Change Paths In Reasoning !
- 1. Change paths in reasoning! Raphael Volz FZI Forschungszentrum Informatik Universität Karlsruhe (TH) Karlsruhe, Germany 11.11.2007
- 2. Proposition We need a consensus benchmark Approximate is better than nothing Tractable languages make speed Incremental reasoning is smart
- 3. We need a consensus benchmark (1) Recent Performance Benchmark @ FZI Note: Joint work with Jürgen Bock, Qiu Ji, Peter Haase, Pellet performance close to Racer, Sesame preformance close to OWLIM Is this evaluation representative???
- 4. We need a consensus benchmark (2) Shortcomings of various benchmarks Source: Timo Weithörner et al., What‘s wrong with owl benchmarks, Proc. of 2nd int. Workshop on Scalable Semantic Web Knowledge Base Systems (SSWS 2006)
- 5. Need for a consensus benchmark (3) Improvements Achieved with TREC Source: E. M. Voorhees, TREC: Improving Information Access through Evaluation, American Society for Information Science and Technology Vol. 32, No. 1 Oct/Nov 2005 Consensus Metric Consensus Data Sets
- 6. Approximate is better than nothing (1) Ontologies in WATSON Corpus Source: Mathieu d‘Acquin et al., Characterizing Knowledge on the Semantic Web with Watson, EON 2007 Workshop, Busan, Korea We need to approximate with OWL DL reasoners
- 7. Approximate is better than nothing (2) Idea derived from FOL work Source: Hitzler and Vrandecic, Resolution-based approximate reasoning for OWL DL, in Y. Gil et al. (Eds.): Proc. of ISWC 2005, LNCS 3729, pp. 383–397, 2005.
- 8. Approximate is better than nothing (3) Idea derived from FOL work Source: Hitzler and Vrandecic, Resolution-based approximate reasoning for OWL DL, in Y. Gil et al. (Eds.): Proc. of ISWC 2005, LNCS 3729, pp. 383–397, 2005.
- 9. Approximate is better than nothing (3) Syntactic approaches ( possibly unsound ) Stuckenschmidt & van Harmelen 2002 Wache, Groot, & Stuckenschmidt 2005 Hitzler & Vrandecic 2005 Groot, Stuckenschmidt & Wache 2005 Hurtado, Poulovassilis, & Wood 2006 Approaches based on Knowledge Compilation Selman & Kautz 1991 Pan & Thomas 2007
- 10. Tractable languages make speed (1) Dez 2003 - DAML.ORG Corpus Source: Mathieu d‘Acquin et al., Characterizing Knowledge on the Semantic Web with Watson, EON 2007 Workshop, Busan, Korea Jul 2007 - WATSON Corpus Source: Raphael Volz, Web Ontology Reasoning with logic databases, dissertation, university of karlsruhe, 2004 Tractable Languages dominate (and will continue to do so)
- 11. Tractable languages make speed (2) Source: Markus Krötzsch, Sebastian Rudolph, and Pascal Hitzler; Complexity of Horn Description Logics Technical Report, Institute AIFB, University of Karlsruhe, 2007 Combined complexity of various DLs 1 1 1 1 3 # Available Reasoners (known to me)
- 12. Incremental reasoning is smart (1) Incremental answers is what we expect on the web Standard expectation for query answering on the web
- 13. Incremental reasoning is smart (2) Alternate possible interpretations of incremental reasoning Maintain state information from previous reasoning cycles when dealing with change to the KB (Pellet interpretation) Provide answers / results incrementally (anytime behaviour) Incremental reasoning saves work Source: C. Halaschek-Wiener et al. Description Logic Reasoning for Dynamic A-Boxes, 2006 DL Workshop, CEUR WS 189
- 14. The ideal Semantic WEB reasoning approach time Quality as a combination of soundness and completness Functional Qualities Measurable quality Recognizable quality Monotonicity Consistency Diminishing returns Interruptability Preemtability Source: Shlomo Zilberstein, Using Anytime Algorithms in Intelligent Systems, AI Magazine, Fall 1996 HAPPY TO DISCUSS WITH YOU !!!