Research map / Image generation / Autoregressive and latent generation
InfoDiffusion: Representation Learning Using Information Maximizing Diffusion Models
Yingheng Wang, Yair Schiff, Aaron Gokaslan, Weishen Pan, Fei Wang, Christopher De Sa, Volodymyr Kuleshov
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 |
|---|---|---|---|
| Use InfoDiffusion instead of DiffAE | Diffusion representation learning · FashionMNIST | Latent Qual.: 0.822 \pm 0.006 -> 0.856 \pm 0.004 | Improved |
| Use InfoDiffusion instead of DiffAE | Diffusion representation learning · CIFAR10 | Latent Qual.: 0.442 \pm 0.004 -> 0.461 \pm 0.003 | Improved |
| Use InfoDiffusion instead of DiffAE | Diffusion representation learning · FFHQ | Latent Qual.: 0.565 \pm 0.001 -> 0.628 \pm 0.002 | Improved |
| Sample z from a learned latent prior instead of the standard prior | InfoDiffusion · FashionMNIST | FID: 22.5 -> 8.1 | Improved |
| Sample z from a learned latent prior instead of the standard prior | InfoDiffusion · CIFAR10 | FID: 36.2 -> 25.7 | Improved |
| Sample z from a learned latent prior instead of the standard prior | InfoDiffusion · FFHQ | FID: 19.8 -> 13.1 | Improved |
| Sample z from a learned latent prior instead of the standard prior | InfoDiffusion · CelebA | FID: 12.1 -> 7.6 | Improved |
| Add guidance on z with w=2 (ablation) | InfoDiffusion with learned latent sampling · FashionMNIST | FID: 8.1 -> 6.8 | Improved |
| Add guidance on z with w=2 (ablation) | InfoDiffusion with learned latent sampling · CIFAR10 | FID: 25.7 -> 21.3 | Improved |
| Add guidance on z with w=2 (ablation) | InfoDiffusion with learned latent sampling · FFHQ | FID: 13.1 -> 9.8 | Improved |
| Add guidance on z with w=2 (ablation) | InfoDiffusion with learned latent sampling · CelebA | FID: 7.6 -> 6.3 | Improved |
| Use the regularization setting (lambda,zeta)=(0.1,0.9) instead of zeta=1 with small lambda (ablation) | InfoDiffusion · CelebA | attribute information in latents | Improved |
Later papers that compared with this method
1 later accepted paper reports InfoDiffusion as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- DisDiff: Unsupervised Disentanglement of Diffusion Probabilistic Models (NeurIPS 2023 · better on 2 of 2 shared results)
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Nested Diffusion Models Using Hierarchical Latent Priors (CVPR 2025 · similarity 0.89)
- Interpretable Diffusion via Information Decomposition (ICLR 2024 · similarity 0.89)
- Isometric Representation Learning for Disentangled Latent Space of Diffusion Models (ICML 2024 · similarity 0.89)
- Diffusion Model with Cross Attention as an Inductive Bias for Disentanglement (NeurIPS 2024 · similarity 0.88)
- Compositional Discrete Latent Code for High Fidelity, Productive Diffusion Models (NeurIPS 2025 · similarity 0.88)
- DisCo-Diff: Enhancing Continuous Diffusion Models with Discrete Latents (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.
