Nonlinear Neural Networks

J. Leo van Hemmen, Reiner Kühn · Physical Review Letters · 1986

A general theory of neural networks with nonlinear synapses is developed. To this end a meanfield model of a novel type is introduced and solved exactly. For suitable nonlinearity, synaptic sign changes may be eliminated altogether without affecting the efficiency of the network. Static noise is easily included.

Read the paper · More papers on PaperTik