Study On Some Issues Of The Modular Neural Networks

Modular neural network is a new learning model based on multi-agents, whose decisions are combined in a fashion of competition and cooperation of a number of artificial neural networks. The essential idea of the learning system is to improve the performance of the overall system by searching the difference of the sub-Neural networks. A great deal of successful applications demonstrates that modular neural network outperforms single neural network in terms of generalization and reliability and undoubtedly provides for us with a new tool for problem-solving. Moreover, various theoretical explanations proposed recently justify the effectiveness of some conventional methods for modular neural network.Having recognized that modular neural network has an enormous potential and bright prospect in application, a large number of researchers plunge themselves into the field, yielding a lot of relevant theories and application achievements. At present modular neural networks is a rathers hot topic in many diverse areas such as pattern recognition, neuro-computation, machine learning, information processing.The research for modular neural network concentrates on two aspects: how to combine the decisions of the component networks, and how to generate the component networks in the entire system. In this paper, the first aspect is mainly studied, a scheme of selective ensemble ispresented, and a new architecture of modular neural networks is presented. At the same time, the robust algorithm of modular neural networks is studied.This paper composes five chapters in all. Chapter 1 expatiates the probability and necessity of the generation of modular aeural networks from the viewpoints of commands, neurobiology and social science. In Chapter 2 a robust learning algorithm of modular neural networks based on the theory of robust regression is presented. Empirical study demonstrates that the robust learning algorithm of MNN has better precision and generalization than both modular neural networks and single neural network with the same robust algorithm, when trained under the contained dada. In Chapter 3, problems of adaptive combination are discussed. In Chapter 4 a new architecture of multiple neural networks based on modularity is presented, which is named hierarchical modular neural networks. Finally in Chapter 5, contributions of this paper are summarized and several issues for future work are indicated.