Research map / Vision-language and multimodal / Cross-modal learning
Robust Multimodal Learning via Representation Decoupling
Shicai Wei, Yang Luo, Yuji Wang, Chunbo Luo
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 decoupled multimodal representation (sample embeddings from learned distribution) (ablation) | SF-MD baseline · CASIA-SURF | average ACER: 7.52 -> 4.63 | Improved |
| add hard combination regularizer on top of DMR (ablation) | DMRNet · CASIA-SURF | average ACER: 4.63 -> 3.58 | Improved |
| estimate distribution on feature map via convolution instead of feature-vector PE or PCME (ablation) | DMR · CASIA-SURF | average ACER: 5.85 -> 4.66 | Improved |
| set distributional regularization weight alpha to 0 (ablation) | DMR · CASIA-SURF | average ACER: 4.66 -> 8.18 | Hurt |
| set alpha too large (1e-1) (ablation) | DMR · CASIA-SURF | average ACER: 4.66 -> 49.76 | Hurt |
| set hard combination weight beta to 0.7 instead of 0.1 or 0.9 (ablation) | DMRNet · CASIA-SURF | average ACER: 4.31 -> 3.58 | Improved |
| use DMRNet instead of ESANet-MD for multimodal segmentation | ESANet · NYUv2 | average mIoU: 42.77 -> 45.08 | Improved |
| use DMRNet instead of ShaSpec on audio-only CREMA-D | ShaSpec · CREMA-D (audio only) | accuracy: 59.86 -> 58.87 | Hurt |
| use DMRNet instead of MMANet on CASIA-SURF | MMANet · CASIA-SURF | average ACER: 3.94 -> 3.58 | Improved |
| use DMRNet instead of ShaSpec and MMANet on Kinetics-Sounds | ShaSpec · Kinetics-Sounds | average accuracy: 51.62 -> 56.01 | Improved |
Closest papers by meaning
- Robust Multimodal Learning With Missing Modalities via Parameter-Efficient Adaptation (TPAMI 2024 · similarity 0.89)
- Missing Modality Prediction for Unpaired Multimodal Learning via Joint Embedding of Unimodal Models (ECCV 2024 · similarity 0.89)
- Amplifying Prominent Representations in Multimodal Learning via Variational Dirichlet Process (NeurIPS 2025 · similarity 0.87)
- Modality-Aware SAM: Sharpness-Aware-Minimization Driven Gradient Modulation for Harmonized Multimodal Learning (NeurIPS 2025 · similarity 0.87)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.87)
- Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness (NeurIPS 2026 · 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.
