一类输入元素非线性连接的反馈式神经网络及其联想能力分析  

On the Associative Ability of a Feedback Neural Network with Nonlinearly Connected Input Elements

作　　者：周青山[1] 邹勇[1] 胡健栋 

机构地区：[1]北京邮电大学电信工程系

出　　处：《电子学报》1996年第7期121-124,共4页Acta Electronica Sinica

基　　金：邮电部切块科研基金

摘　　要：本文研究了输入元素非线性连接的反馈式神经网络。文中以二阶非线性连接为例给出了网络拓扑结构，导出了能够实现模式平移不变识别的学习方法，并借助于等权类的概念把不变识别条件建造于网络权结构之中，同时降低了网络连接复杂度。In this paper,a feedback neural network,of which input elements are nonlinearly connected, is studied. The topology of its second order case is shown,the learning rules which make shift pattern recognition possible are derived. With the help of the idea of the equivalent weight subset,not only the shift pattern invariance conditions are coded in weight matrix but the nonlinearly connected network is made much simpler.

关 键 词：神经网络 模式识别 平移不变性 学习方法 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程] TP391.4[自动化与计算机技术—控制科学与工程]