Research map / Image generation / Fast diffusion sampling
Patronus: Interpretable Diffusion Models with Prototypes
Nina Weng, Aasa Feragen, Siavash Bigdeli
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 patch-based prototype activation vector instead of global semantic latent for guidance | DiffAE/InfoDiff · CelebA | latent AUROC: 0.84 -> 0.87 | Improved |
| use prototype activation vector instead of global latent | InfoDiff · CelebA | TAD: 0.30 -> 0.43 | Improved |
| use prototype activation vector instead of global latent | InfoDiff · CelebA | captured attributes: 3.0 -> 9.0 | Improved |
| use prototype activation vector instead of global latent | InfoDiff · FFHQ | latent AUROC: 0.61 -> 0.92 | Improved |
| use prototype activation vector instead of global latent | DiffAE/InfoDiff · FMNIST | latent AUROC: 0.84 -> 0.82 | Hurt |
| condition generation on prototype activations from test images | InfoDiff · CelebA | FID: 23.6 -> 14.6 | Improved |
| condition generation on prototype activations from test images | DiffAE/InfoDiff · FMNIST | FID: 8.2 -> 14.7 | Hurt |
| condition generation on prototype activations from test images | DiffAE/InfoDiff · FFHQ | FID: 31.2 -> 37.3 | Hurt |
| sample prototype activations with a learned latent diffusion model (unconditional) | Patronus · CIFAR-10 | FID: 32.9 -> 8.0 | Improved |
| add Prototype Distinct Loss (delta 0.5 or 1.0) for 100 extra epochs (ablation) | Patronus · CelebA prototypes | prototype correlation | No clear effect |
Closest papers by meaning
- Trade-Offs in Fine-Tuned Diffusion Models between Accuracy and Interpretability (AAAI 2024 · similarity 0.90)
- DifFRACT: Diffusion Feature Reconstruction and Attribution for Circuit Tracing (NeurIPS 2026 · similarity 0.90)
- Interpretable Diffusion via Information Decomposition (ICLR 2024 · similarity 0.89)
- Nested Diffusion Models Using Hierarchical Latent Priors (CVPR 2025 · similarity 0.89)
- NoiseCLR: A Contrastive Learning Approach for Unsupervised Discovery of Interpretable Directions in Diffusion Models (CVPR 2024 · similarity 0.89)
- Diffusion PID: Interpreting Diffusion via Partial Information Decomposition (NeurIPS 2024 · 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.
