Evolutionary algorithms using a neural network like migration scheme

Abstract

We introduce a multiple subpopulation approach for parallel evolutionary algorithms the migration scheme of which follows a neural network learning like dynamic. It is adapted from the approach of collective learning in self-organizing maps with a more and more separation during time. We succesfully apply this approach to clustering real world data in psychotherapy research and VLSI-design. The advantages of the approach are shown which consist in a reduced communication overhead between the subpopulations preserving a non-vanishing information flow and an improved convergence rate resulting in decreasing computational costs.