Proceedings of 2009 International Workshop on Information Security and Application (IWISA 2009) Qingdao, China, November 2122, 2009 Editors: Feng Gao and Xijun Zhu AP Catalog Number: APPROCCS09CN004 ISBN: 9789525726060 Page(s): 678682 

The Neural Network Based on Three Degree u System Wuyi Fan, Fenhong Guo, and Jinhong Li 
Full text: PDF 
Abstract 

Artificial Neural Network (ANN) is a new kind of computing system with the capacity of selfstudy, which is widely used in many areas, such as pattern recognition, function approximation and credit risk assessment, etc. BP Network is commonly used in a number of occasions, as it can approach arbitrary nonlinear curve. But the speed of convergence is slow, the training time is long, and it often fall into local minimum points in BP network. Because of the shortcomings of BP network, the promotion of BP network is limited. In order to overcome those shortcomings, a new neural network based on three degree U system is proposed in this text. In the arithmetic three degree U orthogonal polynomials is as the activation functions of every hidden neuron in the neural network, and then a new U3 neural network is constructed. Experimental results show that the new network is obviously superior to BP network and Chebyshev network, it can approach original curve rapidly in a short time, and get a better approximation result at the inflection point. 

Index Terms 

The Basic Concept and Composition of Neural Networks, U System and Its Properties, A Neural Network Modeling of U3, Simulation results, Conclusions, Acknowledgment, References 

Copyright @ 2009 ACADEMY PUBLISHER — All rights reserved 