Publication Date
2009
Document Type
Thesis
Committee Members
John Gallagher (Advisor), John Gallagher (Committee Chair), Michael Raymer (Committee Member), Mateen Rizki (Committee Member)
Degree Name
Master of Science in Computer Engineering (MSCE)
Abstract
Evolvable Hardware is an emerging sub-field of evolutionary computation in which evolutionary algorithms are employed to create designs for hardware devices. Recent work has combined continuous time recurrent neural networks with the Mini Population (Minipop) Evolutionary Algorithm to create self-configuring device controllers. Standard Minipop eschews recombination operators due to the belief that they increase the size of an algorithm's on chip implementation without adding significant search power for finding neural network controllers. The focus of this thesis is to challenge that thinking by testing a number of hardware efficient recombination operators against two benchmark problems. We consider variants that recombine at neuron parameter and whole neuron boundaries taking advantage of easily measured neuron output correlation information. Although we conclude that there is no compelling evidence to adopt any of these variants at this time, we have identified interesting opportunities that might be exploited in the future to improve Minipop search over spaces of neurodynamic systems
Page Count
94
Department or Program
Department of Computer Science and Engineering
Year Degree Awarded
2009
Copyright
Copyright 2009, all rights reserved. This open access ETD is published by Wright State University and OhioLINK.
