Research map / Image generation / Fast diffusion sampling
Maximum Likelihood Training of Score-Based Diffusion Models
Yang Song, Conor Durkan, Iain Murray, Stefano Ermon
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 |
|---|---|---|---|
| Train with likelihood weighting instead of the original weighting (ablation) | Score-based diffusion models · CIFAR-10 | NLL: 3.04 -> 2.94 | Improved |
| Train with likelihood weighting instead of the original weighting (ablation) | Score-based diffusion models · ImageNet 32×32 | NLL: 3.84 -> 3.86 | Hurt |
| Add importance sampling to likelihood-weighted training (ablation) | Score-based diffusion models · CIFAR-10 | loss variance after convergence: 98.48 -> 0.068 | Improved |
| Add importance sampling to likelihood-weighted training (ablation) | Score-based diffusion models · ImageNet | loss variance after convergence: 0.51 -> 0.043 | Improved |
| Combine likelihood weighting with importance sampling (ablation) | Score-based diffusion models · CIFAR-10 and ImageNet 32×32 | NLL | Improved |
| Use variational rather than uniform dequantization (ablation) | Score-based diffusion models · CIFAR-10 and ImageNet 32×32 | NLL: 3.16 -> 3.04; 3.90 -> 3.84 | Improved |
| Use a deeper DDPM++ architecture (ablation) | Score-based diffusion models · CIFAR-10 | NLL: 2.94 -> 2.93; 2.84 -> 2.81 | Mixed |
| Use a deeper DDPM++ architecture (ablation) | Score-based diffusion models · ImageNet 32×32 | NLL: 3.80 -> 3.79; 3.79 -> 3.76 | Mixed |
| Use subVP rather than VP SDE with likelihood weighting and importance sampling (ablation) | Score-based diffusion models · ImageNet 32×32 | NLL: 3.80 -> 3.79; 3.79 -> 3.76 | Improved |
| Use subVP rather than VP SDE with likelihood weighting and importance sampling (ablation) | Score-based diffusion models · CIFAR-10 | NLL: 2.83 -> 2.84; 2.80 -> 2.81 | Mixed |
| Add variational dequantization to the ScoreFlow model (ablation) | Score-based diffusion models · CIFAR-10 and ImageNet 32×32 | NLL | Improved |
| Optimize likelihood using likelihood weighting and importance sampling (ablation) | Score-based diffusion models · CIFAR-10 | FID: 3.98 -> 6.03 | Hurt |
Closest papers by meaning
- O(d/T) Convergence Theory for Diffusion Probabilistic Models under Minimal Assumptions (ICLR 2025 · similarity 0.91)
- Dimension-free Score Matching and Time Bootstrapping for Diffusion Models (NeurIPS 2025 · similarity 0.91)
- Adapting to Unknown Low-Dimensional Structures in Score-Based Diffusion Models (NeurIPS 2024 · similarity 0.91)
- Minimax Optimality of Score-based Diffusion Models: Beyond the Density Lower Bound Assumptions (ICML 2024 · similarity 0.90)
- Neural Network-Based Score Estimation in Diffusion Models: Optimization and Generalization (ICLR 2024 · similarity 0.90)
- Training Diffusion-based Generative Models with Limited Data (ICML 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.
