Research map / Vision-language and multimodal / Cross-modal learning
Disentangling for Transfer: Boosting Limited Modalities via Information-Theoretic Regularization and Cross-Modal Reconstruction
Zhiyun Zhang, Yan-Jie Zhou, Yujian Hu, Xiyao Ma, Zhouhang Yuan, Zirui Wang, Hongkun Zhang, Minfeng Xu
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 DANTE stage 2 regularization (no pre-training) vs best distillation baseline | segmentation model · BraTS 2018 ET | Dice (%): 51.23 -> 54.93 | Improved |
| add Stage 1 cross-modal pre-training (ablation) | DANTE · BraTS 2018 ET | Dice (%): 54.93 -> 57.09 | Improved |
| add Stage 1 cross-modal pre-training (ablation) | DANTE · DP2E | Dice (%): 30.10 -> 31.06 | Improved |
| replace cross-modal pre-training with naive self-reconstruction MAE (ablation) | DANTE · BraTS 2018 ET | Dice (%): 54.93 -> 55.26 (naive); 57.09 (cross-modal) | Mixed |
| use cross-modal pre-training vs naive MAE (ablation) | DANTE · BraTS 2018 ET | Dice (%): 55.26 -> 57.09 | Improved |
| add privileged compaction loss L_PIB (ablation) | DANTE · BraTS 2020 ET | Dice (%): +2.2% at lambda1=0.3 | Improved |
| add mutual alignment loss (ablation) | DANTE · BraTS 2020 ET | Dice (%): +2.4% at lambda2=0.5 | Improved |
| increase loss weights beyond optimum (ablation) | DANTE · BraTS ET | Dice (%) | Hurt |
| use UNet-based DANTE instead of pre-trained ViT (ablation) | segmentation backbone · BraTS 2020 ET | Dice (%): 45.15 -> 61.00 | Improved |
| DANTE vs CMFD on BraTS 2020 TC/WT with pre-training | DANTE · BraTS 2020 WT | Dice (%): 91.52 -> 92.12 | No clear effect |
Closest papers by meaning
- Context-Aware Generative Imputation for Robust Multimodal Learning in Missing Modality Scenarios (NeurIPS 2026 · similarity 0.88)
- DiA-gnostic VLVAE: Disentangled Alignment-Constrained Vision Language Variational AutoEncoder for Robust Radiology Reporting with Missing Modalities (AAAI 2026 · similarity 0.87)
- An Information Criterion for Controlled Disentanglement of Multimodal Data (ICLR 2025 · similarity 0.86)
- Multi-modal Vision Pre-training for Medical Image Analysis (CVPR 2025 · similarity 0.86)
- Knowledge Transfer Scaling Laws for 3D Medical Imaging (NeurIPS 2026 · similarity 0.86)
- UniMRSeg: Unified Modality-Relax Segmentation via Hierarchical Self-Supervised Compensation (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.
