Dynamical networks for information processing

Abstract

Coupled evolution of state and topology of dynamical networks is introduced. Due to a well organized tensor structure, the governing equations are presented in a canonical form, and required attractors as well as their basins can be easily placed and controlled. This new class of dynamical networks can represent phenomenological models for self-organization in physics and biology. Applications of these networks to pattern recognition, associative memory, synthesis of models based upon observation data, detection of abnormalities and data compression are discussed. The difference between the proposed dynamical networks and neural networks is emphasized.