Event
Special Seminar: Suvrajeet Sen, "Stochastic Decomposition--A Revival"
Friday, September 21, 2018
10:30 a.m.
1518 Van Munching Hall
Michael Fu
301 405 2241
mfu@umd.edu
Stochastic Decomposition--A Revival
Suvrajeet Sen
Daniel J. Epstein Dept. of ISE
University of Southern California
Los Angeles, Calif.
Abstract
Stochastic Decomposition (SD) is a Stochastic Linear Programming algorithm which is experiencing a resurgence about 30 years after its invention. One can think of SD as a sampled version of decomposition algorithms based on outer linearization, although sampling raises many important issues for asymptotic convergence, size of approximations, stopping rules, and others. Such issues were resolved in the previous millennium! The revival that we are witnessing is due to the growing interest in stochastic programming (SP). Recent results which will be covered in this talk are due to a new generation of students who have taken the state of the art to new heights.
a) Multi-stage Stochastic Decomposition (due to Z. Zhou, 2014)
b) Variance reduction via a new concept known as a "compromise decision" for replicated runs (due to Y. Liu, 2016)
c) Statistical optimality for replicated runs (due to Y. Deng, 2017)
d) Sublinear convergence rate for Stochastic Quadratic-Linear Programming (due to J. Liu, 2018)
e) Randomness in all (second-stage) data elements, except the recourse matrix (due to H. Gangammanavar and Y. Liu, 2018)
After a brief summary of early developments of SD, we plan to discuss the above revival, and present computational results which scale up for real-scale power system dispatch problems. Needless to say, this talk is based on joint research with current and former Ph.D. students.
Biograpny
Suvrajeet Sen served as a program director at NSF where he was responsible for the Operations Research, and the Service Enterprise Engineering programs. At NSF, he also headed the Cyber infrastructure planning activities of the Engineering Directorate. Concurrently with his appointment at NSF, he was a professor of Systems and Industrial Engineering at the University of Arizona.
Professor Sen has served on the editorial board of several journals, including Operations Research as Area Editor for Optimization, and as Associate Editor in INFORMS Journal on Computing, Telecommunications Systems, as well as Operations Research. He is the past-Chair of the INFORMS Telecommunications Section and founded the INFORMS Optimization Section. Professor Sen is a Fellow of INFORMS.
