Abdurakhmon Sadiev

TL;DR: The proposed approach works, at least, like the best existing approaches, but for a special set-up (simplex type constraints and closeness of Lipschitz constants in 1 and 2 norms) the approach reduces n/logn times the required number of oracle calls (function calculations).

TL;DR: Stochastic smooth (strongly) convex-concave saddle-point problems using zeroth-order oracles are solved, and theoretical analysis shows that in the case when the optimization set is a simplex, the authors lose only $\log n$ times in the stochastic convergence term.

TL;DR: In this paper, the outer minimization problem was considered as a minimization with inexact oracle, which is either minimization or a maximization problem, and an inexact variant of Vaydya's cutting-plane method or a variant of accelerated gradient method was used to solve the outer problem.

TL;DR: In this paper, the outer minimization problem was considered as a minimization with an inexact oracle, which is either minimization or maximization problem, and a framework was proposed to solve the outer problem with nonasymptotic complexity.

TL;DR: This work develops a distributed variant of random reshuffling with gradient compression (Q-RR), and shows how to reduce the variance coming from gradient quantization through the use of control iterates, and proposes a variant of Q-RR called Q-NASTYA to have a better fit to Federated Learning applications.