Rule-based reasoning using extended neural logic network

Abstract

Neural Logic Network (NEULONET) are studied in National University of Singapore to incorporate both the pattern processing capability of Multi-layer Perceptrons and the logical inference capability of Boolean Logic Inference Networks within a single frame of neural network environment. In this paper, a few extensions to the NEULONET are proposed. These enhancements to the network structure strengthen its ability to perform rule-based reasonings. The concept of network element (netel) is introduced. With netel, expert system rules may now be easily mapped into rudimentary NEULONETs. The resulting netel knowledge base inherits the semantic meanings of the expert system rules and the learning ability of the connectionist architecture.