A study of evolutionary robustness in stochastically tiled polyominos

Given an evolutionary optimization problem with many possible genotypes for each phenotype this study investigates if the evolved genes for a given phenotype are more robust to point mutation than randomly sampled genes for the same phenotype. This question is addressed using a cellular representation for polyominos in the plane. The evolutionary computation system optimizes for shapes which pack well onto the surface of a torus when dropped at random. For the majority of the evolved phenotypes the evolved genes for a given shape proved to be significantly more robust to point mutation than those sampled at random for that same shape. A few evolved genotypes, however, were not significantly more robust than those sampled at random and in some cases were less robust. These observations are placed in the context of the fitness landscape for the representation. Categories and Subject Descriptors I.2 [Artificial Intelligence]: Miscellaneous General Terms Design Keywords Cellular Repres...

• Type: Conference
• Year: 2005
• Where: GECCO
• Authors: Justin Schonfeld, Daniel A. Ashlock
• Keywords: Cellular Representation, Evolved Genes, GECCO 2005, Point Mutation