Optimal One-Max Strategy with Dynamic Island Models

Abstract

In this paper, we recall the dynamic island model concept, in order to dynamically select local search operators within a multi-operator genetic algorithm. We use a fully-connected island model, where each island is assigned to a local search operator. Selection of operators is simulated by migration steps, whose policies depend on a learning process. The efficiency of this approach is assessed in comparing, for the One-Max Problem, theoretical and ideal results to those obtained by the model. Experiments show that the model has the expected behavior and is able to regain the optimal local search strategy for this well-known problem.