Complexity Compression and Evolution

Compression of information is an important concept in the theory of learning. We argue for the hypothesis that there is an inherent compression pressure towards short, elegant and general solutions in a genetic programming system and other variable length evolutionary algorithms. This pressure becomes visible if the size or complexity of solutions are measured without non-eeective code segments called introns. The built in parsi-mony pressure eeects complex tness functions , crossover probability, generality, maximum depth or length of solutions, explicit parsimony, granularity of tness function, initialization depth or length, and modulariz-ation. Some of these eeects are positive and some are negative. In this work we provide a basis for an analysis of these eeects and suggestions to overcome the negative implications in order to obtain the balance needed for successful evolution. An empirical investigation that supports our hypothesis is also presented.