Research map / Probabilistic and scientific modeling / Variational and generative sampling
Score Correction for Generative Models with Probabilistic Constraints
Shishang Wu, Bingjing Tang, Vinayak Rao
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 DualScore dual-optimized score correction instead of importance resampling or conditional sampling under noisy classifier | EDM diffusion on CelebA (balanced) · CelebA balanced, rho=0.5 | constraint gap L2: 0.090 -> 0.009 | Improved |
| use DualScore instead of IRS with noise-free classifier (rho=0) | EDM diffusion on CIFAR-10 LT/5% (balanced) · CIFAR-10 LT/5% balanced, rho=0 | constraint gap L2: 0.009 -> 0.014 | No clear effect |
| use DualScore instead of IRS under strong classifier noise | EDM diffusion on CelebA (reversed) · CelebA reversed, rho=0.75 | FID to original model: 0.711 -> 127.591 | Hurt |
| use SGLD sampling of the corrected density instead of importance resampling in the dual optimization (ablation) | DualScore · synthetic Gaussian/Laplace with density chasm N(-10,2^2) | constraint gap L2 | Improved |
| reduce SGLD steps per SGD iteration down to 16 (ablation) | DualScore · synthetic 1-d task | constraint gap L2 | No clear effect |
| use a stricter constraint (epsilon = 0) in a regularized NPMLE with small mixture weight r (ablation) | DualScore · NPMLE, p0X=N(0,1) | constraint gap L2 | Hurt |
Closest papers by meaning
- Preconditioned Score-Based Generative Models (IJCV 2025 · similarity 0.89)
- Score-Based Generative Modeling through Stochastic Differential Equations (ICLR 2021 · similarity 0.88)
- Zero-Shot Conditioning of Score-Based Diffusion Models by Neuro-Symbolic Constraints (AAAI 2025 · similarity 0.87)
- On Inference Stability for Diffusion Models (AAAI 2024 · similarity 0.87)
- Beyond Log-Concavity and Score Regularity: Improved Convergence Bounds for Score-Based Generative Models in W2-distance (ICML 2025 · similarity 0.86)
- Sequential Neural Score Estimation: Likelihood-Free Inference with Conditional Score Based Diffusion Models (ICML 2024 · similarity 0.86)
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.
