My thesis develops nonparametric statistical inference methods for comparing black-box predictors, namely sequential forecasters and abstaining classifiers. I leverage techniques from safe, anytime-valid inference (SAVI) and game-theoretic statistics, including test supermartingales, e-processes, and confidence sequences; model and forecast evaluation methods, including proper scoring rules; and nonparametric causal inference.

During my detour from the Ph.D. program (2017-2020), I also developed interest in deep learning (generalization and invariant prediction) as well as its applications to natural language processing (language modeling, understanding, and generation; grammatical error correction; and multilingual dataset construction) and other “discrete” structure problems (drug discovery). I am intrigued by the recent advances in large language models, and I am interested in leveraging statistical approaches to assessing and improving their reliability and robustness.