异联想记忆Hopfield神经网络的模型、算法及性能  

Hetero Associative Memory Hopfield NN Model, Learning Algorithms and Performance

作　　者：姜惠兰[1] 孙雅明[1] 

机构地区：[1]天津大学电气与自动化工程学院,天津300072

出　　处：《系统工程理论与实践》2005年第5期101-107,共7页Systems Engineering-Theory & Practice

基　　金：国家自然科学基金(59877016)

摘　　要：　对联想记忆神经网络(neuralnetworks NN)的特性进行了分析,基于双向联想存储器BAM原理,对自联想记忆HopfieldNN模型进行了扩展,建造了适合于求解模式识别问题的异联想记忆HopfieldNN模型结构.并针对HopfieldNN记忆容量不足的缺陷,对常规的学习算法进行了改进,建立了基于投影原理的伪逆学习算法和广义逆学习算法,提高了HopfieldNN对样本的记忆存储能力.分析表明:本文的研究解决了HopfieldNN应用于异联想记忆模式识别的关键问题.This paper analyzes the performance of associative memory neural network. On the basis of bidirectional associative memory theory, a hetero associative memory Hopfield NN model is built to be used for pattern recognition. For the shortcoming of lack of storage capacity of Hopfield NN this paper improves traditional learning algorithm, proposes pseudo converse and generalized converse learning algorithm, and therefore the capacity of Hopfield NN to store training samples is increased. The simulation and results show that the study of this paper solves the key of Hopfield NN being applied in hetero associative memory pattern recognition.

关 键 词：HOPFIELD NN 联想记忆 记忆容量 模式识别 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]