Research map / Label-efficient and robust learning / Domain adaptation and generalization
MER-DG: Modality-Entropy Regularization for Multimodal Domain Generalization
Yavuz Yarici, Ghassan AlRegib
Read the paperCheck your idea against 137,217 papers
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 |
|---|---|---|---|
| add MER-DG entropy regularizer to standard fusion | Fusion (video+audio) · EPIC-Kitchens multi-source | avg top-1 accuracy: 59.09 -> 62.82 | Improved |
| add MER-DG to Fusion | Fusion (video+audio) · HAC multi-source | avg top-1 accuracy: 63.70 -> 71.43 | Improved |
| add MER-DG to CMRF | CMRF (video+audio) · EPIC-Kitchens multi-source | avg top-1 accuracy: 63.91 -> 64.60 | Improved |
| add MER-DG to CMRF | CMRF (video+flow) · HAC multi-source | avg top-1 accuracy: 72.64 -> 72.72 | No clear effect |
| add marginal-entropy loss only (ablation) | MER-DG on Fusion · EPIC-Kitchens V+A | avg accuracy: 59.09 -> 62.55 | Improved |
| add spectral-entropy loss only (ablation) | MER-DG on Fusion · EPIC-Kitchens V+A | avg accuracy: 59.09 -> 61.19 | Improved |
| combine marginal and spectral entropy losses (ablation) | MER-DG · EPIC-Kitchens V+A | avg accuracy: 62.55 -> 62.82 | Improved |
| add MER-DG to independently trained unimodal encoder (ablation) | Uni-Video · EPIC-Kitchens target domain | accuracy: 59.46 -> 59.51 | No clear effect |
| add MER-DG to fusion training | Fusion video encoder · EPIC-Kitchens target domain | standalone accuracy: 56.53 -> 58.61 | Improved |
| use cross-attention fusion with MER-DG instead of late fusion (ablation) | MER-DG · EPIC-Kitchens V+A | avg accuracy: 59.48 -> 61.13 | Improved |
| use standard regularizers (label smoothing, weight decay, dropout, noise) instead of MER-DG (ablation) | Fusion · EPIC-Kitchens V+A | avg accuracy: 59.09 -> 60.21 (best, label smoothing) | No clear effect |
| add MER-DG for robustness to Gaussian noise or missing audio | Fusion · EPIC-Kitchens V+A corruptions | accuracy drop: -10.94 -> -8.55 (drop audio) | Improved |
Closest papers by meaning
- Cross-modal Representation Flattening for Multi-modal Domain Generalization (NeurIPS 2024 · similarity 0.89)
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 2025 · similarity 0.89)
- Multimodal Unsupervised Domain Generalization by Retrieving Across the Modality Gap (ICLR 2025 · similarity 0.88)
- Beyond Modality Fusion: Deep Ensembles for Multimodal Classification (NeurIPS 2026 · similarity 0.88)
- Are We Making Progress in Multimodal Domain Generalization? A Comprehensive Benchmark Study (NeurIPS 2026 · similarity 0.88)
- On the Out-Of-Distribution Generalization of Large Multimodal Models (CVPR 2025 · similarity 0.87)
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.
