The Effects of Elitism on Spatial Coevolutionary GAs

Elitism is a common concept in the standard Genetic Algorithm (GA), but one that has received little to no research when considered on spatial coevolutionary GAs. This paper attempts to address this by implementing and analyzing elitism mechanisms within a spatial coevolutionary system. To this end we develop a new globally selected local elitism and compare it with the more basic strictly non-local elitism generally seen in standard GAs. To aid in this analysis we also create a new simple function, one-max matching, which incorporates the simplicity of one-max yet is highly multimodal. We find that elitism seems to have a massive positive effect on coevolutionary systems, Heat maps show that the globally selected local elite often cluster within in the spatial network. Finally, surprisingly it took two elite per population in a coevolutionary system to see any benefit from elitism; we don’t know why.