action recognition beyond hand-crafted datasets. To this end we put the process of collecting data on its feet again and start with the annotation of a test set of 250 cooking videos. The training data is then gathered by searching for the respective annotated classes within the subtitles of freely available videos. The uniqueness of the dataset is attributed to the fact that the whole process of collecting the data and training does not involve any human intervention. To address the problem of semantic inconsistencies that arise with this kind of training data, we further propose a semantical hierarchical structure for the mined classes.
△ Less
Submitted 3 June, 2019;
originally announced June 2019.
-
Unsupervised learning of action classes with continuous temporal embedding
Authors:
Anna Kukleva,
Hilde Kuehne,
Fadime Sener,
Juergen Gall
Abstract:
The task of temporally detecting and segmenting actions in untrimmed videos has seen an increased attention recently. One problem in this context arises from the need to define and label action boundaries to create annotations for training which is very time and cost intensive. To address this issue, we propose an unsupervised approach for learning action classes from untrimmed video sequences. To…
▽ More
The task of temporally detecting and segmenting actions in untrimmed videos has seen an increased attention recently. One problem in this context arises from the need to define and label action boundaries to create annotations for training which is very time and cost intensive. To address this issue, we propose an unsupervised approach for learning action classes from untrimmed video sequences. To this end, we use a continuous temporal embedding of framewise features to benefit from the sequential nature of activities. Based on the latent space created by the embedding, we identify clusters of temporal segments across all videos that correspond to semantic meaningful action classes. The approach is evaluated on three challenging datasets, namely the Breakfast dataset, YouTube Instructions, and the 50Salads dataset. While previous works assumed that the videos contain the same high level activity, we furthermore show that the proposed approach can also be applied to a more general setting where the content of the videos is unknown.
△ Less
Submitted 8 April, 2019;
originally announced April 2019.
-
Field-induced instability of the quantum-spin-liquid ground state in the $J_{\rm eff}=\frac{1}{2}$ triangular-lattice compound NaYbO$_2$
Authors:
K. M. Ranjith,
D. Dmytriieva,
S. Khim,
J. Sichelschmidt,
S. Luther,
D. Ehlers,
H. Yasuoka,
J. Wosnitza,
A. A. Tsirlin,
H. Kühne,
M. Baenitz
Abstract:
Polycrystalline samples of NaYbO$_2$ are investigated by bulk magnetization and specific-heat measurements, as well as by nuclear magnetic resonance (NMR) and electron spin resonance (ESR) as local probes. No signatures of long-range magnetic order are found down to 0.3~K, evidencing a highly frustrated spin-liquid-like ground state in zero field. Above 2\,T, signatures of magnetic order are obser…
▽ More
Polycrystalline samples of NaYbO$_2$ are investigated by bulk magnetization and specific-heat measurements, as well as by nuclear magnetic resonance (NMR) and electron spin resonance (ESR) as local probes. No signatures of long-range magnetic order are found down to 0.3~K, evidencing a highly frustrated spin-liquid-like ground state in zero field. Above 2\,T, signatures of magnetic order are observed in thermodynamic measurements, suggesting the possibility of a field-induced quantum phase transition. The $^{23}$Na NMR relaxation rates reveal the absence of magnetic order and persistent fluctuations down to 0.3~K at very low fields and confirm the bulk magnetic order above 2~T. The $H$-$T$ phase diagram is obtained and discussed along with the existing theoretical concepts for layered spin-$\frac{1}{2}$ triangular-lattice antiferromagnets
△ Less
Submitted 1 May, 2019; v1 submitted 23 January, 2019;
originally announced January 2019.
