Research map / Image generation / Fast diffusion sampling
Fast and Scalable Analytical Diffusion
Xinyi Shang, Peng Sun, Jingyu Lin, Zhiqiang Shen
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 |
|---|---|---|---|
| replace full-dataset scan with dynamic golden subset | PCA analytical denoiser · AFHQ | time per step (s) / MSE: 24.896 -> 0.351 s; MSE 0.008 -> 0.007 | Improved |
| replace full-dataset scan with dynamic golden subset | PCA analytical denoiser · ImageNet-1K unconditional | time per step (s): 110.798 -> 2.607 | Improved |
| use unbiased streaming softmax on golden subset instead of biased weighted streaming softmax (ablation) | GolDDiff · CelebA-HQ | r2: 0.818 -> 0.836 | Improved |
| use unbiased streaming softmax on full dataset (PCA Unbiased) instead of golden subset (ablation) | PCA · ImageNet-1K conditional, 100 steps | r2: 0.521 -> 0.411 | Hurt |
| plug GolDDiff into the Kamb denoiser | Kamb · CelebA-HQ | time per step (s): 37.158 -> 1.336 | Improved |
| plug GolDDiff into the Optimal denoiser | Optimal denoiser · AFHQ | r2: -0.713 -> -0.270 | Improved |
| reduce proxy dimension from D to D/16 (ablation) | GolDDiff · ImageNet-1K | r2: 0.494 -> 0.490 | No clear effect |
| reduce proxy dimension to D/256 (ablation) | GolDDiff · ImageNet-1K | r2: 0.490 -> 0.480 | Hurt |
| shrink coarse set mmax or golden subset kmin to extreme lower bound (ablation) | GolDDiff · CIFAR-10, CelebA-HQ, AFHQ | MSE | Hurt |
| use GolDDiff against different neural denoisers (EDM-VP) | GolDDiff vs PCA · CIFAR-10 | r2: 0.594 -> 0.654 | Improved |
| train a neural denoiser with analytical posterior-mean target instead of clean sample (ablation) | CNN on MNIST · MNIST | FID at epoch 30: 20.2 -> 13.4 | Improved |
| use random small subsets (N=10, 100) in high-noise steps (ablation) | PCA analytical denoiser · CIFAR-10 | MSE | Hurt |
Closest papers by meaning
- EVODiff: Entropy-aware Variance Optimized Diffusion Inference (NeurIPS 2025 · similarity 0.90)
- Efficient Denoising Diffusion via Probabilistic Masking (ICML 2024 · similarity 0.90)
- Tracing the Roots: Leveraging Temporal Dynamics in Diffusion Trajectories for Origin Attribution (NeurIPS 2025 · similarity 0.90)
- Test-Time Scaling of Diffusion Models via Noise Trajectory Search (NeurIPS 2025 · similarity 0.90)
- Accelerating Diffusion Models with Parallel Sampling: Inference at Sub-Linear Time Complexity (NeurIPS 2024 · similarity 0.90)
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.90)
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.
