A Niche-MAS Framework for Evolving Artificial Neural Networks

In order to adopt PEAs for evolving neural networks and to implement or simulate them efficiently, we introduce a distributed framework, called NMtENN in this paper. The NMtENN framework is embedded with a general PGA, which guide the evolution of the whole neural network population. But the PGA is not an explicit entity; it carries out by the Niche-MAS, which consists of a number of NAs (Niche Agents) and a CA (Commonweal Agent). By the implementation of NEAT, SANE, and MOBNET, NMfENN is demonstrated to be easily and conveniently employed to realize users' parallel evolutionary algorithms or strategies for evolving neural networks.