Research map / Image generation / Fast diffusion sampling
M-Star: Markovian Projection of Star-Shaped Diffusion for Exponential Family Distributions
François Bertholom, Khalid Oublal
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 Markovian projection to star-shaped diffusion (ablation) | SS-DDPM · ImageNet-64, 8 NFE | FID: 12.45 -> 4.12 | Improved |
| add progressive distillation on top of Markovian projection (ablation) | M-Star · ImageNet-64, 8 NFE | FID: 4.12 -> 4.10 | No clear effect |
| add learned interaction matrix Mt instead of Mt = 0 (ablation) | SS-DDPM · ImageNet-64, 8 NFE | FID: 12.45 -> 4.12 | Improved |
| use learned Mt instead of fixed analytical Mt (ablation) | M-Star · ImageNet-64, 8 NFE | FID: 4.38 -> 4.12 | Improved |
| use learned Mt instead of fixed analytical Mt (ablation) | M-Star · ImageNet-64, 1000 NFE | FID: 1.44 -> 1.42 | No clear effect |
| add Markovian projection (learned Mt) at 1000 NFE (ablation) | SS-DDPM · ImageNet-64, 1000 NFE | FID: 3.87 -> 1.42 | Improved |
| swap progressive distillation for Shortcut conditioning (ablation) | M-Star · ImageNet-64, 8 NFE | FID: 4.12 -> 4.31 | No clear effect |
| use M-Star distillation instead of PD at 8 NFE | PD · ImageNet-64 few-step | FID: 1.70 -> 1.24 | Improved |
| use M-Star instead of Multistep-CD at 8 NFE | Multistep-CD · ImageNet-128 | FID: 2.10 -> 1.49 | Improved |
| distill base teacher to 8-step M-Star student | Base teacher · ImageNet-64 | FID: 1.42 -> 1.24 | Improved |
| distill base teacher to 8-step M-Star student | Base teacher · ImageNet-128 | FID: 1.76 -> 1.49 | Improved |
| distill base teacher to 8-step M-Star student | Base teacher · ImageNet-64 | IS: 84 -> 78 | Hurt |
Closest papers by meaning
- Towards Non-Asymptotic Convergence for Diffusion-Based Generative Models (ICLR 2024 · similarity 0.88)
- Generative Fractional Diffusion Models (NeurIPS 2024 · similarity 0.87)
- Piecewise deterministic generative models (NeurIPS 2024 · similarity 0.87)
- Distributional Diffusion Models with Scoring Rules (ICML 2025 · similarity 0.87)
- A solvable model of learning generative diffusion: theory and insights (NeurIPS 2025 · similarity 0.87)
- Scalable Discrete Diffusion Samplers: Combinatorial Optimization and Statistical Physics (ICLR 2025 · similarity 0.87)
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.
