A Heuristic Chaotic Neural Network: Candidate Model for Perception

Abstract

In this paper a new Chaotic Neural Network (CNN) have been made. This network contains desired number of interacting units and each one has its own chaotic dynamic and strange attractor caused by creating convex hull among output units. Having a special interaction characteristic, the model is able to create enormous different chaotic behaviors. Lyapunov Exponent and phase space plane criteria have been used for demonstrating discrimination between behaviors. Making use of convex hull for trapping generated outputs of each unit in subsequent iteration, its folding characteristic and stretching property of logistic function, emerging of arbitrary number of various strange attractors have been accomplished. Therefore, based on desired criterion, this network is able to assign each strange attractor to each sensory input. In other words the network has the ability of being a candidate for modeling perception.