Dynamical Interference between the Attractors in a Neural Network
G. A. Kohring · Europhysics Letters (EPL) · 1989
The dynamical effects arising from the states having microscopic, O (1/√ N ), overlaps with a nominated configuration are shown to be nonnegligible. For the Hopfield model, the inclusion of these states into the dynamics gives results which agree with the previous thermodynamic calculations. For asymmetric, maximally stable networks, the inclusion of these states predicts a radius of attraction which is in very good agreement with computer simulations.