Research map / Label-efficient and robust learning / Knowledge distillation
Flash-DD: An Ultra Parameter-Efficient Approach to Dataset Distillation
Ruonan Yu, 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 |
|---|---|---|---|
| apply Flash-DD pruned single teacher to RDED | RDED · ImageNet-1K IPC 1 | accuracy (%): 6.6 -> 20.0 | Improved |
| apply Flash-DD pruned single teacher to RDED | RDED · ImageNet-1K IPC 50 | accuracy (%): 56.5 -> 58.5 | Improved |
| apply Flash-DD to SRe2L | SRe2L · ImageNet-1K IPC 10 | accuracy (%): 21.3 -> 44.5 | Improved |
| apply Flash-DD to G_VBSM | G_VBSM · ImageNet-1K IPC 50 | accuracy (%): 51.8 -> 60.3 | Improved |
| apply Flash-DD to SRe2L | SRe2L · Places365-Standard IPC 1 | accuracy (%): 1.3 -> 12.9 | Improved |
| apply Flash-DD to RDED on cross-architecture students | RDED · ImageNet-1K IPC 10, Swin-V2-Tiny | accuracy (%): 17.8 -> 27.5 | Improved |
| apply Flash-DD to RDED on AlexNet student | RDED · Places365-Standard IPC 10, AlexNet | accuracy (%): 10.3 -> 11.3 | Improved |
| replace l1-norm pruning with l2-norm (ablation) | Flash-DD on RDED · ImageNet-1K IPC 10 | accuracy (%): 47.5 -> 47.4 | No clear effect |
| replace l1-norm pruning with Taylor (ablation) | Flash-DD on RDED · ImageNet-1K IPC 50 | accuracy (%): 58.5 -> 59.2 | No clear effect |
| use 2 ensembled pruned teachers instead of best single teacher | Flash-DD · ImageNet-1K IPC 10 | accuracy (%): 47.5 -> 48.6 | Improved |
| use 3 ensembled pruned teachers instead of best single teacher | Flash-DD · ImageNet-1K IPC 10 | accuracy (%): 47.5 -> 48.8 | Improved |
| use ensemble of pruned teachers instead of single model of same 21.65MB | Flash-DD · ImageNet-1K IPC 10 | accuracy (%): 45.0 -> 48.8 | Improved |
Closest papers by meaning
- Rethinking Dataset Distillation for Classification: Do Distilled Sets Outperform Coresets? (NeurIPS 2026 · similarity 0.91)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.90)
- DD-Ranking: Rethinking the Evaluation of Dataset Distillation (NeurIPS 2026 · similarity 0.90)
- A Label is Worth A Thousand Images in Dataset Distillation (NeurIPS 2024 · similarity 0.90)
- GIFT: Unlocking Full Potential of Labels in Distilled Dataset at Near-zero Cost (ICLR 2025 · similarity 0.90)
- Large Scale Dataset Distillation with Domain Shift (ICML 2024 · similarity 0.89)
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.
