Role of noises in neural networks

Abstract

The important role played by noise in retrieving memories in a network is discussed. Two-stage annealing is proposed to retrieve memories stored in a neural network. The network, undergoing two-stage annealing, behaves like a nonergodic system, and the noise helps the network to select memory inside a local region in which the initial stimuli drop. Theoretical and numerical results of two-stage annealing are presented, and some further possible applications are pointed out. (c) 1995 The American Physical Society