-
NeuralNetwork-Viterbi: A Framework for Weakly Supervised Video Learning
Authors:
Alexander Richard,
Hilde Kuehne,
Ahsan Iqbal,
Juergen Gall
Abstract:
Video learning is an important task in computer vision and has experienced increasing interest over the recent years. Since even a small amount of videos easily comprises several million frames, methods that do not rely on a frame-level annotation are of special importance. In this work, we propose a novel learning algorithm with a Viterbi-based loss that allows for online and incremental learning…
▽ More
Video learning is an important task in computer vision and has experienced increasing interest over the recent years. Since even a small amount of videos easily comprises several million frames, methods that do not rely on a frame-level annotation are of special importance. In this work, we propose a novel learning algorithm with a Viterbi-based loss that allows for online and incremental learning of weakly annotated video data. We moreover show that explicit context and length modeling leads to huge improvements in video segmentation and labeling tasks andinclude these models into our framework. On several action segmentation benchmarks, we obtain an improvement of up to 10% compared to current state-of-the-art methods.
△ Less
Submitted 17 May, 2018;
originally announced May 2018.
-
Monovacancy paramagnetism in neutron-irradiated graphite probed by $^{13}$C NMR
Authors:
Z. T. Zhang,
C. Xu,
D. Dmytriieva,
S. Molatta,
J. Wosnitza,
Y. T. Wang,
M. Helm,
Shengqiang Zhou,
H. Kühne
Abstract:
We report on the magnetic properties of monovacancy defects in neutron-irradiated graphite, probed by $^{13}$C nuclear magnetic resonance spectroscopy. The bulk paramagnetism of the defect moments is revealed by the temperature dependence of the NMR frequency shift and spectral linewidth, both of which follow a Curie behavior, in agreement with measurements of the macroscopic magnetization. Compar…
▽ More
We report on the magnetic properties of monovacancy defects in neutron-irradiated graphite, probed by $^{13}$C nuclear magnetic resonance spectroscopy. The bulk paramagnetism of the defect moments is revealed by the temperature dependence of the NMR frequency shift and spectral linewidth, both of which follow a Curie behavior, in agreement with measurements of the macroscopic magnetization. Compared to pristine graphite, the fluctuating hyperfine fields generated by the defect moments lead to an enhancement of the $^{13}$C nuclear spin-lattice relaxation rate $1/T_{1}$ by about two orders of magnitude. With an applied magnetic field of 7.1 T, the temperature dependence of $1/T_{1}$ below about 10 K can well be described by a thermally activated form, $1/T_{1}\propto\exp(-Δ/k_{B}T)$, yielding a singular Zeeman energy of ($0.41\pm0.01$) meV, in excellent agreement with the sole presence of polarized, non-interacting defect moments.
△ Less
Submitted 20 December, 2017; v1 submitted 12 July, 2017;
originally announced July 2017.
-
Recurrent Residual Learning for Action Recognition
Authors:
Ahsan Iqbal,
Alexander Richard,
Hilde Kuehne,
Juergen Gall
Abstract:
Action recognition is a fundamental problem in computer vision with a lot of potential applications such as video surveillance, human computer interaction, and robot learning. Given pre-segmented videos, the task is to recognize actions happening within videos. Historically, hand crafted video features were used to address the task of action recognition. With the success of Deep ConvNets as an ima…
▽ More
Action recognition is a fundamental problem in computer vision with a lot of potential applications such as video surveillance, human computer interaction, and robot learning. Given pre-segmented videos, the task is to recognize actions happening within videos. Historically, hand crafted video features were used to address the task of action recognition. With the success of Deep ConvNets as an image analysis method, a lot of extensions of standard ConvNets were purposed to process variable length video data. In this work, we propose a novel recurrent ConvNet architecture called recurrent residual networks to address the task of action recognition. The approach extends ResNet, a state of the art model for image classification. While the original formulation of ResNet aims at learning spatial residuals in its layers, we extend the approach by introducing recurrent connections that allow to learn a spatio-temporal residual. In contrast to fully recurrent networks, our temporal connections only allow a limited range of preceding frames to contribute to the output for the current frame, enabling efficient training and inference as well as limiting the temporal context to a reasonable local range around each frame. On a large-scale action recognition dataset, we show that our model improves over both, the standard ResNet architecture and a ResNet extended by a fully recurrent layer.
△ Less
Submitted 27 June, 2017;
originally announced June 2017.
