Research map / Label-efficient and robust learning / Zero-/few-shot and meta-learning
MGDD: A Meta Generator for Fast Dataset Distillation
Songhua Liu, Xinchao Wang
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 |
|---|---|---|---|
| Use meta-learned generator instead of training the generator from scratch (ablation) | MGDD · CUB200 IPC 1, 1000 steps | test accuracy (%): 6.73 -> 7.40 | Improved |
| Use meta-learned generator instead of training from scratch (ablation) | MGDD · PACS joint IPC 1, 1000 steps | test accuracy (%): 42.73 -> 43.46 | Improved |
| Adapt the meta generator with the FRePo objective instead of running FRePo for the same steps | MGDD · CIFAR10 IPC 1, 2000 steps | test accuracy (%): 26.8 -> 42.6 | Improved |
| Adapt the meta generator with the DC objective (ablation) | MGDD · CIFAR10 IPC 1, 2000 steps | test accuracy (%): 24.7 -> 34.3 | Improved |
| Adapt the meta generator with the DM objective (ablation) | MGDD · CIFAR10 IPC 1, 2000 steps | test accuracy (%): 25.7 -> 31.4 | Improved |
| Evaluate a single-init adapted generator with a different synthetic initialization (ablation) | MGDD · CIFAR10 IPC 10 | test accuracy (%): 60.76 -> 43.27 | Hurt |
| Use multi-initialization during adaptation (ablation) | MGDD · CIFAR10 IPC 10, different initialization | test accuracy (%): 43.27 -> 60.24 | Improved |
| Add 2x down-sampling parameterization to synthetic data (ablation) | MGDD · CIFAR10 IPC 1 | test accuracy (%): 46.26 -> 48.17 | Improved |
| Add 2x down-sampling parameterization to synthetic data (ablation) | MGDD · CIFAR100 IPC 1 | test accuracy (%): 24.66 -> 27.76 | Improved |
| Train to full convergence instead of 1000 steps, compared with FRePo baseline | MGDD · PACS photo domain IPC 10 | test accuracy (%): 85.40 -> 82.27 | Hurt |
| Remove size-embedding channels from generator (ablation) | MGDD · CIFAR10 unseen IPCs 20-40 | test accuracy (%) | Hurt |
| Replace analytical labels with one-hot labels (ablation) | MGDD · CIFAR10 IPC 10 | test accuracy (%) | Hurt |
Closest papers by meaning
- Large Scale Dataset Distillation with Domain Shift (ICML 2024 · similarity 0.90)
- Embarrassingly Simple Dataset Distillation (ICLR 2024 · similarity 0.89)
- Data Distillation Can Be Like Vodka: Distilling More Times For Better Quality (ICLR 2024 · similarity 0.89)
- Towards Lossless Dataset Distillation via Difficulty-Aligned Trajectory Matching (ICLR 2024 · similarity 0.88)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · similarity 0.88)
- Distilling Long-tailed Datasets (CVPR 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.
