Research map / Probabilistic and scientific modeling / Variational and generative sampling
Autoregressive Score Matching
Chenlin Meng, Lantao Yu, Yang Song, Jiaming Song, Stefano Ermon
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 |
|---|---|---|---|
| train implicit VAE encoder with CSM instead of ELBO | VAE · MNIST latent dim 8 | negative log-likelihood (AIS): 96.74 -> 95.02 | Improved |
| train implicit VAE encoder with CSM instead of ELBO | VAE · MNIST latent dim 16 | negative log-likelihood (AIS): 91.82 -> 88.42 | Improved |
| train implicit VAE encoder with CSM instead of ELBO | VAE · CelebA latent dim 32 | FID: 66.31 -> 62.20 | Improved |
| use CSM instead of sliced score matching (SSM) for the entropy gradient | VAE implicit encoder · MNIST latent dim 8 | negative log-likelihood (AIS): 95.61 -> 95.02 | Improved |
| use CSM instead of SSM | VAE implicit encoder · CelebA | FID: 62.50 -> 62.20 | Improved |
| use CSM instead of Stein or Spectral estimators | VAE implicit encoder · CelebA | FID: 108.84 -> 62.20 | Improved |
| train the autoregressive score model with CSM instead of SSM (SSM-AR) | AR score model for entropy gradient · CelebA | FID: 66.88 -> 62.20 | Improved |
| use CSM instead of original score matching on Gaussians | score matching · multivariate Gaussians, increasing dimension | time per iteration and memory | Improved |
| use CSM instead of sliced score matching | score matching · 100-d Gaussian N(0, 0.1^2 I) | training convergence | Improved |
| use CSM instead of denoising score matching with small sigma | score matching · 100-d Gaussian and MNIST MADE | convergence and NLL | Improved |
| decrease DSM noise level sigma | denoising score matching · MNIST MADE | bias vs convergence speed | Mixed |
| train MADE with CSM (AR-CSM) instead of MLE | MADE with Gaussian densities · MNIST/CIFAR-10/CelebA samples | sample quality | Improved |
Closest papers by meaning
- Expressive Score-Based Priors for Distribution Matching with Geometry-Preserving Regularization (ICML 2025 · similarity 0.86)
- What’s the score? Automated Denoising Score Matching for Nonlinear Diffusions (ICML 2024 · similarity 0.85)
- Neural Network-Based Score Estimation in Diffusion Models: Optimization and Generalization (ICLR 2024 · similarity 0.85)
- Sequential Neural Score Estimation: Likelihood-Free Inference with Conditional Score Based Diffusion Models (ICML 2024 · similarity 0.85)
- Score Matching with Missing Data (ICML 2025 · similarity 0.85)
- Continuous Visual Autoregressive Generation via Score Maximization (ICML 2025 · similarity 0.85)
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.
