Optimising synaptic learning rules in linear associative memories (Q1182329)

A model of associative matrix memory incorporating 4 numbers, defining a learning rule, is considered. The maximization of the signal-to-noise ratio is used as a criterion for optimization of the form of the learning rule of the model. A well known covariance rule is shown to be optimal. Two other rules, previously suggested, are shown to be asymptotically optimal. All three rules, however, are shown to be strongly biologically irrelevant. Some explanation for this discrepancy and further discussion is proposed.