Revisiting Cross-Attention Mechanisms: Leveraging Beneficial Noise for Domain-Adaptive Learning
Zelin Zang, Yehui Yang, Fei Wang, Liangyu Li, Baigui Sun · 2026
Read Revisiting Cross-Attention Mechanisms: Leveraging Beneficial Noise for Domain-Adaptive Learning by Zelin Zang, Yehui Yang, Fei Wang, Liangyu Li, Baigui Sun (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-2603-17474.txt.
Description
Zang and co-authors present DACSM, a framework for unsupervised domain adaptation with cross-attention. Style and scale gaps often break feature alignment between source and target domains. Their design adds controlled noise to attention, pushing the model to track content over style. A second module aligns features across several resolutions at once. On VisDA-2017 the method beat a strong baseline by up to 2.3 points, they report. The truck class gained 5.9 points.
Catalog record
| CREATOR | Zelin Zang, Yehui Yang, Fei Wang, Liangyu Li, Baigui Sun |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 74664 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2603.17474