Masters Theses

Abstract

"Artificial Neural networks (ANNs) have increasingly been applied to different system identification, pattern recognition and function approximation problems. The predominantly used neural network architectures are the feedforward and feedback networks. Multilayer perceptrons and radial basis function networks are among the most commonly used feed-forward architecture neural networks and the Elman and Jordan networks are the most widely used recurrent neural networks. These networks are also called as second generation ANNs.

This thesis focuses on the third generation ANNs which are biologically inspired and operate on the same principles as the living neurons. These are also known as biologically inspired artificial neural networks (BIANNs). Spiking neural networks (SNNs) and polychronous spiking neural networks (PSNs) are among the mostly researched BIANNs. The advantage of BIANNs over traditional ANNs is their ability to scale up. But encoding and decoding information in and out of SNNs is a big challenge. This thesis addresses both the challenges.

Izhikevich's PSN model is used for the experimental simulation purposes in this thesis. Gaussian receptive field (GRF) encoding is used to convert real world data into spike streams which can be used by the SNNs. A novel decoding technique is proposed and implemented in this work which extracts the required information from the PSN spike outputs. The PSN has a lot of internal dynamics and when some information is processed by the PSN, the information gets encrypted along with the PSN's internal dynamics. The proposed decoding technique uses a leaky integrate & fire (LIP) module to accumulate and convert the PSN spikes to real world magnitude values. These values are then fed into a MLP decoder which maps the input to the target output and extracts the required encrypted information from the PSN spikes. The MLP is trained using particle swarm optimization (PSO). The performance of SNNs and the proposed decoding technique in time series estimation is tested on various benchmark time series functions and neuro-identification of system dynamics is tested on a two-area four-machine power system for identifying generator dynamics"--Abstract, page iii.

Advisor(s)

Venayagamoorthy, Ganesh K.

Committee Member(s)

Wunsch, Donald C.
Grant, Steven L.

Department(s)

Electrical and Computer Engineering

Degree Name

M.S. in Electrical Engineering

Sponsor(s)

National Science Foundation (U.S.)

Publisher

Missouri University of Science and Technology

Publication Date

Fall 2012

Pagination

ix, 68 pages

Note about bibliography

Includes bibliographical references (pages 64-67).

Rights

© 2012 Sinchan Roychowdhury, All rights reserved.

Document Type

Thesis - Restricted Access

File Type

text

Language

English

Library of Congress Subject Headings

Decoders (Electronics)
Neural networks (Computer science)
Nonlinear systems -- Automatic control
Synchronous data transmission systems

Thesis Number

T 9984

Print OCLC #

816364442

Electronic OCLC #

909373990

Link to Catalog Record

Electronic access to the full-text of this document is restricted to Missouri S&T users. Otherwise, request this publication directly from Missouri S&T Library or contact your local library.

http://laurel.lso.missouri.edu/record=b9391683~S5

Share

 
COinS