Application of chaotic automata to improve the pattern recognition ability of attractor neural networks

摘要

We present a preprocessor for attractor neural networks consisting of a chaotic cellular automaton. We show that this preprocessor improves on the ability to recognize correlated memories. The preprocessor, although chaotic, is reversible because the patterns have a finite number of neurons and appropriate boundary conditions were chosen. The performance of the present model was checked using numerical simulations in both the RS and the Hopfield models of neural networks. However, this is a very general approach that can be used in any other architecture with different learning rules. [References: 18]