Research map / Vision-language and multimodal / Cross-modal learning
Medical multimodal fusion
Combines clinical and imaging modalities with decomposition, symmetric consistency and graph structure.
166 accepted papers at ICML, NeurIPS, ICLR, CVPR and AAAI (2018–2026) · an approach in Cross-modal learning, keyword cluster “fusion · missing · multi modal”
State History Tried Checklist Preprints Who Export
Where it stands
Not enough benchmark data Fewer than four papers report the same benchmark, so there is no record to follow.
29 papers in 2023–24 and 121 in 2025–26 (at ICML, NeurIPS, ICLR, CVPR and AAAI); its share of the topic grew, from 10.0% to 15.4%.
How this is decided
A benchmark is tracked when at least four papers of this approach report their own method on the same dataset and metric in at least two different years. Its record is the best number reported up to 2024 compared with the best reported in 2025–26: still improving means at least 1% better, flat means at most 0.2% better with at least two papers from 2025–26 trying. Numbers on a different scale from the rest (a fraction among percentages) and far outliers are left out. Results come from different papers with different setups, so they are signals, not head-to-head comparisons. The share of papers is counted at ICML, NeurIPS, ICLR, CVPR and AAAI.
How it got here
Most cited paper of each year
- 2018 · Perception Coordination Network: A Framework for Online Multi-Modal Concept Acquisition and Binding (AAAI 2018 · 5 citations)
- 2019 · Joint Representation Learning for Multi-Modal Transportation Recommendation (AAAI 2019 · 75 citations)
- 2020 · Cross-Modal Subspace Clustering via Deep Canonical Correlation Analysis (AAAI 2020 · 62 citations)
- 2021 · Vectorization and Rasterization: Self-Supervised Learning for Sketch and Handwriting (CVPR 2021 · 63 citations)
- 2022 · Multimodal Dynamics: Dynamical Fusion for Trustworthy Multimodal Classification (CVPR 2022 · 187 citations)
- 2023 · M3AE: Multimodal Representation Learning for Brain Tumor Segmentation with Missing Modalities (AAAI 2023 · 93 citations)
- 2024 · DrFuse: Learning Disentangled Representation for Clinical Multi-Modal Fusion with Missing Modality and Modal Inconsistency (AAAI 2024 · 56 citations)
- 2025 · BSAFusion: A Bidirectional Stepwise Feature Alignment Network for Unaligned Medical Image Fusion (AAAI 2025 · 44 citations)
- 2026 · HFR-MKGC: Hierarchical Fusion Reasoning with MLLMs for Multi-modal Knowledge Graph Completion (AAAI 2026 · 5 citations)
Earliest papers here
- Perception Coordination Network: A Framework for Online Multi-Modal Concept Acquisition and Binding (AAAI 2018 · 5 citations)
- Joint Representation Learning for Multi-Modal Transportation Recommendation (AAAI 2019 · 75 citations)
- DM2C: Deep Mixed-Modal Clustering (NeurIPS 2019 · 23 citations)
What has been tried
75 papers here report 879 measured changes (a component added, removed or swapped, compared with a baseline): helped 77%, no clear effect 4%, hurt 16%, mixed 2%. Papers mostly publish changes that helped, so "helped" is over-represented.
