Sharp Risk Bounds for Early-Stopping in Gaussian Linear Regression
Tobias Wegel, Gil Kur, Patrick Rebeschini · 2025
Read Sharp Risk Bounds for Early-Stopping in Gaussian Linear Regression by Tobias Wegel, Gil Kur, Patrick Rebeschini (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-2503-03426.txt.
Description
Tobias Wegel, Gil Kur, and Patrick Rebeschini pin down how well stopped-early fitting can do on noisy linear problems. Mirror descent is a fitting routine that improves step by step; halting it early acts as a guard against overfitting. The setting is high-dimensional regression with Gaussian noise, over any convex constraint set and any design. The benchmark is the least squares estimator, whose best-known risk bounds rest on a local width measure of the constraint set.
The paper shows those same sharp bounds carry over to the early-stopped routine. The transfer works when the routine's guiding potential meets conditions tied to a size functional of the set. Those conditions let the authors build fresh potentials and judge old ones. Consequences include general tests for when early stopping is minimax optimal, and the tightest known bound for the l1-constrained case.
Catalog record
| CREATOR | Tobias Wegel, Gil Kur, Patrick Rebeschini |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 119 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-03426