The Realization Of Weight Function Under MATLAB

In the traditional neural network learning algorithm (Such as the BP algorithm, the RBF algorithm), Weight is constant, which can not be able to sample information to complete memory. In response to this problem, Professor Zhang Daiyuan who works in Nanjing University of Posts and Telecommunications has proposed the spline weight function neural network algorithm in the monograph "neural network theory and methods", which can achieve accurate memory of the sample characteristics in the learning and training when weights are replaced with weight function. Also, neural network generalization capability is enhanced with the increase of the number of samples in the new network algorithm. But under normal circumstances, it is impossible to get the theoretical weight function by a finite number of known sample points. Approximate weight function can be obtained and the weight function neural network can be constructed by the interpolation method based on the study of Professor Zhang Daiyuan.Approximate weight function can be constructed by the interpolation method in the theoretical core part, which analyzes errors in both cases of multi-input single-output and multiple input multiple output. The analysis showed that the more number of training samples, the smaller the error.This subject mainly realizes is about weight function neural network in the application of MATLAB. Based on the software requirement analysis, design to the network parameter establish neural networks with weight functions.for the collection of samples, the weighted factor input, designed to read a text file and binary file function. The auxiliary function in software, adding weight function neural network sensitivity analysis. At the same time, the software has the help menu and the English version of the software, perfect the software application functions.