- From: David Martin <martin@AI.SRI.COM>
- Date: Thu, 03 Jun 2004 16:35:51 -0700
- To: public-sws-ig <public-sws-ig@w3.org>
apologies for late posting ... Minutes of SWSL teleconference May 13 2004 Attendees: Michael Kifer David Martin Sheila McIlraith Rick Hull Daniela Berardi Steve Battle Bijan Parsia Benjamin Grosof Michael Gruninger Jienwen Su (new member) Review of upcoming F2F agenda: Sunday, 23 May: 1200-1300 Registration 1300-1330 Plenary Session - Administrative Remarks 1330-1500 SWSA and SWSL Sessions 1500-1530 Break 1530-1800 SWSA and SWSL Sessions Monday, 24 May: 0830-0900 Registration 0900-1030 SWSA and SWSL Sessions 1030-1100 Break 1100-1200 SWSA and SWSL Sessions 1200-1300 Lunch 1300-1500 SWSA and SWSL Sessions 1500-1530 Break 1530-1700 SWSA and SWSL Sessions 1700-1800 Plenary Session - Outbriefs 1800 Adjourn Sunday will be use for presentation of the remaining straw proposals. On Monday we are keeping the time open (and unstructured for now) for discussion of the straw proposals. For those who can't come, teleconferencing will be possible. ===== FLOWS (straw proposal) presentation - See presentations slides linked from here: http://www.daml.org/services/swsl/straw-proposals/ [Note: The presentation itself is not captured here, but (most of) the questions, clarifications, and discussion is captured. Apologies for sketchiness.] Sheila presenting (slides 1-12) Benj Q: what does "explicit repn" mean for msgs/dataflow - would generic FOL/rules lang qualify, with just a set of predicates in that? A: yes slide 4 = BG: drawbacks of FOL include: not extensible to nonmon commercial reasoners not well-developed slide 5 - Question: what's the relationship between PSL and first-order logic (FOL)? SM: PSL is all first-order, but with a non-standard model complex actions are first-class objects MK: Golog is a programming language; PSL is not BG: what is "writability" (on the slide)? Answer: "writability" is guidance on how to most effectively express things Comment: PSL is expressed in KIF; not a Web language Comment: PSL is lacking in modularity slide 6 - Note: WSC on slide 6 means WS Composition Q: how is nonmon brought in? A: FLOWS is a language in which other types of reasoning can be projected Comment: circumscription is awkward, starting with FOL SM: use circumscription to understand the semantics (only) BG: predicate completion, stratified logic programs can be captured using circumscription SM: answer completion - work ongoing beyond LP; finding nice solutions to closure issues Q by Benj and Michael K: how do you hope to map nonmon formalisms into FOL? A: one way is to add a 2nd-order circumscription axiom, then show for example that it's equivalent to a predicate completion, a la Reiter approach to solving the frame problem Michael G: the idea is that FLOWS has a first order ontology and first order statements, but reasoning -- e.g., nonmon -- could be in a non-FOL logical KR Benj: i.e., can PROJECT other reasoning, e.g., nonmon, into FLOWS FOL? A: yes, that's well put discussion: can represent predicate completion / stratified LP / answer set programming / othre circumscriptions in this way, using some known results slide 7 - BG: Isn't this proposal somewhat polyannish? SM: yes BG: there are *too many* results; worried that FLOWS doesn't represent any decisions; more guidance will be needed; each task is really hard MG: more emphasis on rep. than reasoning MG: tension between high-expressive and lack of effective reasoning; looking for median: start with expressivenss; look for tractable subcases slide 10 - MG: ontology aspect of FLOWS: identify subset of PSL corresponding to owl-s, extended with messages, dataflow, negotiation (potentially ext of PSL) MG: But PSL has ignored messages RH: We will commit to a particular approach will map concepts from OWL-S into a subset of PSL, plus represent/map some other things like messages into an extension of PSL - have to decide exactly how to do that ontologically Benj Q/obs: isn't there a tension between using generality of FOL to represent existential consequent expressions, e.g., on slide 13, versus keeping things simpler wrt fitting into LP or tractability? E.g., would it be OK here to skolemize? A: yes, there's a tension, we have work to do slide 11 - Michael G. presenting (slides 13-16): slides 13-15 - BG: how much violence would skolemization do to your approach? MG: good question; analogous to considering a prototypcial occurrence; implies reasoning by cases BG: some of these decisions will have important implications re: reasoning, tractability BP: What's the relationship between these descriptions