Introduction to the Special Issue: Variable-Length Representation and Noncoding Segments for Evolutionary Algorithms

Abstract

The huge variation in the number and types of organisms that have evolved in nature is paralleled by an equally impressive diversity in the underlying genetic representation. Variation can be found, not only in the size of genomes, but also in the organization of information and the structure of the genomes. The result of this freedom of expression is that natural evolution is virtually unlimited in the types and complexity of the organisms that can be evolved. An entire new field has emerged in computer science to investigate the evolution of complexity and how it may be applied as problem-solving tools (Fogel, Owens, & Walsh, 1966; Holland, 1975; Koza, 1992; Rechenberg, 1973). Evolutionary computation systems traditionally employ a much more rigid representation of information than natural systems. Though these simpler, fixed representations may be easier to analyze, they also limit the scope and complexity of what may be evolved. We believe that it is time to seriously investigate the impact of more complex, variable representations on artificial evolutionary systems. The goal of this special issue is to examine the issues involved in extending the representational complexity of evolutionary algorithms (EA) and how this affects an EAs ability to evolve problem solutions. In particular, we focus on the ramifications of introducing variable-length representations and noncoding regions to EAs. A simple but significant difference between natural and computational evolutionary systems is the prevalence of variable-size representations in natural systems. In nature, genome sizes range from as little as lo5 base pairs (bp) for algae to as much as 10 bp for some plants and amphibians (Lewin, 1994). Even within phyla, the range of genome sizes varies from twofold in birds and mammals to more than tenfold for plants and amphibians. Variablesize representations lead naturally to the dynamic organization of information (genes) and to the appearance of noncoding or unexpressed segments in the genome. Noncoding DNA or DNA that does not directly contribute to the production of proteins is thought to make up upwards of 90% of some genomes (Bell, 1988; Nei, 1987). Thus, nature is able to vary both the amount and ratio of coding and noncoding regions in a genome, which, in turn, is thought to influence the shuffling or recombination of the coding regions (Gilbert, 1987). A brief review of the genetic process and a discussion of the different types of noncoding DNA is presented by Wu and Lindsay (1996b).