基于小世界体系的投影学习联想记忆模型研究  

A Projection Learning Rule Associative Memory Model Based on Small-World Architecture

作　　者：金燕晖[1] 陈贤富[1] 

机构地区：[1]中国科学技术大学电子科学与技术系,安徽合肥230027

出　　处：《计算机仿真》2009年第4期207-210,共4页Computer Simulation

摘　　要：联想记忆是神经网络的重要应用之一,传统Hopfield网络所用的外积法限制了输入向量的模式,降低了网络的联想性能,而且全互连的结构增加了网络的复杂性,不符合神经生物学观点。针对这些问题,提出一种新型的小世界联想记忆模型,用局部的规则连接和稀疏的长程连接取代全互连结构,降低网络的复杂性,同时引入投影学习规则来提高网络的回忆能力和抗噪能力。通过matlab软件对该模型模拟仿真,并与其他模型进行比较,结果表明,该模型在有效降低网络复杂度的情况下,保持了良好的回忆性能。Associative memory is one of the most important applications of neural networks. But the Outer Product Method used by traditional Hopfield network limits the input modes, decreases the probability of the network. And the full -connected architecture is not logical from the neurobiological viewpoint. Aimed at the problems above, a Projection Learning Rule associative memory model based on Small- World architecture is proposed. The model decreases the complexity with local short- distanced connectivity and sparse long- distanced connectivity, and increases the memory capability and error correction capability by inducting the Projection Learning Rule. Through simulation on the MATLAB software, and comparing with other associative memory model, the results indicated that this scheme can keep good recall capability while decreasing network＇s complexity.

关 键 词：联想记忆 小世界体系 投影学习 回忆 神经网络 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]