Research map / Label-efficient and robust learning / Knowledge distillation
Grounding and Enhancing Informativeness and Utility in Dataset Distillation
Shaobo Wang, Yantai Yang, Guo Chen, Peiru Li, Kaixin Li, Yufa Zhou, Zhaorun Chen, Linfeng Zhang
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 GradNorm scoring instead of loss scoring (ablation) | InfoUtil / RDED · ImageNette (IPC=50) | top-1 accuracy (%): 80.4 -> 85.0 | Improved |
| add GradNorm scoring instead of loss scoring (ablation) | InfoUtil / RDED · ImageNet-1K (IPC=10) | top-1 accuracy (%): 42.0 -> 43.5 | Improved |
| add Shapley attribution cropping on top of GradNorm (ablation) | InfoUtil · ImageNette (IPC=50) | top-1 accuracy (%): 85.0 -> 86.2 | Improved |
| add Shapley attribution cropping on top of GradNorm (ablation) | InfoUtil · ImageWoof (IPC=50) | top-1 accuracy (%): 68.8 -> 69.6 | Improved |
| add noise injection in attribution heatmap (ablation) | InfoUtil · ImageNette (IPC=50) | top-1 accuracy (%): 70.6 -> 86.2 | Improved |
| add noise injection in attribution heatmap (ablation) | InfoUtil · ImageNet-1K (IPC=10) | top-1 accuracy (%): 38.5 -> 44.2 | Improved |
| use Shapley value instead of Grad-CAM attribution (ablation) | InfoUtil · ImageNet-1K (IPC=10) | top-1 accuracy (%): 30.394 -> 43.880 | Improved |
| use early-epoch teacher for soft labels at IPC=1 (ablation) | InfoUtil · ImageWoof (ConvNet, IPC=1) | top-1 accuracy (%) | Improved |
| use fully converged teacher for soft labels at larger IPC (ablation) | InfoUtil · ImageWoof (ConvNet, IPC=10/50) | top-1 accuracy (%) | Improved |
| replace RDED with InfoUtil under same converged teacher | InfoUtil · ImageWoof (ResNet-101, IPC=50) | top-1 accuracy (%): 59.1 -> 67.0 | Improved |
| replace RDED with InfoUtil in cross-architecture eval (ResNet-18 teacher, MobileNet-V2 student) | InfoUtil · ImageNet-1K (IPC=10) | top-1 accuracy (%): 40.4 -> 37.1 | Hurt |
| replace RDED with InfoUtil in cross-architecture eval (VGG-11 teacher, ResNet-18 student) | InfoUtil · ImageNet-1K (IPC=10) | top-1 accuracy (%): 22.7 -> 35.1 | Improved |
Closest papers by meaning
- Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets? (NeurIPS 2026 · similarity 0.91)
- DD-Ranking: Rethinking the Evaluation of Dataset Distillation (NeurIPS 2026 · similarity 0.90)
- Embarrassingly Simple Dataset Distillation (ICLR 2024 · similarity 0.89)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.89)
- A Label is Worth A Thousand Images in Dataset Distillation (NeurIPS 2024 · similarity 0.88)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.88)
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.
