Research map / Vision-language and multimodal / Cross-modal learning
M-IDoL: Information Decomposition for Modality-Specific and Diverse Representation Learning in Medical Foundation Model
Yihang Liu, Longzhen Yang, Jiaxiong Yang, Ying Wen, Lianghua He, Heng Tao Shen
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 MoE projector alone to unified CL (ablation) | Unified CL baseline · APTOS (ablation) | downstream score (%): 90.36 -> 87.97 | Hurt |
| add MoE projector alone to unified CL (ablation) | Unified CL baseline · RSNA (ablation) | downstream score (%): 87.42 -> 88.31 | Mixed |
| add MoE projector alone to unified CL (ablation) | Unified CL baseline · HAM10000 (ablation) | downstream score (%): 92.77 -> 90.86 | Hurt |
| add routing-consistency loss L_route to MoE (ablation) | MoE projector · RSNA (ablation) | downstream score (%): 88.31 -> 91.45 | Improved |
| add routing-consistency loss L_route to MoE (ablation) | MoE projector · APTOS (ablation) | downstream score (%): 87.97 -> 93.46 | Improved |
| add routing-consistency loss L_route to MoE (ablation) | MoE projector · Mitosis (ablation) | downstream score (%): 88.64 -> 91.75 | Improved |
| add intra-modality contrastive loss L_cst (ablation) | MoE + L_route · RSNA (ablation) | downstream score (%): 91.45 -> 93.23 | Improved |
| add intra-modality contrastive loss L_cst (ablation) | MoE + L_route · Mitosis (ablation) | downstream score (%): 91.75 -> 95.62 | Improved |
| add intra-modality contrastive loss L_cst (ablation) | MoE + L_route · OCTID (ablation) | downstream score (%): 98.84 -> 99.34 | Improved |
| add all M-IDoL components (MoE, L_route, L_cst) to unified CL baseline (ablation) | Unified CL baseline · HAM10000 (ablation) | downstream score (%): 92.77 -> 97.96 | Improved |
| increase number of MoE experts beyond one per modality (ablation) | M-IDoL · downstream tasks | performance | No clear effect |
| use M-IDoL (unified, multimodal) instead of modality-specific Panderm on ISIC2016 | Panderm · ISIC2016 (downstream) | ACC (%): 87.88 -> 86.39 | Hurt |
Closest papers by meaning
- KAMP: Knowledge-Anchored Multimodal Pretraining Framework for Medical Image Representation (CVPR 2026 · similarity 0.91)
- Multimodal Patient Representation Learning with Missing Modalities and Labels (ICLR 2024 · similarity 0.87)
- Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · similarity 0.87)
- Flex-MoE: Modeling Arbitrary Modality Combination via the Flexible Mixture-of-Experts (NeurIPS 2024 · similarity 0.86)
- Modalities Contribute Unequally: Enhancing Medical Multi-modal Learning through Adaptive Modality Token Re-balancing (ICML 2025 · similarity 0.86)
- Continual Self-supervised Learning: Towards Universal Multi-modal Medical Data Representation Learning (CVPR 2024 · 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.
