模拟神经网络稳定性研究

本文提出了模拟神经网络稳定性分析的静态等效模型,通过推广特勒根定理,从等效模型导出了物理意义清晰的能量函数。由一阶微分方程组求网络能量函数的方法普适于所有的一阶连续自治系统。文章并利用能量平稳原理对点格神经网络的稳定性质进行了讨论,得到了一些对于网络的电路实现具有一定指导意义的结果。

RESEARCHES ON THE STABILITIES OF ANALOG ELECTRONIC NEURAL NETWORKS

A new method for analysing the stabilities of analog electronic neural networks is presented. The energy functions with clear and definite physical meaning are derived by introducing the static equivalent circuit models, which has expanded the Tellegen Theorem for application on circuit analysis. The method used to derive the energy functions of nets from first order differential equations is valid for all first order continuous autonomous systems. The stability analysis of cellular neural networks is made by the use of stationary cocontent theorem. Some results are instructive for the network implementation on circuits.