Research map / Large language models / Language, safety and interpretability
DC-BENCH: Dataset Condensation Benchmark
Justin Cui, Ruochen Wang, Si Si, Cho-Jui Hsieh
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 |
|---|---|---|---|
| enable best data augmentation at evaluation time (ablation) | DC · CIFAR-10 IPC 10 | test accuracy (%): 44.43 -> 50.99 | Improved |
| enable best data augmentation at evaluation time (ablation) | DSA · CIFAR-10 IPC 10 | test accuracy (%): 43.54 -> 52.96 | Improved |
| enable best data augmentation at evaluation time (ablation) | DM · CIFAR-10 IPC 10 | test accuracy (%): 42.45 -> 47.64 | Improved |
| enable best data augmentation at evaluation time (ablation) | TM · CIFAR-10 IPC 10 | test accuracy (%): 53.49 -> 63.66 | Improved |
| evaluate with augmentation instead of none (ablation) | K-Center · CIFAR-100 IPC 50 | test accuracy (%): 33.61 -> 38.64 | Improved |
| K-Center selection vs DC synthesis under augmentation | K-Center · CIFAR-100 IPC 50 | test accuracy (%): 30.56 -> 38.64 (CIFAR-100) | Improved |
| increase IPC to 200 and above | condensation methods vs random selection · CIFAR-10 | test accuracy (%) | No clear effect |
| transfer condensed data from ConvNet to other architectures | TM · CIFAR-10 IPC 10 | test accuracy (%): 63.66 -> 31.55 | Hurt |
| use condensed data from DC for NAS ranking instead of full data | DC · NAS-Bench-201 CIFAR-10 | Spearman correlation: 0.7487 -> -0.19 | Hurt |
| use K-Center instead of random selection for NAS ranking | K-Center · NAS-Bench-201 CIFAR-10 | Spearman correlation: -0.06 -> 0.11 | Improved |
| initialize synthetic images from K-Center instead of random selection (ablation) | DM · CIFAR-10 | compute to same accuracy / final accuracy: about 30% budget; +1.3% final | Improved |
| scale ResNet152 training data from 300 to 1000 random images per class | ResNet152 · CIFAR-10 | test accuracy (%): 45.59 -> 70.29 | Improved |
Closest papers by meaning
- Ameliorate Spurious Correlations in Dataset Condensation (ICML 2024 · similarity 0.90)
- Elucidating the Design Space of Dataset Condensation (NeurIPS 2024 · similarity 0.90)
- Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets? (NeurIPS 2026 · similarity 0.88)
- Multisize Dataset Condensation (ICLR 2024 · similarity 0.87)
- Fetch and Forge: Efficient Dataset Condensation for Object Detection (NeurIPS 2024 · similarity 0.87)
- CondTSF: One-line Plugin of Dataset Condensation for Time Series Forecasting (NeurIPS 2024 · similarity 0.87)
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.
