Migration of Probability Models Instead of Individuals: An Alternative When Applying the Island Model to EDAs

In this work we experiment with the application of island models to Estimation of Distribution Algorithms (EDAs) in the field of combinatorial optimization. This study is motivated by the success obtained by these models when applied to other meta-heuristics (such as genetic algorithms, simulated annealing or VNS) and by the use of a compact representation of the population that make EDAs through probability distributions. This fact can be exploited during information interchange among islands. Thus, besides discussing the performance of different topologies, in this work we focus our research into three different ways of performing migration: individuals, models and models weighted by fitness. The proposed algorithms are tested by using four well known combinatorial optimization problems as benchmarks.