Actions

Ontolog Forum

Revision as of 15:47, 26 March 2015 by imported>KennethBaclawski
(diff) ← Older revision | Latest revision (diff) | Newer revision → (diff)

OntologySummit2015 Decision Making in Different Domains

Goal

To explore several approaches to automated inference in applications ranging from complex event processing and situation awareness to manufacturing.

Case Studies

Insights and Lessons Learned

  • A little semantics goes a long way Joseph Kopena
    • Potential stakeholders primarily interested in basic taxonomies
  • Fairly difficult to get developers without KR experience up to speed Joseph Kopena
    • Project apps didn't get to point of utilizing capabilities for collaboration, versioning, etc., offered by the underlying model
  • Evaluation of KR systems is extremely difficult Joseph Kopena
    • Performance is non-trivial but fairly straightforward
      • Sidenote: What's hard for network may not be hard for KR, and vice versa
    • Testing actual effectiveness and value requires complex yet realistic scenarios, revolves around metrics that are difficult to quantify
  • SPARQL and RDF model aren't quite the right tools for this task Joseph Kopena
    • SPARQL great for querying the KB, less ideal for fetching objects
      • Apps want all the metadata about content, resulting in massive queries
    • RDF+SPARQL cumbersome when working with dynamic data
  • Automated reasoning is difficult! Michael Grüninger
    • Some queries could not be answered in the time limit.
    • Approaches are necessary for dealing with this problem.
  • Manufacturing processes are complex. Michael Grüninger
    • Objects flow through a sequence of processes, and at any point in a process plan, there are multiple activities that can possibly occur next.
    • Process plans may also be nondeterministic.
    • A first-order process ontology can be used to create smart objects that can reason about the manufacturing processes in which the object participates.
    • Eventually, smart items could be dynamically self-routed through the various process plans.
  • Complex event processing can benefit from semantics Adrian Paschke
    • Event data becomes declarative knowledge while conforming to an underlying formal semantics
    • Reasoning over situations and states by event processing agents
    • Better understanding of the relationships between events
    • Declarative knowledge-based processing of events and reactions to situations
  • The Pragmatic Web consists of the tools, practices and theories describing why and how people use information. Adrian Paschke
    • In contrast to the Syntactic Web and Semantic Web the Pragmatic Web is not only about form or meaning of information, but about interaction which brings about e.g. understanding and commitments. Pragmatic Web Site
  • Decisions are made within a context
    • Situation Assessment is central to information fusion Ken Baclawski
    • Situation Awareness is fundamental part of event processing Adrian Paschke
  • Decisions are made within a process
  • Decisions require a variety of processing techniques Ken Baclawski William Murdoch
    • Data mining
    • Inductive reasoning
    • Abductive reasoning
    • Probabilistic reasoning
  • Decision Making in Question Answering William Murdoch
    • Choosing answers to questions!
      • IBM Watson generates many candidate answers
      • For each answer, how confident are we that the answer is right?
    • Deciding whether to answer
      • Based on how confident we are that the answer is right
      • Based on cost/benefit of right answers and wrong answers
    • Deciding how many answers to provide
    • Deciding whether to hedge
  • Question Answering vs Decision Making
    • Question Answering presumes that there is a unique correct answer to a question and the purpose of decision support is to find the answer.
    • Decision-making can be regarded as the cognitive process resulting in the selection of a belief or a course of action among several alternative possibilities. Wikipedia page on decision-making
      • In such a process, there is no unique answer or even necessarily a correct answer that one could verify.
    • Question Answering and Decision Making have much in common but a system designed for one is not usually suitable for the other.

Additional Resources