Research map / Image generation / Fast diffusion sampling
Score-Based Generative Modeling through Stochastic Differential Equations
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the deeper NCSN++ architecture with the continuous objective (ablation) | NCSN++ with VE SDE · CIFAR-10 | FID: 2.45 -> 2.20 | Helped |
| use the continuous objective without increasing depth (ablation) | NCSN++ with VE SDE · CIFAR-10 | FID: 2.45 -> 2.38 | Helped |
| use a continuous objective and sub-VP rather than VP (ablation) | DDPM++ · CIFAR-10 uniformly dequantized test | NLL (bits/dim): 3.16 -> 3.02 | Helped |
| increase depth and use sub-VP rather than VP (ablation) | DDPM++ cont. · CIFAR-10 uniformly dequantized test | NLL (bits/dim): 3.16 -> 2.99 | Helped |
| use sub-VP rather than VP (ablation) | DDPM++ cont. · CIFAR-10 | FID: 2.55 -> 2.61 | Mixed |
| increase network depth (ablation) | DDPM++ cont. (sub-VP) · CIFAR-10 uniformly dequantized test | NLL (bits/dim): 3.02 -> 2.99 | Helped |
| add one corrector step per predictor step (ablation) | VE SDE predictor-only sampler · CIFAR-10 | FID: 4.79 ± .07 -> 3.60 ± .02 | Helped |
| add one corrector step per predictor step (ablation) | VP SDE predictor-only sampler · CIFAR-10 | FID: 3.21 ± .02 -> 3.18 ± .01 | Helped |
| use reverse diffusion rather than ancestral sampling (ablation) | VE SDE predictor-only sampler · CIFAR-10 | FID: 4.98 ± .06 -> 4.79 ± .07 | Helped |
| use reverse diffusion rather than ancestral sampling (ablation) | VP SDE predictor-only sampler · CIFAR-10 | FID: 3.24 ± .02 -> 3.21 ± .02 | Helped |
| use corrector-only sampling instead of predictor-only sampling (ablation) | VE SDE · CIFAR-10 | FID: 4.88 ± .06 -> 20.43 ± .07 | Hurt |
| use corrector-only sampling instead of predictor-only sampling (ablation) | VP SDE · CIFAR-10 | FID: 3.24 ± .02 -> 19.06 ± .06 | Hurt |
Closest papers by meaning
- Score-based Generative Modeling through Stochastic Evolution Equations in Hilbert Spaces (NeurIPS 2023 · similarity 0.93)
- A Variational Perspective on Diffusion-Based Generative Models and Score Matching (NeurIPS 2021 · similarity 0.91)
- Noise Conditional Variational Score Distillation (ICML 2025 · similarity 0.89)
- Nearest Neighbour Score Estimators for Diffusion Generative Models (ICML 2024 · similarity 0.89)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.89)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.88)
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.