Diploidy and Dominance in Artificial Genetic Search
Robert E. Smith
Department of Engineering Mechanics, University of Alabama,
Tuscaloosa, Alabama 35487, USA
David E. Goldberg
Department of General Engineering,
University of Illinois at Urbana-Champaign,
Urbana, Illinois 61801, USA
Abstract
Genetic algorithms (GAs) continue to receive increased attention as general-purpose methods for search and optimization. GAs are attractive for search in complex spaces where the functional relationships between parameters and objective function values are of unknown, arbitrary mathematical character. Despite their generally robust character, typical GAs are known to fare poorly on functions that vary with time, where the goal is to track nonstationary optima. It has been theorized that in natural genetics, diploidy and dominance increase the survivability of species in environments that vary with time. This paper examines the effects of diploid representations and dominance operators in genetic algorithms applied to nonstationary search problems. Experimental results indicate that these additions greatly increase the efficacy of GAs in time-varying environments. This increased performance is made possible by abeyant recessive alleles. Analytical arguments show that these recessive alleles increase population diversity without the disruptive effects of high mutation rates, thus allowing the GA to renew its search process as the problem varies with time. Analysis also reveals that abeyant recessives are sensitive to past environmental conditions, and can therefore act as a form of distributed, probabilistic memory of environmental conditions that occur periodically. Final sections discuss extensions and implications of this work, including multi-locus dominance under genic control, intrachromosomal dominance, and how diploidy may affect the inherently noisy GA search process, if this noise is viewed as a nonstationary aspect of the objective function.
