Über uns
I'm a technical co-founder of Emotech & hold a PhD in Computer Science from UCL. My…
Activity
-
Many major #cloud providers have Generative AI offerings today. But what makes #Oracle best suited to meet the Generative AI and AI…
Many major #cloud providers have Generative AI offerings today. But what makes #Oracle best suited to meet the Generative AI and AI…
Liked by Jan Medvesek, Ph.D.
-
As AI continues to transform industries, Oracle is innovating by offering its customers the most powerful and flexible AI infrastructure available in…
As AI continues to transform industries, Oracle is innovating by offering its customers the most powerful and flexible AI infrastructure available in…
Liked by Jan Medvesek, Ph.D.
-
Today, I had the privilege of speaking at the IoT & AI Challenge in Dubai about how generative AI can transform government operations. I emphasized…
Today, I had the privilege of speaking at the IoT & AI Challenge in Dubai about how generative AI can transform government operations. I emphasized…
Liked by Jan Medvesek, Ph.D.
Erleben Sie
Bildung
Publications
-
Low-overhead Gaussian-Process Training for Indoor Positioning Systems
IPIN2015 - Indoor Positioning and Indoor Navigation; Banff, Alberta, Canada
Despite considerable research efforts over the past decades, indoor localization remains a difficult problem. Although inertial dead reckoning approaches yield accurate results over short periods, their position error typically scales cubically with time. Alternate methods of localization often use radio measurements between mobile and fixed points, either through fingerprinting or range-based methods. However, the accuracy of such approaches is inherently limited by the complex non-linear…
Despite considerable research efforts over the past decades, indoor localization remains a difficult problem. Although inertial dead reckoning approaches yield accurate results over short periods, their position error typically scales cubically with time. Alternate methods of localization often use radio measurements between mobile and fixed points, either through fingerprinting or range-based methods. However, the accuracy of such approaches is inherently limited by the complex non-linear effect of clutter and noise on signal propagation. Consequently, substantial training overhead is often required to characterize the spatial evolution of these signal patterns to improve accuracy. This work presents a low-overhead, practical method for characterizing non-linear radio signal propagation for subsequent use in indoor positioning and navigation systems. Specifically, we propose an inertial dead-reckoning based method for building Gaussian process maps that describe the predicted error distributions of a given signal as it varies across space. We further analyze the accuracy of these maps and position estimate techniques based thereon as a function of the inertial training data and the number of known correction points required.
Other authors -
Gaussian process inference approximation for indoor pedestrian localisation
Electronics Letters (Volume:51 , Issue: 5 )
Clutter has a complex effect on radio propagation, and limits the effectiveness of deterministic methods in wireless indoor positioning. In contrast, a Gaussian process (GP) can be used to learn the spatially correlated measurement error directly from training samples, and build a representation from which a position can be inferred. A method of exploiting GP inference to obtain measurement predictions from within a pose graph optimisation framework is presented. However, GP inference has a…
Clutter has a complex effect on radio propagation, and limits the effectiveness of deterministic methods in wireless indoor positioning. In contrast, a Gaussian process (GP) can be used to learn the spatially correlated measurement error directly from training samples, and build a representation from which a position can be inferred. A method of exploiting GP inference to obtain measurement predictions from within a pose graph optimisation framework is presented. However, GP inference has a run-time complexity of O(N3) in the number of training samples N, which precludes it from being called in each optimiser iteration. The novel contributions of this work are a method for building an approximate GP inference map and an O(1) bi-cubic interpolation strategy for sampling this map during optimisation. Using inertial, magnetic, signal strength and time-of-flight measurements between four anchors and a single mobile sensor, it is shown empirically that the presented approach leads to decimetre precision indoor pedestrian localisation.