-
Action Sets: Weakly Supervised Action Segmentation without Ordering Constraints
Authors:
Alexander Richard,
Hilde Kuehne,
Juergen Gall
Abstract:
Action detection and temporal segmentation of actions in videos are topics of increasing interest. While fully supervised systems have gained much attention lately, full annotation of each action within the video is costly and impractical for large amounts of video data. Thus, weakly supervised action detection and temporal segmentation methods are of great importance. While most works in this are…
▽ More
Action detection and temporal segmentation of actions in videos are topics of increasing interest. While fully supervised systems have gained much attention lately, full annotation of each action within the video is costly and impractical for large amounts of video data. Thus, weakly supervised action detection and temporal segmentation methods are of great importance. While most works in this area assume an ordered sequence of occurring actions to be given, our approach only uses a set of actions. Such action sets provide much less supervision since neither action ordering nor the number of action occurrences are known. In exchange, they can be easily obtained, for instance, from meta-tags, while ordered sequences still require human annotation. We introduce a system that automatically learns to temporally segment and label actions in a video, where the only supervision that is used are action sets. An evaluation on three datasets shows that our method still achieves good results although the amount of supervision is significantly smaller than for other related methods.
△ Less
Submitted 17 May, 2018; v1 submitted 2 June, 2017;
originally announced June 2017.
-
Weakly Supervised Action Learning with RNN based Fine-to-coarse Modeling
Authors:
Alexander Richard,
Hilde Kuehne,
Juergen Gall
Abstract:
We present an approach for weakly supervised learning of human actions. Given a set of videos and an ordered list of the occurring actions, the goal is to infer start and end frames of the related action classes within the video and to train the respective action classifiers without any need for hand labeled frame boundaries. To address this task, we propose a combination of a discriminative repre…
▽ More
We present an approach for weakly supervised learning of human actions. Given a set of videos and an ordered list of the occurring actions, the goal is to infer start and end frames of the related action classes within the video and to train the respective action classifiers without any need for hand labeled frame boundaries. To address this task, we propose a combination of a discriminative representation of subactions, modeled by a recurrent neural network, and a coarse probabilistic model to allow for a temporal alignment and inference over long sequences. While this system alone already generates good results, we show that the performance can be further improved by approximating the number of subactions to the characteristics of the different action classes. To this end, we adapt the number of subaction classes by iterating realignment and reestimation during training. The proposed system is evaluated on two benchmark datasets, the Breakfast and the Hollywood extended dataset, showing a competitive performance on various weak learning tasks such as temporal action segmentation and action alignment.
△ Less
Submitted 9 October, 2017; v1 submitted 23 March, 2017;
originally announced March 2017.
-
Increasing stripe-type fluctuations in $A$Fe$_{2}$As$_{2}$ ($A$ = K, Rb, Cs) superconductors probed by $^{75}$As NMR spectroscopy
Authors:
Z. T. Zhang,
D. Dmytriieva,
S. Molatta,
J. Wosnitza,
S. Khim,
S. Gass,
A. U. B. Wolter,
S. Wurmehl,
H. -J. Grafe,
H. Kühne
Abstract:
We report $^{75}$As nuclear magnetic resonance measurements on single crystals of RbFe$_{2}$As$_{2}$ and CsFe$_{2}$As$_{2}$. Taking previously reported results for KFe$_{2}$As$_{2}$ into account, we find that the anisotropic electronic correlations evolve towards a magnetic instability in the $A$Fe$_{2}$As$_{2}$ series (with $A$ = K, Rb, Cs). Upon isovalent substitution with larger alkali ions, a…
▽ More
We report $^{75}$As nuclear magnetic resonance measurements on single crystals of RbFe$_{2}$As$_{2}$ and CsFe$_{2}$As$_{2}$. Taking previously reported results for KFe$_{2}$As$_{2}$ into account, we find that the anisotropic electronic correlations evolve towards a magnetic instability in the $A$Fe$_{2}$As$_{2}$ series (with $A$ = K, Rb, Cs). Upon isovalent substitution with larger alkali ions, a drastic enhancement of the anisotropic nuclear spin-lattice relaxation rate and decreasing Knight shift reveal the formation of pronounced spin fluctuations with stripe-type modulation. Furthermore, a decreasing power-law exponent of the nuclear spin-lattice relaxation rate $(1/T_{1})_{H\parallel{ab}}$, probing the in-plane spin fluctuations, evidences an emergent deviation from Fermi-liquid behavior. All these findings clearly indicate that the expansion of the lattice in the $A$Fe$_{2}$As$_{2}$ series tunes the electronic correlations towards a quantum critical point at the transition to a yet unobserved, ordered phase.
