Manela, Mauro; (1994) Contributions to the Theory and Applications of Genetic Algorithms. Doctoral thesis (Ph.D), UCL (University College London).

Text Contributions_to_the_theory_an.pdf Download (4MB)

Abstract

The thesis consists of a series of studies around the central theme of Genetic Algorithms (GAs), with particular emphasis on its application in the domain of distributed artificial intelligence (DAI). The theoretical part of the thesis starts from results derived by A. D. Bethke for binary alphabets, and from the suggestion that epistasis variance (borrowed from population genetics) is a reasonable measure to judge the suitability of a given representation for exploration by genetic search. It is suggested that these seemingly unrelated topics can be tied together with the help of abstract group theory. This provides a general framework within which further theoretical generalisations may be derived. Within this framework results derived for binary strings are extended to any non-binary alphabet. Also, issues of representation and different encodings are discussed and studied, and suggestions about how to modify simple genetic algorithms are reported. By means of a practical application to fermentation-process data, it is illustrated that a suitable change of representation can overcome the problem of dealing with search spaces that have non-simple boundaries determined by constraints. Also, it is demonstrated that optimisation over parameters that are not obviously suitable for "classical" treatments of optimisation can be conducted simultaneously with exploitation of the classical variables. The design of autonomous agents and a framework to discuss basic design issues within the scope of applicability of GAs are introduced. The usefulness of GAs as tools to help DAI designers is demonstrated by a systematic exploration of the domain of pursuit games, traditionally used as testbeds for comparing alternative architectures. A GA is used both to optimise low-level architectural features of agents and to help evaluate the quality of pursuit games as testbeds for DAI. As a consequence of this exploration, new features (e.g. "boredom", "listening rate") of potential value for general agent architectures are introduced.

Download activity - last month

Export as CSVXMLJSON

Download activity - last 12 months

Export as CSVJSONXML

Downloads by country - last 12 months

Export as XMLJSONCSV

Archive Staff Only

View Item