5/11/23 | 4:15pm | E25-111

Jose Blanchet

Professor of Management Science and Engineering
Stanford

Abstract:

The objective of distributionally robust decision-making is to facilitate sound decisions under uncertain circumstances when the "training" environment (or "assumed model") differs from the actual context in which the decision will be implemented. This is prevalent in scenarios with highly non-stationary environments or when training must occur in a simulated environment due to various constraints. DRO formulations rely on min-max games, where a manager engages in a game against a fictitious adversary introduced for a systematic "what-if-my-model-is-wrong" analysis. This methodology has a well-established history in the economics literature and control.

Fundamentally, there are two ways in which a probabilistic model can be inaccurate: either the likelihoods are incorrect, the actual outcomes are misspecified, or both. Historically, these types of model inaccuracies have been addressed separately. This talk aims to examine these issues from a unified perspective by employing the theory of optimal transport with martingale constraints. Our approach recovers most forms of DRO formulations (and introduces new ones). Additionally, if time permits, we delve into the broader implications of dynamic decision-making under uncertainty.

This talk is based on joint work with Nick Bambos, Daniel Kuhn, Jiajin Li, Sirui Lin, Kyriakos Lotidis, and Bahar Tahksen.  

Bio: Jose Blanchet is a Professor of Management Science and Engineering (MS&E) at Stanford. Prior to joining MS&E, he was a professor at Columbia (Industrial Engineering and Operations Research, and Statistics, 2008-2017), and before that he taught at Harvard (Statistics, 2004-2008). Jose is a recipient of the 2010 Erlang Prize and several best publication awards in areas such as applied probability, simulation, operations management, and revenue management. He also received a Presidential Early Career Award for Scientists and Engineers in 2010.