Abstract

Hopfield presented a simple model to describe the storage of 1-D binary signals in a neural network. In this model, signals are stored in the interconnections between the processing elements and not in the elements themselves. The model is capable of retrieving a stored signal even if presented with very distorted versions. The retrieval operation requires a series of matrix-vector multiplications followed by simple threshold operations. An extension of this model which enables the network to process 2-D m-ary signals is proposed. The amount of calculation required remains largely unchanged.

© 1987 Optical Society of America