延时神经网络对方波信号的记忆

Memory of Square Wave Signals in the Time Delay Neural Networks

在输入信号方波的刺激下,网络具有如下的动力学特征:如果输入的方波的周期是网络延时的整数份之一倍,则在输入信号停止后,网络的总活性还能保持着以输入方波的周期(节律)进行振荡;如果这一关系不能满足,则在方波停止输入后,网络的总活性是不能再以方波的节律进行振荡,甚至振荡的图像是混乱的.通过引入同步度的概念,对这一现象进行了较为深入的分析.

The role that different square wave signals play in the dynamics of a time delay network of McCulloch-Pitts elements have been investigated using computer simulations. The different cycle (positive and negative) square wave signals induce the following network features: signals, whose cycles exactly divide the delay time, give rise to system memory; on the other hand, signals, whose cycles have not the numerical relationship mentioned above with or are larger than the delay time, give rise to disordered integral activity of the network or patterns which can not exhibit step of the inputs after the inputs are eliminated. This phenomenon is analyses by means of the degree of synchronization.