A neural architecture for fast rule matching

Abstract

This paper describes a simple neural architecture that can be used to match rules in knowledge based systems. The approach allows very large numbers of rules to be searched and matched using simple neural correlation matrix memories. The architecture is specifically designed to cope with inputs that may contain errors or be incomplete. Because the neural architecture is based on binary inputs and binary weights it is particularly applicable to fast operation on standard computers as well as specialized hardware. The paper describes the current implementation of the system, its advantages compared to other methods and the motivation that led to its design.