LLaVA-NeuMT: Selective Layer-Neuron Modulation for Efficient Multilingual Multimodal Translation
Jingxuan Wei, Caijun Jia, Qi Chen, Yujun Cai, Linzhuang Sun, Xiangxiang Zhang, Gaowei Wu, Bihui Yu · 2025
Read LLaVA-NeuMT: Selective Layer-Neuron Modulation for Efficient Multilingual Multimodal Translation by Jingxuan Wei, Caijun Jia, Qi Chen, Yujun Cai, Linzhuang Sun, Xiangxiang Zhang, Gaowei Wu, Bihui Yu (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-2507-18940.txt.
Description
Jingxuan Wei and a team of co-authors taught a vision model to translate in many tongues at low cost. Their paper, from twenty twenty-five, deals with multimodal translation, where a picture clears a text's fog. Two tongue work is strong. Many tongue work fights cross tongue bleed and weak sharing plans. Their frame, LLaVA-NeuMT, splits tongue own traits from tongue free ones.
A layer pick step finds the layers that count most for each tongue pair. A neuron step then wakes just the units a tongue needs. Readers in machine translation get a lean path to many tongue work with sight in the loop.
Catalog record
| CREATOR | Jingxuan Wei, Caijun Jia, Qi Chen, Yujun Cai, Linzhuang Sun, Xiangxiang Zhang, Gaowei Wu, Bihui Yu |
|---|---|
| YEAR | 2025 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 33 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2507-18940