Research map / Label-efficient and robust learning / Recognition and long-tail
Improved Distribution Matching for Dataset Condensation
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 |
|---|---|---|---|
| Replace DM with IDM's partitioning, enriched model sampling, and class-aware regularization | DM dataset condensation · CIFAR-10, 1 Img/Cls | accuracy: 26.0±0.8 -> 45.6±0.7 | Helped |
| Replace DM with IDM's full method | DM dataset condensation · CIFAR-100, 10 Img/Cls | accuracy: 29.7±0.3 -> 45.1±0.1 | Helped |
| Add model sampling (ablation) | DM · CIFAR-100, 10 Img/Cls | accuracy: 29.7±0.3 -> 32.0±0.5 | Helped |
| Add model sampling (ablation) | DM · ImageNet Subset, 10 Img/Cls | accuracy: 11.9±0.3 -> 10.8±0.7 | Hurt |
| Add class-aware distribution regularization after model sampling (ablation) | DM + Model Sampling · CIFAR-100, 10 Img/Cls | accuracy: 32.0±0.5 -> 34.3±0.3 | Helped |
| Add class-aware distribution regularization after model sampling (ablation) | DM + Model Sampling · ImageNet Subset, 10 Img/Cls | accuracy: 10.8±0.7 -> 15.6±0.4 | Helped |
| Add partitioning and expansion augmentation (ablation) | DM + Model Sampling + Distribution Regularization · CIFAR-100, 10 Img/Cls | accuracy: 34.3±0.3 -> 45.1±0.1 | Helped |
| Add partitioning and expansion augmentation (ablation) | DM + Model Sampling + Distribution Regularization · ImageNet Subset, 10 Img/Cls | accuracy: 15.6±0.4 -> 17.1±0.6 | Helped |
| Use 2×2 instead of 1×1 partitioning (ablation) | IDM augmentation · ImageNet Subset, 1 Img/Cls | accuracy: 5.4±0.1 -> 11.2±0.5 | Helped |
| Increase partition size from 2×2 to 3×3 (ablation) | IDM augmentation · ImageNet Subset, 1 Img/Cls | accuracy: 11.2±0.5 -> 10.7±0.7 | Hurt |
| Use IDM instead of DM for condensed data construction | CIFAR-100 continual-learning memory · CIFAR-100, 5-step continual learning | final performance: 33.8% -> 39.3% | Helped |
| Use IDM instead of DM for condensed data construction | CIFAR-100 continual-learning memory · CIFAR-100, 10-step continual learning | final performance: 33.7% -> 39.3% | Helped |
Closest papers by meaning
- M3D: Dataset Condensation by Minimizing Maximum Mean Discrepancy (AAAI 2024 · similarity 0.89)
- Elucidating the Design Space of Dataset Condensation (NeurIPS 2024 · similarity 0.89)
- Exploiting Inter-sample and Inter-feature Relations in Dataset Distillation (CVPR 2024 · similarity 0.87)
- Ameliorate Spurious Correlations in Dataset Condensation (ICML 2024 · similarity 0.86)
- Expressive Score-Based Priors for Distribution Matching with Geometry-Preserving Regularization (ICML 2025 · similarity 0.85)
- Multisize Dataset Condensation (ICLR 2024 · similarity 0.85)
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.