Source-free Semantic Regularization Learning for Semi-supervised Domain Adaptation
Xinyang Huang, Chuang Zhu 2, Ruiying Ren, Shengjie Liu, and Tiejun Huang · 2025
Read Source-free Semantic Regularization Learning for Semi-supervised Domain Adaptation by Xinyang Huang, Chuang Zhu 2, Ruiying Ren, Shengjie Liu, and Tiejun Huang (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-2501-01126.txt.
Description
Xinyang Huang and co-authors write a 2025 paper on adapting vision models to new data. A model trained in one setting often drops in skill when the scene or style changes. Their setting gives only a few labeled examples in the new domain and no access to the old training data.
The proposed method, called SERL, fine-tunes the old model using three forms of guidance drawn from the new data itself. One part sharpens how the model tells classes apart. One part uses easy cases to teach from hard cases. One part keeps new guesses in line with past ones to limit error. Tests on three standard data sets put the method at the top of its class.
Catalog record
| CREATOR | Xinyang Huang, Chuang Zhu 2, Ruiying Ren, Shengjie Liu, and Tiejun Huang |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 56 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2501-01126