Associative Memories that can form Hypotheses: Phase coded Network Architectures

摘要: Nonlinear associative memories as realized, e. g., by Hopfield nets are characterized attractor type dynamics. When fed with a starting pattern they converge to exactly one of the stored patterns which is supposed be most similar one. These systems cannot render hypotheses classification, i.e. several possible answers given classification problem. Inspired C. von der Malsburg’s correlation theory brain function we extend conventional neural network architectures introducing additional dynamical variables, so-called phases, for each formal neuron in net. The phases measure detailed correlations activities neglected architectures. Using simple selforganizing networks based on feature map algorithms present an memory that actually capable forming classification.