Saptarshi Roy
Assistant Professor of Statistics and Data Science
Research Interests
Professor Roy's research interests lie in the intersection of theoretical statistics and modern ML problems, including GenAI, privacy, and reinforcement learning. His research primarily aims to reduce the theory-practice gap by developing statistically grounded methodologies to address contemporary problems in the ML landscape. His research has been featured at competitive global venues such as NeurIPS, ICML, AISTATS, and in other prestigious journals.
Email
Prior to joining MBZUAI, Professor Roy was a joint Postdoctoral Fellow in the Computer Science department and Statistics & Data Science department at the University of Texas at Austin. There, he worked on the theoretical aspects of generative flow models, which were supported by the Institute for Foundations of Machine Learning and the Institute for Emerging CORE Methods of Data Science. Prior to that, he completed his Ph.D. in Statistics from the University of Michigan, Ann Arbor. Apart from research, Professor Roy has also mentored students in doing research.
- Ph.D. in Statistics, University of Michigan, Ann Arbor, USA
- M.Stat. in Statistics, Indian Statistical Institute, India
- B.Stat. in Statistics, Indian Statistical Institute, India
- "Silver reviewer" for ICML, 2026
- Data Challenge Expo winner (Professional division) at JSM, 2025
- Honorary mention for “Best Thesis Award 2024”
- Best Poster Award in UmichSML Workshop, 2022
- Certificate of Merit from National Board of Higher Mathematics for outstanding performance in Indian National Mathematical Olympiad, 2013
- Roy, S., Rinaldo, A., & Sarkar, P.: "Low-Dimensional Adaptation of Rectified Flow: A New Perspective through the Lens of Diffusion and Stochastic Localization", arXiv preprint arXiv:2601.15500, 2026.
- Roy, S., Tewari, A., & Zhu, Z.: "High-dimensional variable selection with heterogeneous signals: A precise asymptotic perspective", Bernoulli, 31(2), 1206-1229, 2025.
- Roy, S., Chakraborty, S., & Basu, D.: "FLIPHAT: Joint Differential Privacy for High Dimensional Linear Bandits", in International Conference on Artificial Intelligence and Statistics (pp. 2359-2367), PMLR, April 2025.
- Chakraborty, S., Roy, S., & Tewari, A.: "Thompson sampling for high-dimensional sparse linear contextual bandits", in International Conference on Machine Learning (pp. 3979-4008) PMLR, July 2023.
- Roy, S., Bansal, V., Sarkar, P., & Rinaldo, A.: "On the convergence and straightness of flow matching models", in The 29th International Conference on Artificial Intelligence and Statistics.