V-Zero: Self-Improving Multimodal Reasoning with Zero Annotation
Han Wang, Yi Yang, Jingyuan Hu, Minfeng Zhu, Wei Chen · 2026
Read V-Zero: Self-Improving Multimodal Reasoning with Zero Annotation by Han Wang, Yi Yang, Jingyuan Hu, Minfeng Zhu, Wei Chen (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-2601-10094.txt.
Description
Han Wang, Yi Yang, Jingyuan Hu, Minfeng Zhu, and Wei Chen teach sight and text models with no marked data. Such models have grown strong at tasks that mix image and text. But top grades rest on vast marked sets. Those sets cost much time and pay to build.
The team made V-Zero, a post train plan that needs no marks. It uses raw images with no tags at all. The model makes its own tasks and checks from those images. It then learns from its own work. The plan aims to lift skill while it cuts the need for hand marked sets.
Catalog record
| CREATOR | Han Wang, Yi Yang, Jingyuan Hu, Minfeng Zhu, Wei Chen |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 35 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2601-10094