Beyond Short Steps in Frank-Wolfe Algorithms
David Martínez-Rubio, Sebastian Pokutta
Abstract
We introduce novel techniques to enhance Frank-Wolfe algorithms by leveraging function smoothness beyond traditional short steps. Our study focuses on Frank-Wolfe algorithms with step sizes that incorporate primal-dual guarantees, offering practical stopping criteria. We present a new Frank-Wolfe algorithm utilizing an optimistic framework and provide a primal-dual convergence proof. Additionally, we propose a generalized short-step strategy aimed at optimizing a computable primal-dual gap. Interestingly, this new generalized short-step strategy is also applicable to gradient descent algorithms beyond Frank-Wolfe methods. Empirical results demonstrate that our optimistic algorithm outperforms existing methods, highlighting its practical advantages.
BibTeX
@inproceedings{
martinez-rubio2026beyond,
title={Beyond Short Steps in Frank-Wolfe Algorithms},
author={David Mart{\'\i}nez-Rubio and Sebastian Pokutta},
booktitle={The Fourteenth International Conference on Learning Representations},
year={2026},
url={https://openreview.net/forum?id=HBmZlcD8Ue}
}