IW Meeting 2009-01-20

From Inference Web

Jump to: navigation, search

agenda

  • updates on combine algorithm (Geoff, Cynthia)
    • self improvement
    • self improvement on root
  • next move
  • meeting time?

notes

  • self improvement is possible: optimization previously did not replace one node with another from the SAME proof.

some proofs have repeated nodes (like EP's example proofs with false, false, false). that is sound, but ugly proof - proof where a node repeat several times, while it is yet a DAG, but node could repeat

Li: we may need to run two-step work: (i) clean up each proof, and (ii)combine proof

  • previous code never replaced the root node of a proof, now it can be replaced with a node from WITHIN the proof but NOT from the outside.
  • next move - node equivalence
Paulo: for equivalence semantics, we use some simple heuristics, and we may add a few more user-asserted heuristics
Deb: we should not get too much with equivalence stuff for its complexity
  • next move - best quality
    • we need literature survey, and people survey to get the best metrics for evaluating the quality of a proof mainly using proof structure property (similar to ontology evaluation)
      • node, step, axiom per problem per reasoner
    • another angle is to measure the quality of a collection of alternative proofs? (maybe trust)

todo

  • todo - question to cynthia - do we clean up proofs individually first before we do the combinations between two or more proofs?
  • todo - paulo and cynthia discuss the pml encoding and paulo send email to the group with the outcome
  • todo - geoff gather input on other quality metrics. reasonable metrics for comparing two proofs
  • todo - we keep meeting the same time and come back in two weeks.
  • todo - paulo demo probIt. send screen shot for DAG
Personal tools
Navigation