Hopfield连续神经网络新的综合和学习方法

本文对设计一个全局稳定的Ｈｏｐｆｉｅｌｄ连续时间联想神经网络提出一种新的综合方法，用本文提出的方法综合的网络不仅能保证每一个要求的模式全部记忆，而且存贮在网络的稳定平衡点上的每一个记忆具有很好的吸引域特性。从最小误差参数估计的观点导出求解网络联接矩阵的学习方法具有快速收敛特性。本文还提出了利用反馈网络电路结构实现网络综合的方案。一个零自联接联想网络的综合实例证实了本文提出的方法的有效性和实用性。

A NEW SYNTHESIS AND LEARNING ALGOITHM FOR HOPFIELD CONTINUOUS-TIME NEURAL NETWORK

A new synthesis for designing a globally stable Hopfield continuous-t,me associative neural network is proposed in the paper. The associative network synthesized with the proposed method not only can successfully store every desired pattern, but also have a good attraction property of each stored memory in the equilibria of the network. Derived from the standpoint of minimum error parameter estimation,a fast learning algorithm for solving interconnection matrix W of the network is of fast convergence property.A schematic circuit architecture of feedback network to implement network synfhesis is also suggested in the paper. A specific example of synthesis of a zero auto-interconnection Hopfield associative network is presented to demonstrate the efficient performances of the network synthesized.