PATTERN SELECTIVITY AND BINARY-SYNAPSE NEURAL NETWORKS

Using the Rau, Wong and Sherrington model of the storage of a pair of correlated patterns by an optimized synaptic neural network, we argue that all maximally-stable neural networks will show high selectivity b etween correlated memorized patterns. The dilute-retrieval phase diagr am is calculated specifically for a network with binary-valued synapse s, and is shown to be qualitatively identical to that for the spherica lly-constrained synapses of the original work.