Feature-Space Planes Searcher: A Universal Domain Adaptation Framework for Interpretability and Computational Efficiency
Zhitong Cheng+, Yiran Jiang+, Yulong Ge, Yufeng Li, Zhongheng Qin, Rongzhi Lin, Jianwei Ma · 2025
Read Feature-Space Planes Searcher: A Universal Domain Adaptation Framework for Interpretability and Computational Efficiency by Zhitong Cheng+, Yiran Jiang+, Yulong Ge, Yufeng Li, Zhongheng Qin, Rongzhi Lin, Jianwei Ma (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-2508-18693.txt.
Description
Zhitong Cheng and co-authors adapt vision models to new domains by moving only the decision borders, in a 2025 paper. Models trained on one data set sag on another, and the standard cure, fine-tuning the whole network, costs memory and scrambles features unpredictably. Their analysis finds big pretrained models keep tidy geometry in feature space: classes cluster and separate across domains. What shifts is mainly where the borders fall.
Their framework searches for fresh border planes in that frozen space, hence the name, Feature-Space Planes Searcher. Features are extracted once, offline, so whole data sets fit in one computation pass. The approach keeps the process interpretable and cheap. Evaluations show competitive adaptation without touching the encoder.
Catalog record
| CREATOR | Zhitong Cheng+, Yiran Jiang+, Yulong Ge, Yufeng Li, Zhongheng Qin, Rongzhi Lin, Jianwei Ma |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 60 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2508-18693