Optimizing Decision Quality with Contract Algorithms (1995) [14 citations — 5 self]
Popular Tags
No tags have been applied to this document.
BibTeX | Add To MetaCart
@INPROCEEDINGS{Zilberstein95optimizingdecision,author = {Shlomo Zilberstein},
title = {Optimizing Decision Quality with Contract Algorithms},
booktitle = {Fourteenth International Joint Conference on Artificial Intelligence},
year = {1995},
pages = {1576--1582}
}
Years of Citing Articles
Abstract:
shlomocs.umass.edu Contract algorithms offer a tradeoff between output quality and computation time, provided that the amount of computation time is determined prior to their activation. Originally, they were introduced as an intermediate step in the composition of interruptible anytime algorithms. However, for many real-time tasks such as information gathering, game playing, and a large class of planning problems, contract algorithms offer an ideal mechanism to optimize decision quality. This paper extends previous results regarding the meta-level control of contract algorithms by handling a more general type of performance description. The output quality of each contract algorithm is described by a probabilistic (rather than deterministic) conditional performance profile. Such profiles map input quality and computation time to a probability distribution of output quality. The composition problem is solved by an efficient off-line compilation technique that simplifies the run-time monitoring task. I Decision making with contract algorithms The wide performance variability of artificial intelligence techniques, most notably in search and knowledgebased systems, has been a major obstacle in applying these techniques to real-time environments. This problem led to the development of a variety of approxima-tion techniques such as anytime algorithms [Dean and Boddy, 1988; Horvitz, 1987], design-to-time [Garvey and Lesser, 1993] and various progressive reasoning methods [Mouaddib and Zilberstein, 1995]. It is by now well understood that a successful system must trade off decision quality for computation time. Anytime algorithms in particular offer a simple means by which a system can monitor and maximize its overall utility. Contract algorithms are a special type of anytime algorithms that was introduced in order to simplify the *URL:
