图样间联想光学神经网络模型存贮分析及两层异联想识别模型的构造

ANALYSIS ON THE STORAGE CAPACITY OF IPA MODEL AND THE CONSTRUCTION OF A TWO LAYER HETERO ASSOCIATION OPTICAL NEURAL NETWORK MODEL

本文详细讨论了图样间联想网络的最大存贮容量 ,给出了实现图样间异联想的两个充分条件 .在此基础上 ,利用改进的图样间异联想算法构造了两层异联想模型 (THA)用于图样识别 ,网络判辨率与恢复率较图样间自联想识别均有很大提高 ;且其互连权矩阵更加简单稀疏并可平面化 。

The utmost storage capacity of the interpattern association (IPA) neural network is discussed in this paper,and two sufficient terms of realizing interpattern heteroassociation are pointed out as a further step.Next a two layer heteroassociation model (THA) is proposed by using the modified IHA algorithm,and the recognition of 10 digital numbers by THA has shown much improved performances compared with the IPA model in both identifying and retrieving patterns.Moreover,THA owns sparser weight matrixes (IWM) with 1 D interconnections,which brings more convenience to optical implementation.