IW Meeting 2011-03-31

From Inference Web

Jump to: navigation, search


Meeting info


  • geoff and cynthia update <---
  • desires for telecon from group
  • yangfan problem for pml
  • tim's use of pml
  • logistics - skype overhead - XXX



We are moving back to the land line for conference calls. We will pick up on Skype when Paulo's students start attending regularly.

proof combining

7 replacements (still manual) by manual inspect, a possible improvements seen, but ran into a naming issue. still remaining: automate doing manual is good b/c good for finding bugs in process.

draft for workshop paper will be provided on 14th of April (it is DUE 2nd of May)

Cynthia expects to finish automation in next couple of days.

proof combining - scientific combination

geoff expecting paulo's written description so he has a gut feeling.

Paulo: using FRBR could be a good use of proof combination. frbr - http://archive.ifla.org/VII/s13/wgfrbr/index.htm Functional Requirements for Bibliographic Records

as long as we know how to say conclusions are SIMILAR or EQUAL.

not same language, format, or places.

book Moby Dick. we both talk about same thing. I read it in English, Paulo heard it on an audio CD. We used different items (his library's and mine)

Paulo: mapping PML to FRBR:

Item               < - source
Manifestation      < - 
Expression         < - all three into information
Work               < -

Geoff (the THREE requirements to get proof combination): DAG notion of equivalence between nodes (Paulo: relaxed to similarity with threshold) measure of DAG quality.

two DAGs "in the right context" - trying to prove the same theorem

all DAGs need to be in the same context, but they could be from different.

coming to the same conclusion (all man are created equal) two conclusions derived from two different sets of antecedents

e.g. distance from earth to the sun. proof trace for each of three results.

e.g. Japan's earthquake 8.9 vs 9.0 we know that the two proofs have to come to the same result.

  having two proofs

we are still waiting for paulo and geoff to call in Conference Dial-in Number: (605) 475-4850

(when joining) Participant Access Code: 852017#

Imagine two ways of computing the distance from Miami to New York. Method 1: Take distance from Miami to Raleigh, and then add the distance from Raleigh to NY. Method 2: Take distance from Miami to Raleigh, then add the distance from Raleigh to Washington, then add the distance from Washington to NY. If both methods agree on the distance from Miami to Raleigh (equivalent nodes in the DAG), but maybe they used different ways to compute that distance (different history in that sub-DAG), then that node in Method 1 can be replaced by the equivalent node in Method 2.

"right' according to context - am I driving or flying or digging through the earth?

In the C&G view, we might consider the nodes to be the segments of the route, and two segments are "equivalent" if they have the same end points. Then even if the distances are different for the two methods, their nodes are equivalent because they have the same end points, and hence we can do a combination. That would produce a third "method", with a different complete distance from Miami to NY.

multiple proofs builds confidence.

Introducing Yangfan

searching for possible errors in data.

user finding possible error, how use PML to help them track possible reasons. (IPAW paper discusses crowdsourcing and incorporate domain experts)

e.g. application in cloud computing - everything is a service. if we have doubt in that, can PML be used to help identify the source of the problem?

PML encodes how one gets an answer, but how to start selecting where to start looking within the full trace.

Yangfan wants to help reduce effort of users to identify the problem.


graph connectivity analysis vs. uncertainty of a quantitative value.

Yangfan was motivated by James' IPAW (?) paper. As a layperson, onemay not be sure a value is correct. How to help the user understand the derivation to know what caused it?

should I just check all steps one by one? or is there a way to prioritize what should be looked at to avoid wasting time looking at every step?

given a triple asserting something we diasgree with, what is the step that is to "blame" for introducing it?

Paulo: science is about getting an answer with the (useful) right level of uncertainty.

"most uncertainty"

users know how to decide what to look into in their domain.

Personal tools