△ Less
Submitted 7 March, 2018; v1 submitted 2 March, 2017;
originally announced March 2017.
-
Defect-induced magnetism in SiC probed by nuclear magnetic resonance
Authors:
Z. T. Zhang,
D. Dmytriieva,
S. Molatta,
J. Wosnitza,
Yutian Wang,
M. Helm,
Shengqiang Zhou,
H. Kühne
Abstract:
We give evidence for intrinsic, defect-induced bulk paramagnetism in SiC by means of $^{13}$C and $^{29}$Si nuclear magnetic resonance (NMR) spectroscopy. The temperature dependence of the internal dipole-field distribution, probed by the spin part of the NMR Knight shift and the spectral linewidth, follows a Curie law and scales very well with the macroscopic DC susceptibility. In order to quanti…
▽ More
We give evidence for intrinsic, defect-induced bulk paramagnetism in SiC by means of $^{13}$C and $^{29}$Si nuclear magnetic resonance (NMR) spectroscopy. The temperature dependence of the internal dipole-field distribution, probed by the spin part of the NMR Knight shift and the spectral linewidth, follows a Curie law and scales very well with the macroscopic DC susceptibility. In order to quantitatively analyze the NMR spectra, a microscopic model based on dipole-dipole interactions was developed. The very good agreement between these simulations and the NMR data establishes a direct relation between the frequency distribution of the spectral intensity and the corresponding real-space volumes of nuclear spins. The presented approach by NMR can be applied to a variety of similar materials and, thus, opens a new avenue for the microscopic exploration and exploitation of diluted bulk magnetism in semiconductors.
△ Less
Submitted 12 March, 2017; v1 submitted 24 November, 2016;
originally announced November 2016.
-
Weakly supervised learning of actions from transcripts
Authors:
Hilde Kuehne,
Alexander Richard,
Juergen Gall
Abstract:
We present an approach for weakly supervised learning of human actions from video transcriptions. Our system is based on the idea that, given a sequence of input data and a transcript, i.e. a list of the order the actions occur in the video, it is possible to infer the actions within the video stream, and thus, learn the related action models without the need for any frame-based annotation. Starti…
▽ More
We present an approach for weakly supervised learning of human actions from video transcriptions. Our system is based on the idea that, given a sequence of input data and a transcript, i.e. a list of the order the actions occur in the video, it is possible to infer the actions within the video stream, and thus, learn the related action models without the need for any frame-based annotation. Starting from the transcript information at hand, we split the given data sequences uniformly based on the number of expected actions. We then learn action models for each class by maximizing the probability that the training video sequences are generated by the action models given the sequence order as defined by the transcripts. The learned model can be used to temporally segment an unseen video with or without transcript. We evaluate our approach on four distinct activity datasets, namely Hollywood Extended, MPII Cooking, Breakfast and CRIM13. We show that our system is able to align the scripted actions with the video data and that the learned models localize and classify actions competitively in comparison to models trained with full supervision, i.e. with frame level annotations, and that they outperform any current state-of-the-art approach for aligning transcripts with video data.
△ Less
Submitted 19 June, 2017; v1 submitted 7 October, 2016;
originally announced October 2016.
