Abstract

When the Hopfield neural network is extended to deal with a 2-D image composed of N×N pixels, the weight interconnection is a fourth-rank tensor with N⁴ elements. Each neuron is interconnected with all other neurons of the network. For an image, N will be large. So N⁴, the number of elements of the interconnection tensor, will be so large as to make the neural network's learning time (which corresponds to the precalculation of the interconnection tensor elements) too long. It is also difficult to implement the 2-D Hopfield neural network optically.

© 1989 Optical Society of America