Pattern recognition and associative memory as dynamical processes in nonlinear systems

The authors present a formalism for associative memory and pattern recognition performed by the time evolution of a dynamical system. The patterns are treated as multicomponent vectors, as well as continuous functions in space and time. Equations of motion are derived from a nonlinear potential and transformed to a low-dimensional subspace, where an appropriate form for neural nets is given. The example of rotated patterns shows how the formalism works in that case.<>