61. Last-Iterate Convergence of Extragradient-based Methods
Invited abstract in session WE-2: Recent advances in computer-aided analyses of optimization algorithms I, stream Conic optimization: theory, algorithms and applications.
Wednesday, 14:10 - 15:50Room: M:O
Authors (first author is the speaker)
| 1. | Eduard Gorbunov
|
| Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) | |
| 2. | Adrien Taylor
|
| Inria/ENS | |
| 3. | Samuel Horvath
|
| Mohamed bin Zayed University of Artificial Intelligence (MBZUAI) | |
| 4. | Nicolas Loizou
|
| Department of Computer Science and Operations Research, Mila - Quebec Artificial Intelligence Institute & University of Montreal | |
| 5. | Gauthier Gidel
|
| Computer Science, Mila, Université de Montréal |
Abstract
Extragradient (EG) and Optimistic Gradient (OG) methods are among the most popular methods for solving monotone Lipschitz variational inequalities. However, despite the long history of these methods, the questions about their last-iterate convergence were resolved very recently. In this talk, we will discuss several results about the last-iterate convergence of EG and OG under Lipschitzness and the monotonicity of the operator. In particular, we will focus on the proofs of these results that were obtained via computer. Extensions to the case of negative comonotone operators will also be considered.
Keywords
- Complexity and efficiency of optimization algorithms
Status: accepted
Back to the list of papers