Distributed associative memories for high-speed symbolic reasoning

This paper briefly introduces a novel symbolic reasoning system based upon distributed associative memories which are constructed from correlation matrix memories (CMM). The system is aimed at high-speed rule-based symbolic operations. It has the advantage of very fast rule matching without the long training times normally associated with neural-network-based symbolic manipulation systems. In particular, the network is able to perform partial matching on symbolic information at high speed. As such, the system is aimed at the practical use of neural networks in high-speed reasoning systems. The paper describes the advantages and disadvantages of using CMM and shows how the approach overcomes those disadvantages. It then briefly describes a system incorporating CMM.