[New blog post!] A gentle, (mostly) non-technical introduction to game-theoretic statistics, e-values, and my recent paper with Aaditya Ramdas on combining evidence across asymmetric information sets. Originally posted in the UChicago Data Science Institute Blog.

We develop anytime-valid inference methods for estimating & testing the time-varying mean score difference between two sequential forecasters. Published in Operations Research.

We develop a statistical inference approach for counterfactually evaluating and comparing abstaining classifiers using nonparametric causal inference. Published in NeurIPS.

We address an intriguing challenge for anytime-valid sequential inference that arises when combining evidence processes constructed using different information sets. Under review.

We formalize different notions of linear representations in LLMs (e.g., linear probes and steering vectors) and unify those notions by identifying an inner product that encodes semantic structure. Published in ICML 2024.

Building upon our recent work, we formalize the representations of categorical concepts in LLMs as convex polytopes and further show that hierarchical relations between concepts are represented as orthogonality. Accepted to ICLR 2025 for oral presentation.