The modularity in freeform evolving neural networks

In this paper, we validate whether the network modularity can emerge, and the evolution performance can be improved by varying the environment or evolution process under a more freeform artificial evolution. Previous studies have demonstrated that the modular structure naturally arisen as a response of the variations on environment and selection process, however, since the models they used were relatively simple and with some biasing constraints, the results may lack of generality. In contrast, we evolve more freeform neural networks to address this issue, and an artificial tracer method was employed to quantify the modularity. A series of varying scenarios have been experimented, the results show that the evolution performance have been improved in most cases, however, the modularity never appeared among those scenarios. A further experiment shows that our method has the potentials to produce modular networks but the more advanced methods are still needed to encourage the emergence of modularity on the complex questions.