权值可直接算出的Legendre神经网络之隐节点数自动确定  

Number Determination of Hidden-layer Neurons in Weights-directly-determined Legendre Neural Network

作　　者：张雨浓[1] 刘巍[2] 蔡炳煌[1] 肖秀春[1] 

机构地区：[1]中山大学电子与通信工程系,广东广州510275 [2]中山大学软件学院,广东广州510275

出　　处：《小型微型计算机系统》2009年第7期1298-1301,共4页Journal of Chinese Computer Systems

基　　金：国家自然科学基金项目(60775050)资助;中山大学科研启动费;后备重点课题资助

摘　　要：隐神经元数目的确定在神经网络学习过程中有着重要的意义.然而,目前,还没有相应的确定性理论指导隐神经元数的设计.针对Legendre前向神经网络,在基于伪逆的权值直接确定法的基础上构造出一种神经网络隐节点数自动确定的算法.仿真结果显示该隐节点数自动确定算法能较快地找到最简化结构Legendre前向神经网络的隐节点数.Number determination of hidden-layer neurons is very significant in the study of neural networks＇ learning procedure. There are many algorithms for determining the number of hidden-layer neurons at present, but few of them can obtain the precise number deterministically. Based on a weights-direct-determination method, we propose in this paper an automaticdetermination algorithm for the number of hidden-layer neurons used in Legendre neural network. Computer-simulation results substantiate that the proposed algorithm could find the optimal number of hidden-layer neurons quickly and precisely.

关 键 词：Legendre前向神经网络 隐节点数自动确定 权值直接确定 伪逆 

分 类 号：TP241[自动化与计算机技术—检测技术与自动化装置]