Results from the Gardner-Derrida-Mottishaw theory of associative memory.

General computable formulas for overlap and the other parameters are derived from the Gardner-Derrida-Mottishaw theory of associative memory on the Little-Hopfield model. The overlap is expressed in terms of integral of many-dimensional Gaussian functions. A method of approximation is developed to make numerical computation easy.

Analysis of the Recalling Processes of Associative Memory by an Integral Representation of the Sign Function.

The recalling processes of associative memory proposed by Amari and Maginu are analyzed without any assumption on asymptotic distributions of noise using an integral representation of the sign function and by introducing extra variables. General procedures for calculating mean of direction cosine a(t) between a certain memory s(1) and the state x(t) at time t are developed and the exact formulas for E[a(t)] are obtained up to t = 3. It is shown explicitly that E[a(3)] involves non-Gaussian effects caused by some self-coherences.