Approximate reasoning and prototypical knowledge

AbstractPrototypical knowledge plays an important role in many representation formalisms, particularly in those used to implement diagnostic expert systems. The aim of this paper is to present a general technique for performing approximate reasoning in systems based on prototypical knowledge representation. We extend some of the formalisms currently used in knowledge representation to the case of uncertain knowledge by defining a more flexible mechanism for representing admissible values of a given feature (via fuzzy logic), by designing general evaluation mechanisms for the match between prototypical knowledge and incomplete and/or partial data, and by introducing the notion of relevance in order to avoid the choice between strictly necessary conditions and strictly sufficient conditions