Smoothing the Shift: Towards Stable Test-Time Adaptation under Complex Multimodal Noises
Zirun Guo, Tao Jin · 2025
Read Smoothing the Shift: Towards Stable Test-Time Adaptation under Complex Multimodal Noises by Zirun Guo, Tao Jin (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-2503-02616.txt.
Description
Zirun Guo and Tao Jin steady a model's on-the-fly adaptation when several senses go bad at once, in a 2025 paper. Test-time adaptation retunes a model on unlabeled new data, but mixed corruption across image, sound, and text, plus missing channels, can wreck the knowledge it started with.
Their method first smooths the sample stream with a spread-based filter, so sudden shifts arrive gently. It then leans on single-sense features to pick confident, information-rich samples for tuning. A sharing term pulls the senses' information into agreement, cutting discrepancies. The combined design targets what the authors name wild multimodal adaptation, where every corruption arrives blended. Robust tuning matters for field robots and phones that cannot phone home for retraining.
Catalog record
| CREATOR | Zirun Guo, Tao Jin |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 49 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2503-02616