Evolving neural induction regular language using combined evolutionary algorithms

This paper proposes a new algorithm called combined evolutionary algorithm (CEA) to train a neural network, and demonstrates its use in inducing the finite state automata task. This algorithm evolves neural networks by incorporating the ideas of evolutionary programming (EP) and real coded genetic algorithms (RCGA) into evolution strategies (ESs). Simultaneously, we add the local competition into the CEA in order to reduce the complexity and maintain the diversity. This algorithm is able to balance the exploration anti exploitation dynamically. We implement CEA and experiment on seven benchmark problems of regular language. The results indicate that the CEA is a powerful technique to construct neural networks.