Meta-level Control of Approximate Reasoning: A Decision Theoretic Approach

Shlomo Zilberstein. Meta-level Control of Approximate Reasoning: A Decision Theoretic Approach. Proceedings of the Eighth International Symposium on Methodologies for Intelligent Systems (ISMIS), 114-123, Charlotte, North Carolina, 1994.

Abstract

This paper describes a novel methodology for meta-level control of approximate reasoning. We show that approximate reasoning supported by anytime algorithms offers a simple means by which an intelligent system can trade-off decision quality for deliberation cost. Such a tradeoff is an essential capability of almost every intelligent system. The model exploits probabilistic knowledge about the environment and about the performance of each component in order to optimally manage computational resources. An off-line knowledge compilation technique and a run-time monitoring process guarantee that the system is performing the ``right'' amount of thinking in a well-defined sense. The paper concludes with a brief description of two successful applications.

Bibtex entry:

@inproceedings{Zismis94,
  author	= {Shlomo Zilberstein},
  title		= {Meta-level Control of Approximate Reasoning: A Decision 
                   Theoretic Approach},
  booktitle     = {Proceedings of the Eighth International Symposium on
                   Methodologies for Intelligent Systems},
  year		= {1994},
  pages		= {114-123},
  address       = {Charlotte, North Carolina},
  url		= {http://rbr.cs.umass.edu/shlomo/papers/Zismis94.html}
}

shlomo@cs.umass.edu
UMass Amherst