An Analysis of Migration Strategies in Island-Based Multimemetic Algorithms

Multimemetic algorithms (MMAs) are memetic algorithms that explicitly represent and evolve memes (computational representations of problem solving methods) as a part of solutions. We consider an island-based model of MMAs and provide a comparative analysis of six migrant selection strategies and two migrant replacement operators. We use a test suite of four hard pseudoboolean functions to examine qualitative behavioral differences at the genetic and memetic level, and provide a sound statistical analysis of performance. The results indicate the choice of migrant selection operator is more important than that of migrant replacement, and that policies based on fitness or pure genetic diversity do not compare favorably to more holistic strategies.