-
Charge fluctuations and superconductivity in organic conductors: the case of $β"$-(BEDT-TTF)$_2$SF$_5$CH$_2$CF$_2$SO$_3$
Authors:
G. Koutroulakis,
H. Kühne,
H. H. Wang,
J. A. Schlueter,
J. Wosnitza,
S. E. Brown
Abstract:
A $^{13}$C NMR study of the normal and superconducting states of the all-organic charge-transfer salt $β"$-(BEDT-TTF)$_2$SF$_5$CH$_2$CF$_2$SO$_3$ is presented. We find that the normal state is a charge-ordered metal configured as vertical stripes, produced by a combination of 1/4-filling, correlations, and a polar counterion sublattice. The NMR properties associated with the superconducting state…
▽ More
A $^{13}$C NMR study of the normal and superconducting states of the all-organic charge-transfer salt $β"$-(BEDT-TTF)$_2$SF$_5$CH$_2$CF$_2$SO$_3$ is presented. We find that the normal state is a charge-ordered metal configured as vertical stripes, produced by a combination of 1/4-filling, correlations, and a polar counterion sublattice. The NMR properties associated with the superconducting state are consistent with gap nodes and singlet pairing, and therefore similar to other organic superconductors. Quite distinct, however, is the absence of evidence for low-energy antiferromagnetic spin fluctuations for $T>T_c=4.5$ K. Both aspects are discussed in the context of a proposal that the pairing in this compound is driven by charge fluctuations.
△ Less
Submitted 22 January, 2016;
originally announced January 2016.
-
Microscopic study of the Fulde-Ferrell-Larkin-Ovchinnikov state in an all-organic superconductor
Authors:
G. Koutroulakis,
H. Kühne,
J. A. Schlueter,
J. Wosnitza,
S. E. Brown
Abstract:
Quasi-two dimensional superconductors with sufficiently weak interlayer coupling allow magnetic flux to penetrate in the form of Josephson vortices for in-plane applied magnetic fields. A consequence is the dominance of the Zeeman interaction over orbital effects. In the clean limit, the normal state is favored over superconductivity for fields greater than the paramagnetic limiting field, unless…
▽ More
Quasi-two dimensional superconductors with sufficiently weak interlayer coupling allow magnetic flux to penetrate in the form of Josephson vortices for in-plane applied magnetic fields. A consequence is the dominance of the Zeeman interaction over orbital effects. In the clean limit, the normal state is favored over superconductivity for fields greater than the paramagnetic limiting field, unless an intermediate, inhomogeneous state is stabilized. Presented here are nuclear magnetic resonance (NMR) studies of the inhomogeneous (FFLO) state for $β^{\prime \prime}$-(ET)$_2$SF$_5$CH$_2$CF$_2$SO$_3$. The uniform superconductivity-FFLO transition is identified at an applied field value of 9.3(0.1) T at low temperature ($T=130$ mK), and evidence for a possible second transition between inhomogeneous states at $\sim11$ T is presented. The spin polarization distribution inferred from the NMR absorption spectrum compares favorably to a single-Q modulation of the superconducting order parameter.
△ Less
Submitted 11 November, 2015;
originally announced November 2015.
-
An end-to-end generative framework for video segmentation and recognition
Authors:
Hilde Kuehne,
Juergen Gall,
Thomas Serre
Abstract:
We describe an end-to-end generative approach for the segmentation and recognition of human activities. In this approach, a visual representation based on reduced Fisher Vectors is combined with a structured temporal model for recognition. We show that the statistical properties of Fisher Vectors make them an especially suitable front-end for generative models such as Gaussian mixtures. The system…
▽ More
We describe an end-to-end generative approach for the segmentation and recognition of human activities. In this approach, a visual representation based on reduced Fisher Vectors is combined with a structured temporal model for recognition. We show that the statistical properties of Fisher Vectors make them an especially suitable front-end for generative models such as Gaussian mixtures. The system is evaluated for both the recognition of complex activities as well as their parsing into action units. Using a variety of video datasets ranging from human cooking activities to animal behaviors, our experiments demonstrate that the resulting architecture outperforms state-of-the-art approaches for larger datasets, i.e. when sufficient amount of data is available for training structured generative models.
△ Less
Submitted 17 March, 2016; v1 submitted 7 September, 2015;
originally announced September 2015.
