Distribution of Internal Fields and Dynamics of Neural Networks
Rolf D. Henkel, Manfred Opper · Europhysics Letters (EPL) · 1990
A new approach to treat the temporal development of fully connected neural networks is introduced. It is based on a non-Gaussian ansatz for the distribution of internal fields, incorporating memory effects. For the network with projector couplings recursion relations for the parameters of the field distribution are derived, which yield remanence effects as well as basins of attractions in good agreement with numerical simulations.