The Structure of Evolutionary Exploration: On Crossover, Buildings Blocks, and Estimation-Of-Distribution Algorithms

Correlations between alleles after selection are an important source of information. Such correlations should be exploited for further search and thereby constitute the building blocks of evolutionary exploration. With this background we analyze the structure of the offspring probability distribution, or exploration distribution, for a simple GA with mutation only and a crossover GA and compare them to Estimation-Of-Distribution Algorithms (EDAs). This will allow a precise characterization of the structure of exploration w.r.t. correlations in the search distribution for these algorithms. We find that crossover transforms, depending on the crossover mask, mutual information between loci into entropy. In total, it can only decrease such mutual information. In contrast, the objective of EDAs is to estimate the correlations between loci and exploit this information during exploration. This may lead to an effective increase of mutual information in the exploration distribution, what we define correlated exploration.