-
Cooking in the kitchen: Recognizing and Segmenting Human Activities in Videos
Authors:
Hilde Kuehne,
Juergen Gall,
Thomas Serre
Abstract:
As research on action recognition matures, the focus is shifting away from categorizing basic task-oriented actions using hand-segmented video datasets to understanding complex goal-oriented daily human activities in real-world settings. Temporally structured models would seem obvious to tackle this set of problems, but so far, cases where these models have outperformed simpler unstructured bag-of…
▽ More
As research on action recognition matures, the focus is shifting away from categorizing basic task-oriented actions using hand-segmented video datasets to understanding complex goal-oriented daily human activities in real-world settings. Temporally structured models would seem obvious to tackle this set of problems, but so far, cases where these models have outperformed simpler unstructured bag-of-word types of models are scarce. With the increasing availability of large human activity datasets, combined with the development of novel feature coding techniques that yield more compact representations, it is time to revisit structured generative approaches.
Here, we describe an end-to-end generative approach from the encoding of features to the structural modeling of complex human activities by applying Fisher vectors and temporal models for the analysis of video sequences.
We systematically evaluate the proposed approach on several available datasets (ADL, MPIICooking, and Breakfast datasets) using a variety of performance metrics. Through extensive system evaluations, we demonstrate that combining compact video representations based on Fisher Vectors with HMM-based modeling yields very significant gains in accuracy and when properly trained with sufficient training samples, structured temporal models outperform unstructured bag-of-word types of models by a large margin on the tested performance metric.
△ Less
Submitted 17 March, 2016; v1 submitted 25 August, 2015;
originally announced August 2015.
-
Critical exponents and intrinsic broadening of the field-induced transition in NiCl$_2$$\cdot$4SC(NH$_2$)$_2$
Authors:
E. Wulf,
D. Hüvonen,
R. Schönemann,
H. Kühne,
T. Herrmannsdörfer,
I. Glavatskyy,
S. Gerischer,
K. Kiefer,
S. Gvasaliya,
A. Zheludev
Abstract:
The field-induced ordering transition in the quantum spin system NiCl$_2$$\cdot$4SC(NH$_2$)$_2$ is studied by means of neutron diffraction, AC magnetometry and relaxation calorimetry. The interpretation of the data is strongly influenced by a finite distribution of transition fields in the samples, which was present but disregarded in previous studies. Taking this effect into account, we find that…
▽ More
The field-induced ordering transition in the quantum spin system NiCl$_2$$\cdot$4SC(NH$_2$)$_2$ is studied by means of neutron diffraction, AC magnetometry and relaxation calorimetry. The interpretation of the data is strongly influenced by a finite distribution of transition fields in the samples, which was present but disregarded in previous studies. Taking this effect into account, we find that the order-parameter critical exponent is inconsistent with the BEC universality class even at temperatures below 100 mK. All results are discussed in comparison with previous measurements and with recent similar studies of disordered Ni(Cl$_{1-x}$Br$_x$)$_2$$\cdot$4SC(NH$_2$)$_2$.
△ Less
Submitted 15 December, 2014; v1 submitted 1 December, 2014;
originally announced December 2014.
-
Multigap superconductivity in locally non-centrosymmetric SrPtAs: An As nuclear quadrupole resonance investigation
Authors:
F. Brückner,
R. Sarkar,
M. Günther,
H. Kühne,
H. Luetkens,
T. Neupert,
A. P. Reyes,
P. L. Kuhns,
P. K. Biswas,
T. Stürzer,
D. Johrendt,
H. -H. Klauss
Abstract:
We report detailed $^{75}$As-NQR investigations of the locally non-centrosymmetric superconductor SrPtAs. The spin-lattice relaxation studies prove weakly coupled multi-gap superconductivity. The Hebel-Slichter peak, a hallmark of conventional superconductivity, is strongly suppressed, which points to an unconventional superconducting state. The observed behavior excludes a superconducting order p…
▽ More
We report detailed $^{75}$As-NQR investigations of the locally non-centrosymmetric superconductor SrPtAs. The spin-lattice relaxation studies prove weakly coupled multi-gap superconductivity. The Hebel-Slichter peak, a hallmark of conventional superconductivity, is strongly suppressed, which points to an unconventional superconducting state. The observed behavior excludes a superconducting order parameter with line nodes and is consistent with proposed $f$-wave and chiral $d$-wave order parameters.
