HYSTERESIS IN NEURAL NETWORKS

The zero-temperature dynamics of neural networks is used to study hist ory-dependent effects in random magnets. Random-field as well as rando m-bond Ising models are shown to exhibit hysteresis. The hysteresis lo ops of the random-bond model have a staircase structure while those of the random-field model are relatively smooth. We discuss the relevanc e of neural networks in the context of random-bond models, and the bea ring the neural network dynamics has on the spectrum of relaxation tim es in a hysteretic system.