摘要
自联想记忆神经网络具有类似于大脑的记忆和联想的特性 .在Hopfield网络的理论基础上 ,提出了一种反馈型的自联想记忆神经网络 .和Hopfield模型不同的是 ,这种神经网络增加了一个隐含层来扩大网络的存储容量 ,并采用局部相连的拓扑结构来代替全相连 ,从而减少了计算复杂度 .在网络的学习过程中 ,各神经元之间的权值按照学习规则不断地进行调整 ,使回忆后的输出结果更加接近期望输出 .
The Associative memory network has the function of memory like human's brain. Based on the theory of the Hopfield Network, a new network model is presented, which is called Recurrent Associative Memory Network. Different from the Hopfield Model, it adds a hidden layer into the network that can enlarge the capacity of network. And the structure of this model takes the form of local-connection instead of full-connection in order to reduce the complexity of computation. During the course of learning patterns, the value of weight between neural cells has to adjust continually in order to make the output close to the expected pattern.
出处
《宁波大学学报（理工版）》
CAS
2004年第3期274-278,共5页
Journal of Ningbo University：Natural Science and Engineering Edition
基金
浙江省科技厅重点科研项目 (2 0 0 3C2 10 0 9)
宁波市科技攻关项目支持
