Measuring Inconsistency in Probabilistic Knowledge Bases
Resumo
In AI, inconsistency measures have been proposed as a way to manage inconsistent knowledge bases. This work investigates inconsistency measuring in probabilistic logic. We show that previously existing rationality postulates for inconsistency measures in probabilistic knowledge bases are themselves incompatible and introduce a new way of localising inconsistency to reconcile these postulates. We then show the equivalence between distance-based inconsistency measures, from the AI community, and incoherence measures, from philosophy, that are based on the disadvantageous gambling behaviour entailed by incoherent probabilistic beliefs (via Dutch books). This provides a meaningful interpretation to the former and efficient methods to compute the latter.
Referências
de Finetti, B. (1930). Problemi determinati e indeterminati nel calcolo delle probabilità. Rendiconti Reale Accademia dei Lincei, 6:367–373.
Halpern, J. (2003). Reasoning about uncertainty.
Hansen, P. and Jaumard, B. (2000). Probabilistic satisfiability. Handbook of Defeasible Reasoning and Uncertainty Management Systems: Algorithms for uncertainty and defeasible reasoning, page 321.
Hunter, A. and Konieczny, S. (2008). Measuring inconsistency through minimal inconsistent sets. In 11th International Conference on Principles of Knowledge Representation and Reasoning (KR’08), pages 358–366.
Nilsson, N. (1986). Probabilistic logic. Artificial Intelligence, 28(1):71–87.
Potyka, N. (2014). Linear programs for measuring inconsistency in probabilistic logics. In Fourteenth International Conference on Principles of Knowledge Representation and Reasoning (KR-14). AAAI.
Schervish, M., Seidenfeld, T., and Kadane, J. (2002). Measuring incoherence. Sankhyā: The Indian Journal of Statistics, Series A, pages 561–587.
Thimm, M. (2013). Inconsistency measures for probabilistic logics. Artificial Intelligence, 197:1–24.
