Invariant perception for symmetry related patterns

A powerful tool for handling configuration-space symmetries is presented. The authors have succeeded in devising a map that allows any neural network dynamics to be implemented on whole classes of equivalence under an arbitrary symmetry group. In certain maps part of the convergence of a pattern toward a memory is accomplished by the map itself. The authors believe this approach could eventually permit the whole process of recognition to be done at this stage for a wide enough radius of attraction.<>