‘Quantum’ Parallel computation with neural networks

Correlation matrix memories have been successfully applied to many domains. This work implements a production system put forward in [Austin, 2003], to demonstrate its viability as an efficient rule-chaining process. Background information on rule-chaining and CMMs is given, followed by a review of the proposed production system. Throughout the iterative development process, experimentation is performed in order to investigate the effects of changing the properties of vectors used in this system. The results show that generating vectors using the algorithm proposed in [Baum, 1988] with a weight close to log2 of the vector length provides the highest storage capacity. The simple system implemented in this work performs rule-chaining effectively. This leads to the conclusion that the proposed production system is viable, and that this area warrants further work.