My research is focused on automated reasoning in Artificial Intelligence, particularly in the areas of search, constraint-based reasoning and reasoning under uncertainty.

My ongoing focus is on constraint processing, a field which unifies themes cutting across many traditional areas in Artificial Intelligence. A variety of techniques have been developed for processing different kinds of constraint expressions and are being applied to diverse tasks such as vision, design, diagnosis, truth maintenance, scheduling, spatio-temporal reasoning, logic programming, and user interface. Many of these methods were incorporated into constraint programming languages which enhance practical applications substantially.

Since most reasoning tasks are computationally intractable, my primary approach is to devise methods through the understanding and exploitation of tractable reasoning tasks. My previous works on greedy problems, the mechanical generation of heuristics, the identification of tractable constraint models via topological decompositions, and the establishment of boundaries of local computations have been driven by this principal concern. With my students, I analyze algorithms both analytically and empirically using real life applications such as scheduling, planning, and diagnosis.

In the past decade I extended my research to general graphical models and especially to reasoning under uncertainty using Bayesian networks. We introduced the two unifying algorithmic frameworks of bucket-elimination and AND/OR search which capture the most common styles of human reasoning (e.g., Inference, and conditioning). Bucket elimination unifies dynamic programming for combinatorial optimization with algorithms for theorem proving, logic programs, temporal reasoning, probabilistic reasoning and planning under uncertainty. AND/OR search allows exploiting problem decomposition during search and is the basis to many recent algorithmic advances in graphical models. Within these two frameworks we develop efficient exact and approximate algorithms with potential impact across many computational disciplines.