A cellular neural network learning the pseudorandom behaviour of a complex system

Abstract

This paper presents the development of a Cellular Neural Network (CNN) architecture that is capable of learning the behaviour of a Cellular Automaton (CA) operating under local rule 30. Such a CA rule models the complex behaviour of a random system. The CNN was trained using the Levenberg-Marquardt approximation to Newton's method and convergence was achieved very fast. The proposed CNN was able to generalize efficiently and it can be used as a pseudorandom number generator. The CNN architecture proposed in this paper is especially suited to VLSI implementation due to its inherent regularity, modularity and parallelism and also, due to the locality of interconnections.