Pattern Formation of Synaptic Connections in a Generalized Hebb-Type Learning Model

Abstract

On the basis of a generalized Hebb-type learning model we discuss pattern formation of synapses in neural network system. We propose a generalized Hebb-type learning equation in which connection weight of synapse has an arbitrary potential function. Especially much attention is paid for the case where potential has double stable points. Computer simulation is performed upon this model and it is found that this neural network model has a merit that system keeps learned patterns after input signals are stopped. The correspondence of our model with physiological observations is discussed.