IW Meeting 2006 Nov 30

From Inference Web

Jump to: navigation, search


  • GILA
    • explanation paper:
      • execution trace as explanation: (i) rule representation: PDDR/InferenceML or business rule (OWL-S, RIF, ?) (ii) the acutal input/output/precondition/effect of one application of a rule (iii) execution trace contains applications of rules, now how can we reverse engineer the description of the input/output of rule?
      • decision trace as explanation: (i) how to capture decision rules, in what granularity? (ii) how to show usage of knowledge, esp informal domain knowledge? (iii) can we categorize information manipulation process to achieve better representation. (iv) how to represent learned knowledge?
        • In particular, how to explain machine learning process, when there are a lot of examples as input. Maybe we can generate some templates.
        • Q: what is the benefit to have an ontology to represent learning process? Will there be a reasoning engine, in what details and how the ontology will be used?
      • how explanation will be used? how to make sure explanations can be easily consumed by human and machines. vasco suggested use data-mining ontology and mentioned Tom Dietterich's paper
        • users care the result
        • users care what knowledge, what algorithm has been used
        • do we need to reproduce/verify the learning process using explanation.
        • explanation is declarative representation. If use to explain machine learning, that indicate that machine learning algorithm can be represent/execute in declarative way. Will such explanation be sudo code (script) and can be run on virtual machine?
    • follow-up constraint learning
      • constraint and preference ontology and inference engine requirements
    • consolidate GIL interface
      • add hasExplanation to Solution
      • how to capture problem solution?
    • follow-up LTML and OWL-QL with Yale/BBN
    • follow-up 4D spatio-temporal ontology and consolidation of ACO ontology
  • TAMI
    • follow-up broken link for tabulator
    • follow-up generating PML from scheme
    • follow-up scenario 4 demo
    • follow-up Data Purpose Algebra (DPA) paper
  • PML2
    • pml2 ontology with detailed examples and usage
    • pml2 validator
    • pml-pml2 mapping
    • plan pml2 extension to capture different types of traces
  • CALO
    • site visit

action items

Personal tools