Optimal Population Size in Island Model Genetic Algorithms

American Proverb: Two’s company, three’s a crowd. Genetic Algorithm Proverb: One’s a hill climber and a thousand’s random search. Population size is one of the key parameters affecting the success of genetic algorithms (GAs). Assuming a limited number of fitness evaluations (the most time-intensive factor in virtually all optimization problems), there exists an optimal population size for a genetic algorithm for a given application. Intuitively, a GA with population size one is a hill climber and a GA with maximal population size performs random search. Somewhere in between lies the sweet spot. The Island Model GA divides a single population into semi-isolated subpopulations connected by migration. On the extreme of high migration, the subpopulations function as a single large population. On the extreme of no migration, the subpopulations mights as well be independent runs of smaller population size GAs. Somewhere in between lies the sweet spot. In this paper we propose to explore the dynamics of optimal population size as a function of migration in island model GAs.