Research map / Image generation / Fast diffusion sampling
A Continuous Time Framework for Discrete Denoising Models
Andrew Campbell, Joe Benton, Valentin De Bortoli, Tom Rainforth, George Deligiannidis, Arnaud Doucet
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 |
|---|---|---|---|
| add 10 corrector steps per predictor step at the end of reverse sampling (t<0.1T) (ablation) | tauLDR · CIFAR10 | FID: 8.10 -> 3.74 | Improved |
| add 10 corrector steps per predictor step (ablation) | tauLDR · CIFAR10 | Inception Score: 8.74 -> 9.49 | Improved |
| add 10 corrector steps per predictor step (ablation) | tauLDR · CIFAR10 | ELBO (bits/dim): -3.59 -> -3.59 | No clear effect |
| use continuous-time tauLDR-0 instead of discrete-time D3PM Gauss | discrete-state CIFAR10 diffusion · CIFAR10 | FID: 7.34 -> 8.10 | Hurt |
| use continuous-time tauLDR-0 instead of discrete-time D3PM Gauss | discrete-state CIFAR10 diffusion · CIFAR10 | Inception Score: 8.56 -> 8.74 | Improved |
| use continuous-time tauLDR-0 instead of discrete-time D3PM Gauss | discrete-state CIFAR10 diffusion · CIFAR10 | ELBO (bits/dim): - 3.44 -> -3.59 | Hurt |
| use uniform forward rate instead of birth/death rate (ablation) | tauLDR-0 · Lakh pianoroll | Hellinger distance: 0.3928 -> 0.3765 | Improved |
| use uniform forward rate instead of birth/death rate (ablation) | tauLDR-0 · Lakh pianoroll | Proportion of outliers: 0.1316 -> 0.1106 | Improved |
| add 2 corrector steps per predictor step (ablation) | tauLDR Uniform · Lakh pianoroll | Hellinger distance: 0.3765 -> 0.3762 | No clear effect |
| replace discrete-time D3PM Uniform with continuous-time tauLDR-0 Uniform | music generation · Lakh pianoroll | Hellinger distance: 0.3839 -> 0.3765 | Improved |
| use more corrector steps under fixed sampling budget (ablation) | tauLDR · CIFAR10 | FID | Mixed |
Closest papers by meaning
- Assessing the quality of denoising diffusion models in Wasserstein distance: noisy score and optimal bounds (NeurIPS 2025 · similarity 0.88)
- Why Masking Diffusion Works: Condition on the Jump Schedule for Improved Discrete Diffusion (NeurIPS 2025 · similarity 0.88)
- Bridging discrete and continuous state spaces: Exploring the Ehrenfest process in time-continuous diffusion models (ICML 2024 · similarity 0.88)
- Simple Guidance Mechanisms for Discrete Diffusion Models (ICLR 2025 · similarity 0.88)
- Convergence of Score-Based Discrete Diffusion Models: A Discrete-Time Analysis (ICLR 2025 · similarity 0.88)
- Particle Denoising Diffusion Sampler (ICML 2024 · similarity 0.88)
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.
