A Hybrid Trainable Rule-based System

In this paper we introduce a new formalism for rule speciication that extends the behaviour of a traditional rule based system and allows the natural development of hybrid trainable systems. The formalism in itself allows a simple and concise speciication of rules and lends itself to the introduction of symbolic rule induction mechanisms (example-based knowledge acquisition) as well as artiicial neural networks. In the paper we describe such a formalism and four increasingly powerful mechanisms for rule induction. The rst one is based on a truth-table representation; the second is based on a form of example based learning; the third on feed-forward artiicial neural nets; the fourth on genetic algorithms. Examples of systems based on these hybrid paradigms are presented and their advantages with respect to traditional approaches are discussed.