Seminar • Machine Learning • Distributionally Robust Machine Learning

Please note: This seminar will take place in DC 1304.

Shiori Sagawa, PhD candidate
Department of Computer Science, Stanford University

Machine learning systems are powerful, but they can fail due to distribution shifts: mismatches in the data distribution between training and deployment. Distribution shifts are ubiquitous and have real-world consequences: models can fail on subpopulations (e.g., demographic groups) and on new domains unseen during training (e.g., new hospitals).

In this talk, I will discuss my work on building machine learning systems that are robust to distribution shifts. First, to mitigate subpopulation shifts, I extend the classical framework of distributionally robust optimization (DRO), which trains models that are certifiably robust to a family of shifts via minimax optimization. My work enables DRO on modern neural networks and on real-world shifts by addressing the challenges of scalable optimization and overparameterization and by proposing new DRO formulations appropriate for real-world shifts. Second, I develop a benchmark and an algorithm for domain shifts. My work demonstrates the importance of anchoring algorithm development to real-world shifts: on our WILDS benchmark of real-world distribution shifts, existing methods fail, despite their successes on prior benchmarks with synthetic shifts. Furthermore, I show that existing methods fail because their key assumption does not hold on real-world shifts, and based on this insight, propose a method that achieves state-of-the-art performance on multiple WILDS datasets. Altogether, my algorithms have mitigated a wide range of distribution shifts in the wild, from subpopulation shifts in language modeling to domain shifts in wildlife monitoring and histopathology.

Bio: Shiori Sagawa is a final-year PhD Candidate in Computer Science at Stanford University, advised by Percy Liang. Her research focuses on algorithms for building reliable machine learning systems. She has been recognized with the Stanford Graduate Fellowship and an Apple Scholars in AI/ML PhD Fellowship. Prior to her PhD, she received her B.A. in Computer Science and Molecular and Cell Biology from UC Berkeley, and she worked at D. E. Shaw Research.