Storage capacity of a neural network with state-dependent synapses. (English) Zbl 0842.68063

Summary: The storage capacity of the Hopfield network is limited to \(p/N \leq \alpha_c = 0.138\) \((p =\) number of patterns, \(N =\) number of neurons), beyond which the contribution of weakly correlated patterns surpasses that of the desired pattern. This contribution can be eliminated by introducing a threshold: a pattern correlation below this threshold is simply set to zero in the synapses. We solve the mean-field equations and derive the critical value of the threshold required to stabilize \(p = \alpha N\) patterns with \(\alpha > \alpha_c\).


68T05 Learning and adaptive systems in artificial intelligence
Full Text: DOI