Model Confidence
Kathryn Blackmond Laskey
Department of Systems Engineering
George Mason UniversityPresented to Summer Institute on Probability in
Artificial Intelligence
University of Oregon, July, 1994
An intelligent agent uses a model of the world to reason about the effects
of its actions and decide on the best course of action. Except on the simplest
and least interesting problems, the agent's model of the world may be incorrect.
Intelligent age nts need to be robust to model misspecification and to
be able to reason appropriately with fallible models. This presentation
describes methods for representing information about confidence in a model,
methods for revising assessments of model confidenc e based on performance
of the model, and approaches to reasoning with fallible models.
Contents
