Research map / Image generation / Fast diffusion sampling
Learning to Generate Samples from Noise through Infusion Training
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 a transition operator using target infusion | Generative modeling on MNIST · MNIST test | Parzen-window-based estimator of lower bound on average test log-likelihood: 282±2 -> 312±1.7 | Helped |
| Train a transition operator using target infusion | Generative modeling on CIFAR-10 · CIFAR-10 | Inception score: 4.36 ± .06 -> 4.62 ± .06 | Helped |
| Replace diagonal covariance output with isotropic Gaussian output (ablation) | Infusion training · MNIST test | log-likelihood (IS estimate): 1836.27±0.551 -> 413.297±0.460 | Hurt |
| Set infusion rate to 0.3 rather than 0.0 (ablation) | Infusion training with 1 step · MNIST test | lower bound: 824.34 -> 1209.39 | Helped |
| Set infusion rate to 0.9 rather than 0.3 (ablation) | Infusion training with 1 step · MNIST test | lower bound: 1209.39 -> -161.37 | Hurt |
| Set infusion rate to 0.1 rather than 0.0 (ablation) | Infusion training with 5 steps · MNIST test | lower bound: 823.81 -> 1448.66 | Helped |
| Set infusion rate to 0.02 rather than 0.0 (ablation) | Infusion training with 10 steps · MNIST test | lower bound: 824.42 -> 1389.12 | Helped |
| Set infusion rate to 0.01 rather than 0.0 (ablation) | Infusion training with 15 steps · MNIST test | lower bound: 824.50 -> 1351.03 | Helped |
| Use a high infusion rate (0.06) with 15 steps (ablation) | Infusion training · MNIST test | lower bound: 1351.03 -> -941.78 | Hurt |
| Increase the number of denoising steps while using an appropriately lower infusion rate (ablation) | Infusion training · MNIST sample quality | qualitative sample quality | Helped |
| Use an increasing rather than constant infusion rate (ablation) | Infusion training with 15 steps · MNIST sample quality | qualitative sample quality | Helped |
| Use a high infusion rate (ablation) | Infusion training · MNIST test | overfitting | Hurt |
Closest papers by meaning
- Is Noise Conditioning Necessary for Denoising Generative Models? (ICML 2025 · similarity 0.87)
- Your Diffusion Model is Secretly a Noise Classifier and Benefits from Contrastive Training (NeurIPS 2024 · similarity 0.87)
- Denoising Task Difficulty-based Curriculum for Training Diffusion Models (ICLR 2025 · similarity 0.87)
- Analysis of Learning a Flow-based Generative Model from Limited Sample Complexity (ICLR 2024 · similarity 0.87)
- Sample-specific Noise Injection for Diffusion-based Adversarial Purification (ICML 2025 · similarity 0.87)
- Evaluating the design space of diffusion-based generative models (NeurIPS 2024 · similarity 0.87)
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.