论文部分内容阅读
在一个神经网络的单元中,我们考虑公式μx=—x+p+WF(x),式中x=x(t)是一个矢量,它的输入表示电荷强度。W为一个衡量突触权的矩阵,F是一个非线性函数,P是一个输入的矢量(常数或随时间缓慢变化的量)。如果变换式WF(x)是收敛的,则系统有唯一的平衡曲线,其变化是渐近稳定的,因此该网络表现为一个稳定的编码器,其对输入的稳态响应与网络起始状态无关。我们假定特征值W和
In a unit of a neural network, we consider the formula μx = -x + p + WF (x), where x = x (t) is a vector whose input represents the charge intensity. W is a matrix of synaptic weights, F is a non-linear function, and P is an input vector (constant or slowly varying over time). If the transformation WF (x) is convergent, the system has a unique balance curve, the change is asymptotically stable, so the network behaves as a stable encoder that compares the steady-state response of the input with the initial state of the network Nothing to do We assume eigenvalues W and