Anytime AND/OR Best-First Search for Optimization in Graphical Models

Natalia Flerova, Radu Marinescu, Rina Dechter

Decision: conferencePoster

Abstract:

Depth-first search schemes are known to be more cost-effective for
solving graphical models tasks than Best-First Search schemes. In this
paper we show however that anytime Best-First algorithms recently
developed for path-finding problems, can fare well when applied
to graphical models. Specifically, we augment best-first schemes
designed for graphical models with such anytime capabilities and
demonstrate their potential when compared against one of the most
competitive depth-first branch and bound scheme. Though Best-First
search using weighted heuristics is successfully used in many domains,
the crucial question of weight parameter choice has not been systematically studied and presents an interesting machine learning
problem.