综合联想记忆神经网络的外积取等准则

Outer Product Equation(OPE) Learning Rule for Designing Associative Memory Neural Networks

本文提出了一个新的联想记忆设计准则，即外积取等准则，它具有外积和准则的所有优点。由外积取等准则设计出的联想记忆网络能够存储任意给定的训练模式，即对于训练模式的数目和它们之间相关性的强弱没有限制。外积取等准则可用来定量地评价记忆模式向量各分量对于记忆模式分类或识别的重要性。由外积取等准则设计出的网络的连接权值只取１、０或－１，因而网络易于光学实现。计算机实验结果充分说明了外积取等准则的有效性。

A new design rule for associative memory.called outer product equation(OPE) rule,is proposed. The rule has all the advantages of outer product sum(OPS) rule.The associative memory network designed by OPE rule can store arbitrary given training patterns,no matter what number of them and correlation among them.The importance of components of memoried patterns for classification and recognition of them can be quantitatively evaluated by OPE rule.The weight connection is 1,0,or-1,so the network designed by OPE rule can be optically implemented easily.Computer experimental results confirm the efficiency of OPE rule.