Retrieval process of an associative memory with nonmonotonic input-output function

The retrieval process of an associative memory with a general input-output function is studied by means of a signal-to-noise ratio analysis. A set of recursion relations for macroscopic variables is derived to describe the time development of the network. By taking the equilibrium limit of the recursion relations, it is found that a certain type of nonmonotonic input-output relation of a single neuron yields a more enhanced memory capacity than the conventional monotonic relation. This result, combined with Monte Carlo simulations, confirms the prediction of Morita et al., and reveals new aspects of the retrieval process