Theme preservation and the evolution of representation

The identification of mechanisms by which constraints on phenotypic variability are tuned in nature, and the implementation of these mechanisms in Evolutionary Algorithms (EAs) carries the promise of making EAs less “wasteful”. The constraints on phenotypic variability are determined by the way genotypic variability maps to phenotypic variability. This in turn is determined by the way that phenotypes are represented genotypically. We use a formal model of an EA to show that when some part of the genome is mutated with a much lower probability than some other part, representations used to search the phenotype space - and hence the constraints on phenotypic variability - can themselves be thought to evolve. Specifically, we formally analyze a class of mutationonly fitness proportional evolutionary algorithms and show that these evolutionary algorithms implicitly implement what we call subrepresentation evolving multithreaded evolution. These EAs conduct second-order search over a ...

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Keki M. Burjorjee, Jordan B. Pollack
• Keywords: Evolutionary Algorithms, GECCO 2005, Phenotypic Variability, Way Genotypic Variability