Research map / Image generation / Fast diffusion sampling
CountsDiff: A diffusion model on the natural numbers for generation and imputation of count-based data
Renzo Soatto, Anders Hoel, Greycen Ren, Shorna Alam, Stephen Bates, Nikolaos P. Daskalakis, Caroline Uhler, Maria Skoularidou
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 |
|---|---|---|---|
| generalise FI p-schedule to continuous time (ablation) | CountsDiff (CIFAR-10) · CIFAR-10 | FID: 5.73 -> 5.44 | Improved |
| add guidance 1.0 and attrition 0.01 to FI Continuous (ablation) | CountsDiff FI · CIFAR-10 | FID: 5.44 -> 5.20 | Improved |
| raise guidance to 2.0 and attrition to 0.02 (FI Continuous) (ablation) | CountsDiff FI · CIFAR-10 | FID: 5.20 -> 9.71 | Hurt |
| add guidance 1.0 and attrition 0.01 to cosine schedule (ablation) | CountsDiff cosine · CIFAR-10 | FID: 5.76 -> 5.26 | Improved |
| raise guidance to 2.0 and attrition to 0.02 (cosine) (ablation) | CountsDiff cosine · CIFAR-10 | FID: 5.26 -> 11.55 | Hurt |
| add guidance 1.0 and attrition to cosine schedule (ablation) | CountsDiff cosine · CIFAR-10 | Inception Score: 9.29 -> 9.85 | Improved |
| use cosine schedule instead of FI continuous (unguided) (ablation) | CountsDiff · CIFAR-10 | Inception Score: 9.09 -> 9.29 | Improved |
| use cosine schedule instead of FI continuous (unguided) (ablation) | CountsDiff · CIFAR-10 | FID: 5.44 -> 5.76 | Hurt |
| average 5 samples instead of 1 (ablation) | CountsDiff · fetus atlas 50% MCAR | RMSE: 1.42 -> 1.17 | Improved |
| average 5 samples instead of 1 (ablation) | CountsDiff · fetus atlas 50% MCAR | log(scFID): -9.18 -> -7.60 | Hurt |
| use CountsDiff 1-sample instead of Blackout | Blackout · fetus atlas 50% MCAR | log(scFID): -7.37 -> -9.18 | Improved |
| use CountsDiff 5-sample instead of Blackout | Blackout · fetus atlas 25% MNAR | RMSE: 0.88 -> 0.58 | Improved |
Closest papers by meaning
- Diffusion Transformers for Imputation: Statistical Efficiency and Uncertainty Quantification (NeurIPS 2025 · similarity 0.88)
- Flow Matching for Count Data (NeurIPS 2026 · similarity 0.88)
- Rethinking the Diffusion Models for Missing Data Imputation: A Gradient Flow Perspective (NeurIPS 2024 · similarity 0.88)
- TabDiff: a Mixed-type Diffusion Model for Tabular Data Generation (ICLR 2025 · similarity 0.88)
- DiffPuter: Empowering Diffusion Models for Missing Data Imputation (ICLR 2025 · similarity 0.87)
- Discretizing Continuous Time Series for Imputation with Masked Diffusion Training (NeurIPS 2026 · 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.
