Neutrality, Robustness, and Evolvability in Genetic Programming

Abstract

Redundant mapping from genotype to phenotype is common in evolutionary algorithms, especially in genetic programming (GP). Such a redundancy can lead to neutrality, where mutations to a genotype may not alter its phenotypic outcome. The effects of neutrality can be better understood by quantitatively analysing its two observed properties, i.e., robustness and evolvability. In this study, we examine a compact Linear GP algorithm, characterize its entire genotype, phenotype, and fitness networks, and quantitatively measure robustness and evolvability at the genotypic, phenotypic, and fitness levels. We investigate the relationship of robustness and evolvability at those different levels. We use an ensemble of random walks and hill climbs to study how robustness and evolvability and the structure of genotypic, phenotypic, and fitness networks influence the evolutionary search process.