Date of Award

7-7-2020

Publication Type

Master Thesis

Degree Name

M.A.Sc.

Department

Electrical and Computer Engineering

First Advisor

Majid Ahmadi

Keywords

Pattern Recognition, Spiking Neural Networks

Rights

info:eu-repo/semantics/openAccess

Creative Commons License

Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License
This work is licensed under a Creative Commons Attribution-Noncommercial-No Derivative Works 4.0 License.

Abstract

Deep learning believed to be a promising approach for solving specific problems in the field of artificial intelligence whenever a large amount of data and computation is available. However, tasks that require immediate yet robust decisions in the presence of small data are not suited for such an approach. The superior performance of the human brain in specific tasks like pattern recognition in comparison to traditional neural networks convinced neuroscientists to introduce a biologically plausible model of the neuron, which is known as spiking neurons. In opposition to conventional neuron, spiking neurons use a short electrical pulse known as a spike to transfer the information. The complexity and dynamic of these neurons allow them to perform complex computational tasks. However, training a spiking neural network does not follow the rule of conventional ANN, and we need to devise new methods of training that are compatible with the unsupervised nature of these networks. This thesis aims to investigate the unsupervised approaches of training spiking networks using spike time-dependent plasticity (STDP) and assess their performance on real-world machine learning applications like handwritten digit recognition.

Share

COinS