Loading Events

AI Seminar | Women in Computing

Confounding-Robust Policy Learning under Sequentially Exogenous Unobserved Confounders

Angela ZhouAssistant Professor, Data Sciences and OperationsUniversity of Southern California
3725 Beyster BuildingMap


https://umich.zoom.us/j/92216884113 (password: UMichAI)

Meeting Signup

Sign up to meet with Prof. Zhou here.


Angela is an Assistant Professor at University of Southern California Marshall School of Business in Data Sciences and Operations. Her research interests are in statistical machine learning for data-driven sequential decision making under uncertainty, causal inference, and the interplay of statistics and optimization. She is particularly interested in applications-motivated methodology with guarantees in order to bridge method and practice. She was a co-program chair for ACM conference on Equity and Access in Algorithms, Mechanisms, and Optimization (EAAMO).


We study the problem of learning single-timestep and sequential personalized decision policies from observational data while accounting for possible unobserved confounding. Previous approaches, which assume unconfoundedness (no unobserved confounders affect both the treatment assignment as well as outcome) can lead to policies that introduce harm rather than benefit when some unobserved confounding is present. Instead, because policy value and regret may not be point-identifiable, we study a robust method that minimizes the worst-case estimated regret of a candidate policy against a baseline policy over an uncertainty set for propensity weights that controls the extent of unobserved confounding. Our uncertainty sets are superpopulation versions of sensitivity analysis in causal inference. We prove generalization guarantees that ensure our policy is safe when applied in practice and in fact obtains the best possible uniform control on the range of all possible population regrets that agree with the possible extent of confounding. We develop efficient algorithmic solutions to compute this minimax-optimal policy. In the single-timestep setting, we assess and compare our methods on synthetic and semisynthetic data; including a case study on personalizing hormone replacement therapy based on observational data in which we illustrate our results on a randomized experiment. Similar ideas about robustness extend to more complicated decision settings, such as obtaining bounds in the infinite-horizon setting or finding robust policies in the finite-horizon setting, but require assumptions of sequentially exogenous unobserved confounders. With an emphasis on practicality, we provide an orthogonalized robust fitted-Q-iteration algorithm for the finite-horizon setting.

This AI Seminar is sponsored by LG AI Research.


AI Lab

Faculty Host

Maggie Makar