Download presentation
Presentation is loading. Please wait.
Published byHelena Hodges Modified over 9 years ago
1
McGuinness Oct 17, 2002 1 Eurasian Pygmy Owl -Glaucidium passerinum – picture Romek Mikusek Explanation BOF Moderator: Deborah McGuinness
2
McGuinness Oct 17, 2002 2 BOF Goals n Identify needs for explanation/proof work from the daml community n Identify who has plans to work on DAML+OIL/OWL explanation/proof work n Generate list of actionable items. n Significant issues n Recommendations/plan of action n Discussion of “good” proofs (explanations)
3
McGuinness Oct 17, 2002 3 Motivation n Trust disclosure – trust inference rules, premises, recency, inference engine, … n Interoperability – multiple owls interacting, proof composition, … n Proof reuse – individual reuse, individual refinement, group reuse/refinement….
4
McGuinness Oct 17, 2002 4 Issues n Variable granularity (lcf, pruning, etc) n Degree of annotation for human readability (human paraphrase in addition to machine readability) n Agents should be able to verify proofs n Proofs should be “nestable” and “queryable and/or reexecutable” n Proof language should be ubiquitous n Proofs should be incremental n Confidence in proof steps should be expressible n Daml-compliant inference engines should respond to client requests with “reasonable explanation” in the daml language n Identifying rules (naming,…)
5
McGuinness Oct 17, 2002 5 Issues, continued n System needs to be extensible with respect to inference rules, … n Should include black box algorithms with trust annotation on black box n What is trust? Trust of inference rules, agent (might have additional granularity), source. Look at solutions such as delegated trust in n3 n Proofs with true but not useful information- need techniques for pruning n Are there techniques like Google’s reverse links that can help? n If you want a “good explanation” that may impact the proof spec. And what is a “good explanation” n Where do ground facts ground out (what granularity) n Provenance or other annotations on information
6
McGuinness Oct 17, 2002 6 Plans/People n W3C – Contact: Berners-Lee, Connolly,… l Cwm will handle explanation and validation sometime n Stanford – Contact: McGuinness l http://www.ksl.stanford.edu/projects/daml/Proof/ l DAML+OIL/OWL specification of proofs, examples, challenges… l Implementation of explanation/proof browser for proofs/inference webs l JTP reasoner is being made compatible with proof spec n Cycorp – Contact: Steve Reed l Explanation implementation of Stanford's design initial test subsumption, l why assertion NOT assertable and make recommendations n Agfa - Contact: Jos de Roo
7
McGuinness Oct 17, 2002 7 Plans/People cont. n Teknowledge - Contact – Adam Pease l Proof pruning, coordination n UWF/IMHC- Contact - Pat Hayes l Designing proofs for good explanation n Northeastern University - Contact - Mitch Kokar l Ontology for inconsistencies in DAML n get pointer from pat on lcf…. n McGuinness will maintain list – send mail to dlm@ksl.stanford.edu to update. dlm@ksl.stanford.edu
8
McGuinness Oct 17, 2002 8 Actionable items n Build and maintain list of contacts on explanation work on RDF- compliant systems – McGuinness n Build a test ontology and set of test cases l Possible domains – wine ontology, …. n Draft DAML+OIL/owl spec for shareable proofs and architecture n Obtain comments on draft spec for shareable proofs - - Karlsruhe, RKF(SRI, KM, Northwestern, Boeing, …), Cycorp, … n Interoperability tests (at least Stanford and Cycorp) n List of heuristics for pruning/presenting explanations
9
McGuinness Oct 17, 2002 9 updates to dlm@ksl.stanford.edu Barn Owl Tyto alba; picture Andy Harmer
Similar presentations
© 2025 SlidePlayer.com. Inc.
All rights reserved.