Research map / Vision-language and multimodal / Cross-modal learning
Calibrating Multimodal Learning
Huan Ma, Qingyang Zhang, Changqing Zhang, Bingzhe Wu, Huazhu Fu, Joey Tianyi Zhou, Qinghua Hu
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 CML regularization to CPM-Nets (ablation) | Type I multimodal classifier · YaleB | VRR (%): 39.15±4.97 -> 15.05±1.12 | Improved |
| Add CML regularization to CPM-Nets (ablation) | Type I multimodal classifier · TUANDROMD | Accuracy: 84.66±0.43 -> 85.20±0.81 | Improved |
| Add CML regularization to MIWAE (ablation) | Type II multimodal classifier · TUANDROMD | Accuracy: 86.32±0.85 -> 88.69±0.99 | Improved |
| Add CML regularization to MMTM (ablation) | Type III multimodal classifier · NYUD2 | Accuracy: 66.89±0.85 -> 68.09±0.68 | Improved |
| Add CML regularization to MMTM (ablation) | Type III multimodal classifier · SUN-RGBD | Accuracy: 62.11±0.31 -> 62.78±0.32 | Improved |
| Add CML regularization (ablation) | CPM-Nets, modality 1 corrupted · CUB, Gaussian noise ε=0.5 | Accuracy: 71.17±9.14 -> 81.11±4.37 | Improved |
| Add CML regularization (ablation) | CPM-Nets, modalities 1 and 2 corrupted · CUB, Gaussian noise ε=0.5 | Accuracy: 72.50±11.14 -> 81.39±6.43 | Improved |
| Use a strict confidence-difference penalty instead of penalizing only confidence increases after modality removal (ablation) | CML multimodal classifier · Partial-modality confidence estimation | Predictive confidence | Hurt |
| Add CML regularization (ablation) | Type I multimodal classifier · CUB | NLL: 20.49±0.30 -> 20.53±0.46 | No clear effect |
| Add CML regularization (ablation) | Type I multimodal classifier · CUB | Accuracy: 87.00±4.36 -> 88.33±4.05 | No clear effect |
| Add CML regularization (ablation) | Type II multimodal classifier · CUB | Accuracy: 92.33±1.11 -> 94.50±1.71 | No clear effect |
Closest papers by meaning
- Rethinking Multimodal Learning from the Perspective of Mitigating Classification Ability Disproportion (NeurIPS 2025 · similarity 0.87)
- Fusion or Confusion? Multimodal Complexity Is Not All You Need (NeurIPS 2026 · similarity 0.87)
- Balancing Multimodal Training Through Game-Theoretic Regularization (NeurIPS 2025 · similarity 0.86)
- Beyond Modality Fusion: Deep Ensembles for Multimodal Classification (NeurIPS 2026 · similarity 0.86)
- Classifier-guided Gradient Modulation for Enhanced Multimodal Learning (NeurIPS 2024 · similarity 0.86)
- MIDAS: Misalignment-based Data Augmentation Strategy for Imbalanced Multimodal Learning (NeurIPS 2025 · 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.
