Abstract

We consider neural networks with feedback as nonlinear dynamical systems that can perform complex computations of the kind encountered in learning, optimization, and cognition at high speed especially when implemented in appropriate analog hardware. Progress in imbedding concepts of nonlinear dynamics in photonic hardware to produce neurocomputers with learning and cognitive ability, and the potential for using electron trapping materials that furnish unique photon storage function in such systems are discussed. Bifurcation in synchronously updated networks which support more than one type of attractor is described and proposed as mechanism for feature binding and cognition

© 1991 Optical Society of America