Research map / Vision-language and multimodal / Cross-modal learning
Inference-Time Dynamic Modality Selection for Incomplete Multimodal Classification
Siyi Du, Xinzhe Luo, Declan O'regan, Chen Qin
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 inference-time dynamic modality selection (DyMo) over recovered modalities (ablation) | MoPoE-recovered multimodal network · PolyMNIST (missing rate 0.8) | accuracy: 84.21 -> 96.61 | Improved |
| add dynamic modality selection (ablation) | DyMo · MST (missing {M,T}) | accuracy: 80.73 -> 85.31 | Improved |
| add dynamic modality selection (ablation) | DyMo · DVM (full table missing) | accuracy: 88.36 -> 93.14 | Improved |
| use iterative rather than simultaneous selection (ablation) | DyMo · PolyMNIST (missing rate 0.8) | accuracy: 94.33 -> 94.50 | Improved |
| add intra-class similarity calibration of reward (ablation) | DyMo · PolyMNIST (missing rate 0.8) | accuracy: 94.50 -> 96.61 | Improved |
| add intra-class similarity calibration of reward (ablation) | DyMo · CAD and Infarction (missing table) | AUC: 72.20 -> 71.02 (CAD, full table missing) | Hurt |
| add incomplete-modality simulation training (ablation) | MultiAE, MoPoE, M3Care, MAP · PolyMNIST (missing rate 0.8) | accuracy: MultiAE 84.39 -> 89.86; MoPoE 79.84 -> 87.06 | Improved |
| replace MoPoE with CMVAE or MMVAE+ recovery (ablation) | DyMo · PolyMNIST (missing rate 0.8) | accuracy: 96.61 -> 97.69 (CMVAE) | No clear effect |
| use squared Euclidean instead of cosine distance (ablation) | DyMo · PolyMNIST, MST, DVM | accuracy: 96.61 vs 96.81 (PolyMNIST 0.8) | No clear effect |
| DyMo vs M3Care with simulation training | DyMo · DVM full table missing | accuracy: 93.43 vs 93.14 | No clear effect |
Closest papers by meaning
- Facilitating Multimodal Classification via Dynamically Learning Modality Gap (NeurIPS 2024 · similarity 0.88)
- Beyond Modality Fusion: Deep Ensembles for Multimodal Classification (NeurIPS 2026 · similarity 0.87)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.87)
- Measure Less, Know More: Self-Supervised Test-Time Feature Acquisition (NeurIPS 2026 · similarity 0.87)
- Fusion or Confusion? Multimodal Complexity Is Not All You Need (NeurIPS 2026 · similarity 0.86)
- Continual Multimodal Contrastive Learning (NeurIPS 2025 · similarity 0.86)
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.
