built in 1995 protocols: HTTP FTP LDAP
  resulting hetrogeneity in result reporting/error handling

QoS critical to commercial adoption

In 1995, HTTP was not a good choice for a global grid -- GETs and PUTs on funny URLs with concatonated arguments.

WSDL's abstraction of endpoint implementation is very useful

Web Service's view of the universe is fundamentally static.
  Registered with UDDI,
  Operation is fundamentally stateless.
Open GRID Services Architecture worries about creation and management of transient services.
  Services considered stateful.

We have a set of service description conventions, some of which are being fed back into Web Services (ref meeting in Paris).

Semantic Web tech can help track data deriviation through manipulations.

[EGP - does GRID speak BGP?]

Matchmaking protocol - match classified adds


kessleman: cache static parts of the query and just pass on the dynamic parts?
mark: haven't thought about that.


12:15 - Brian McBride

Q: Ian Horrocks: big diff between tolerant and oblivious
A: didn't mean to imply that tolerance and precision/formalism are inconsistent. I'm just concearned that the formalism is standing in the way.

Q: what about local web communities?
A: keep doing stuff in the local communities but ask the global questions.

Q: Kesselman: When we started the Global Grid project, we rigorously followed the 80/20 rule. The folks who said "Everything needs to be glaobal and atomic" are all dead now.

15:00 - Bringing together semantic web and web services
  - Xpath ontologies stored in MDL (meaning definition language) files.

Q: You uses classes and types in MDL. what if you use XmlSchema instead of XML?
A: XmlSchema can only represent one view of one domain. want to combine views of two domains. XSLT is also not sufficient as long as RDF is so incompatible with RDF.

15:20 - Concurrent Executiopn Semantics of DAML-S with Subtypes

[Q: given a precondition, how do i find a service that can satisfy it?]

Q: linking DAML-S ontologies is done with subsumtion rules.
A: trying to import the utility of DAML+OIL

Q Benjamin Grosof: Is there a way to simplify the ?
A: there are a number of differences:
     no variables
     
Q: why not choose BTSB or label transition system that allows one to model more complex systems
A: our lang is based heavily on pi-calc and CCS.

15:40 - Semantic Matching of Web Service Capabilities
(working with Taka?)

Requestor -> Provider - not much distinction made as Requestor may be a Provider for another.
  [and how about actual collaboration?]

Soft Matching engine built on a DAML+OIL reasoner and uses subsumption

Honesty - a provider would want to say "i take any input and generate any output."
  managed by feedback scoring against false positives.

Output requested should be a subset of advertised outputs.

Browsing Schedules - An Agent-Based Approach to Navigating the Semantic Web
presented by Terry Payne

Q: interest in onology translation problem. how can you relate a speaker and a person together?
A: not trying to anything clever, just translating ontologies.

16:45 - Peter Patel-Schneider
treat OWL as an XML language
  parsed into triples, but they are separate.

[introduces dark triples]

Q: Can't you just use a well founded model?
A: paradox comes from rdf:type or from the logic layer.

Q: Hendler: I differ with these assumptions.
 You take the layer cake much more strenuously than Tim or anyone.
 How about some logic ontologies not to be mixed with others?
A: you describe a dark triples approaceh where the tripels lose their meaning at higher levels.

Q: Jeremy: I, of course, disagree with you.
 I need to have all triples in the domain of discouse.
 How about we mark areas off-limits for current discussion and work on the areas we know about?
A: This is dark triples.
Q: Jeremy: so when it's a problem, we give it back to the mathmeticians. We don't stop everything.

Q: Benjamin G: I find most of this discussion irrelevent. Treat it as syntactic encoding.
A: You propose all RDF  be dark triples. I want to get more use than that. RDF is good for part of it.

Q: What about simple XML?
A: That's my next talk.

17:15 - Notions of Indistinguishability

ALCI: DAML+OIL subset: booleans + oneOf + hasType

Q: Your equiv does not take into account order of ref'd docs.
A: Very often order does not matter.

Q: should probably add a temporal dimension.

Q: Jeremy: your slide had green dots in different cycles. Then you define position based on the (constrained) ability of the language.
A: just a handy definition and lang constraint.
[what was this about? the introduced language definition of "position" was counter intuitive but just accept it as a term and see what you can do with it.]

17:35 Representing Disjunction and Quantifiers

Q: you use types to represent the connectives

Q: you have RDF with its own model theoretical semantics and you add more expressions with their own semantics. Are there interactions?
A: [didn't really get this one - sorry]

17:55 - Nonmonotonic Rule Sustems - Grigoris

defaults imply non-monotonicity
exception rules stronger than defaults - defeasible reasoning.

Q: [dispute about poly or linear computation time]

18:10 - SquishQL - Andy Seaborne

Q: why don't they use datalog?
A: Can remove pieces from a complex system or build up from a simple system to get one with the desired expressivity. We are doing the latter.

Q: Why no RDFS query lang?
A: RDQL does to inference.
   RDFS entails virtual triples.

Q: this lang seems to make deep joins. does it scale?
A: yes if you choose your queries well.
 I've been working on sparse data sets.
 Need to work on hash join algorythms.