Selection pressure and performance in spatially distributed evolutionary algorithms

Recent studies of spatially distributed EAs have formally characterized the selection pressure induced by various selection strategies applied to local neighborhoods of various sizes and shapes. These analyses provide us with the ability to predict the expected behavior of the local neighborhood EAs. In this paper we empirically validate these predictions using the domain of function optimization. We demonstrate the various ways selection pressure can be varied in a spatially distributed EA and show that, from a performance point of view, no optimal selection pressure can be defined since it also depends on the fitness landscape of the problem being solved. Our results suggest that it may be possible to adaptively tune selection pressure by varying a single parameter, the neighborhood radius.