Utility-Based Abstraction and Categorization

Eric Horvitz

Medical Computer Science Group
Knowledge Systems Laboratory
Stanford University
Stanford, California 94305

Adrian Klein

Center for the Study of Language and Information
Ventura Hall
Stanford University
Stanford, California 94305

Author Email: horvitz@microsoft.com

Click here to access postscript file.

Click here to access pdf file.


We take a utility-based approach to categorization. We construct genralizations about events and actions by considering losses associated with failing to distinguish among detailed distinctions in a decision model. The utility-based methods transform detailed states of the world into more abstract categories comprised of disjunctions of the states. We show we can cluster distinctions into groups of distinctions at progressively higher levels of abstraction, and describe rules for decision making with the abstractions. The techniques introduce a utility-based perspective on the nature of concepts, and provide a means of simplifying decision models used in automated reasoning systems. We demonstrate the techniques by describing capabilities and output of TUBA, a program for utility-based abstraction.

Keywords: Categorization, decision theory, planning under bounded resources, Bayesian reasoning, abstraction

In: Proceedings of Ninth Conference on Uncertainty in Artificial Intelligence, Washington DC, July 1993. pages 128-135. Morgan Kaufmann: San Francisco.