SILO: Worst-case generation via minimax optimization in Wasserstein space

Abstract:

Generative models such as normalizing flows and diffusion processes have transformed how we represent complex, high-dimensional data. In this talk, I introduce a framework for worst-case generation formulated as a minimax optimization problem in Wasserstein space. This perspective reveals a unified view of risk-induced generation and distributional robustness, showing that worst-case distributions arise as perturbed (pushforwarded) distributions by a neural transport mapping. The formulation can be interpreted as a Stackelberg game between two players, namely the adversarial transport map (generator) and the model parameters (learner). I will present a single-loop gradient-descent–ascent (GDA) scheme involving transport maps in a functional space with convergence guarantees under mild regularity conditions when the min-max problem possibly lacks convex-concave structure, and illustrate how the framework enables risk-aware or risk-induced worst-case data generation for robust learning and decision-making.

Bio:

Yao Xie is the Coca-Cola Foundation Chair and Professor in the H. Milton Stewart School of Industrial and Systems Engineering at the Georgia Institute of Technology and Associate Director of the Machine Learning Center (ML@GT). She received her Ph.D. in Electrical Engineering (minor in Mathematics) from Stanford University and was previously a Research Scientist at Duke University. Her research lies at the intersection of statistics, machine learning, and optimization, focusing on developing statistically powerful and computationally efficient methods for high-dimensional, sequential, and spatio-temporal data with strong theoretical guarantees and real-world impact. She is a Member of Cohort 2026 in the New Voices in Sciences, Engineering, and Medicine program of the U.S. National Academies and the IEEE Information Theory Society Distinguished Lecturer for 2026–2027, and her honors include the NSF CAREER Award (2017), INFORMS Wagner Prize Finalist (2021), INFORMS Gaver Early Career Award (2022), and C.W.S. Woodroofe Award (2024). She serves as an Associate Editor for IEEE Transactions on Information Theory, Journal of the American Statistical Association–Theory and Methods, The American Statistician, Operations Research, Annals of Applied Statistics, Sequential Analysis, and INFORMS Journal on Data Science, and as an Area Chair for NeurIPS, ICML, and ICLR, and Senior Program Committee Member for AAAI.