Optical associative memory model based on neural networks having variable interconnection weights.

A new model is proposed for a content-addressable memory (CAM) based on neural networks. Like the previous Hopfield model, the information is stored in the structure of the network and the read-out procedure may be implemented in the form of an optical vector-matrix multiplier. This model introduces intermediate layers of interneurons between the neuron layers and a dependence of the interconnection weights to a given neuron on the previous history of the neuron. The storage prescription allows each matrix element to have three values instead of only two as in the previous Hopfield model. This more complex model gives better results than the Hopfield model.

Compensation of edge effect in neuron-like layer nets with local feedbacks.

One or two-dimensional layer nets with local feedback between neuron-like elements were investigated. The mutual influence between elements of the net is based on the principle of lateral inhibition. A properly adapted and modified Z-transform method applied to a difference equation describing the function of the net allowed us to characterize the properties and dynamics of the net and to define its stability region. Finite dimensions of the net cause "reflections" of signals from the edges of the structure. This makes the detection of pattern difficult or even impossible and therefore the problem of compensation of the edge-effects appears. Several methods of compensation, involving discrete or continuous change of coupling weights, are described. A comparison of the behavior of a compensated and uncompensated net, as modelled on a digital computer, shows the advantages of the compensation.