EXTERNAL STIMULI IN OPTIMIZED ATTRACTOR NEURAL NETWORKS FOR SPARSELY CODED PATTERNS

Abstract

Attractor neural networks (ANN) have aroused wide interest as a model for the cortex because of the many features that they share in common, such as their high degree of connectivity, non-local information storage ability, tolerance to damage, and feedback. In this paper a further biologically motivated modification is examined, namely looking at a more realistic model of the input side of such networks. The usual assumption about external stimuli attractor neural network into an initial state. Once this impinging stimulus has prepared the network in an initial state, it disappears and the network is left to evolve according to its own dynamics, a comprehensive review of which is given in.