Date of Award
9-27-2023
Publication Type
Thesis
Degree Name
M.A.Sc.
Department
Electrical and Computer Engineering
Keywords
Artificial Intelligence;Artificial Neural Networks;Brain-Inspired;Machine Learning;Reinforcement Learning;Spiking Neural Networks
Supervisor
Majid Ahmadi
Supervisor
Arash Ahmadi
Rights
info:eu-repo/semantics/openAccess
Creative Commons License
This work is licensed under a Creative Commons Attribution 4.0 International License.
Abstract
Over the past two decades, Spiking Neural Networks (SNNs), as the third generation of Artificial Neural Networks, have gained widespread use due to their ability to closely mimic human brain neuron behavior. Being spike-based and resembling biological neurons, SNNs have demonstrated superior energy efficiency compared to conventional counterparts. However, to fully harness their potential, effective training methods are essential. Despite significant progress in the field of biological implementation of SNNs, there remains a need for research to identify the most optimal learning approach. To address this, we employed Policy-based stochastic reinforcement learning to train a spiking neural network using Izhikevich neurons. Our designed network was evaluated on the Freeway ATARI game provided by OpenAI gym. This research delves into the application of RL techniques to Spiking Neural Networks, offering promising results and demonstrating their potential for real-time applications. While the outcomes show that SNNs exhibit promise, achieving the highest accuracy requires further investigation when compared to traditional Artificial Neural Networks (ANNs). Nonetheless, the findings highlight the potential of SNNs in various domains and motivate continued exploration to enhance their performance and utilization.
Recommended Citation
Zamani, Seyede Narjes, "A Reinforcement Learning Algorithm for Training a Spiking Neural Network Agent" (2023). Electronic Theses and Dissertations. 9276.
https://scholar.uwindsor.ca/etd/9276