Dear Colleagues,
I am delighted to announce the following LANCS Initiative seminar, to be given at
Lancaster University in Lecture Theatre 6, Lancaster University Management School, at 10.00am on Tuesday 3 February 2009
and at
The University of Nottingham in Room D.LT3, Exchange Building, Jubilee Campus at 2.30pm on Wednesday 4 February 2009
Approximate Dynamic Programming for High-Dimensional Applications
Warren B. Powell
Department of Operations Research and Financial Engineering
Princeton University
Princeton, NJ 08544
www.castlelab.princeton.edu
Stochastic resource allocation problems produce dynamic programs with state, information and action variables with thousands or even millions of dimensions, a characteristic we refer to as the "three curses of dimensionality." Classical techniques in approximate dynamic programming solve only part of the problem. We show how the use of the post-decision state variable allows us to break a problem into three distinct components: simulation, deterministic optimization and statistical learning. This strategy decomposes problems over time, allowing us to use commercial packages to solve even large-scale integer programming problems, producing solutions that are optimal at a point in time (given a value function approximation), with approximate (but high quality) solutions over time. The talk will also describe how we have overcome the challenge of estimating value function approximations for complex resource allocation problems, and the critical, but often overlooked, choice of stepsize rule. We summarize the state of what we can prove theoretically, and then demonstrate the power of these methods in the context of several industrial problems and energy resource planning.
Professor Powell is an INFORMS Fellow and director of CASTLE Laboratory, founded by him in 1990 to reflect an expanding research programme into dynamic resource management. He is author of the recent book 'Approximate Dynamic Programming'.
Colleagues from outside Lancaster and Nottingham are very welcome to attend. Information and advice on local arrangements are available from
Rosemary Hindley
LANCS Initiative Administrator,
Lancaster University, Tel: 01524 592937
Email: [log in to unmask]
Website: http://www.lancs-initiative.ac.uk
And
Debby Pitchfork
Nottingham University, Tel: 0115 8466543
Email: [log in to unmask]
Best wishes, Kevin Glazebrook
Director, The LANCS Initiative
Prof Kevin Glazebrook
Dept of Management Science/Maths & Statistics
Lancaster University
Tel: 01524 592697
Fax: 01524 594020
|