Research map / Image generation / Autoregressive and latent generation
Latent Diffusion Inversion Requires Understanding the Latent Space
Mingxing Rao, Bowen Qu, Daniel Moyer
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 |
|---|---|---|---|
| filter out the 40% least-influential latent dimensions (by decoder pullback metric) before computing the attack norm | SimA · CIFAR-10 | AUC: 89.10 -> 91.26 | Improved |
| filter out the 40% least-influential latent dimensions | Loss attack · CIFAR-10 | AUC: 73.28 -> 75.87 | Improved |
| filter out the 40% least-influential latent dimensions | PIA · CelebA | TPR@1%FPR: 9.41 -> 21.13 | Improved |
| filter out the 40% least-influential latent dimensions | SimA · Pokemon (fine-tuned SD) | TPR@1%FPR: 20.38 -> 52.04 | Improved |
| filter out the 40% least-influential latent dimensions | SimA · MS-COCO (fine-tuned SD) | AUC: 93.71 -> 96.86 | Improved |
| filter out the 40% least-influential latent dimensions | SimA · Flickr (fine-tuned SD) | TPR@1%FPR: 2.59 -> 3.49 | Improved |
| drop 40% of dimensions randomly instead of by influence ranking (ablation) | SimA · CIFAR-10 | AUC: 89.10 -> 86.44 | Hurt |
| drop 40% of dimensions randomly (ablation) | SimA · ImageNet-1K | TPR@1%FPR: 3.87 -> 4.16 | Mixed |
| select samples in high decoder-distortion quartiles | SimA attack · CIFAR-10, ImageNet-1K | AUC by quartile | Improved |
| high-frequency suppression of denoised images (Lian et al.) applied to latent diffusion | Stable Diffusion MIA · Stable Diffusion | MIA performance | Hurt |
Closest papers by meaning
- On the Edge of Memorization in Diffusion Models (NeurIPS 2025 · similarity 0.91)
- Gradient-free Decoder Inversion in Latent Diffusion Models (NeurIPS 2024 · similarity 0.90)
- Understanding diffusion models requires rethinking (again) generalization (NeurIPS 2026 · similarity 0.90)
- Probing the Latent Hierarchical Structure of Data via Diffusion Models (ICLR 2025 · similarity 0.89)
- Isometric Representation Learning for Disentangled Latent Space of Diffusion Models (ICML 2024 · similarity 0.89)
- Why Diffusion Models Don’t Memorize: The Role of Implicit Dynamical Regularization in Training (NeurIPS 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.