and services? What's the relationship to golog? MG: buy_product is a service process descriptions can be considered generalizations of golog programs constructs in golog are macros for temporal formulas - which can be directly specified in PSL Q: How to represent equivalence between 2 different services? A: equiv. relation would be yet another macro in golog SM: FLOWS is nicer than golog from a formal perspective golog didn't go deeply into model-theoretic semantics slide 15 - Benj comment: overall, I'm worried that for putting out standards proposals, we want to be incremental up from what we know how to do in a scalable and practical way, including scalability from viewpoint of authoring (of service description knowledge) not just of computation - Michael G: we started up from the scenarios - Benj: those aren't necessarily the scenarios that we should weight most heavily, if we don't know how to do those scenarios in a scalable practical way Michael K Q: will these service descriptions be generated from programming code, i.e., from executables? A: good question! but don't want to be restricted to executable, of course Michael K: suppose we have a language that is executable and do simulation of execution in that -- rather than more general reasoning? wrt tractability: Benj: we should identify what are the already known tractable cases Michael G: yes Benj Q: what about FSAs, for example -- when are they tractable? Rick A: yes, we're looking at that; often exponential time; but worst-case exponential time sometimes is practically tractable, e.g., Description Logic techniques in Daniela's approach Sheila A: also depends whether the exponential parameter is big in practice, e.g., classical planning technique application results today; e.g., in WS Compositon argue often are working in very broad but very shallow trees -- which keeps the practical search space small enough Rick Hull presenting (slides 17-end): slide 17 - PSL approach allows to quantify over complex activities can embody FSAs into psl slide 19 - BG: head existentials might be problematic RH: there are some constructs that don't naturally go into LP framework; we may not need all of LP BG: some tasks use LP, but reachability, e.g., might be beyond LP BG: ontological commitments in FOL - that's nice, but commitments to sublanguages will be needed (to be useful and taken seriously) MG: FT usecase - realistic scenarios call for expressive approach BG: historically FOL efforts haven't gotten very far: authoring scalability should start with what we know how to do; then work towards more generaility. beware scenarios that require reasoning that's not well-understood MG: this is not a lang looking for problems - we are starting from use cases MK: how do you envision these specs (slides 15, 19)? statements about a C program? executable? RH: expressions or assertions MG: The goal here is to reason about services; not embed SWSL into implementation (so not necessarily executable) MK: could be both executable and support reasoning RH: requirement that everything reducible to an exec. spec isn't good e.g. CDL is not executable. SWSL should embrace everything CDL can express. MK: FOL a non-starter; looking for subsets what if we have an executable language to permit simulation? MG is simulation more tractable? MK: yes MG, SM, BG: simulation is another kind of reasoning BG: It's hard to sell "tractable subsets to be named later" MG: tractability can talk about tractability of different *theories* (class of processes) BG: FSAs: reformulated, will be tractable MG: weakness - computational subclasses not identified Q: under what circumstances are FSA (BS) tractable RH: exp. time algo. is pragmatically tractable worst-case is exponential pragmatic question: heuristics SM: planning is pspace or np-complete but large problems are handled very quickly search space for WS composition tends to be small BG: deep KR - XML - presentation The Who: Bijan Q/issue: who's going to write/author SWSL descriptions? Rick A: the kind of person who will write WS Choreography Bijan Q: are there more people working with process algebras than with FOL for process specifications? Michael K: way more are familiar with FOL altogether Michael G: in case of PSL, it's the s/w vendors Michael G: for SWSL, will it be service providers writing SWSL? Benj: we can call this issue: conceptual familiarity for target authors; David: let's identify issues for discussion at F2F Benj: who is the target author community would be a great one, I think that should include service providers not just s/w vendors
Received on Thursday, 3 June 2004 19:34:11 UTC