Abstract

"Evolutionary Algorithms (EAs) are a class of algorithms inspired by biological evolution. EAs are applicable to a wide range of problems; however, there are a number of parameters to set in order to use an EA. The performance of an EA is extremely sensitive to these parameter values; setting these parameters often requires expert knowledge of EAs. This prevents EAs from being more widely adopted by nonexperts. Parameter control, the automation of dynamic parameter value selection, has the potential to not only alleviate the burden of parameter tuning, but also to improve performance of EAs on a variety of problem classes in comparison to employing fixed parameter values. The science of parameter control in EAs is, however, still in its infancy and most published work in this area has concentrated on just a subset of the standard parameters. In particular, the control of offspring size has so far received very little attention, despite its importance for balancing exploration and exploitation. This thesis introduces three novel methods for controlling offspring size: Self- Adaptive Offspring Sizing (SAOS), Futility-Based Offspring Sizing (FuBOS), and Diversity-Guided Futility-Based Offspring Sizing (DiGFuBOS). EAs employing these methods are compared to each other and a highly tuned, fixed offspring size EA on a wide range of test problems. It is shown that an EA employing FuBOS or DiGFuBOS performs on par with the highly tuned, fixed offspring size EA on many complex problem instances, while being far more efficient in terms of fitness evaluations. Furthermore, DiGFuBOS does not introduce any new user parameters, thus truly alleviating the burden of tuning the offspring size parameter in EAs"--Abstract, page iii.

Advisor(s)

Tauritz, Daniel R.

Committee Member(s)

Frank, Ronald L.
Wilkerson, Ralph W.

Department(s)

Computer Science

Degree Name

M.S. in Computer Science

Publisher

Missouri University of Science and Technology

Publication Date

Summer 2009

Pagination

ix, 52 pages

Rights

© 2009 André Chidi Nwamba, All rights reserved.

Document Type

Thesis - Open Access

File Type

text

Language

English

Library of Congress Subject Headings

Evolutionary computation
Evolutionary programming (Computer science)
Parameter estimation -- Mathematical models

Thesis Number

T 9544

Print OCLC #

471874637

Electronic OCLC #

429917135

Share

 
COinS