kamyabkm

Building-a-computational-model-of-the-visual-system-to-investigate-temporal-harmonics-in-modulated-v

Peripheral vision is an important aspect when it comes to studying vision as it makes up the majority of the visual field. Therefore when one part of the visual system such as the central vision is impaired, it is crucial to be able to use different techniques to use the remaining of the visual field for all daily tasks. Disorders such as age-related macular degeneration suffer from losing central vision, hence, these patients evolve new eye movements and strategies to use the remaining of the visual field for different tasks. Meanwhile, visual scientists attempt to develop new methods to further improve the use of the remaining vision in visual disorders. Some of these approaches include manipulating visual targets’ dynamics by introducing temporal modulations that can create temporal harmonics to widen the spatio-temporal range of the stimuli and improve peripheral acuity. In this study we attempt to build a model of the visual system in order to investigate whether temporal harmonics are indeed involved in improving peripheral acuity in two specific approaches; subsampling the stimuli and contrast polarity reversal of the stimuli. We also made comparisons between the model estimations and human data to draw further interpretations about temporal harmonics’ effects on vision. We found that the model replicates certain aspects of the visual system such as the threshold of spatial and temporal frequencies and visibility is better in unmodulated stimuli. The model is a suitable foundation that can be expanded on by future experimentation and to reach a more complete representation of the visual system.

MLiS2

Playing-Blackjack-with-Reinforcement-Learning

We have employed several techniques as Monte Carlo Exploration, Q-Learning, DeepQ-Learning or Policy Gradient in order to solve two proposed versions of the Blackjack game. Defining and constructing the environments according to a particular set of rules has been part of the project too.

Processing-of-Sounds-by-Population-Spikes

The primary auditory cortex plays a crucial role in localising and processing complex sounds. The tonotopic organisation of A1 is mapped such that neurons with similar frequencies are oriented in separate iso-frequency bands (i.e., tonotopic axis). In this study, we replicate a model whereby each iso-frequency column is described by a recurrent neural network with short-term synaptic depression. In particular, this network generates population spikes, in which a group of neurons synchronously fire for a short time period. To mimic the tonotopic organisation of A1, frequency bands are inter-connected where population spikes will disseminate from one column to neighbouring columns depending on the input’s characteristics. Consequently, incoming sounds are processed through sequences of population spikes that contain distinct information about simple and complex tones. Furthermore, we demonstrate the effect of noisy sensory input to the propagation and response of population spikes. We replicate frequency tuning curves of iso-freqeuncy columns, which rely on the strength of intracortical inhibitory and excitatory connections. We also show that consecutive two-tone stimuli exhibit forward masking which are contingent on spatial and temporal properties of the inputs, alongside PS-based encoding scheme for processing complex sounds (i.e., different amplitude and frequencies).

Programming-an-Autonomous-Driving-Car

This repository consists of examples of codes used to programme an autonomous driving car on a Raspberry Pi for a Machine Learning assessment. It also includes the report and the presentation as PDF files with examples of the datasets used.

YOLO-convert-txt-2-xml

darknet txt to darkflow xml

kamyabkm/Building-a-computational-model-of-the-visual-system-to-investigate-temporal-harmonics-in-modulated-v

Peripheral vision is an important aspect when it comes to studying vision as it makes up the majority of the visual field. Therefore when one part of the visual system such as the central vision is impaired, it is crucial to be able to use different techniques to use the remaining of the visual field for all daily tasks. Disorders such as age-related macular degeneration suffer from losing central vision, hence, these patients evolve new eye movements and strategies to use the remaining of the visual field for different tasks. Meanwhile, visual scientists attempt to develop new methods to further improve the use of the remaining vision in visual disorders. Some of these approaches include manipulating visual targets’ dynamics by introducing temporal modulations that can create temporal harmonics to widen the spatio-temporal range of the stimuli and improve peripheral acuity. In this study we attempt to build a model of the visual system in order to investigate whether temporal harmonics are indeed involved in improving peripheral acuity in two specific approaches; subsampling the stimuli and contrast polarity reversal of the stimuli. We also made comparisons between the model estimations and human data to draw further interpretations about temporal harmonics’ effects on vision. We found that the model replicates certain aspects of the visual system such as the threshold of spatial and temporal frequencies and visibility is better in unmodulated stimuli. The model is a suitable foundation that can be expanded on by future experimentation and to reach a more complete representation of the visual system.

kamyabkm/MLiS2

kamyabkm/Playing-Blackjack-with-Reinforcement-Learning

We have employed several techniques as Monte Carlo Exploration, Q-Learning, DeepQ-Learning or Policy Gradient in order to solve two proposed versions of the Blackjack game. Defining and constructing the environments according to a particular set of rules has been part of the project too.

kamyabkm/Processing-of-Sounds-by-Population-Spikes

The primary auditory cortex plays a crucial role in localising and processing complex sounds. The tonotopic organisation of A1 is mapped such that neurons with similar frequencies are oriented in separate iso-frequency bands (i.e., tonotopic axis). In this study, we replicate a model whereby each iso-frequency column is described by a recurrent neural network with short-term synaptic depression. In particular, this network generates population spikes, in which a group of neurons synchronously fire for a short time period. To mimic the tonotopic organisation of A1, frequency bands are inter-connected where population spikes will disseminate from one column to neighbouring columns depending on the input’s characteristics. Consequently, incoming sounds are processed through sequences of population spikes that contain distinct information about simple and complex tones. Furthermore, we demonstrate the effect of noisy sensory input to the propagation and response of population spikes. We replicate frequency tuning curves of iso-freqeuncy columns, which rely on the strength of intracortical inhibitory and excitatory connections. We also show that consecutive two-tone stimuli exhibit forward masking which are contingent on spatial and temporal properties of the inputs, alongside PS-based encoding scheme for processing complex sounds (i.e., different amplitude and frequencies).

kamyabkm/Programming-an-Autonomous-Driving-Car

This repository consists of examples of codes used to programme an autonomous driving car on a Raspberry Pi for a Machine Learning assessment. It also includes the report and the presentation as PDF files with examples of the datasets used.

kamyabkm/YOLO-convert-txt-2-xml

darknet txt to darkflow xml