Plug, Play, and Fortify: A Low-Cost Module for Robust Multimodal Image Understanding Models
Siqi Lu, Wanying Xu, Yongbin Zheng, Wenting Luan, Peng Sun, Jianhang YaoNational University of Defense Technology {lusiqi9803, wanying_xu, zybnudt}@nudt.edu.cn{luanwentingnudt, sunpeng, yaojianhang23}@nudt.edu.cn · 2026
Read Plug, Play, and Fortify: A Low-Cost Module for Robust Multimodal Image Understanding Models by Siqi Lu, Wanying Xu, Yongbin Zheng, Wenting Luan, Peng Sun, Jianhang YaoNational University of Defense Technology {lusiqi9803, wanying_xu, zybnudt}@nudt.edu.cn{luanwentingnudt, sunpeng, yaojianhang23}@nudt.edu.cn (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-2602-22644.txt.
Description
Siqi Lu and five co-authors harden multi-input models against missing inputs. When one input drops out, scores often crash. The team traces the weakness to lopsided learning: the model leans on favoured inputs and starves the rest. Their insight is that this skew can be read in wave patterns.
A ratio score first gauges input preference from wave-domain traits. Guided by that score, a weight-sharing module then re-balances each branch's share during training, pushing learning toward a fuller picture. The module plugs into varied bases, from convolutional nets to vision transformers. It gave steady gains across tasks and input mixes, even lifting leading methods built for the missing-input problem.
Catalog record
| CREATOR | Siqi Lu, Wanying Xu, Yongbin Zheng, Wenting Luan, Peng Sun, Jianhang YaoNational University of Defense Technology {lusiqi9803, wanying_xu, zybnudt}@nudt.edu.cn{luanwentingnudt, sunpeng, yaojianhang23}@nudt.edu.cn |
|---|---|
| YEAR | 2026 |
| COLLECTION | PAPERS |
| FORMAT | TXT |
| FILE SIZE | 90 KB |
| VERSION | 1.0 |
| LICENSE | http://creativecommons.org/licenses/by/4.0/ |
| RIGHTS | CC BY |
| SOURCE | arXiv |
Source: https://arxiv.org/abs/2602-22644