Research map / Vision-language and multimodal / Cross-modal learning
ICYM2I: The illusion of multimodal informativeness under missingness
Young Sang Choi, Vincent Jeanselme, Pierre Elias, Shalmali Joshi
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 |
|---|---|---|---|
| apply IPW correction to training and evaluation | naive complete-case analysis · Bitwise logic AND (MAR) | AUROC of X1 model: 0.66 -> 0.83 | Improved |
| apply IPW correction | naive complete-case analysis · Bitwise logic OR (MAR) | AUROC of X1 model: 0.95 -> 0.85 | Improved |
| apply IPW correction to PID | naive complete-case analysis · Bitwise logic AND (MAR) | Unique information 1: 0.44 -> 0.03 | Improved |
| apply IPW correction to PID | naive complete-case analysis · Bitwise logic XOR (MAR) | Complementary information: 0.62 -> 0.96 | Improved |
| apply IPW correction to PID | naive complete-case analysis · Structural heart disease detection | Unique information of CXR: about 5% -> 1.8% | Improved |
| apply IPW correction to PID | naive complete-case analysis · Structural heart disease detection | Shared information: 0.10 -> 0.48 | Improved |
| apply IPW correction | naive complete-case analysis · UR-FUNNY (70% missingness) | AUROC text: 0.61 -> 0.66 | Improved |
| apply IPW correction | naive complete-case analysis · Hateful Memes (70% missingness) | AUROC text: 0.68 -> 0.67 | No clear effect |
| correct both training and evaluation instead of one (ablation) | IPW correction · simulation | RMSE to oracle AUC | Improved |
| apply IPW under MCAR | IPW correction · Bitwise logic (MCAR) | AUROC / PID | No clear effect |
| apply IPW under MNAR | IPW correction · Bitwise logic AND (MNAR) | Unique information 1: 0.45 -> 0.45 | No clear effect |
| add CXR to ECG | ECG model · Structural heart disease detection | AUROC: 0.82 -> 0.83 | No clear effect |
Closest papers by meaning
- Do More Modalities Always Help? A Geometric Perspective on Missing-Modality Robustness (NeurIPS 2026 · similarity 0.88)
- Balancing Multimodal Training Through Game-Theoretic Regularization (NeurIPS 2025 · similarity 0.87)
- Asymmetric Reinforcing Against Multi-Modal Representation Bias (AAAI 2025 · similarity 0.87)
- Fusion or Confusion? Multimodal Complexity Is Not All You Need (NeurIPS 2026 · similarity 0.87)
- RedCore: Relative Advantage Aware Cross-Modal Representation Learning for Missing Modalities with Imbalanced Missing Rates (AAAI 2024 · similarity 0.86)
- SynIB: Informational Bottleneck for Maximizing Synergy in Multimodal Learning (NeurIPS 2026 · 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.
