The Effectiveness of the Simplicity in Evolutionary Computation

摘要

Current research in Evolutionary Computation concentrates on proposing more and more sophisticated methods that are supposed to be more effective than their predecessors. New mechanisms, like linkage learning (LL) that improve the overall method effectiveness, are also proposed. These research directions are promising and lead to effectiveness increase that cannot be questioned. Nevertheless, in this paper, we concentrate on a situation in which the simplification of the method leads to the improvement of its effectiveness. We show situations when primitive methods, like Random Search (RS) combined with local search, can compete with highly sophisticated and highly effective methods. The presented results were obtained for an up-to-date, practical, NP-complete problem, namely the Routing and Spectrum Allocation of Multicast and Unicast Flows (RSA/MU) in Elastic Optical Networks (EONs). None of the considered test cases is trivial. The number of solutions possible to encode by an evolutionary method is large.