An examination is made of some properties of the asynchronous binary neural network popularized by J. J. Hopfield (1982). While the computational appeal of this network is primarily due to its utility as a content addressable memory, the collective behavior of the neurons resembles additional properties of biological memories. These properties are demonstrated when the training of the network is accomplished using adaptive algorithms, which update the weights and thresholds of the neurons based on the changing characteristics of the environmental patterns. Two algorithms are presented here. The authors demonstrate that gradual learning in the network is superior to mass practice; overlearning slows the rate of forgetting, and a 'saving score' due to past knowledge speeds the relearning of forgotten patterns.
|Original language||English (US)|
|Number of pages||3|
|State||Published - 1987|
All Science Journal Classification (ASJC) codes