Research map / Image generation / Fast diffusion sampling
Dataset Distillation as Pushforward Optimal Quantization
Hong Ye Tan, Emma Slade
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 k-means cluster-count weights to the training loss on distilled data (ablation) | D4M-style clustering distillation · ImageNet-1K classes (distilled latents) | Wasserstein-2 distance: -15.7% | Improved |
| add k-means cluster weights to training loss | D4M · ImageNet-1K IPC 10 (ResNet-18) | top-1 accuracy (%): 27.9 -> 33.1 | Improved |
| add cluster weights | D4M · ImageNet-1K IPC 50 (ResNet-18) | top-1 accuracy (%): 55.2 -> 56.2 | Improved |
| add cluster weights | D4M · ImageNet-1K IPC 200 (ResNet-101) | top-1 accuracy (%): 68.1 -> 68.6 | Improved |
| add cluster weights, student MobileNet-V2 with ResNet-18 teacher | D4M · ImageNet-1K IPC 50 cross-arch | top-1 accuracy (%): 47.9 -> 52.1 | Improved |
| add cluster weights, Swin-T student | D4M · ImageNet-1K IPC 50 cross-arch | top-1 accuracy (%): 58.1 -> 57.4 | Hurt |
| use DDOQ clustering instead of RDED patch images at IPC 10 | RDED · ImageNet-1K IPC 10 (ResNet-18) | top-1 accuracy (%): 42.0 -> 33.1 | Hurt |
| swap UNet LDM for DiT backbone in DDOQ | DDOQ · ImageNet-1K IPC 10 (ResNet-18) | top-1 accuracy (%): 33.1 -> 53.0 | Improved |
| use latent clustering init instead of IGD guidance on DiT | Minimax-IGD · ImageNet-1K IPC 10 (DiT backbone) | top-1 accuracy (%): 46.2 -> 53.0 | Improved |
| use latent clustering init instead of IGD guidance on DiT | Minimax-IGD · ImageWoof IPC 10 | accuracy (%): 43.3 -> 48.8 | Improved |
| use latent clustering init instead of IGD guidance on DiT | Minimax-IGD · ImageNette IPC 50 | accuracy (%): 82.3 -> 79.8 | Hurt |
Closest papers by meaning
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.93)
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.91)
- Taming Diffusion for Dataset Distillation with High Representativeness (ICML 2025 · similarity 0.90)
- Efficient Dataset Distillation via Minimax Diffusion (CVPR 2024 · similarity 0.90)
- Q-DiT: Accurate Post-Training Quantization for Diffusion Transformers (CVPR 2025 · similarity 0.89)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · 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.