Other authorsSee publication -
Supporting situated STEM learning - TouchDevelop Integration of the UCL Engduino over Bluetooth
PROMOTO '14 Proceedings of the 2nd Workshop on Programming for Mobile & Touch
Teaching programming is fast becoming a fundamental learning practice in schools for Science, Technology, Engineering and Maths (STEM) subjects. Through interactive learning with tangible devices such as Lego Mindstorms, students and teachers can explore ways to collect data, analyse and illustrate core principles in STEM subjects. The majority of such educational programmable devices require the use of a PC to program them. This short paper reports on a development project that demonstrates a…
Teaching programming is fast becoming a fundamental learning practice in schools for Science, Technology, Engineering and Maths (STEM) subjects. Through interactive learning with tangible devices such as Lego Mindstorms, students and teachers can explore ways to collect data, analyse and illustrate core principles in STEM subjects. The majority of such educational programmable devices require the use of a PC to program them. This short paper reports on a development project that demonstrates a STEM situated and wireless method to enable students to learn pro-gramming. It allows students and teachers to create STEM exper-iments though only a Windows phone with TouchDevelop and a programmable device solution. This focuses their Constructionist learning applied to STEM learning when situated with program-ming capabilities via a Bluetooth based library to operate the UCL Engduino, a customised teaching apparatus based on the Arduino platform. We demonstrate a process for integrating teaching devices with TouchDevelop to expand on pedagogical techniques for both student programmers and educators in programming
Other authorsSee publication
Projects
-
ITS - Indoor Tracking System
During my PhD research work I have developed Indoor Tracking System (ITS) platform. Including hardware design, embedded systems, data gathering systems and research analyses. The main contribution was incorporation of machine learning (ML) techniques into a non-linear least square solver (NLLS), that solves a localisation problem. The problem is solved iteratively with fusing together many sensor sources and sensor models. The post-processing method archived better than half a meter accuracy in…
During my PhD research work I have developed Indoor Tracking System (ITS) platform. Including hardware design, embedded systems, data gathering systems and research analyses. The main contribution was incorporation of machine learning (ML) techniques into a non-linear least square solver (NLLS), that solves a localisation problem. The problem is solved iteratively with fusing together many sensor sources and sensor models. The post-processing method archived better than half a meter accuracy in cluttered indoor environments. Gaussian process ML method was used to learn spatial evolution of signal patterns and Google's Ceres solver was used to solve pose-graph optimization problem.
-
Android applications
Android applications on Google Play:
Photo Album:
The Globus Photo Album is an innovative presentation of photo albums. It is designed to make your everyday photo experience one of a kind. It includes a connection of your albums on Facebook. It supports downloading, creating or editing photo albums from Facebook.
Keys: Facebook API, OpenGL, Sqlite
https://play.google.com/store/apps/details?id=com.jinj.globusphotoviewer
Code optimizations:
The application is…Android applications on Google Play:
Photo Album:
The Globus Photo Album is an innovative presentation of photo albums. It is designed to make your everyday photo experience one of a kind. It includes a connection of your albums on Facebook. It supports downloading, creating or editing photo albums from Facebook.
Keys: Facebook API, OpenGL, Sqlite
https://play.google.com/store/apps/details?id=com.jinj.globusphotoviewer
Code optimizations:
The application is essentially a presentation of code optimizations. Pure JAVA code is not always the best choice, for example, Vectorized NEON code runs almost 100 times faster than Java code.Five different types of code are implemented in the application.
Keys: NDK, NEON, Vectorization, Optimization
https://play.google.com/store/apps/details?id=com.jinj.electricalfields
Simple game:
The game is nothing special, expect that it was in the top 3 downloads in our country after publishing. A great idea at the right moment.
https://play.google.com/store/apps/details?id=com.janm.volitvegame
Honors & Awards
-
Techcrunch Disrupt London 2015
Techcrunch
We were selected as on of the 14 teams to compete at Techcrunch Disrupt London event out of 500 startups applied.
-
Microsoft Indoor Localization Competition - IPSN 2015
Microsoft Research
We attended the Microsoft Indoor Competition in Seattle and won 7th place in our category, with 12th place overall out of 24 competing teams (48 teams applied initially).
-
AD futura scholarship for PhD research abroad in 2014/2015
AD Futura - The Slovene Human Resources and Scholarship Fund
-
AD futura scholarship for PhD research abroad in 2013/2014
AD Futura - The Slovene Human Resources and Scholarship Fund
September 2013
-
AD futura scholarship for PhD research abroad in 2012/2013
AD Futura - The Slovene Human Resources and Scholarship Fund
September 2012
Languages
-
Englisch
-
-
Slovenian
-
More activity by Jan
-
Join us at Oracle #CloudWorld 2024 for the latest announcements, real-world lessons, and expert insights on AI. Our Chairman and CTO Larry Ellison…
Join us at Oracle #CloudWorld 2024 for the latest announcements, real-world lessons, and expert insights on AI. Our Chairman and CTO Larry Ellison…
Liked by Jan Medvesek, Ph.D.
-
As part of our 6th anniversary celebrations, we held a fireside chat featuring our Group CEO, Peng Xiao, and our Group CTO, Kiril Evtimov yesterday.…
As part of our 6th anniversary celebrations, we held a fireside chat featuring our Group CEO, Peng Xiao, and our Group CTO, Kiril Evtimov yesterday.…
Liked by Jan Medvesek, Ph.D.
-
It's truly humbling to see the incredible adoption of Kumo.AI across various AI use cases! With Kumo.AI, building and operationalizing AI models…
It's truly humbling to see the incredible adoption of Kumo.AI across various AI use cases! With Kumo.AI, building and operationalizing AI models…
Liked by Jan Medvesek, Ph.D.
Other similar profiles
Gemeinsame Artikel erkunden
We’re unlocking community knowledge in a new way. Experts add insights directly into each article, started with the help of AI.
Explore More