Papers by Lanxin Bi
Two Challenges, One Solution: Robust Multimodal Learning through Dynamic Modality Recognition and Enhancement (2025.findings-emnlp)
Copied to clipboard
| Challenge: | Existing methods require full-modality data during training phase or require explicit annotations to detect missing modalities. |
| Approach: | They propose a Dynamic modality Recognition and Enhancement for Adaptive Multimodal fusion framework that directs selective reconstruction of missing or underperforming modalities. |
| Outcome: | The proposed framework outperforms several baseline and state-of-the-art models on three benchmark datasets. |