TITLE: Planning under Uncertainty: Going Forward
PRESENTER: Ugur Kuter, http://www.cs.umd.edu/~ukuter/, email@example.com
AFFILIATION:Department of Computer Science Department of Computer Science, University of Maryland, http://www.cs.umd.edu/
DATE: Friday 7th October 2005
PLACE: CSE Seminar Room, Level 1, K17
Most planning problems of practical importance require reasoning
under uncertainty for generating a course of action that achieves
the underlying objectives. Planning algorithms designed to work under
conditions of uncertainty typically have large efficiency problems
due to the need to explore all or most of the state space.
In classical planning, many techniques have been developed for controlling
the search in large state spaces of planning problems. These techniques
have been especially successful in classical forward-chaining planners.
This talk describes ways to exploit those techniques, originally developed
for classical forward planners, in planning under uncertainty. We discuss
the theoretical properties of the new algorithms we developed using this
approach, and present experimental evaluations of these algorithms. Our
results demonstrate the conditions under which the new algorithms can
solve planning problems exponentially faster than the existing ones
developed for planning under uncertainty.
BIOGRAPHY OF SPEAKER:
Ugur Kuter is interested in Artificial Intelligence (AI), and in
particular, in automated planning and decision making under conditions
of with several sources of uncertainty such as nondeterminism, time,
probabilities, utilities, and incomplete information. He got his
undergraduate and masters degrees from the Computer Engineering
Department, Middle East Technical University, Ankara, Turkey. He is
currently working with Prof. Dana Nau at University of Maryland on his
PHD training, and is expecting to graduate in Spring 2006.
Van Hai Ho