An empirical investigation into correlation functions in a spatially-dispersed evolutionary algorithm

Abstract

Spatially-structured populations play an important role in controlling selection pressure in evolutionary algorithms. The imposing space on a evolving population has traditionally been biased toward the underlying architecture that the population is executed on. The spatially dispersed evolutionary algorithm (sdEA) is an attempt to model population structures incorporating more probabilistic measures into the construction of demes. One important component of the sdEA in determining demes is the correlation function. This paper introduces three new correlation functions into the realm of the sdEA and compares their resultant behaviours on four differing test cases. Initial results indicate that the design of a correlation function should bias deme construction to small areas in the population surface.