On a model of associative memory with huge storage capacity (Q2410291): Difference between revisions

This is an investigation on how more patterns in an associative neural network can be stored. One mainly refers to the generalized Hopfield model introduced in the paper of [\textit{D. Krotov} and \textit{J. J. Hopfield}, ``Dense associative memory for pattern recognition'', in: Advances in neural information processing systems. Red Hook: Curran Associates, Inc. 1172--1180 (2016)]. In their model, Krotov and Hopfield introduced a more general, asynchronous updating rule, based on a smooth function \(F(x)\), the interaction function. In the present article, there is first a discussion on the number of stored patterns in the case of a polynomial interaction \(F(x)= x^n\). In the case of a generalized Hopfield model with the dynamics being defined by means of an exponential interaction function \(F(x)=e^x\), one proves that the model has an exponential storage capacity in the number of neurons.