Research Agenda
I'm interested in solving hard combinatorial problems, with an emphasis on planning and scheduling problems, combining techniques form Computer Science (CS), Artificial Intelligence (AI), and Operations Reserach (OR). Currently my work has focused on studying the role of randomization in computation, characterization of the distribution profiles of randomized algorithms and consequences for algorithm design. In this work, I study so-called heavy-tailed distributions that characterize complete randomized search methods. A promising way of exploiting heavy-tailed behavior is by using restart strategies or by combining a suite of search methods into a portfolio, running on a distributed compute cluster. It can be shown that such strategies dramatically reduce the expected overall computational cost, thereby allowing us to solve large, previously unsolved planning and scheduling problems. Using a recent equipment grant, we are constructing a 150 node compute cluster with a high-speed communication network to further evaluate and study our randomized approaches.
A pictorial view of my research interests.
Research statement Postscript
PDF [ Brief-synopsis]
American
Association for the Advancement of Science(AAAS) - Annual
Meeting (2005):
Check
the Science News article by Ivars Peterson "Completing Latin
Squares"
Science
Daily (1999) gives a journalistic view of Heavy-tailed phenomena.
Creating Intelligence gives a nice overview of AI at Cornell.
I'm in the process of organizing some information for this web site on the following issues (for a sneak preview):
Quasigroup Completion Problem - A Structured Benchmark Domain
Heavy-Tailed Phenomena in Combinatorial Search
Algorithm Portfolio Design
Planning and Scheduling