chapter  8
26 Pages

Overview: Dynamical Perspectives on Neural Networks: Paul Smolensky

The computational perspectives of Part I focus on the function relating input activity to output activity, and how properties of this function are determined by the internal properties of the neural network machine. In the dynamical perspectives of Part II, attention shifts to the temporal or dynamic properties of activity (and, to a lesser extent, of weights) in networks that evolve over an extended period of time.