ABSTRACT
This chapter examines various extensions and generalizations of Hopfield model. It considers straightforward variations, including the modifications needed to cope with correlated patterns. Then, the chapter treats networks of continuous-valued units with dynamics described by differential equations. It sketches a couple of hardware implementations of Hopfield-like networks, one electrical and one optical. The chapter discusses models that recall sequences of states, so that the usual point attractors are replaced by limit cycles or by more complicated dynamical trajectories. It discusses how the associative memory properties of the basic Hopfield model carry over to various more complicated situations. In many cases the results are qualitatively similar to the simple case, but changes occur in quantitative values such as the capacity ac=pmax/N. This chapter considers the perturbation of the connection strengths away from those given by the Hebb prescription. This may be of practical importance when trying to build a network with connections of limited precision or range.
