Research Bio

Nikita Zhivotovskiy is a tenure-track Assistant Professor in the Department of Statistics at UC Berkeley whose research lies at the interface of mathematical statistics, learning theory, and high-dimensional probability. He studies the fundamental question of what can be learned from data under minimal assumptions, and which structural features of a model or algorithm determine its performance. His work develops non-asymptotic, distribution-free guarantees that clarify when learning procedures generalize reliably and when no method can do better, producing sharp finite-sample bounds together with matching lower bounds. This includes results on distribution-free guarantees beyond uniform convergence, principled complexity measures for learning and prediction, and conceptually simple procedures that remain valid beyond idealized modeling assumptions. His research has appeared in journals such as Annals of Statistics, Bernoulli, Probability Theory and Related Fields, Electronic Journal of Probability, and Journal of the European Mathematical Society, and at conferences such as the Conference on Learning Theory and the IEEE Symposium on Foundations of Computer Science.

 

Research Expertise and Interest

mathematical statistics, applied probability, statistical learning theory