Research map / Image generation / Fast diffusion sampling
Variational Diffusion Models
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 |
|---|---|---|---|
| learn noise schedule to minimize VLB estimator variance (ablation) | continuous-time VDM · CIFAR-10 | Var(BPD): 6.35 -> 0.53 | Helped |
| increase number of diffusion steps T (ablation) | discrete-time VDM · CIFAR-10 | BPD: 4.31 -> 2.66 | Helped |
| train in continuous time instead of discrete time, evaluated with few steps (ablation) | VDM · CIFAR-10, Teval=10 | BPD: 4.31 -> 7.54 | Hurt |
| train in continuous time instead of discrete time, evaluated with many steps (ablation) | VDM · CIFAR-10 | BPD: 2.66 -> 2.65 | Helped |
| add Fourier features (ablation) | VDM · CIFAR-10 | test BPD | Helped |
| add Fourier features (ablation) | PixelCNN++ · CIFAR-10 | test BPD | No clear effect |
| learn SNR endpoints instead of fixed Ho et al. schedule (ablation) | VDM with Fourier features · CIFAR-10 | test BPD | Helped |
| use weighted diffusion loss w(SNR) instead of unweighted VLB (ablation) | VDM · CIFAR-10 | FID: 7.41 -> 4.0 | Helped |
| VDM vs prior best likelihood model | density estimation · CIFAR-10 no aug | BPD: 2.80 -> 2.65 | Helped |
Closest papers by meaning
- Information Theoretic Learning for Diffusion Models with Warm Start (NeurIPS 2025 · similarity 0.91)
- DiffEnc: Variational Diffusion with a Learned Encoder (ICLR 2024 · similarity 0.91)
- Diffusion Priors for Variational Likelihood Estimation and Image Denoising (NeurIPS 2024 · similarity 0.91)
- Neural Diffusion Models (ICML 2024 · similarity 0.91)
- A Variational Perspective on Solving Inverse Problems with Diffusion Models (ICLR 2024 · similarity 0.91)
- Diffusion Models With Learned Adaptive Noise (NeurIPS 2024 · similarity 0.91)
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.