ppy/chat-transcript_20100722b_edited.txt // PeterYim: . Welcome to the Ontolog Invited Speaker Presentation - Professor IanHorrocks - Thu 2010.07.22 (2EQX) * Session Host: Dr. LeoObrst (Ontolog; MITRE) (2F37) * Invited Speakers: Professor IanHorrocks (University of Oxford) (2F3 * Presentation Title: "Scalable Ontology-Based Information Systems" (2F39) Please refer to details on session page at: http://ontolog.cim3.net/cgi-bin/wiki.pl?ConferenceCall_2010_07_22 . anonymous morphed into IanHorrocks anonymous morphed into JulitaBermejoAlonso anonymous11 morphed into Mike Hewett anonymous1 morphed into Peter Chan anonymous2 morphed into Jess Turner anonymous3 morphed into ElisaKendall IanHorrocks: Slides are at: http://ontolog.cim3.net/file/resource/presentation/IanHorrocks_20100722/Scalable_Ontology-Based_Information_Systems--IanHorrocks_20100722.pdf anonymous3 morphed into Sina Madani anonymous6 morphed into FabianNeuhaus anonymous1 morphed into GuoqianJiang anonymous7 morphed into Mikhail Soutchanski anonymous1 morphed into PierluigiMiraglia anonymous6 morphed into AmandaVizedom anonymous31 morphed into BobbinTeegarden anonymous3 morphed into Cui Tao anonymous4 morphed into Peter Wagner anonymous6 morphed into Melanie Courtot anonymous1 morphed into PavithraKenjige anonymous7 morphed into Sudarsan Rachuri anonymous4 morphed into Donald Kretz ArturoSanchez: LOL JoelBender: The place is going to the dogs! IanHorrocks: As a DL person I am well used to that sort of response anonymous11 morphed into Mark Johnson anonymous5 morphed into Peter Bruhn Andersen anonymous1 morphed into ElizabethFlorescu anonymous6 morphed into MyCoyne JohnSinger: trying to download the pdf, but its going slow PierluigiMiraglia: Hi Amanda & Peter AmandaVizedom: Hello PierLuigi, good to see you here. anonymous6 morphed into Rob Hausam anonymous7 morphed into MarciaZeng anonymous3 morphed into RaviSharma FrankChum morphed into FrankChum anonymous5 morphed into SimonSpero ToddSchneider: Peter, Ian is fading in and out. ArturoSanchez: OK here ... JohnSinger: ok here Michael Riben: ok here Clarence Dillon: ok here. Mike Hewett: Sounds pretty good to me. A little weak. But no fading. anonymous1 morphed into Lynn Leitte anonymous4 morphed into Bruce Peoples anonymous3 morphed into Jae Hyun Lee Michael Riben: what slide is he on? EricChan: slide 12 Michael Riben: okay..thanks anonymous1 morphed into Anantha Narayanan ArturoSanchez: @Prof. Horrocks: Although the ontologies are not what the systems are supposed to do, but what the data the systems will use are all about. Do you agree with this? AmandaVizedom: Comment for discussion later, perhaps: Ian, you've just remarked (on slide 12) that you see the low complexity of DLs as desirable from a performance perspective, and from a perspective I might paraphrase as tractability of representation (degree to which tiny changes ripple out in changes not easily anticipated). This may be true on very small, narrow domain, controlled representation spaces. In my experience, however, the actual result of KR language simplicity is more complex, and often the opposite. In complex, RW problem-driven representation, a lower-expressivity language is *at best* like a logical language with only 3 minimal operators. In theory, you can represent whatever you need; in practice, doing so may require extraordinarily complex expressions and mental contortions. In the ontology case, lower expressivity has a significant cost, increasing with number of collaborators or users and with scale of task. The more ontologists have to stretch the language extra-ordinarily, the worse performance becomes and the more problems arise for collaboration and interoperation. Have you considered this cost of low-expressivity, as in DLs? Have you seen DLs used for very large scale and/or broad domain real world cases? BillHogan: perhaps it's only an artifact of my medical training, but I don't think of all heart diseases as being a subtype of vascular disease. FrankChum: There are s lot of business cases for the O&G industry in ontology-based information system! Sudarsan Rachuri: How to we define a notion of quality for ontology? CoryCasanave: These are good real world examples. As you said, the real