An Overview of Modularity in Artificial Evolutionary Systems

Modularity in biological systems exists on many levels, however it is very difficult to define or quantify. Loosely, evolutionary modularity is the compartmentalization of genotype-phenotype interactions such that pleiotropy is suppressed among genes that do not serve like functions. Evolutionary modularity is attributed to evolvability, specialization, and canalization in genetic systems. These effects, though conceptually separate, are highly interrelated, and often share a bijective relationship with modularity (e.g. an increase in the selection pressure for evolvability also increases modularity as a side effect). Little is known about how genetic modularity arises in biology, however recent research points to processes such as gene duplication, secondary selection, and symbiogenesis as possible factors. Since modularity seems to play such an important role in biology, it should also be studied with respect to neuro-evolution. In particular, since traditional neuro-evolution methods do not employ anything more complex than a one-to-one genotype-phenotype map, how can simplicity be maintained when modularity is introduced in the system?