Research map / Image generation / Fast diffusion sampling
Learnability-Guided Diffusion for Dataset Distillation
Jeffrey A. Chan-Santiago, Mubarak Shah
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 learnability guidance to DiT (ablation) | DiT · ImageNette ResNetAP-10 IPC50 (component ablation) | top-1 accuracy (%): 75.6 -> 82.5 | Improved |
| add deviation guidance for diversity (ablation) | LGD · ImageNette ResNetAP-10 IPC50 (component ablation) | top-1 accuracy (%): 82.5 -> 84.4 | Improved |
| add learnability ranking for candidate selection (ablation) | LGD · ImageNette ResNetAP-10 IPC50 (component ablation) | top-1 accuracy (%): 84.4 -> 85.6 | Improved |
| increase candidates per position from kappa=1 to kappa=10 (ablation) | LGD · ImageNette ResNetAP-10 IPC50 (candidate count) | top-1 accuracy (%): 84.4 -> 85.6 | Improved |
| set reference weight omega to 0.0 (no reference model regularization) (ablation) | LGD · ImageNette (IPC 20) | top-1 accuracy (%) | Hurt |
| increase reference weight omega to 2.0 (ablation) | LGD · ImageNette IPC50 | top-1 accuracy (%) | Hurt |
| remove deviation guidance (gamma=0 vs gamma=50) (ablation) | LGD · ImageNette IPC50 | top-1 accuracy (%): 82.5 -> 84.7 | Improved |
| use guidance strength lambda=15 instead of 5 (ablation) | LGD · ImageNette IPC50 | top-1 accuracy (%): 83.7 -> 84.4 | Improved |
| use guidance strength lambda=15 instead of 25 (ablation) | LGD · ImageNette IPC50 | top-1 accuracy (%): 83.1 -> 84.4 | Improved |
| seed with IGD instead of random real images (ablation) | LGD · ImageNette ResNetAP-10 IPC50 | top-1 accuracy (%): 83.8 -> 84.4 | Improved |
| add learnability guidance on Minimax backbone (vs Minimax-IGD) | Minimax · ImageWoof ResNet-18 IPC50 | top-1 accuracy (%): 63.4 -> 66.2 | Improved |
| build dataset incrementally with LGD instead of DiT | DiT · ImageNette IPC50 (incremental) | top-1 accuracy (%): 75.6 -> 85.2 | Improved |
Closest papers by meaning
- Unlocking Dataset Distillation with Diffusion Models (NeurIPS 2025 · similarity 0.92)
- MGD$^3$ : Mode-Guided Dataset Distillation using Diffusion Models (ICML 2025 · similarity 0.92)
- D^4: Dataset Distillation via Disentangled Diffusion Model (CVPR 2024 · similarity 0.92)
- Taming Diffusion for Dataset Distillation with High Representativeness (ICML 2025 · similarity 0.90)
- Boost Self-Supervised Dataset Distillation via Parameterization, Predefined Augmentation, and Approximation (ICLR 2025 · similarity 0.89)
- Efficient Dataset Distillation via Minimax Diffusion (CVPR 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.
