ABSTRACT
The John Hopfield model is in fact an academic exercise—instructive from a theoretical point of view, but not very useful for practical applications. This chapter describes variants of the Hopfield model and analogous techniques to overcome some of the limitations. It also describes different ways to increase the capacity of the network, and discusses the effects of weakening certain hypotheses on the connectivity which are incompatible with neurobiological observations. The chapter also discusses variants which have sequences of references as attractors. The constraint of symmetrical connections imposed by the Hebb rule is difficult to imagine in a natural system such as the brain. Recall that the perceptron algorithm leads to an asymmetrical connection matrix. The methods based on the algebraic approach are more powerful and better adapted to the concrete problems of recognition and associative memory than the Hopfield model.
