基于稀疏RAM的N-tuple神经网络模型  

N-tuple Neural Network Model with Sparse RAM

作　　者：彭宏京[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：《计算机科学》2001年第11期37-39,16,共4页Computer Science

基　　金：国家自然科学基金(编号:69973021)

摘　　要：1.引言 N-tuple结构首先由Bledsoe ＆ Browing(1959)提出[1],并由英国帝国理工学院的Alexsander教授等人通过用RAM式神经元代替常规的神经元,利用N元(N-tuple)分解技术和可扩展的并行体系结构设计并实现了一个单层自适应模式识别系统(WISARD)[2],使其具有大规模并行分布处理能力,以及通用性和自适应性.适合于大维数输入模式和大样本集类识别问题,并在工业零件和人脸识别方面获得了成功的应用.该系统的学习是通过样本的一次提交完成的,因此速度快,识别具有实时性且对已学样本具有完全的识别精度.但不足是:(1)存在过泛化现象,造成误识;(2)因学习算法的限制,RAM神经元的空间存在饱和问题;(3)对每类模式分配一个分类器,硬件成本高;(4)对N元分解方法的敏感性;(5)在模式识别中,由于同类数据分布的凝聚性,常规RAM中只有很小比例的存储单元被使用,造成存储空间的很大浪费.WISARD system,which needs many hardware by virtue of every class having one discriminator device, is a N-tuple classifier with RAM without any ability of approximating continue functions. Kolcz. A et al. have presented a regression algorithm for the N-tuple neural network (NTNN) to be used to interpolate arbitrary functions by canceling any discriminator concept and derivated a N-tuple kernel function to be applied to the general regression neural network (GRNN). However, when N is a very large number,many storages will be invalid and a hash technology has to be used. This paper presents a novel adaptive pattern recognition system-N-Tuple neural network model with sparse RAM which can be applied to pattern recognition as well as function approximation tasks. It is a general model to some extent in which both NTNN and SDM can be regarded as a special case. Finally,experiments have shown the ability of approximating functions for this novel model.

关 键 词：N-tuple神经网络 学习算法 RAM 人脸识别 模式识别 

分 类 号：TP183[自动化与计算机技术—控制理论与控制工程] TP391.41[自动化与计算机技术—控制科学与工程]