Biological neurons use spikes to process and learn temporally dynamic inputs in an energy and com... more Biological neurons use spikes to process and learn temporally dynamic inputs in an energy and computationally efficient way. However, applying the state-of-the-art gradient-based supervised algorithms to spiking neural networks (SNN) is a challenge due to the non-differentiability of the activation function of spiking neurons. Employing surrogate gradients is one of the main solutions to overcome this challenge. Although SNNs naturally work in the temporal domain, recent studies have focused on developing SNNs to solve static image categorization tasks. In this paper, we employ a surrogate gradient descent learning algorithm to recognize twelve human hand gestures recorded by dynamic vision sensor (DVS) cameras. The proposed SNN could reach 97.2% recognition accuracy on test data.
This article presents a brief explanation of research interests and team description of MRL SPL t... more This article presents a brief explanation of research interests and team description of MRL SPL team intending to participate in RoboCup 2013 Standard Platform League. This team description includes various sub sections such as software structure, perception, global modeling and localization, behavior control and dynamic head motion, and locomotion control. Related published articles, future research topics, and research experiences are also provided here in this paper.
In the RoboCup Standard Platform League (SPL), NAO biped robots are used for all teams in competi... more In the RoboCup Standard Platform League (SPL), NAO biped robots are used for all teams in competitions. The robots have two on-board directional cameras and should perform fully autonomous, which requires precise data. The debugging tools always play critical role in developing reliable algorithms and calibrating sensors. In this paper we present a debugger and visualizer for vision of a standard platform league robot. This tool can be utilized for running off-line image processing algorithms aside calibrating the vision parameters like camera offsets and color lookup table. It also provides a very simple connection manager for transferring data with multiple robots and simulator.
Biological neurons use spikes to process and learn temporally dynamic inputs in an energy and com... more Biological neurons use spikes to process and learn temporally dynamic inputs in an energy and computationally efficient way. However, applying the state-of-the-art gradient-based supervised algorithms to spiking neural networks (SNN) is a challenge due to the non-differentiability of the activation function of spiking neurons. Employing surrogate gradients is one of the main solutions to overcome this challenge. Although SNNs naturally work in the temporal domain, recent studies have focused on developing SNNs to solve static image categorization tasks. In this paper, we employ a surrogate gradient descent learning algorithm to recognize twelve human hand gestures recorded by dynamic vision sensor (DVS) cameras. The proposed SNN could reach 97.2% recognition accuracy on test data.
This article presents a brief explanation of research interests and team description of MRL SPL t... more This article presents a brief explanation of research interests and team description of MRL SPL team intending to participate in RoboCup 2013 Standard Platform League. This team description includes various sub sections such as software structure, perception, global modeling and localization, behavior control and dynamic head motion, and locomotion control. Related published articles, future research topics, and research experiences are also provided here in this paper.
In the RoboCup Standard Platform League (SPL), NAO biped robots are used for all teams in competi... more In the RoboCup Standard Platform League (SPL), NAO biped robots are used for all teams in competitions. The robots have two on-board directional cameras and should perform fully autonomous, which requires precise data. The debugging tools always play critical role in developing reliable algorithms and calibrating sensors. In this paper we present a debugger and visualizer for vision of a standard platform league robot. This tool can be utilized for running off-line image processing algorithms aside calibrating the vision parameters like camera offsets and color lookup table. It also provides a very simple connection manager for transferring data with multiple robots and simulator.
Uploads
Papers by Aref Moqadam Mehr