A comparative study of sparse associative memories (Q315647)

One starts with the presentation of some neural models of associative memories with sparse information: the Hopfield model, the Amari model, the Willshaw model and the Gripon-Berrou (GB) model. The purpose is a comparative study of these models regarding the influence of their parameters on the storage capacity. In the section ``Wrong messages and a first bound on the storage capacity'' the question of the right order for the storage capacity is discussed. In the section ``Stability and error correction'', one concentrates on lower and upper bounds for the number of patterns that can be stored in these models such that stability is achieved and errors in the input can be corrected. In the section ``Dynamical properties of the models'', the convergence of the proposed dynamics is investigated. The paper ends with simulation results.