Methods for Deep Transfer Learning and Knowledge Transfer in the NeuCube Brain-Inspired Spiking Neural Network

Date
2021
Authors
Tan, Yongyao
Supervisor
Kasabov, Nikola
Doborjeh, Maryam
Item type
Thesis
Degree name
Master of Computer and Information Sciences
Journal Title
Journal ISSN
Volume Title
Publisher
Auckland University of Technology
Abstract

With the increasing number of computational systems based on continuous streams of information, progressively learning and accommodating new knowledge in a more efficient manner becomes a long-standing challenge. This thesis proposes methods employing a Brain-Inspired Spiking Neural Network (BI-SNN) architecture for transfer learning scenarios. The proposed transfer learning approaches were experimentally validated using a benchmark brain data related to upper limb movement. The results showed that the proposed methods have the capability to effectively learn new knowledge by retaining and reusing previously learned knowledge, resulting in a better accuracy of classification (up to 88.89%) when compared with non-transfer learning methods. Further, a new deep knowledge representation approach is proposed and developed, which allows extracting spatial temporal rules from deep knowledge, enabling a better interpretation of learning patterns in the SNN models and evolution trace of knowledge during transfer learning.

Description
Keywords
Incremental learning , Transfer learning , Spatio-temporal EEG data , Deep knowledge representation , Spiking neural networks , Explainable AI , Human movements
Source
DOI
Publisher's version
Rights statement
Collections