Synaptically distributed memory vs. synaptically localized memory

Abstract

We clarify that the only essential difference between the two major categories of unsupervised learning rules discussed in theories of artificial neural networks-the competitive learning and the Hebbian learning rules-is that lateral inhibition is present in the former and is absent in the later. We demonstrate analytically that a competitive learning neural network, which has synaptically localized memory, shows better tolerance over noise in training patterns in comparison with the Hopfield neural network, which uses a Hebbian-type learning rule without any lateral inhibition and has synaptically distributed memory. >