BEGIN:VCALENDAR
VERSION:2.0
PRODID:-//Memento EPFL//
BEGIN:VEVENT
SUMMARY:Locally Robust Models for Optimization under Tail-based Data Imbal
 ance
DTSTART:20230623T140000
DTEND:20230623T150000
DTSTAMP:20260406T214718Z
UID:bff9e9e02e93246c65b24ecfa6b40169df73cf266d27617731ed9934
CATEGORIES:Conferences - Seminars
DESCRIPTION:Professor Karthyek Murthy  \nAbstract: Several problems in da
 ta-driven decision-making and risk management suffer from data imbalance\,
  a term referring to settings where a small fraction of data has an outsiz
 ed impact on estimating one or more decision-making criteria. Due to the p
 aucity of relevant samples\, such problems are usually approached with the
  “estimate\, then optimize” workflow involving a model estimation from
  data in the first step before plugging in the trained model to solve vari
 ous downstream optimization tasks. As biases due to model selection\, miss
 pecification\, and overfitting to in-sample data are difficult to avoid in
  the first-step estimated model\, particularly more so in settings affecte
 d by data imbalance\, we construct novel locally robust optimization formu
 lations in which the first-step estimation has no effect\, locally\, on th
 e optimal solutions obtained. We show that this local insensitivity transl
 ates to improved out-of-sample performance freed from the first-order impa
 ct of model errors introduced in the first-step estimation.\n\nA key ingre
 dient in achieving this local robustness is a novel debiasing procedure th
 at adds a non-parametric bias correction term to the objective. The debias
 ed formulation retains convexity\, and the imputation of the correction te
 rm relies only on a non-restrictive large deviations behavior conducive fo
 r transferring knowledge from representative data-rich regions to the data
 -scarce tail regions suffering from imbalance. The bias correction gets de
 termined by the extent of model error in the estimation step and the speci
 fics of the stochastic program in the optimization step\, thereby serving 
 as a scalable “smart-correction" step bridging the disparate goals in es
 timation and optimization.\n\nBio sketch: Karthyek Murthy serves as an Ass
 istant Professor in Singapore University of Technology & Design. His resea
 rch interests lie in data-driven Operations Research. Prior to joining SUT
 D\, he was a postdoctoral researcher in Columbia University IEOR departmen
 t and he received his PhD in Tata Institute of Fundamental Research\, Mumb
 ai. His research has been recognised with 2021 INFORMS Junior Faculty Foru
 m (JFIG) Paper competition (Third place) and 2019 WSC Best Paper Award. Ka
 rthyek serves as an Associate Editor for the INFORMS journal Stochastic Sy
 stems and as a judge for the INFORMS Nicholson student paper competition.
LOCATION:ODY 4 03 https://plan.epfl.ch/?room==ODY%204%2003
STATUS:CONFIRMED
END:VEVENT
END:VCALENDAR