faculty.gif (20410 bytes)
choices.gif (4488 bytes)

Bart Selman

Associate Professor

PhD Univ. of Toronto, 1991

I am interested in the development of new formalisms and methods for artificial intelligence (AI) by combining a sound theoretical approach with a principled experimental component. The focus of my research is on compute-intensive methods for AI. Traditionally, general search and reasoning is avoided in AI largely by explicitly incorporating large amounts of domain-specific knowledge. While a knowledge-intensive approach has been successful in certain domains, like automatic diagnosis, in other areas, such as planning or general reasoning, progress has been disappointing. However, recent advances in general search and reasoning methods combined with faster hardware and better implementations provide strong evidence that a

bart.tif (274156 bytes)
compute-intensive approach is not only suitable for dealing with the combinatorial nature of many AI formalisms but may be required to supplement domain-specific knowledge—especially considering the knowledge-acquisition bottleneck in terms of encoding highly specific domain knowledge.

I am interested in fast, general, reasoning and search methods, with an emphasis on stochastic procedures—a promising recent development for solving computationally hard problems. I also investigate various sources of complexity in hard problems, using theoretical and experimental methods. This work explores interesting connections between computer science, AI, and statistical physics. In addition, I study issues in problem representation, including the robustness of encodings, abstraction, compilation, and approximation methods. These issues are critical to the successful application in realistic domains of reasoning and search methods. In terms of applications, I am particularly interested in challenge problems from planning, knowledge representation, machine learning, and data mining. Our planning system, BlackBox, developed jointly with Henry Kautz of AT&T Labs, is one of the current fastest general-purpose planning systems.

I also pursue applications in areas outside AI, such as operations research and software/protocol verification.


NSF Faculty Early Career Development Award

University Activities

  • PhD Admissions Committee

  • Coordinator: AI seminar series

  • Member: Field of Cognitive Studies

Professional Activities

  • Editor: J. Artificial Intelligence Research (JAIR); Annals of Mathematics and Artificial Intelligence (guest editor)

  • Editorial Board: Constraints: An International J.

  • Program Committee: Fifteenth Nat. Conf. Artificial Intelligence (AAAI 98); Thirteenth Biennial European Conf. Artificial Intelligence (ECAI 98); Sixth Int. Conf. Principles of Knowledge Representation and Reasoning (KR 98); Fourth Int. Conf. Artificial Intelligence Planning Systems (AIPS 98); Eighth Int. Conf. Artificial Intelligence: Mehodology, Systems, Applications (AIMSA 98); Symp. Abstraction, Reformulation and Approximation (SARA 98); Constraint Programming (CP 97)

  • Organizing Committee: Planning as Combinatorial Search (AIPS 98 Worshop); Recommender Systems (AAAI 98 Workshop); Software Tools for Developing Agents (AAAI 98 Workshop); Constraints and Agents (AAAI 97 Workshop); Empirical Methods in AI (ECAI 97 Workshop)

  • Co-Chair: Tutorial Forum, AAAI 98

Referee/Reviewer: Science, Artificial Intelligence J., JACM; J. Automated Reasoning; NSF review panel


Scaling properties of constraint-based planners. Fourth Int. Conf. Artificial Intelligence Planning Systems, Pittsburgh, PA, June 1998.

Heavy-tailed phenomena in combinatorial search. Georgia Tech, Atlanta, GA, Oct. 1997.

Compute-intensive methods for artificial intelligence. Four-hour tutorial. Fourteenth Nat. Conf. Artificial Intelligence, New Providence, RI, Aug. 1997 (with Henry Kautz).

Compute-intensive methods for knowledge representation and reasoning. DARPA Young Investigators Meeting, New Providence, RI, Aug. 1997.

Evidence for invariants in local search. Fourteenth Nat. Conf. Artificial Intelligence, New Providence, RI, Aug. 1997.

Algorithm portfolio design. AT&T Labs, Florham Park, NJ, July 1997.


Randomization in backtrack search: Exploiting heavy-tailed profiles for solving hard scheduling problems. Proc. Fourth Int. Conf. Artificial Intelligence Planning Systems (AIPS-98), Pittsburgh, PA (June 1998), 208-213 (with C. Gomes, K. McAloon, and C. Tretkoff).

The role of domain-specific knowledge in the planning as satisfiability framework. Proc. Fourth Int. Conf. Artificial Intelligence Planning Systems (AIPS-98), Pittsburgh, PA (June 1998), 181-189 (with H. Kautz).

Greedy local search. In MIT Encyclopedia of the Cognitive Sciences, S. Russell and M. Jordan (Eds.), Cambridge: MIT Press, 1998.

Heavy-tailed probability distributions in combinatorial search. Proc. Constraint Programming (CP 97), Linz, Austria (Oct. 1997) (with C. Gomes and N. Crato).

Ten challenges in propositional reasoning and search. Proc. Fifteenth Int. Conf. Artificial Intelligence (IJCAI 97), Nagoya, Japan (Aug. 1997) (with H. Kautz and D. McAllester).

Problem structure in the presence of perturbations. Proc. Fourteenth Nat. Conf. Artificial Intelligence (AAAI 97), New Providence, RI, (Aug. 1997) (with C. Gomes).

Evidence for invariants in local search. Proc. Fourteenth Nat. Conf. Artificial Intelligence (AAAI 97), New Providence, RI (Aug. 1997) (with D. McAllester and H. Kautz).

Algorithm portfolio design: theory vs. practice. Proc. Thirteenth Conf. Uncertainty in AI, New Providence, RI (Aug. 1997) (with C. Gomes).

Referralnet: combining social networks and collaborative filtering. Comm. ACM 40, 3 (1997), 63-65 (with H. Kautz and M. Shah).

The hidden web. Artificial Intelligence Magazine 18, 2 (1997), 27-36 (with H. Kautz and M. Shah).


Message filtering techniques. US patent no. 5619648, 1997 (with L.M. Cabale, H.A. Kautz and A.E. Milewski).

Methods and apparatus for constraint satisfaction. US patent no. 5636328, 1997 (with H.A. Kautz).