IW Meeting 2011-02-17

From Inference Web

Jump to: navigation, search


Meeting info


  • Tim, Jim
  • Deborah (via cell in car; no internet)
  • Paulo
  • Cynthia
  • Geoff
  • Nick (regrets)
  • Patrick, Stephen (regret)


(administrative week)

  • Update regarding the W3C Provenance Task Force (D->P)
  • Google group requests (T->.)
  • Tim not attending 24th


Last week: Cynthia did data for set theory and one of process control TPTP problem. Geoff found one optimization to make Cynthia revealed bug, Geoff fixed IDV this morning. requested data for ...140+2 (?)

(review) last week's TODO: run single step combo on set theory example THEN: do multi step hill climbin combining with AB <--- still next to do next. THEN multi step on set theroy and process control et al. THEN: on suite of examples THEN write paper

TODO: Cynthia to make combining operation.

Paulo: From proof combination on logical proofs. We assume if we can compare two conclusions, we can replicate anywhere else. (Geoff confirms). Replicating in another domain, but which one? Paulo playing with USGS's data encoded in NetCDF (c.f. common data model). He sees easy way to get scientific data from one schema that has been used to derived other things. Common format -> common fields/variables (and same resolution). We should be ablet to say two data from different places are the same thing. Or come up with mechanism to say they are similar enough. If we have proofs, we have a mechanism to replicate the combining in scientific applications.

Geoff: Key is to provide Cynthia with data that identifies two data similar "enough".

Paulo: Has been wrestling with finding the scientific application that has the data amenable to proof combination. Different resolutions makes it challenging. They have conventions in science that can aid combining and connecting. It will be hard to capture the proofs.

Paulo: Concrete example: How many data have temperature data in them in the US? Temperature in Miami at this resolution. (from 5 sources is doable wiht current technology). But what do you do with those temperatures. How to increase the level of scientific insights? Interdisciplinary work and connecting their data.

Goeff: All that is needed: 1) Encode into PML 2) similar enough 3) Run Cynthia's proof combining algorithm

Tim: Is this a use case?

Paulo: Peter's paper. Abduction. TODO: what paper?

TODO: which use case did Paulo suggest to give us a few weeks ago?

Deborah has form that she and Peter use regularly.

Paulo: Lightning datasets. 4/5 variables. (wind direction) both might have pressure and temperature. Free using netCDF and HDF. Make assumptions that "temp" is "temp" across datasets. (mentions opendap - item later in agenda).

TODO: Deborah and Tim will look at Paulo's use case.

Paulo: We need to figure out how to reuse Geoff and Cynthia's mechanisms.

WC3 task force update

Nothing to report.

Yolanda Gil is leading

Paulo is waiting to hear.

OPM Flavors

Paulo wants to know what we want to learn.

Jim: they have made distinction between conceptual model and logical model. MANY logical models (i.e. flavors)

Paulo: re conceptual vs. logical model. has not seen OPM as a logical model. what can we learn by understanding the different logical models? we actually see a bunch of conceptual models that are different. everything you do can be OPM. not everything you do can be PML.

Tim: all of these are materials to consider in next steps.

Paulo: versioning, relationship of agents to derivation trace.

Paulo: Philosophy has handled a lot of this. We should not reinvent. e.g. Agency. Causality and explanation are basically the same concept. you cannot talk about causality without talking about agency. to understand derivation, we need to reach out to philosophy. e.g. version control. (library community).


is a web of data. you reuse the web architecture (client/server). instead of web sites on server, you have data sets. in request/response: incubator group was discussing explaining HTTP protocol and how to describe provenance of it. Paulo: provenance is part of a question / answering framework. where answer is answer plus explanation. e.g. web documents but sci data over web, the analogue is opendap. in same way have have response back, you should get response + explanation. a zillion ways to add provenance to opendap. opendap allows adding attributes to any dataset. many tradeoffs. at dataset level? at record level? both levels? what are the things yoiu want an explanation for? when it was added? opendap is a place we should tackle first.

Paulo: What were Peter/Patrick/Stephan thoughts?

you get data, but you get the PML of the data as well.

TODO: table until we get one of them on phone. TODO: Tim to invite them to IW meeting to discuss their opendap+provenance experiences. TODO: Paulo to follow up with talking points for discussion.

documentation lacks why things in place, what are reusable components, general description of data model. just a bunch of recipes, no general description. PML is opposite (general description but few recipes)


Paulo: Cyber share grand opening 8 April. TODO: Paulo to send agenda to Deborah re. cyber share.

Template. Thanks for interest in PML< we are happy to have new members. TODO: Tim to add that to the page.

Personal tools