一类神经网络算法的渐近行为

A Asymptotic Behaviov of Neural Networks Algorithm

研究了一类连续型全反馈人工神经网络的稳定性,通过求解简单的一维微分方程,给出了这类神经网络系统的解的解析表达式,利用解的解析表达式证明了当A的特征值0l,至少有一个特征值为负时,网络的一切从Rn中的单位超球内出发的解收敛于原点或单位超球内的一点.

In this paper the problem of artificial neural network model is studied. The differential equation of an artificial neural network model is solved by solving a simple one imension differential equation. Applying the solution of the neural network model, we proved that all solutions of a network from the unit hypersphere over Rn converge to the origin or a point in the sphere, when all of A's eigenvalues are less than or equal to zero as well as not all of the eigenvalues equal to zero.