Research map / Architectures and efficiency / RNNs and normalization
Dataset Distillation as Data Compression: A Rate-Utility Perspective
Youneng Bao, Yiping Liu, Zhuo Chen, Yongsheng Liang, Mu Li, Kede Ma
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 joint rate-utility optimization to trajectory matching | TM · ImageNet Nette (128x128, bpc<=192kB) | accuracy (%): 51.4 -> 76.5 | Improved |
| add joint rate-utility optimization to gradient matching | GM · ImageNet subsets avg, GM loss (bpc<=192kB) | accuracy (%): 28.1 -> 49.5 | Improved |
| add joint rate-utility optimization to distribution matching | DM · ImageNet subsets avg, DM loss (bpc<=192kB) | accuracy (%): 25.7 -> 59.1 | Improved |
| TM-RUO versus DDiF averaged over architectures | TM-RUO · ImageNet subsets avg, cross-architecture (bpc<=192kB) | accuracy (%): 45.2 -> 50.7 | Improved |
| TM-RUO versus DDiF on AlexNet (Nette to Squawk average) | TM-RUO · AlexNet cross-architecture | accuracy (%): 49.3 -> 46.7 | Hurt |
| TM-RUO versus DDiF on ViT variant | TM-RUO · ViT cross-architecture | accuracy (%): 43.5 vs 43.4 | No clear effect |
| TM-RUO versus DDiF on Woof (TM loss) | TM-RUO · ImageNet Woof (bpc<=192kB) | accuracy (%): 42.9 -> 46.0 | Improved |
| GM-RUO versus DDiF on Woof | GM-RUO · ImageNet Woof, GM loss | accuracy (%): 35.2 -> 33.1 | Hurt |
| increase spc from 1 to 102 in TM-RUO (ablation) | TM-RUO · ImageNet Nette | accuracy (%): 49.2 -> 76.5 | Improved |
| TM-RUO at 94 kB versus best baseline on CIFAR-10 | TM-RUO · CIFAR-10 | accuracy (%): 74.0 -> 77.5 | Improved |
| TM-RUO at 53 kB versus baseline at 120 kB on CIFAR-100 | TM-RUO · CIFAR-100 | accuracy (%): 46.1 -> 49.2 | Improved |
| raise storage budget from low to high bpc shifts bit allocation toward explicit latents (ablation) | TM-RUO · Nette bit allocation | share of implicit bits: 74% -> 7% | Mixed |
Closest papers by meaning
- DD-Ranking: Rethinking the Evaluation of Dataset Distillation (NeurIPS 2026 · similarity 0.89)
- Embarrassingly Simple Dataset Distillation (ICLR 2024 · similarity 0.88)
- Large Scale Dataset Distillation with Domain Shift (ICML 2024 · similarity 0.88)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.88)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.88)
- A Label is Worth A Thousand Images in Dataset Distillation (NeurIPS 2024 · 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.
