Research map / Image generation / Fast diffusion sampling
Tackling the Generative Learning Trilemma with Denoising Diffusion GANs
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 |
|---|---|---|---|
| Use multimodal conditional GAN denoising with four steps | Diffusion image generation · CIFAR-10 | Time (s): 2000× faster than predictor-corrector sampling | Helped |
| Use four-step multimodal denoising rather than the cited diffusion baselines | Denoising Diffusion GAN · CIFAR-10 | Recall: 0.57 vs 0.57 | No clear effect |
| Reduce denoising to one step (ablation) | Denoising Diffusion GAN · CIFAR-10 | Recall: 0.57 -> 0.19 | Hurt |
| Increase denoising steps from four to eight (ablation) | Denoising Diffusion GAN · CIFAR-10 | FID: 3.75 -> 4.36 | Hurt |
| Increase denoising steps from four to two (ablation) | Denoising Diffusion GAN · CIFAR-10 | IS: 9.63 -> 9.80 | Helped |
| Replace iterative denoising with one-shot GAN plus diffusion augmentation (ablation) | Denoising Diffusion GAN · CIFAR-10 | FID: 3.75 -> 13.2 | Hurt |
| Predict x(t-1) directly instead of predicting x0 and sampling the posterior (ablation) | Denoising Diffusion GAN · CIFAR-10 | FID: 3.75 -> 6.03 | Hurt |
| Predict perturbation noise instead of predicting x0 and sampling the posterior (ablation) | Denoising Diffusion GAN · CIFAR-10 | FID: 3.75 -> 8.04 | Hurt |
| Add a latent variable to enable multimodal denoising (ablation) | Denoising Diffusion GAN · CIFAR-10 | FID: 20.6 -> 3.75 | Helped |
| Add a latent variable to enable multimodal denoising (ablation) | Denoising Diffusion GAN · CIFAR-10 | Recall: 0.42 -> 0.57 | Helped |
| Use four-step denoising rather than a one-shot GAN (ablation) | Denoising Diffusion GAN · CIFAR-10 | Recall: 0.19 -> 0.57 | Helped |
| Use four-step multimodal denoising rather than StyleGAN2 with ADA | Denoising Diffusion GAN · CIFAR-10 | Recall: 0.49 -> 0.57 | Helped |
Closest papers by meaning
- You Only Sample Once: Taming One-Step Text-to-Image Synthesis by Self-Cooperative Diffusion GANs (ICLR 2025 · similarity 0.91)
- Revisiting Diffusion Models: From Generative Pre-training to One-Step Generation (ICML 2025 · similarity 0.90)
- Denoising Diffusion Bridge Models (ICLR 2024 · similarity 0.89)
- Bridging Data Gaps in Diffusion Models with Adversarial Noise-Based Transfer Learning (ICML 2024 · similarity 0.89)
- Denoising Task Difficulty-based Curriculum for Training Diffusion Models (ICLR 2025 · similarity 0.89)
- One Step Diffusion via Shortcut Models (ICLR 2025 · similarity 0.89)
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.