About

I am a Reader in Mathematics at the University of York working on the mathematical foundations of modern artificial intelligence. My research combines ideas from statistical learning theory, empirical process theory and functional analysis to develop rigorous mathematical models for understanding and improving machine learning systems.

Modern AI systems are extraordinarily capable but difficult to analyse. I am interested in developing tractable mathematical surrogate models that capture their essential behaviour and allow us to analyse and improve them. My recent work explores this perspective in the context of compute-efficient AI, including model compression and efficient representations.

Recent News

• 2026: Compressed Empirical Measures (in finite dimensions), published in Annales Henri Lebesgue.
• 2026: Senior Area Chair, NeurIPS.
• 2026: Area Chair, ICML.
• 2026: Invited seminar, Université de Versailles.
• 2026: Designed and taught the MAGIC PhD course Mathematical Foundations of AI.
• 2025: Estimating the Mixing Coefficients of Geometrically Ergodic Markov Processes, published in IEEE Transactions on Information Theory.

Selected Publications

• S. Grünewälder, Compressed Empirical Measures (in finite dimensions),  Annales Henri Lebesgue, 2026.
• S. Grünewälder and A. Khaleghi, Estimating the Mixing Coefficients of Geometrically Ergodic Markov Processes, IEEE Transactions on Information Theory, 2025. 
• S. Page and S. Grünewälder,  The Goldenshluger-Lepski Method for Constrained Least-Squares Estimators over RKHSs, Bernoulli, 2021.
• C. Pike-Burke, S. Agrawal, C. Szepesvári and S. Grünewälder, Bandits with Delayed Anonymous Feedback, International Conference on Machine Learning (ICML), 2018.