【摘要】 <正> 目前有许多神经网络模型以 Hebb 规则做为学习的基础.在 Hebb 规则中不包含神经元的几何结构信息.但对点数相同的二维点阵和一维点阵,大脑易于记忆二维点阵,因为它有更多的几何结构特征.Kohonen在中论述了生物神经网络的结构特征.本文在Hopfield 网络中引入几何结构信息,对其联想记忆过程做了理论分析和实验研究,试图研究单层神经网络的结构特征对联想记忆的影响,并探讨实现局域连接的可能性.1 Hopfieid 神经网络模型及结构信息的引入Hopfield在文中提出了一个神经网络离散模型.此模型中包括 N 个神经元,它们更多还原

【Abstract】 In order to explore the structral features of neural net works and the approaches to local interconnection,geometrical structural information is introduced to the Hopfield neural network model which is applied to associa- tive memory.The dynamies of the recalling is theoretically analyzed and experimentally studied.The results show that the geometrical structural infor- mation is helpless to the associative memory of monolayered neural networks, furthermore,it makes the error probability increased.Therefore,new approa- ches have to be explored to obtain a locally interconnected structure which can make use of the geometrical structural information of the stored patterns.更多还原