Approximate Reasoning Using Anytime Algorithms

Abstract

The complexity of reasoning in intelligent systems makes it undesirable, and sometimes infeasible, to find the optimal actionin every situation since the deliberation process itself degrades the performance of the system. The problem is then to constructintelligent systems that react to a situation after performing the {\textquotedblleft}right{\textquotedblright} amount of thinking. It is by now widely acceptedthat a successful system must trade off decision quality against the computational requirements of decision-making. Anytimealgorithms, introduced by Dean, Horvitz and others in the late 1980{\textquoteright}s, were designed to offer such a trade-off. We have extendedtheir work to the construction of complex systems that are composed of anytime algorithms. This paper describes the compilationand monitoring mechanisms that are required to build intelligent systems that can efficiently control their deliberation time.We present theoretical results showing that the compilation and monitoring problems are tractable in a wide range of cases,and provide two applications to illustrate the ideas.