Beyond Short Steps in Frank-Wolfe Algorithms
David Martínez-Rubio, Sebastian Pokutta · 2025
Read Beyond Short Steps in Frank-Wolfe Algorithms by David Martínez-Rubio, Sebastian Pokutta (2025) in the Tower of Babel, a free online library of public-domain works preserved in clean, searchable text. No sign-up. The full text is served as a plain-text file at a predictable URL, where text is present: https://antarctic-labs.com/tower-of-babel/arxiv-2501-18773.txt.
Description
David Martínez-Rubio and Sebastian Pokutta present new ways to make a classic optimization method work better, in a research paper from 2025. The method, called Frank-Wolfe, solves problems step by step and is used when each step must stay simple. Past work often used short steps that lean on how smooth the target function is.
The authors add an optimistic version of the method, which uses past steps to plan the next one. They prove that this version closes the gap between the current answer and the best answer at a good rate. They also give a broader short-step rule based on a gap that a computer can check. Tests show the new method beats older ones in practice, and the step rule also helps plain gradient descent.
Catalog record
| CREATOR | David Martínez-Rubio, Sebastian Pokutta |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 162 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-18773