Abstract:For min-max optimization and variational inequalities problems (VIP) encountered in diverse machine learning tasks, Stochastic Extragradient (SEG) and Stochastic Gradient Descent Ascent (SGDA) have emerged as preeminent algorithms. Constant step-size variants of SEG/SGDA have gained popularity, with appealing benefits such as easy tuning and rapid forgiveness of initial conditions, but their convergence behaviors are more complicated even in rudimentary bilinear models. Our work endeavors to elucidate and quantify the probabilistic structures intrinsic to these algorithms. By recasting the constant step-size SEG/SGDA as time-homogeneous Markov Chains, we establish a first-of-its-kind Law of Large Numbers and a Central Limit Theorem, demonstrating that the average iterate is asymptotically normal with a unique invariant distribution for an extensive range of monotone and non-monotone VIPs. Specializing to convex-concave min-max optimization, we characterize the relationship between the step-size and the induced bias with respect to the Von-Neumann's value. Finally, we establish that Richardson-Romberg extrapolation can improve proximity of the average iterate to the global solution for VIPs. Our probabilistic analysis, underpinned by experiments corroborating our theoretical discoveries, harnesses techniques from optimization, Markov chains, and operator theory.

Stochastic Bregman extragradient algorithm with line search for stochastic mixed variational inequalities

An accelerated stochastic extragradient-like algorithm with new stepsize rules for stochastic variational inequalities

On quasi-inertial double extragradient-type iterations for variational inequalities and Bregman relatively asymptotically nonexpansive mapping

Extension of the subgradient extragradient algorithm for solving variational inequalities without monotonicity

Subgradient extragradient method with double inertial steps for quasi-monotone variational inequalities

An accelerated subgradient extragradient algorithm for solving bilevel variational inequality problems involving non-Lipschitz operator

Double inertial extragradient algorithms for solving variational inequality problems with convergence analysis

New Extragradient-Type Methods for Solving Variational Inequalities

Mann-type extragradient algorithm for solving variational inequality and fixed point problems

On the convergence properties of non-Euclidean extragradient methods for variational inequalities with generalized monotone operators

Convergence of Extragradient SVRG for Variational Inequalities: Error Bounds and Increasing Iterate Averaging

A stochastic two-step inertial Bregman proximal alternating linearized minimization algorithm for nonconvex and nonsmooth problems

Tseng's extragradient method with double projection for solving pseudomonotone variational inequality problems in Hilbert spaces

A Bregman Proximal Stochastic Gradient Method with Extrapolation for Nonconvex Nonsmooth Problems

Stochastic Methods in Variational Inequalities: Ergodicity, Bias and Refinements

The extragradient algorithm with inertial effects for solving the variational inequality

Viscosity extragradient algorithms with double inertial extrapolation steps for quasimonotone variational inequalities and fixed point problems

A subgradient extragradient algorithm with inertial effects for solving strongly pseudomonotone variational inequalities

Modified Extragradient Methods for Solving Variational Inequalities

Extragradient method with feasible inexact projection to variational inequality problem

Weak and strong convergence results for solving monotone variational inequalities in reflexive Banach spaces