Representing and learning conditional information in possibility theory

Abstract

Conditionals (if-then-rules, default rules) are most important objects in knowledge representation and commonsense reasoning. Due to their non-classical nature, however, they are not easily dealt with. In this paper, we present a new approach to represent conditionals inductively in a possibilistic framework. The algebraic theory which underlies this approach proves to guarantee a most appropriate handling of conditional information. Moreover, this novel conditional theory is a very fundamental one, in that it can also be applied to guide possibilistic belief revision and gives rise to a new methodology to learn rules from data. © Springer-Verlag 2001.