Research map / Image generation / Fast diffusion sampling
Skipping the Zeros in Diffusion Models for Sparse Data Generation
Phil Ostheimer, Mayank Nagda, Andriy Balinskyy, Gabriel Vicente Rodrigues, Jean Radig, Carl Herrmann, Stephan Mandt, Marius Kloft, Sophie Fellenz
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 |
|---|---|---|---|
| use SED instead of dense DDPM for sparse calorimeter images | SED · Muon Signal | Wasserstein PT: 220.32 -> 16.31 | Improved |
| use SED instead of domain-specific SARM | SED · Muon Background | Wasserstein PT: 12.61 -> 9.63 | Improved |
| use SED instead of latent DM with same parameter count | SED · Muon Signal | Wasserstein PT: 28.20 -> 16.31 | Improved |
| add post-hoc thresholding to dense DDPM (ablation) | DDPM · Muon Signal | Wasserstein PT: 220.32 -> 24.22 | Improved |
| add post-hoc thresholding to LDM (ablation) | LDM · Muon Signal | Wasserstein PT: 28.20 -> 43.66 | Hurt |
| add post-hoc thresholding to LDM (ablation) | LDM · Tabula Muris | SCC: 0.87 -> 0.26 | Hurt |
| use SED instead of scDiffusion | SED · Tabula Muris | MMD: 1.53 -> 0.54 | Improved |
| use SED instead of dense LDM on scRNA | SED · Tabula Muris | SCC: 0.87 -> 0.76 SCC; 5.82 -> 0.54 MMD | Mixed |
| skip zeros so compute scales with non-zeros as dimensionality grows | SED · scRNA with added zero genes | relative FLOPs | Improved |
| use SED instead of DDPM sampling on highly sparse data | SED · Muon Signal | sampling time (ms per batch): 453.1 -> 23.5 | Improved |
| use SED on less sparse Fashion-MNIST | SED · Fashion-MNIST (50% sparse) | sampling time (ms per batch): LDM 77.7 vs SED 354.6 | Hurt |
| mean pooling vs special [CLS]-like token in encoder (ablation) | SAVAE encoder · latent representation | performance | No clear effect |
Closest papers by meaning
- Sparsely Supervised Diffusion (NeurIPS 2026 · similarity 0.87)
- Simplified and Generalized Masked Diffusion for Discrete Data (NeurIPS 2024 · similarity 0.87)
- Diffusion Transformers for Imputation: Statistical Efficiency and Uncertainty Quantification (NeurIPS 2025 · similarity 0.86)
- EVODiff: Entropy-aware Variance Optimized Diffusion Inference (NeurIPS 2025 · similarity 0.86)
- Theory on Score-Mismatched Diffusion Models and Zero-Shot Conditional Samplers (ICLR 2025 · similarity 0.86)
- Learning Single Index Models with Diffusion Priors (ICML 2025 · similarity 0.86)
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.
