Categorization and generalization in the Hopfield model
Miguel C. Branchtein, Jeferson J. Arenzon · Journal de Physique I · 1992
The capability of the Hopfield model in generalizing concepts from examples is studied through numerical simulation. Noisy examples are taught and we test if the network grasps the underlying concepts. Several parameters are modified along the simulation: the number of concepts and examples and the initial noise in the examples. It is obtained the first order transition from the retrieval phase to the categorization phase predicted by the mean field theory.