DYNAMICAL PHASE-TRANSITIONS IN THE LITTLE-HOPFIELD MODEL

The time evolution of the distance between two random initial configur ations subjected to the same thermal noise is used to study dynamical phase transitions in attractor neural networks trained by the Hebb rul e. Numerical results are given for fully connected architectures, wher eas, in the dilute case, both analytical and numerical outcomes are pr ovided and a good agreement is shown to exist between the two sets of results.