Research map / Probabilistic and scientific modeling / Variational and generative sampling
Wasserstein Convergence Guarantees for a General Class of Score-Based Generative Models
Xuefeng Gao, Hoang M. Nguyen, Lingjiong Zhu
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 polynomial VP noise schedule beta=(b+at)^5 instead of linear | DDPM VP · CIFAR-10 | FID: 11.26 -> 9.64 | Improved |
| use exponential VP noise schedule ab^t instead of linear | DDPM VP · CIFAR-10 | FID: 11.26 -> 9.98 | Improved |
| use linear schedule instead of constant beta | DDPM VP · CIFAR-10 | FID: 17.46 -> 11.26 | Improved |
| increase polynomial order rho from 5 to 10 (ablation) | DDPM VP · CIFAR-10 | FID: 9.64 -> 10.27 | Hurt |
| increase polynomial order rho from 2 to 5 (ablation) | DDPM VP · CIFAR-10 | FID: 9.77 -> 9.64 | Improved |
| use VP-SDE instead of VE-SDE with exponential schedule | diffusion SGM · CIFAR-10 | FID: 22.11 -> 9.64 | Improved |
| use polynomial g with rho=10 instead of exponential g in VE-SDE | NCSN VE · CIFAR-10 | FID: 22.11 -> 99.89 | Hurt |
| use constant g instead of exponential g in VE-SDE | NCSN VE · CIFAR-10 | FID: 22.11 -> 461.42 | Hurt |
| increase polynomial order rho from 2 to 10 in VE-SDE (ablation) | NCSN VE · CIFAR-10 | FID: 369.51 -> 99.89 | Improved |
| use deeper network (more residual blocks) with rho=5 polynomial schedule | DDPM VP · CIFAR-10 | FID: 9.64 -> 8.20 | Improved |
| use polynomial rho=5 instead of linear schedule in deeper networks | DDPM deep VP · CIFAR-10 | FID: 9.22 -> 8.20 | Improved |
| use deeper network for VE exponential schedule | NCSN VE · CIFAR-10 | FID: 22.11 -> 20.00 | Improved |
Closest papers by meaning
- Global Well-posedness and Convergence Analysis of Score-based Generative Models via Sharp Lipschitz Estimates (ICLR 2025 · similarity 0.90)
- Score-based generative models are provably robust: an uncertainty quantification perspective (NeurIPS 2024 · similarity 0.90)
- Algorithm- and Data-Dependent Generalization Bounds for Diffusion Models (NeurIPS 2025 · similarity 0.89)
- Approximation and Generalization Abilities of Score-based Neural Network Generative Models for Sub-Gaussian Distributions (NeurIPS 2025 · similarity 0.89)
- Beyond Log-Concavity and Score Regularity: Improved Convergence Bounds for Score-Based Generative Models in W2-distance (ICML 2025 · similarity 0.88)
- A Non-Asymptotic Convergent Analysis for Scored-Based Graph Generative Model via a System of Stochastic Differential Equations (ICML 2025 · 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.
