ROBUSTNESS AGAINST RANDOM DILUTION IN ATTRACTOR NEURAL NETWORKS

Abstract

We study the robustness of attractor neural networks against random disruption of a fraction of the synaptic couplings. For the maximally stable network (MSN), we determine the effect of different degrees of dilution on the overall storage capacity, the size of the basins of attraction and the attractor overlap. Comparison with corresponding results for the Hopfield model indicates that, although MSN is more robust at low degrees of dilution, the Hopfield network becomes more robust at high dilution.