On Meme Self-Adaptation in Spatially-Structured Multimemetic Algorithms

Abstract

Multimemetic algorithms (MMAs) are memetic algorithms that explicitly exploit the evolution of memes, i.e., non-genetic expressions of problem-solving strategies. We consider a class of MMAs in which these memes are rewriting rules whose length can be fixed during the run of the algorithm or self-adapt during the search process. We analyze this self-adaptation in the context of spatially-structured MMAs, namely MMAs in which the population is endowed with a certain topology to which interactions (from the point of view of selection and variation operators) are constrained. For the problems considered, it is shown that panmictic (i.e., non-structured) MMAs are more sensitive to this self-adaptation, and that using variable-length memes seems to be a robust strategy throughout different population structures.