Nikita Zhivotovskiy

Junior Research Scientist

Research Interests

I am interested in problems appearing in the analysis of classic models of statistics and statistical learning. This includes the mathematics behind the sample complexity analysis for different learning algorithms in classiffcation and regression, general dependence on noise and adaptivity as well as combinatorics of VC classes, PAC learning, and the sample compression. I am also interested in theoretical guarantees for the learning in a presence of unlabeled data (semi-supervised and transductive learning) under minimal restrictions on the model.



S. Hanneke, N. Zhivotovskiy. Localization of VC Classes: Beyond Local Rademacher Complexities. To appear in Theoretical Computer Science. Short conference version in Algorithmic Learning Theory, LNCS, arXiv:1606.00922.


N. Zhivotovskiy. Combinatorial Bounds of Overfitting with Sub-logarithmic Order of Growth. Proceedings of MIPT, pp. 42 – 54.

G. Blanchard, I. Tolstikhin, N. Zhivotovskiy. Permutational Rademacher Complexity: a New Complexity Measure for Transductive Learning. Algorithmic Learning Theory, Lecture Notes in Computer Science, arXiv:1505.02910.