Anytime Inference in Probabilistic Logic Programs with TP-CompilationShow others and affiliations
2015 (English)In: Proceedings of 24th International Joint Conference on ArtificialIntelligence (IJCAI) / [ed] Qiang Yang; Michael Wooldridge, AAAI Press, 2015, p. 1852-1858Conference paper, Published paper (Refereed)
Abstract [en]
Existing techniques for inference in probabilisticlogic programs are sequential: they first computethe relevant propositional formula for the query ofinterest, then compile it into a tractable target rep-resentation and finally, perform weighted modelcounting on the resulting representation. We pro-poseTP-compilation, a new inference techniquebased on forward reasoning.TP-compilation pro-ceeds incrementally in that it interleaves the knowl-edge compilation step for weighted model countingwith forward reasoning on the logic program. Thisleads to a novel anytime algorithm that provideshard bounds on the inferred probabilities. Fur-thermore, an empirical evaluation shows thatTP-compilation effectively handles larger instances ofcomplex real-world problems than current sequen-tial approaches, both for exact and for anytime ap-proximate inference.
Place, publisher, year, edition, pages
AAAI Press, 2015. p. 1852-1858
Series
IJCAI International Joint Conference on Artificial Intelligence, ISSN 1045-0823
National Category
Computer and Information Sciences
Identifiers
URN: urn:nbn:se:oru:diva-91829ISBN: 9781577357384 (print)OAI: oai:DiVA.org:oru-91829DiVA, id: diva2:1554815
Conference
24th International Joint Conference on Artificial Intelligence (IJCAI 2015), Buenos Aires, Argentina, July 25-31, 2015
2021-05-172021-05-172021-05-18Bibliographically approved