Delay node

The delay operation can be used to model dynamics. The node operates on time-dependent signals. It transforms the inputs $s(1), s(2), \dots , s(T)$ into outputs $s_0, s(1), s(2), \dots , s(T-1)$ where $s_0$ is a scalar parameter that provides a starting distribution for the dynamical process. The symbol $z^{-1}$ in the rightmost subfigure of Fig. 2 illustrating the delay node is the standard notation for the unit delay in signal processing and temporal neural networks Haykin98. Models containing the delay node are called dynamic, and the other models are called static.