An investigation into the structure of genomes within an evolution that uses embryogenesis

Evolutionary algorithms that use embryogenesis in the creation of individuals have several desirable qualities. Such algorithms are able to create complex, modular designs which can scale well to large problems. However, the inner workings of developmental algorithms have not been investigated as thoroughly as their direct-encoding counterparts. More precisely, it would be beneficial to look at how the rules used during embryogenesis evolve alongside the phenotypes they produced. This paper reports on such an investigation into the evolution of a rule set for the growth of an artificial neural network, and identifies several aspects that are desirable for the genomes of a developmental evolutionary algorithm.