Neural nets for correlated and non-binary patterns: feedback from current pattern to neuron response and threshold

We analyse the mechanism by which Hopfield algorithm suppresses all patterns but one. We modify it by adapting the neuron response function and the threshold in each node to the global current pattern and to overall correlation among memory patterns. Such feedback lets the method recognize strongly correlated memories and operate on non-binary neurons (that is, those with more states than just 0,1).