Contributions to reasoning under uncertainty in a possibilistic framework
Our work is related to knowledge representation and reasoning in Artificial Intelligence. Uncertainty and inconsistency are two aspects that affect data and knowledge in many domains. Despite the multitude of formalisms and approaches proposed to represent and reason with uncertain, incomplete or partially inconsistent information, there are still several open problems when it comes to using these approaches in practice. The main objective of our work is to make contributions through compact and flexible possibilistic representations. At the level of representation, we have proposed flexible extensions to possibilistic graphical models and possibilistic knowledge bases, in particular to representations based on imprecise quantification of uncertainty with intervals or sets. At the level of reasoning, we have studied conditioning and inference in these extended possibilistic representations and we have proposed efficient syntactic counterparts. Finally, we have proposed many contributions to reasoning with prioritized and partially inconsistent information and we have illustrated this in two applications. The first one concerns the querying of heterogeneous and large databases with assertional parts affected by uncertainties and possibly conflicts, while the second one, in the domain of computer security, concerns the revision of the predictions of a classifier to comply with some constraints and objectives of the domain. The presentation will end with some conclusions and perspectives for future work.
- Salem Benferhat (Supervisor), University Professor, CRIL - University of Artois, France
- Lluís Godo Lacasa (Rapporteur), Research Director, Artificial Intelligence Research Institute IIIA - CSIC, Spain
- Gabriele Kern-Isberner (Rapporteur), University Professor, Technische Universität Dortmund, Germany
- Pierre Marquis (Reviewer), University Professor, CRIL - University of Artois, France
- Philippe Leray (Reviewer), University Professor, LS2N - University of Nantes, France
- Henri Prade (Rapporteur), Research Director, Toulouse Institute of Computer Science, France