1: \begin{abstract}

2: We develop new adaptive algorithms for variational inequalities with

3: monotone operators, which capture many problems of interest, notably

4: convex optimization and convex-concave saddle point problems. Our

5: algorithms automatically adapt to unknown problem parameters such

6: as the smoothness and the norm of the operator, and the variance of

7: the stochastic evaluation oracle. We show that our algorithms are

8: universal and simultaneously achieve the optimal convergence rates

9: in the non-smooth, smooth, and stochastic settings. The convergence

10: guarantees of our algorithms improve over existing adaptive methods

11: by a $\Omega(\sqrt{\ln T})$ factor, matching the optimal non-adaptive

12: algorithms. Additionally, prior works require that the optimization

13: domain is bounded. In this work, we remove this restriction and give

14: algorithms for unbounded domains that are adaptive and universal.

15: Our general proof techniques can be used for many variants of the

16: algorithm using one or two operator evaluations per iteration. The

17: classical methods based on the ExtraGradient/MirrorProx algorithm

18: require two operator evaluations per iteration, which is the dominant

19: factor in the running time in many settings.

20: \end{abstract}

21: