Research map / Vision-language and multimodal / Cross-modal learning
Multi-Head Modularization to Leverage Generalization Capability in Multi-Modal Networks
Jun-Tae Lee, Hyunsin Park, Sungrack Yun, Simyung Chang
AI-extracted. The measured effects below were read by AI from the paper's full text (no summary yet). Numbers and claims can be wrong or missing context, so check the paper before you cite or build on it.
Measured effects reported in the paper
| What was changed | Where | Measured | Effect |
|---|---|---|---|
| Apply multi-head modularization to the AVE baseline | AVE baseline · AVE | accuracy (%): 2.9% | Improved |
| Apply multi-head modularization to Bas-Net | Bas-Net · THUMOS14 | average mAP (%): 1.8% | Improved |
| Apply multi-head modularization to MulT | MulT · CMU-MOSEI | Acc7 (%): 49.1 -> 50.4 | Improved |
| Apply multi-head modularization to MulT | MulT · CMU-MOSEI | Acc2 (%): 80.6 -> 80.9 | Improved |
| Apply multi-head modularization to MulT | MulT · CMU-MOSEI | F1 (%): 81.0 -> 81.4 | Improved |
| Remove module-freezing during iterative multi-head training (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 77.2 | Hurt |
| Use 2 folds instead of the selected 4 folds (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 76.0 | Hurt |
| Use 3 folds instead of the selected 4 folds (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 77.8 | Hurt |
| Increase the fold count from 4 to 5 (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 78.0 | Hurt |
| Use one head and one head-specific training set (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 75.5 | Hurt |
| Use one head with four changing head-specific training sets (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 75.6 | Hurt |
| Use four heads with one shared head-specific training set (ablation) | MHM on AVE · AVE | accuracy (%): 78.5 -> 76.0 | Hurt |
Closest papers by meaning
- Breaking Neural Network Scaling Laws with Modularity (ICLR 2025 · similarity 0.87)
- Cross-modal Representation Flattening for Multi-modal Domain Generalization (NeurIPS 2024 · similarity 0.86)
- ReconBoost: Boosting Can Achieve Modality Reconcilement (ICML 2024 · similarity 0.86)
- A Closer Look at Multimodal Representation Collapse (ICML 2025 · similarity 0.85)
- MODA: MOdular Duplex Attention for Multimodal Perception, Cognition, and Emotion Understanding (ICML 2025 · similarity 0.85)
- Connect, Collapse, Corrupt: Learning Cross-Modal Tasks with Uni-Modal Data (ICLR 2024 · similarity 0.85)
Writing a paper in this area? scime draft is a free LaTeX editor with AI included, real-time co-authors and a check of your references and venue rules before you submit. It also imports OpenAI Prism projects.