world and our understanding is not perfect. Can you speak to the fragility of description logics in the face of a less than perfect ontology. FrankChum: Does OWL 2 has Full/Lite DL for complexity and computability? LeoObrst: Ian, along the lines of Amanda's question, some believe that there are 2 problems related to KR expressiveness: 1) having a rich enough expressiveness in your modeling/KR language to express what you need to for your domain, and 2) having an efficient enough representation to do automated reasoning in near real time. And so folks think that you need: A) a FOL expressive language for representing your ontology, B) a knowledge compilation process to transform (A) to (C), probably losing information, and C) an efficient runtime representation that enables fast automated reasoning. Some say that DLs try to shoehorn the (1) and (2) language into the same language, and so there is no B. AmandaVizedom: Another question about the DL expressivity trade-off: How common is it, really, to have a sufficiently complex and demanding case that completeness and decidability are going to be serious issues, and yet *not* need to create constraints that terminate queries and inference well before the "some finite amount of time" in which decidable systems can guarantee results? ToddSchneider: Ian, how do implementations of tableau algorithms make use of logical independence? PeterYim: we've skipped from slide 31 to slide 50 ... now at slide#51 PeterYim: skipping to slide#68 PeterYim: skipping to slide#72 MichaelGruninger: What is the citation for the experiments in slide 68? What were the queries/reasoning problems? Were all of the ontologies written in EL? PeterYim: @Michael, if you can locate the literature Ian referred us to, would you kindly paste it onto this chat board, please MichaelGruninger: @Peter: The citation I was looking for is: Consequence-Driven Reasoning for Horn SHIQ Ontologies by Yevgeny Kazakov ... In Proceedings of the 21st International Conference on Artificial Intelligence (IJCAI 2009). Pages 2040-2045. , 2009. PeterYim: @MichaelGruninger - thank you, Michael PeterYim: skipping to slide#89 LeoObrst: Does saturation include quick computation of subsumption ala Ait-Kaci et al's methods, i.e.,optimal hierarchical encoding of types in bit encodings, prime numbers, etc.? LeoObrst: oops. Mike Hewett: What slide is Ian on? ArturoSanchez: 89 PeterYim: @JimDisbrow ... we are holding off Q&A until the prepared presentation is through ... you can try capturing your question on the chat-board as a placeholder for now anonymous morphed into JimDisbrow AmandaVizedom: OWL 2 question for later, unless you answer it anyway before then: Slide #53, the profiles are distinguished by whether they are optimized for size of ontology or size of data sets. But of course, size is not the only major distinction in application contexts. Prior versions of OWL have proven much more or less usable depending on a variety of other application context characteristics, e.g.: complexity and variety of relationships to be represented; whether use is within a single domain/community or crosses multiple domains/communities; complexity of reasoning required (mostly subsumption or not, for example); whether the ontology is used to represent instance-level data or to represent, at a meta-level, the semantics of data stored elsewhere, or both; whether the ontology must support machine usability, human usability, or both, and so on. Were these and other features of the (increasingly varied!) variety of ontology applications considered in the development of OWL 2? If so, what were some of the effects on the final form of the language? PeterYim: will the "anonymous" participants please click on the "Settings" button (top center) and morph into your real name, so we can tell who you are, and properly attribute your contributions anonymous1 morphed into KathyEllis StavrosMacrakis: Is OWL 2 QL expressive enough to handle arbitrary-length chains, e.g. "A is a direct or indirect ancestor of B"? PierluigiMiraglia: Question on QL: is the optimization expected for general inference engines, or only if reasoning is delegated to RDBMS? I.e., do I see a gain if I use OWL QL in a Pellet or FaCT based