Combining modular neural networks developed by evolutionary algorithm

The evolutionary approach to artificial neural networks has been developing rapidly in recent years and shows great possibility as a powerful tool. However, most evolutionary neural networks use the simple node as a building block to evolve and select the one network producing the best result after evolution. In this paper, we present concepts and methodologies for evolutionary modular neural networks, which boost the overall performance by combining several potential networks which have emerged during the course of the evolution. Experimental results with the problem of the recognition of handwritten numerals shows the possibility of combining a number of characteristic networks from a gene pool.