OPTIMAL SIGNALING IN ATTRACTOR NEURAL NETWORKS

In a recent paper we presented a methodological framework describing t he two-iteration performance of Hopfield-like attractor neural network s with history-dependent Bayesian dynamics. We now extend this analysi s in a number of directions: input patterns applied to small subsets o f neurons, general connectivity architectures and more efficient use o f history. We show that the optimal signal (activation) function has a slanted sigmoidal shape, and provide an intuitive account of activati on functions with a non-monotone shape. This function endows the analy tical model with some properties characteristic of cortical neurons' f iring.