A model of associative memory based on adaptive feature-detecting cells

The self-organizing model of feature-detecting cells of S. Amari and A. Takeuchi (Biol. Cybern., vol.29, p.127-36, 1978) is expanded by adding an inhibitory cell with a nonlinear characteristic. The generalized model can form the feature-detecting cells of each pattern, even if the patterns have an inclusion. It is shown that, by varying only a learning parameter in the model, various classifications depending on the similarity can be performed. A model for associative memory based on adaptive feature-detecting cells is proposed. The model is composed of three subsystems: a self-organizing system of feature-detecting cells, a unifying system, and a self-associative system. It has a feedback mechanism on the recalling process. The learning and recalling procedures are described, and the properties of the model are studied. The behavior of the model is examined through computer-simulated experiments