Gradient-Guided Annealing for Domain Generalization
Aristotelis Ballas, Christos Diou · 2025
Read Gradient-Guided Annealing for Domain Generalization by Aristotelis Ballas, Christos Diou (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-2502-20162.txt.
Description
Aristotelis Ballas and Christos Diou warm up neural networks by annealing them toward agreement in a 2025 paper. Models must generalize to data distributions never seen in training, and the authors observe that the earliest training steps decide much of that ability. When domains pull gradients in conflicting directions, training settles into minima that miss the shared features.
Their Gradient-Guided Annealing repeatedly perturbs and resettles the parameters early on, hunting for spots where every domain's gradient points the same way. Starting from such aligned ground, models land in more robust minima. On five standard benchmarks the method alone matched or beat the state of the art, and it lifted older methods further when combined.
Catalog record
| CREATOR | Aristotelis Ballas, Christos Diou |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 58 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2502-20162