Research
Diagnosing Model Performance Under Distribution Shift
Tiffany Cai, Steve Yadlowsky, Hongseok Namkoong
Under revision at Operations Research; presented at FORC 2023, INFORMS 2023
Summary
When a model performs poorly out of distribution, how do we understand why performance became worse? We attribute change in model performance across distributions to X shifts and Y\|X shifts.Posterior Sampling via Autoregressive Generation
Kelly Wang Zhang*, Tiffany Cai*, Hongseok Namkoong, Daniel Russo
Forthcoming
To be presented at ICLR 2024 Workshop: Generative Models for Decision Making
Summary
We recast the problem of principled decision-making under uncertainty (Thompson Sampling) as autoregressive sequential modeling, trained via loss minimization.Constrained Learning for Causal Inference and Semiparametric Statistics
Tiffany Cai*, Yuri Fonseca*, Kaiwen Hou, Hongseok Namkoong
Forthcoming
Presented at CODE@MIT 2023
Summary
We recast the problem of creating asymptotically efficient estimators for the average treatment effect as constrained optimization.Tutorial: Modeling and Exploiting Data Heterogeneity under Distribution Shifts
Jiashuo Liu, Tiffany Cai, Peng Cui, Hongseok Namkoong
Presented at NeurIPS 2023