Continuous Mimetic Evolution

Abstract

There exists no memory of biologic evolution besides the individuals themselves. Indeed, the biologic milieu can change and a previously unfit action or individual can come to be more fit; it would be most dangerous to rely on the memory of the past. This contrasts with artificial evolution most often considering a fixed milieu: the generation of an unfit individual previously explored is only a waste of time. This paper aims at constructing a memory of evolution, and using it to avoid such fruitless explorations. A new evolution scheme, called mimetic evolution , gradually constructs two models along evolution, respectively memorizing the best and the worst individuals of the past generations. Standard crossover and mutation are replaced by mimetic mutation : individuals are attracted or repelled by these models. Mimetic evolution is extended from binary to continuous search spaces. Results of experiments on large-sized problems are detailed and discussed.