Research map / Image generation / Fast diffusion sampling
On the Robustness of Langevin Dynamics to Score Function Error
Daniel Yiming Cao, August Y. Chen, Karthik Sridharan, Yuchen Wu
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 |
|---|---|---|---|
| run Langevin dynamics with an adversarial L2-accurate score estimate instead of the true score | Langevin dynamics · Gaussian target, d=50, ||mu||=49.5 | KL of samples: 8.39 -> 3.08 × 104 | Hurt |
| initialize Langevin at samples used to train the score estimate instead of fresh samples (ablation) | Langevin dynamics with learned score · Gaussian target N(1, 2I), d=50 | KL / Wasserstein of samples | Hurt |
| initialize from fresh target samples vs N(0, I) (vanilla) (ablation) | Langevin dynamics with learned score · Gaussian target, d=50 | KL of samples | No clear effect |
| initialize from fresh samples or training samples instead of N(0, I) vanilla (ablation) | Langevin dynamics with learned score · two-component Gaussian mixture, d=25 | Wasserstein distance | Improved |
| use fresh samples (not training samples) for data-based initialization (ablation) | Langevin dynamics with learned score · Gaussian mixture, d=25 | Wasserstein distance | Improved |
Closest papers by meaning
- Score-based generative models are provably robust: an uncertainty quantification perspective (NeurIPS 2024 · similarity 0.88)
- Preconditioned Langevin Dynamics with Score-based Generative Models for Infinite-Dimensional Linear Bayesian Inverse Problems (NeurIPS 2025 · similarity 0.88)
- Sampling by averaging: A multiscale approach to score estimation (NeurIPS 2025 · similarity 0.88)
- Nearest Neighbour Score Estimators for Diffusion Generative Models (ICML 2024 · similarity 0.88)
- Dimension-free Score Matching and Time Bootstrapping for Diffusion Models (NeurIPS 2025 · similarity 0.88)
- 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.
