Search

Biological organisms have the ability to develop novel phenotypes in response to environmental changes. When several traits are evolved simultaneously or as a result of one another, we talk of coevolution. Cellular Automata (CAs) have been successfully used to artificially evolve problem specific update functions. The resulting CAs are, however, much slower and more sensitive to perturbations than those with an evolved underlying topology and fixed uniform update rule. Unfortunately, these are not nearly as accurate, and suffer from scaling up the total number of cells. We propose a hybrid paradigm that simultaneously coevolves the supporting network and the update functions of CAs. The resulting systems combine the higher fitness and performance of the update evolution and the robustness properties and speed of the topology evolution CAs. Moreover, these systems seem to perform better as the size of the CA scales up, where as single-feature evolution systems are negatively impacted. Coevolution in CAs is an interesting tradeoff between the two single trait evolutions.

In biological organisms, a single genotype may map to several phenotypes and vice-versa. This many-to-many relationship is believed to be a major drive of the phenotypic robustness and genotypic evolvability found in all life forms. Given the inherent complexity of the genotype-to-phenotype (G2P) mappings, we use cellular automata (CAs) as rudimentary proxies for biological organisms. CA models have the same many-to-many G2P mappings, and their sensitivity to initial conditions allows the same genotype to differentiate into different phenotypes. We use a bipartite network to study the G2P landscape, and its projections in either space. The degree distributions of the network and its projections are all heavy-tailed, denoting the presence of highly connected hubs, implying that increased robustness is supported by the network structure. We also show a strong correlation between the phenotype’s complexity and its robustness. We analyze the relationships between the robustness and the evolvability both at the genotypic and phenotypic level. Although we use different computational models, our results agree with those of previous similar studies, and with observations in biological organisms.