On the Locality of Representations

It is well known that using high-locality representations is important for efficient evolutionary search. This paper discusses how the locality of a representation influences the difficulty of a problem when using mutation-based search approaches. The results show that high-locality representations do not change problem difficulty. In contrast, low-locality representations randomize the search process and make problems that are phenotypically easy for mutation-based search more difficult and phenotypically difficult problems more easy. 1 Metrics and Locality When considering representations it must be distinguished between phenotypes xp and genotypes xg. Thus, an optimization problem can be decomposed into two parts. The first maps the genotypic space Φg to the phenotypic space Φp, and the second maps Φp to the fitness space R: fg(xg) : Φg → Φp, fp(xp) : Φp → R, where the overall optimization problem is defined as f = fp ◦ fg = fp(fg(xg)). The genotype-phenotype mapping...

• Type: Conference
• Year: 2003
• Where: GECCO
• Authors: Franz Rothlauf
• Keywords: GECCO 2003, Mutation-based Search, Mutation-based Search Approaches, Optimization Problem