Adapting in the Dark: Efficient and Stable Test-Time Adaptation for Black-Box Models
Yunbei Zhang, Shuaicheng Niu, Chengyi Cai, Feng Liu, Jihun Hamm · 2026
Read Adapting in the Dark: Efficient and Stable Test-Time Adaptation for Black-Box Models by Yunbei Zhang, Shuaicheng Niu, Chengyi Cai, Feng Liu, Jihun Hamm (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-2604-15609.txt.
Description
Zhang and four colleagues introduce BETA, a framework for test-time adaptation of black-box models reachable only through APIs. A lightweight local white-box steering model creates a tractable gradient path, with prediction harmonization, consistency regularization, and prompt-learning filtering enabling stable adaptation with no extra API calls. On ImageNet-C it gains 7.1 percent on ViT-B/16 and 3.4 percent on CLIP, and matches zeroth-order optimization at 250 times lower cost on a commercial API.
Catalog record
| CREATOR | Yunbei Zhang, Shuaicheng Niu, Chengyi Cai, Feng Liu, Jihun Hamm |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 91029 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2604.15609