Seminars

ME/IE 8773-8774
INDUSTRIAL ENGINEERING SERIES (1130 ME)
Topic: Industrial Engineering
Host: Diwakar Gupta

Stochastic Dynamic Optimization with Model Uncertainty and Learning

by

J. George Shanthikumar
Department of Industrial Engineering and Operations Research
University of California at Berkeley
Berkeley, CA 94720-1777

Wednesday, April 27, 2005
3:30-5:00 p.m.
Room 1130 ME
Coffee and cookies will be available at 3:15 p.m. in Room 1130 ME before the seminar

Abstract:

In decision and control in operations research (OR), one may formulate a deterministic optimization model and solve it. For example in inventory control or production planning, it is not uncommon in the past to assume that the demand is known in advance. In many situations, however, only an estimate of the demand may be known, but the actual demand will be unknown. Hence, one of the recent emphasis in OR has been to formulate stochastic models for these problems. Almost all such models assume that the full probabilistic characterization of the models will be known at the time of implementing the solution. However, in reality, only an estimate, not the true probabilistic characterization will be known. In this talk we will demonstrate that such an assumption may lead to inefficient solution to the real problem. Hence, we will formulate decision and control problems under uncertainty by a collection of models (that is, a model with model uncertainty) and solve it to obtain robust solutions. Specifically, we will develop min-max robust formulations to revenue management and inventory control problems. The relationship between this robust formulation and exponential utility will be established. Alternative objectives, such as min-max-regret, competitive ratio, etc., for other possible robust solutions will be discussed as well. Notions of learning under model uncertainty will be discussed and compared to reinforcement learning and statistical learning. This is a joint work with members of the Berkeley IEOR Group on Model Uncertainty & Learning. Current group members are: Professors Andrew E.B. Lim and J. George Shanthikumar and Students: Onur Filiz, Ankit Jain and Thaisiri Watewai.

Bio:

J. George Shanthikumar received the B.Sc. degree in mechanical engineering from the University of Sri Lanka, Peradeniya, and the M.A.Sc. and Ph.D. degrees in industrial engineering from the University of Toronto, Canada. He is Professor of Industrial Engineering and Operations Research at the University of California, Berkeley. His research interests are in integrated interdisciplinary decision making, production systems modeling and analysis, queueing theory, reliability, scheduling, stochastic processes, simulation and supply chain management. He has written or written jointly over 250 papers on these topics. He is a coauthor (with John A. Buzacott) of the book Stochastic Models of Manufacturing Systems and a coauthor (with Moshe Shaked) of the book Stochastic Orders and Their Applications. Dr. Shanthikumar received the E.O.E. Pereira Gold Medal as the outstanding student graduating from the College of Engineering, University of Sri Lanka. He was granted the Canadian Commonwealth Scholarship during 1975-1979 for his studies towards the M.A.Sc. and Ph.D. degrees at the University of Toronto. He is (or was) a member of the editorial boards of the IEEE Transactions on Automation Sciences and Engineering, IIE Transactions, International Journal of Flexible Manufacturing Systems, Journal of Discrete Event Dynamic Systems, Journal of Production and Operations Management, Operations Research, Operations Research Letters, OPSEARCH, Probability in the Engineering and Informational Sciences, and Queueing Systems: Theory and Applications. Dr. Shanthikumar has extensively consulted for various companies like IBM, NTT (Japan), Bellcore, Safeway, and KLA-Tencor and through KLA-Tencor consulted for AMD, IBM, Intel, Motorola, Toshiba, Fujitsu, TSMC and UMC.

Informal Faculty Luncheon: Wednesday, April 27, 2005, 12:00 noon. Meet in 1100 ME and walk to lunch with other faculty. Prof. Shanthikumar will be able to attend.