Associative memory based on sparsely encoded Hopfield-like neural network

Informational and dynamic properties of sparsely encoded Hopfield-like neural network performing the functions of autoassociative memory are investigated analytically and by computer simulation. It is shown that the informational capacity and the processing rate monotonically increase if the sparseness increases. In contradiction to this, the size of the attraction basins and the recall quality initially change nonmonotonically. An optimal sparseness exists when the information extracted from the network due to correction of destroyed stored patterns are maximal.