联想神经网络异步工作方式下收敛性能的研究

STUDY ON CONVERGENT PROPERTIES OF ASSOCLATIVE NEURAL NETWORK UNDER ASYNCHRONOUS UPDATING OPERATION

本文对具有投影学习算法的Hopfield型二值神经网络在异步工作方式下收敛性能进行了比较全面的分析讨论,采用非能量函数的方法证明了网络具有演化收敛到网络平衡点的特性;推导了计算网络收敛所需瞬态工作步数上界的估算公式;分析并给出了计算平衡点(记忆模式)收敛半径的算法.

In this paper, the convergent properties of Hopfield-type associative neural network with projectlearning rule under asynchronous update operation has been studied. Using non-energy function method, we haveproved the network having the property converging to one of its equilibrium points. A formula for estimatingthe upper bound of update steps required for convergence and a algorithm for calculating the convergent radius ofthe equilibrium point have been derived.