△ Less
Submitted 20 December, 2013;
originally announced December 2013.
-
Progressive slowing down of spin fluctuations in underdoped LaFeAsO$_{1-x}$F$_x$
Authors:
F. Hammerath,
U. Gräfe,
H. Kühne,
P. L. Kuhns,
A. P. Reyes,
G. Lang,
S. Wurmehl,
B. Büchner,
P. Carretta,
H. -J. Grafe
Abstract:
The evolution of low-energy spin dynamics in the iron-based superconductor LaFeAsO$_{1-x}$F$_x$ was studied over a broad doping, temperature, and magnetic field range (x = 0 - 0.15, T up to 480K, H up to 30T) by means of As nuclear magnetic resonance (NMR). An enhanced spin-lattice relaxation rate divided by temperature, 1/T1T, in underdoped superconducting samples (x = 0.045, 0.05 and 0.075) sugg…
▽ More
The evolution of low-energy spin dynamics in the iron-based superconductor LaFeAsO$_{1-x}$F$_x$ was studied over a broad doping, temperature, and magnetic field range (x = 0 - 0.15, T up to 480K, H up to 30T) by means of As nuclear magnetic resonance (NMR). An enhanced spin-lattice relaxation rate divided by temperature, 1/T1T, in underdoped superconducting samples (x = 0.045, 0.05 and 0.075) suggests the presence of antiferromagnetic spin fluctuations, which are strongly reduced in optimally-doped (x = 0.10) and completely absent in overdoped (x = 0.15) samples. In contrast to previous analysis, Curie-Weiss fits are shown to be insufficient to describe the data over the whole temperature range. Instead, a BPP-type model is used to describe the occurrence of a peak in 1/T1T clearly above the superconducting transition, reflecting a progressive slowing down of the spin fluctuations down to the superconducting phase transition.
△ Less
Submitted 11 July, 2013;
originally announced July 2013.
-
Dynamics of a Heisenberg spin chain in the quantum critical regime: NMR experiment versus effective field theory
Authors:
H. Kuehne,
A. A. Zvyagin,
M. Guenther,
A. P. Reyes,
P. L. Kuhns,
M. M. Turnbull,
C. P. Landee,
H. -H. Klauss
Abstract:
A comprehensive comparison between the magnetic field- and temperature-dependent low frequency spin dynamics in the antiferromagnetic spin-1/2 Heisenberg chain (AFHC) system copper pyrazine dinitrate, probed via the 13C-nuclear magnetic resonance (NMR) relaxation rate 1/T1, and the field theoretical approach in the Luttinger liquid (LL) regime has been performed. We have found a very good agreemen…
▽ More
A comprehensive comparison between the magnetic field- and temperature-dependent low frequency spin dynamics in the antiferromagnetic spin-1/2 Heisenberg chain (AFHC) system copper pyrazine dinitrate, probed via the 13C-nuclear magnetic resonance (NMR) relaxation rate 1/T1, and the field theoretical approach in the Luttinger liquid (LL) regime has been performed. We have found a very good agreement between the experiment and theory in the investigated temperature and field range. Our results demonstrate how strongly the quantum critical point affects the spin dynamics of Heisenberg spin chain compounds.
△ Less
Submitted 23 July, 2010; v1 submitted 21 July, 2010;
originally announced July 2010.
