Daniil Vankov

Daniil Vankov

Optimization, Game Theory, LLM inference

Publications

You can also find my articles on my Google Scholar profile.

Selected Publications

Generalized Smooth Stochastic Variational Inequalities: Almost Sure Convergence and Convergence Rates

Published in Transactions on Machine Learning Research (accepted with J2C certification, top 10 % contribution), 2025

We show that clipped first-order methods converge a.s. and in expectation to a solution of stochastic variational inequalities without relying on standard smoothness or boundedness assumptions.

Recommended citation: Vankov, Daniil, Angelia Nedić, and Lalitha Sankar. "Generalized smooth stochastic variational inequalities: Almost sure convergence and convergence rates." Transactions on Machine Learning Research (2025).
Download Paper

Optimizing ($L_0, L_1$)-Smooth Functions by Gradient Methods

Published in ICLR 2025, 2025

Our analysis significantly improves the best-known complexity bounds of gradient method for convex optimization under generalized smoothness and establishes new best-known guarantees for accelerated gradient method.

Recommended citation: Vankov, Daniil, Anton Rodomanov, Angelia Nedich, Lalitha Sankar, and Sebastian U. Stich. "Optimizing ($L_0, L_1$)-Smooth Functions by Gradient Methods." In ICLR. 2025.
Download Paper

Generalized smooth variational inequalities: Methods with adaptive stepsizes

Published in International Conference on Machine Learning (ICML) 2024, 2024

We introduce clipped first-order methods for solving non-monotone generalized smoothn variational inequalities, establishing the first convergence guarantees for gradient descent-ascent, extragarient, and optimistic methods that match or improve upon standard smooth rates. We also propose extragradient method with backtracking line-search that matches the same rates wihtout additional knoledge on the parameters.

Recommended citation: Vankov, Daniil, Angelia Nedich, and Lalitha Sankar. "Generalized smooth variational inequalities: Methods with adaptive stepsizes." In Forty-first International Conference on Machine Learning. 2024.
Download Paper