application? JohnSinger: how is IBM involved in this? ArturoSanchez: @Prof. Horrocks: can you mention classes of problems for which OWL 2 is particularly good for, and also classes of problems for which OWL 2 is particularly bad for? Sudarsan Rachuri: How to we define a notion of quality for ontology? AmandaVizedom: Slide #96: Indeed, the infrastructure mentioned regarding privacy and information hiding is partially overlapping with infrastructure needed for a variety of kinds of KR and management of ontology content. Provenance issues are important in many interoperability applications; cross-community uses may also call for considerable capture of knowledge relevant to dynamically selecting relevant ontology content. Many people have resorted to reification of triples for this. Others have built complicated containers for pre-defined contexts and ways of moving between them. None of that has been great. Does OWL 2 significantly improve on this? StavrosMacrakis: I have to get off the call -- PierluigiMiraglia will follow up on my question. MyCoyne: hello ArturoSanchez: Great session. Thank you! ToddSchneider: I also need to leave for another meeting around 3 PM EDT. Sudarsan Rachuri: Thanks for the excellent presentation. Bye AmandaVizedom: Will ask this, but capturing to chat as well: Excellent point about mismatch between OWL "constraint" behavior and what most people think of as "constraint-checking." Any comment on use of SPARQL and other query languages to off-load some of these features and requirements to another level? PeterYim: please hold off "OWL 2" specific questions until our two "OWL 2" sessions next Thursday and the following one CoryCasanave: Follow up on fragility: Your response assumes an ontology under a single authority - in a federated knowledge base this is not practical. Can DL be used in an open, federated environment like the web. AmandaVizedom: Follow-up to Cory's question: Indeed, I think much of the high payoff work in semantic interoperability -- even across communities and data sources within a single, large enterprise -- is a federated, not singly-controlled environment. Ontologies can be and have been used to enable more and better information sharing in such cases. But this requires considerable richness of representation to make contextual assumptions explicit, as well as considerable provenance and other metadata. This is a significant motivation for my earlier questions. CoryCasanave: My mike did not seem to work; I am wondering how Ian responds to Benjamin Grosof's assertion that a rule system is more capabile in an onology that may have local inconsistancies, and that such inconsistencies are inevitable - even in a selected subset of "the web". AmandaVizedom: Strong second to Brian's follow-up about reasoning about provenance. Annotations are not sufficient, and one does indeed need to represent relevant provenance details and reason about them as relevant to specific uses or users. The folks creating the data in once source are not going to conveniently note that their data doesn't meet the standards of my need. But an intermediate infrastructure can reason about whether their data meets my provenance requirements. JoelBender: An outstanding session, thank you! CoryCasanave: Thanks you so much! FrankChum: Great session. Thank you!!! AmandaVizedom: Thank you, Prof. Horrocks. I look forward to the next two sessions. PeterYim: Great talk ... thank you VERY much, Professor Horrocks PeterYim: Thank you all for your participation PeterYim: Look forward to having all of you back here in the next two weeks, when Professor Horrocks will lead the two "OWL 2" sessions, see http://ontolog.cim3.net/cgi-bin/wiki.pl?ConferenceCall_2010_07_29 http://ontolog.cim3.net/cgi-bin/wiki.pl?ConferenceCall_2010_08_05 ... drop me a note (rsvp to ) if you are coming and haven't registered for those sessions yet, please PeterYim: @ALL: if what we do at ONTOLOG aligns well with your professional interest, join the community if you are not already a member of Ontolog - see: http://ontolog.cim3.net/cgi-bin/wiki.pl?WikiHomePage#nid1J IanHorrocks: Thanks to everyone for listening. Look forward to talking to you again next week. PeterYim: - session ended 12:31pm PDT - //