-
Low frequency spin dynamics in the quantum magnet copper pyrazine dinitrate
Authors:
H. Kuehne,
M. Guenther,
S. Grossjohann,
W. Brenig,
F. J. Litterst,
A. P. Reyes,
P. L. Kuhns,
M. M. Turnbull,
C. P. Landee,
H. -H. Klauss
Abstract:
The S=1/2 antiferromagnetic Heisenberg chain exhibits a magnetic field driven quantum critical point. We study the low frequency spin dynamics in copper pyrazine dinitrate (CuPzN), a realization of this model system of quantum magnetism, by means of $^{13}$C-NMR spectroscopy. Measurements of the nuclear spin-lattice relaxation rate $T_1^{-1}$ in the vicinity of the saturation field are compared wi…
▽ More
The S=1/2 antiferromagnetic Heisenberg chain exhibits a magnetic field driven quantum critical point. We study the low frequency spin dynamics in copper pyrazine dinitrate (CuPzN), a realization of this model system of quantum magnetism, by means of $^{13}$C-NMR spectroscopy. Measurements of the nuclear spin-lattice relaxation rate $T_1^{-1}$ in the vicinity of the saturation field are compared with quantum Monte Carlo calculations of the dynamic structure factor. Both show a strong divergence of low energy excitations at temperatures in the quantum regime. The analysis of the anisotropic $T_1^{-1}$-rates and frequency shifts allows one to disentangle the contributions from transverse and longitudinal spin fluctuations for a selective study and to determine the transfer of delocalized spin moments from copper to the neighboring nitrogen atoms.
△ Less
Submitted 23 March, 2010;
originally announced March 2010.
-
Quantum critical dynamics of a S = 1/2 antiferromagnetic Heisenberg chain studied by 13C-NMR spectroscopy
Authors:
H. Kuehne,
H. -H. Klauss,
S. Grossjohann,
W. Brenig,
F. J. Litterst,
A. P. Reyes,
P. L. Kuhns,
M. M. Turnbull,
C. P. Landee
Abstract:
We present a 13C-NMR study of the magnetic field driven transition to complete polarization of the S=1/2 antiferromagnetic Heisenberg chain system copper pyrazine dinitrate Cu(C_4H_4N_2)(NO_3)_2 (CuPzN). The static local magnetization as well as the low-frequency spin dynamics, probed via the nuclear spin-lattice relaxation rate 1/T_1, were explored from the low to the high field limit and at te…
▽ More
We present a 13C-NMR study of the magnetic field driven transition to complete polarization of the S=1/2 antiferromagnetic Heisenberg chain system copper pyrazine dinitrate Cu(C_4H_4N_2)(NO_3)_2 (CuPzN). The static local magnetization as well as the low-frequency spin dynamics, probed via the nuclear spin-lattice relaxation rate 1/T_1, were explored from the low to the high field limit and at temperatures from the quantum regime (k_B T << J) up to the classical regime (k_B T >> J). The experimental data show very good agreement with quantum Monte Carlo calculations over the complete range of parameters investigated. Close to the critical field, as derived from static experiments, a pronounced maximum in 1/T_1 is found which we interpret as the finite-temperature manifestation of a diverging density of zero-energy magnetic excitations at the field-driven quantum critical point.
△ Less
Submitted 13 August, 2009; v1 submitted 14 April, 2008;
originally announced April 2008.
-
Dewetting of thin films on heterogeneous substrates: Pinning vs. coarsening
Authors:
Lutz Brusch,
Heiko Kuehne,
Uwe Thiele,
Markus Baer
Abstract:
We study a model for a thin liquid film dewetting from a periodic heterogeneous substrate (template). The amplitude and periodicity of a striped template heterogeneity necessary to obtain a stable periodic stripe pattern, i.e. pinning, are computed. This requires a stabilization of the longitudinal and transversal modes driving the typical coarsening dynamics during dewetting of a thin film on a…
▽ More
We study a model for a thin liquid film dewetting from a periodic heterogeneous substrate (template). The amplitude and periodicity of a striped template heterogeneity necessary to obtain a stable periodic stripe pattern, i.e. pinning, are computed. This requires a stabilization of the longitudinal and transversal modes driving the typical coarsening dynamics during dewetting of a thin film on a homogeneous substrate. If the heterogeneity has a larger spatial period than the critical dewetting mode, weak heterogeneities are sufficient for pinning. A large region of coexistence between coarsening dynamics and pinning is found.
△ Less
Submitted 10 December, 2001;
originally announced December 2001.
