摘要
在神经网络基础理论中,有一个这样的问题:怎样用三层网络实现对任意连续函数的映射,本文利用Kohonen提出的Self-OrganizingFeatureMap竞争型网络以及区域的2M-分割概念,解决了用具有有限个隐层单元的三层网络以任意精度逼近一个连续函数的问题,具体的网络结构和学习算法已给出.
There is such a problem in the Neural Network basic that is how to realize an arbitary continuous function by a three-layered network.In this paper,by using the Self-Organizing Feature Map competitive network presented by Kohonen in 1981,and 2M-partition definition of domain,a new method has been discussed to solve the problem of how to approximate arbitrary continuous function defined on a hypercube by utilizing a.network with definite hidden unit and a satisfactory conclusion has been obtained.The network architecture and learning algorithm is given in detail.
出处
《河北工学院学报》
1994年第4期7-11,共5页
Journal of Hubei Polytechnic University
关键词
神经网络
逼近
三层网络
连续函数
Neural Network,Approximate,Three-layered Network,Self-Organizing Feature Map,Continuous Function,Hidden unit