MODULAR NEURAL NETWORK AND ITS PERFORMANCE
模块神经网络及其性能

A multi-modular neural network model, which is different from BP network for mapping y=f(x),is proposed. The architecture and parallel dynamics equations of the network are given,and the stability of dynamics is also proven. Through establishing learning algorithm. We have proven that the network can perform mapping or input vector pair (x,y) to associative output vector z. It is the most inportant that both the pattern series varing with the and statics patterns can be stored sitmutaneously, thus the economic principle of memory is proposed. In addition, a dynamics learning algorithm is given and have proven its convergence, computer simulation confirms fully theoretical results. Finally, Some possible applications are discussed.