Research map / Architectures and efficiency / Attention and transformers
Dual Decomposition of Convex Optimization Layers for Consistent Attention in Medical Images
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 dual-decomposed consistent attention | ResNet-50 explanations · FDG-PET-MUAB, Block 3 | AUC: 21.52 -> 32.48$\pm$1.81 | Helped |
| add dual-decomposed consistent attention | ResNet-50 explanations · FDG-PET-MUAB, Block 4 | AUC: 21.90 -> 27.59$\pm$1.23 | Helped |
| add dual-decomposed consistent attention | CNN with U-Net encoder explanations · BraTS18, Block 3 | mAP: 58.68 -> 71.82$\pm$4.2 | Helped |
| add dual-decomposed consistent attention | CNN with U-Net encoder explanations · BraTS18, Block 4 | mAP: 55.66 -> 76.16$\pm$3.8 | Helped |
| add the consistency term to two attention units (ablation) | ResNet-50 explanations · FDG-PET-MUAB, Block 3 | AUC: 21.23$\pm$0.72 -> 32.48$\pm$1.81 | Helped |
| add the consistency term to two attention units (ablation) | ResNet-50 explanations · FDG-PET-MUAB, Block 4 | AUC: 22.11$\pm$0.37 -> 27.59$\pm$1.23 | Helped |
| add the consistency term to two attention units (ablation) | CNN with U-Net encoder explanations · BraTS18, Block 3 | mAP: 19.01$\pm$12.3 -> 71.82$\pm$4.2 | Helped |
| use only the input-nearer attention unit instead of dual consistent attention (ablation) | CNN with U-Net encoder explanations · BraTS18, Block 3 | mAP: 33.03$\pm$11.4 -> 71.82$\pm$4.2 | Helped |
| use only the output-nearer attention unit instead of dual consistent attention (ablation) | CNN with U-Net encoder explanations · BraTS18, Block 4 | mAP: 34.17$\pm$9.0 -> 76.16$\pm$3.8 | Helped |
| replace dual reparameterization with a direct KL consistency term (ablation) | CNN with U-Net encoder explanations · BraTS18, Block 4 | mAP: 45.78$\pm$7.21 -> 76.16$\pm$3.8 | Helped |
| replace dual reparameterization with the reverse direct KL consistency term (ablation) | CNN with U-Net encoder explanations · BraTS18, Block 4 | mAP: 45.56$\pm$7.14 -> 76.16$\pm$3.8 | Helped |
| add dual-decomposed consistent attention | CNN with U-Net encoder explanations · BraTS18, Block 3 | IoU@0.90: 34.20 -> 32.75$\pm$2.3 | Hurt |
Closest papers by meaning
- MedLIME: A Distribution-Aligned and Evidence-Supported Framework for Medical Saliency Explanations (CVPR 2026 · similarity 0.87)
- Unraveling Attention via Convex Duality: Analysis and Interpretations of Vision Transformers (ICML 2022 · similarity 0.85)
- Uncertainty-Aware Attention for Reliable Interpretation and Prediction (NeurIPS 2018 · similarity 0.85)
- Interpretable and Fine-Grained Visual Explanations for Convolutional Neural Networks (CVPR 2019 · similarity 0.85)
- Advancing Multiple Instance Learning with Continual Learning for Whole Slide Imaging (CVPR 2025 · similarity 0.84)
- EMCAD: Efficient Multi-scale Convolutional Attention Decoding for Medical Image Segmentation (CVPR 2024 · similarity 0.83)
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.