基于小世界体系的指数自联想记忆模型研究  

Exponential Auto-Associative Memories Based on Small-World Architecture

作　　者：王敏[1] 陈松灿[1] 

机构地区：[1]南京航空航天大学计算机科学与工程系,南京210016

出　　处：《计算机学报》2005年第12期1988-1992,共5页Chinese Journal of Computers

基　　金：国家自然科学基金(60271017);江苏省自然科学基金(BK2004001);江苏省青蓝工程基金;留学回国人员基金资助.

摘　　要：全互连结构是大多数神经联想记忆模型采用的连接方式.然而,无论从神经生物学的观点还是从硬件实现角度出发,在确保局部连接占主体地位的情况下,最小化神经元间的连接长度是合乎逻辑的.由Watts和Strogatz提出的小世界体系(SWA)为解决这一问题提供了一条有趣途径.借鉴这一具有一定神经生物学特点的小世界体系,构建出一个基于小世界体系的指数自联想记忆模型(SWEAM),其在结构上更接近人脑的生物学特性,同时在存储容量和抗噪声性能上优于现有的同类联想记忆模型.因采用了SWA,SWEAM在工艺实现上也更容易.Up to now, most of the associative memory models are the neural networks with full connectivity. However, from both the neurobiological viewpoint and the hardware implementation perspective, it is logical to minimize the length of inter-neural connections under the circumstances that the connectivity of the networks is predominantly local. Recently, Watts and Stroga-tz have proposed the ＂small-world architecture＂（SWA）, which provides us an interesting approach to address this issue. On the basis of SWA, a new exponential auto-associative memory model called SWEAM is proposed. It is closer to human brain in structure. And it has larger storage capacity and better error correction capability than the existing counterpart of the same class. Thanks to the introduction of SWA, SWEAM can be more easily realized in hardware.

关 键 词：小世界体系 自联想记忆 存储容量 回忆 神经网络 

分 类 号：TP18[自动化与计算机技术—控制理论与控制工程]