Research map / Probabilistic and scientific modeling / Variational and generative sampling
Score-based Generative Modeling in Latent Space
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 the score-based prior in latent space with an LSGM VAE | CIFAR-10 · CIFAR-10 | FID: 43.18 -> 2.10 | Helped |
| Train the score-based prior in latent space with an LSGM VAE | CelebA-HQ-256 · CelebA-HQ-256 | FID: 30.87 -> 7.22 | Helped |
| Generate samples with latent-space ODE sampling instead of original SGM sampling | CelebA-HQ-256 · CelebA-HQ-256 | sampling time: 44.6 min. -> 0.07 min.; 3.91 min. -> 0.07 min. | Helped |
| Use LSGM for binary image modeling | Dynamically binarized OMNIGLOT · Dynamically binarized OMNIGLOT | NLL: 88.29 -> ≤87.79 | Helped |
| Train end-to-end rather than freezing the VAE encoder and decoder in stage two (ablation) | LSGM · CIFAR-10 (end-to-end ablation) | FID: 9.00 -> 5.19 | Helped |
| Train end-to-end rather than freezing the VAE encoder and decoder in stage two (ablation) | LSGM · CIFAR-10 (end-to-end ablation) | NELBO: 3.03 -> 2.98 | Helped |
| Add mixed Normal and neural score parameterization (ablation) | LSGM · CIFAR-10 (mixed-score ablation) | FID: 34.71 -> 7.60 | Helped |
| Add mixed Normal and neural score parameterization (ablation) | LSGM · CIFAR-10 (mixed-score ablation) | NELBO: 3.39 -> 3.29 | Helped |
| Use likelihood-weighted prior training rather than unweighted or reweighted training (ablation) | LSGM · CIFAR-10 | NELBO: 2.96; 2.97; 2.98; 2.99; 3.03; 3.04 | Helped |
| Use unweighted or reweighted score-prior training rather than likelihood weighting (ablation) | LSGM · CIFAR-10 | FID: 5.39; 4.99 | Helped |
| Use geometric VPSDE for likelihood-weighted training instead of linear VPSDE (ablation) | LSGM · CIFAR-10 | NELBO: 2.97 -> 2.96 | Helped |
| Use importance sampling for time instead of uniform time sampling (ablation) | LSGM with likelihood weighting · CIFAR-10 | training objective variance | Helped |
Closest papers by meaning
- Approximation and Generalization Abilities of Score-based Neural Network Generative Models for Sub-Gaussian Distributions (NeurIPS 2025 · similarity 0.89)
- Stabilize the Latent Space for Image Autoregressive Modeling: A Unified Perspective (NeurIPS 2024 · similarity 0.88)
- Beyond Log-Concavity and Score Regularity: Improved Convergence Bounds for Score-Based Generative Models in W2-distance (ICML 2025 · similarity 0.87)
- Supervised Score-Based Modeling by Gradient Boosting (AAAI 2025 · similarity 0.86)
- Algorithm- and Data-Dependent Generalization Bounds for Diffusion Models (NeurIPS 2025 · similarity 0.86)
- A Non-Asymptotic Convergent Analysis for Scored-Based Graph Generative Model via a System of Stochastic Differential Equations (ICML 2025 · 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.