A Modular Neural Network Simulation System for Teaching/Research Based on .NET Framework

Abstract

Modular neural network is a learning paradigm where a collection of neural networks are jointly used for solving a problem. In this work ire develop a modular neural network simulation system based on .NET framework, applying a new training method named expert in one thing and good at many in constructing modular neural network. This method divides the training set into a certain number of subsets, each of which can be selected to train individual sub-network with different performance criterion. An empirical study demonstrates that the proposed method obtains significant generalization ability. In the though of Object-Oriented Programming, the simulation system ire have developed based on .NET framework provides a powerful neurocomputing functions and convenient graphical user interface to aid the study of the knowledge of modular neural network.