Advanced data mining methods

Abstract

Anytime algorithms (e.g., Dean and Boddy, 1988, Russell and Wefald, 1991, Zilberstein, 1996) are algorithms, which offer a trade-off between the solution quality and the computational requirements of the search process. The approach is known under a variety of names, including flexible computation, resource bounded computation, just-in-time computing, imprecise computation, design-to-time scheduling, or decision-theoretic meta-reasoning. All of these methods attempt to find the best answer possible given operational constraints. According to Zilberstein (1996), the desired properties of anytime algorithms include the following: measurable solution quality, which can be easily determined at run time; monotonicity (quality is a non-decreasing function of time); consistency of the quality w.r.t. computation time and input quality; diminishing returns of the quality over time; interruptibility of the algorithm (from here comes the term anytime); and preemptability with minimal overhead.