Size of the domains of attraction in the Hopfield model
A C C Coolen, Harm J. J. Jonker, Th.W. Ruijgrok · Physical Review A · 1989
For Hopfield's neural network model we compute the size of the domains of attraction of p stored patterns, i.e., the fraction of all initial microscopic states that will lead to pattern reconstruction. For p<5 this can be done analytically by solving the evolution equations for the overlaps. For p\ensuremath{\ge}5 we solve the evolution equations numerically. We consider only the case where the number of patterns stored is much smaller than the number of neurons.