Abstract: Inferring causal relationships is critical to reliable decision-making. However, traditional modeling assumptions that allow adjusting prediction models to learn counterfactuals rarely hold in practice. Observed decisions depend on unrecorded confounding variables, user behavior shifts across space and time, and marginalized demographic groups are severely underrepresented in typical datasets. As an example, among 10,000+ cancer clinical trials the National Cancer Institute funds, fewer than 5% of participants are non-white.
We analyze the sensitivity of causal findings against two ubiquitous violations of traditional assumptions: the presence of unobserved confounders and subpopulation shifts. Our worst-case approach guards against brittle findings that are invalidated by even small distributional shifts. In both randomized and observational studies, we develop semiparametric methods that allow the flexible use of large-scale machine learning models. Our estimator enjoys central limit behavior—oracle rates—even when ML-based estimates of nuisance parameters converge at slower rates. On real datasets, we demonstrate that our estimator guards against brittle findings that are invalidated by unobserved confounders and subpopulation shifts.
This talk will be based on the following papers [1, 2, 3].
Bio: Hongseok Namkoong is an Assistant Professor in the Decision, Risk, and Operations division at Columbia Business School and a member of the Columbia Data Science Institute. His research interests lie at the interface of machine learning, operations research, and causal inference, with a particular emphasis on developing reliable learning methods for decision-making problems. Hong is a recipient of several awards and fellowships, including best paper awards at the Neural Information Processing Systems conference and the International Conference on Machine Learning (runner-up), and the best student paper award from the INFORMS Applied Probability Society. He received his Ph.D. from Stanford University where he was jointly advised by John Duchi and Peter Glynn, and worked as a research scientist at Facebook Core Data Science before joining Columbia.