Grammatical Development of Evolutionary Modular Neural Networks

Abstract

Evolutionary algorithms have shown a great potential to develop the optimal neural networks that can change the architectures and learning rules according to the environments. In order to boost up the scalability and utilization, grammatical development has been considered as a promising encoding scheme of the network architecture in the evolutionary process. This paper presents a preliminary result to apply a grammatical development method called L-system to determine the structure of a modular neural network that was previously proposed by the authors. Simulation result with the recognition problem of handwritten digits indicates that the evolved neural network has reproduced some of the characteristics of natural visual system, such as the organization of coarse and fine processing of stimuli in separate pathways.