关联优化存储下Hopfield网络的临界存储  

CRITICAL CAPACITY OF HOPFIELD NEURAL NETWORKS WITH OPTIMAL STORAGE

作　　者：郭东辉[1] 陈振湘[1] 刘瑞堂[1] 吴伯僖[1] 郑利明[1] 郑利龙 

机构地区：[1]厦门大学物理系,香港城市大学电子工程系

出　　处：《计算机学报》1997年第1期77-81,共5页Chinese Journal of Computers

基　　金：国家自然科学基金

摘　　要：本文通过对Ｈｏｐｆｉｅｌｄ神经网络存储相关信息的联想稳定性分析，给出了选取存储样本的优化规则，并得出Ｈｏｐｆｉｅｌｄ神经网络在优化存储规则下各存储样本均能纠一错（即最低联想）的临界存储容量约为０．５Ｎ，比随机选取存储样本的容量（约０．１４Ｎ）要强得多．Based on the analysis of associative stability of Hopfield neural networksstored with correlative patterns, the optimal rule of choosing stored patterns is presented in this paper for improving the associative stability and the storage capacity.By the optimal storage rule, the critical storage capacity that each stored patterncan correct one bit error in every bit of the stored pattern is about 0. SN, which ismuch better than the capacity about 0. 14N of storing stochastic patterns.

关 键 词：神经网络 优化存储 存储容量 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]