Architecture for high speed evolutionary learning

This paper discusses the architecture for high speed learning of neural networks (NN) using genetic algorithms (GA). The proposed architecture prevents local minimum by using the GA characteristics by holding several individual populations for a population-based search and achieves high speed processing adopting dedicated hardware. To keep general purposes equal software, the proposed architecture can be flexible genetic operations on GA and introduced both Heaviside function and Sigmoid function on NN. Furthermore, the proposed architecture is proposed not only the pipeline at evaluation phase on NN, but also the pipeline on the whole at evolutionary phase on GA. Simulation results evaluating the proposed architecture are shown to achieve 750 times the speed on average compared with software processing.