Research map / Probabilistic and scientific modeling / Variational and generative sampling
Nonparametric Score Estimators
Yuhao Zhou, Jiaxin Shi, Jun 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 curl-free kernels instead of diagonal kernels | Nonparametric score estimators · High-dimensional grid distributions | score estimation accuracy | Improved |
| Use curl-free kernels instead of diagonal kernels | WAE score estimators · MNIST and CelebA, high latent dimensions | negative log-likelihood / FID: 92.05 vs 114.57; 50.66 vs 76.85 | Improved |
| Use curl-free kernels instead of diagonal kernels | Nonparametric score estimators · Low-dimensional grid distributions | score estimation accuracy | No clear effect |
| Use ν-method iterative regularization | Curl-free score estimation · MNIST WAE, latent dim 64 | negative log-likelihood: 91.48 vs 93.06 | Improved |
| Use ν-method iterative regularization | Curl-free score estimation · CelebA WAE, latent dim 64 | FID: 50.16 vs 72.68 | Improved |
| Use ν-method iterative regularization | Curl-free score estimation · MNIST WAE, latent dim 8 and 32 | negative log-likelihood: 97.28 vs 96.98; 90.94 vs 89.06 | Hurt |
| Use ν-method iterative regularization | Curl-free score estimation · CelebA WAE, latent dim 8 and 32 | FID: 72.02 vs 69.72; 52.86 vs 49.93 | Hurt |
| Approximate KEF with the conjugate gradient solver | KEF · MNIST WAE, latent dim 128 | time per epoch: 201.1s -> 13.7s | Improved |
| Replace full KEF with low-rank NKEF | KEF · MNIST WAE, latent dims 32, 64, and 128 | negative log-likelihood: 90.93 -> 92.29; 91.58 -> 92.82; 92.40 -> 94.14 | Hurt |
| Increase the NKEF low-rank setting from NKEF₂ to NKEF₄ or NKEF₈ | NKEF · MNIST WAE, latent dim 32 | negative log-likelihood: 92.29 -> 91.19 -> 90.86 | Mixed |
| Increase the NKEF low-rank setting from NKEF₂ to NKEF₄ | NKEF · CelebA WAE, latent dims 32 and 64 | FID: 53.92 -> 54.54; 51.16 -> 50.76 | Mixed |
Closest papers by meaning
- SD-KDE: Score-Debiased Kernel Density Estimation (NeurIPS 2025 · similarity 0.86)
- Neural Network-Based Score Estimation in Diffusion Models: Optimization and Generalization (ICLR 2024 · similarity 0.85)
- Minimax Optimality of Score-based Diffusion Models: Beyond the Density Lower Bound Assumptions (ICML 2024 · similarity 0.85)
- Sobolev Space Regularised Pre Density Models (ICML 2024 · similarity 0.85)
- Beyond Log-Concavity and Score Regularity: Improved Convergence Bounds for Score-Based Generative Models in W2-distance (ICML 2025 · similarity 0.85)
- ConjNorm: Tractable Density Estimation for Out-of-Distribution Detection (ICLR 2024 · similarity 0.84)
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.
