I am an assistant professor at University of Michigan’s Department of Biostatistics, working in cancer, immunology, and their intersection (e.g. CAR-T therapy), but I am generally driven to solve problems that will help people.
My work focuses on improving healthcare through:
- developing predictive methods with emphasis on interpretability (often inspired by and leveraging advances in AI),
- creating well-documented, user-friendly software, and
- teaching statistics for biomedical applications in a friendly, accessible way.
I completed my PhD in Biomedical Data Science at Stanford, where I was advised by Rob Tibshirani. I also hold a BA in Mathematics and an MS in Data Science from New College of Florida, where I was advised by Pat McDonald and Gary Kalmanovich.
Before my PhD, I led the math content team at Wolfram|Alpha. Notably, we developed Step-by-step Solutions.
Want to collaborate? Please reach out at ercr@umich.edu.
News
New preprint: Interpretable AI with local distillation, with Yiling Huang and Snigdha Panigrahi. A black-box teacher guides a sparse linear model at each query point; randomized refits identify stable features and patient subgroups.
Talk, Sept 8, 2026: JRSS Series B Joint Editors’ invited session, RSS International Conference, Bournemouth. Speaking on Pretraining and the lasso.
Selected publications
Preprints
Craig, Huang, Panigrahi. Interpretable AI with local distillation. arXiv, 2026. arXiv · code
Craig, Tibshirani. Supervised learning pays attention. arXiv, 2025. arXiv
Published
Craig, Pilanci, Le Menestrel, Narasimhan, Rivas, Gullaksen, Dehghannasiri, Salzman, Taylor, Tibshirani. Pretraining and the lasso. Journal of the Royal Statistical Society, Series B, 2026. doi
Zaslavsky, Craig, et al. Disease diagnostics using machine learning of B cell and T cell receptor sequences. Science, 387(6736), 2025. Co-first author. doi
Craig, Keyes, et al. Annotation-free discovery of disease-relevant cells in single-cell datasets. Science Advances, 11(35), 2025. doi
Hamilton, Craig, et al. CAR19 monitoring by peripheral blood immunophenotyping reveals histology-specific expansion and toxicity. Blood Advances, 8(12), 2024. doi
Full list of publications, talks, and grants in my CV and on Google Scholar.
Software
ptLasso. Pretraining for the lasso — transfer learning for sparse, interpretable linear models. Video and code tutorials!
MMIL. Mixture models for multiple-instance learning: find disease-relevant cells from patient-level labels.
sweetspot. Find and assess treatment-effect sweet spots in clinical trials.
Contact