반복영역 바로가기
주메뉴로 바로가기
좌측메뉴로 바로가기
본문으로 바로가기

Home News&ActivitySeminars

Seminars

프린트페이스북

[Abstract Changed] 학과 세미나 안내 (12/20, 노스웨스턴 대학 한어진 박사과정/ Seminar Notice : Dec. 20th at 4pm, Doctoral Candidate Han Uh Jin of Northwestern University)

2018.12.17 16:33

 

Dear ISysE Professors,                      

 

Head Taesik Lee invites you to attend the ISysE Dept. seminar to be held on Dec. 20th, 2018 at the multimedia room starting at 4:00 p.m.                                   

               

DATE & TIME: Thursday, Dec. 20th at 4PM

                

VENUE: E2 Bld., Multimedia Room(#2501)

 

TITLE: Performance Bounds for Multistage Distributionally Robust Optimization

                               

SPEAKER : 한어진 (Doctoral Candidate Han Uh Jin, Industrial Engineering at Northwestern University)     

                 

Abstract

 

While single-stage optimization problems under uncertainty can be solved in a tractable fashion, their multistage counterparts reveal themselves to be challenging.

Since such problems naturally describe many practical settings, bounds on the solutions serve to assess the overall performance.

Such bounds have been characterized by using stochastic optimization when the uncertainty is given by a distribution, or by using robust optimization,

when the uncertainty can be described by sets. This paper takes a first step towards using multistage distributionally robust optimization (DRO) with uncertainties

on the distributions in order to obtain such bounds. We address the common challenges in dimensionality for multistage problems under moment uncertainty

by proposing an adaptable approach, where at each stage the set of possible uncertainties is partitioned into an optimally sized subset containing the point of symmetry

and the complement subset. This partitioning enables a closed form solution for the performance bounds of two-stage DRO problems, which we then generalize to multistage settings.

We apply the results to an inventory management problem of one of India’s largest pharmacy retailer and demonstrate that our proposed policies significantly outperform alternative policies

when the assumed and the realized cost parameters deviate. This approach is general and, hence, applicable to a broad range of real-world problems.                                                                                                

                          

We look forward to your attendance at our colloquium and encourage you to forward this invitation to other colleagues you believe may be interested in attending.    

With kind regards,

List