Enhancing generalization with multi-modal relations

Open Webpage

Mengxuan Wang, Xingchen Shen, Jingyao Wang, Lingyu Si, Hongjie Jiang

Published: 2026, Last Modified: 04 Nov 2025Inf. Fusion 2026EveryoneRevisionsBibTeXCC BY-SA 4.0
Abstract: Highlights•Unified representation omits the specific modality information.•Various modalities data share certain level similarity.•Modal relations re-weight multimodal specific predictive heads can facilitate generalization.•Detailed theoretical analysis and rigorous proof can support our method.