A Recursive Domain- and Objective -Adaptive Frank-Wolfe Algorithm
2026
Read A Recursive Domain- and Objective -Adaptive Frank-Wolfe Algorithm (2026) 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-2606-03383.txt.
Description
Kaiser and Sutter extend the Frank-Wolfe algorithm to problems where the domain and goal must be learned. Frank-Wolfe minimizes a convex function over a convex compact set, with no projections. In data-driven settings, both the set and the function start unknown. Their recursive variant weaves estimators into the optimization itself.
At each step, the algorithm refines its estimates of the domain and the objective from data. It then takes a Frank-Wolfe step on the current estimates. The projection-free trait of the classic method survives. Convergence guarantees show the error scales with how good the learned estimators are. Two experiments back the theory. Behavior matches the classic algorithm run with exact knowledge of domain and objective. Compute cost falls a lot at the same time. The work suits problems where the true feasible set is itself a guess.
Catalog record
| YEAR | 2026 |
|---|---|
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 271 KB |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2606.03383