Ablations in recent papers
| What was changed | Measured on | Result | Effect | Paper |
|---|---|---|---|---|
| add JSD alignment and orthogonality losses | MIMIC-IV/MIMIC-CXR matched subset · PRAUC | 0.446 (0.411, 0.501) -> 0.450 (0.426, 0.498) | Helped | DrFuse: Learning Disentangled Representation for Clinical Multi-Modal Fusion with Missing Modality and Modal Inconsistency (AAAI 2024 · 56 citations) |
| Include MDF-FR fully rather than Setting A (no MFRH swapping or L_ce2) | MRI-PET · Q_CV | 99.2 -> 67.7 | Helped | BSAFusion: A Bidirectional Stepwise Feature Alignment Network for Unaligned Medical Image Fusion (AAAI 2025 · 44 citations) |
| add L1 regularisation and a self-normalising network layer | KIRP · validation loss | Helped | HEALNet: Multimodal Fusion for Heterogeneous Biomedical Data (NeurIPS 2024 · 39 citations) | |
| use Coupled Mamba fusion instead of cross-attention | CMU-MOSEI · F1-Score | 84.5 -> 85.5 | Helped | Coupled Mamba: Enhanced Multimodal Fusion with Coupled State Space Model (NeurIPS 2024 · 37 citations) |
| Add DNS to randomly initialized backbone features | Liver multiphase CT, Pre-contrast ← Venous & Arterial · DSC | +11.12 DSC | Helped | Modality-Agnostic Structural Image Representation Learning for Deformable Multi-Modality Medical Image Registration (CVPR 2024 · 35 citations) |
| add the Modality-Adaptive Module | VQAv2 · score | 76.2 -> 76.8 | Helped | mPLUG-Owl2: Revolutionizing Multi-modal Large Language Model with Modality Collaboration (CVPR 2024 · 29 citations) |
| Add modality tokenization (MT) | DB15K · MRR | 35.48 -> 37.72 | Helped | Tokenization, Fusion, and Augmentation: Towards Fine-grained Multi-modal Entity Representation (AAAI 2025 · 26 citations) |
| Replace UMB token merging with pooling | BraTS 2020 · WT DSC (%) | 82.92 -> 87.27 | Hurt | TMFormer: Token Merging Transformer for Brain Tumor Segmentation with Missing Modalities (AAAI 2024 · 26 citations) |
| Add EMD to DWC | FashionIQ Dress · R@10 | 23.55 -> 32.67 | Helped | Dynamic Weighted Combiner for Mixed-Modal Image Retrieval (AAAI 2024 · 20 citations) |
| add cross-modal reconstruction | BraTS2023-PED · Dice Score (%) | 72.52% to 75.16% | Helped | Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · 17 citations) |
| Add CRM to the shared Mamba baseline | PCLT20K-TE · IoU | 62.12 -> 62.70 | Helped | Cross-Modal Interactive Perception Network with Mamba for Lung Tumor Segmentation in PET-CT Images (CVPR 2025 · 12 citations) |
| Use COCO caption calibration data | MMMU · accuracy | 36.6 -> 38.7 | Helped | MBQ: Modality-Balanced Quantization for Large Vision-Language Models (CVPR 2025 · 11 citations) |
What reviewers will expect
Datasets and baselines named in the 51 papers of this approach from 2025–2026 that have a full-text summary. Leaving out the common ones is a frequent review question.
Datasets
- DB15K · 4 of 51 papers
- BraTS2018 · 4 of 51 papers
- MIMIC-IV · 4 of 51 papers
- MKG-W · 3 of 51 papers
- MKG-Y · 3 of 51 papers
- BraTS2020 · 3 of 51 papers
- Event · 2 of 51 papers
- MIMIC-III · 2 of 51 papers
Baselines
- IMF · 3 of 51 papers
- TransE · 3 of 51 papers
- DistMult · 3 of 51 papers
- ComplEx · 3 of 51 papers
- RotatE · 3 of 51 papers
- IKRL · 3 of 51 papers
- TBKGC · 3 of 51 papers
- TransAE · 3 of 51 papers
Recent preprints
Loading recent arXiv preprints…
Who publishes on it
Institutions on the most papers of this approach since 2023 (of 150 papers; from OpenAlex and the papers' affiliations).
- Zhejiang University · 5 papers
- University of Science and Technology of China · 4 papers
- The Hong Kong Polytechnic University · 3 papers
- University of Pennsylvania · 3 papers
- Meta · 3 papers
- Carnegie Mellon University · 3 papers
- University of Cambridge · 3 papers
- Zhengzhou University · 3 papers
- EPFL · 3 papers
- Hong Kong Baptist University · 2 papers
Use it
Download BibTeX of 32 key papersOpen the topic in the interactive map
Get alerts. Describe your idea in scime atlas and save it (free scime account): you get a weekly email when new preprints land close to it.
Related approaches
Same keyword cluster
- Multimodal fusion with missing modalities (208 papers)
- Multi-modality image fusion (194 papers)
- Multimodal sentiment analysis (106 papers)
- Multi-modal object re-identification (40 papers)
Elsewhere in the topic
- Unimodal-to-multimodal representation transfer (145 papers)
- Probabilistic embeddings for cross-modal retrieval (108 papers)
- Contrastive vision-language alignment (104 papers)
- Audio-visual speech separation (76 papers)
- Audio-visual source localization (67 papers)
- Cross-modal hashing (45 papers)
- Self-supervised audio-visual representation learning (44 papers)
Benchmark numbers, ablations and baseline numbers are extracted by AI from each paper's full text and can be wrong; the paper is the authority. Data: results index of 2026-11-02, citation counts from OpenAlex (2026-10-09).