Research map / Image generation / Fast diffusion sampling
Secure Inference for Diffusion Models via Unconditional Scores
Jaeyun Song, Geondo Park, Uigyu Kim, Joonhyung Park, Eunho Yang
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 |
|---|---|---|---|
| add unconditional-score guidance | Loose-Approx Stable Diffusion · MSCOCO | FID: 15.52 -> 14.20 | Improved |
| add unconditional-score guidance | Loose-Approx Stable Diffusion · Flickr8k | FID: 23.25 -> 21.47 | Improved |
| add unconditional-score guidance | Precise-Approx Stable Diffusion · MSCOCO | FID: 13.28 -> 12.45 | Improved |
| add unconditional-score guidance | CipherDM · Binary MNIST | F1-Score: 0.0820 -> 0.6000 | Improved |
| relax polynomial approximations (fewer segments, lower degree) | Stable Diffusion · MSCOCO | FID: 13.28 -> 15.52 | Hurt |
| increase guidance learning rate eta from 0.05 to 0.20 (ablation) | Loose-Approx + Ours · Flickr8k | FID: 21.85 -> 21.25 | Improved |
| add PAG test-time guidance | Loose-Approx · Flickr8k | FID: 23.25 -> 36.72 | Hurt |
| add FKS test-time scaling | Loose-Approx · Flickr8k | FID: 23.25 -> 24.42 | Hurt |
| add unconditional-score guidance to PCM few-step sampling | Loose-Approx + PCM (4 steps) · MSCOCO | FID: 53.80 -> 36.69 | Improved |
| use exact text encoder and VAE decoder (ablation) | Loose-Approx · MSCOCO | FID: 15.52 -> 14.22 | Improved |
| add guidance when encoder/decoder are exact (ablation) | Loose-Approx + Exact-Enc-Dec · MSCOCO | FID: 14.22 -> 13.10 | Improved |
Closest papers by meaning
- RAPID: Retrieval Augmented Training of Differentially Private Diffusion Models (ICLR 2025 · similarity 0.88)
- Toward effective protection against diffusion-based mimicry through score distillation (ICLR 2024 · similarity 0.87)
- Privacy-Preserving Low-Rank Adaptation Against Membership Inference Attacks for Latent Diffusion Models (AAAI 2025 · similarity 0.87)
- DRAG: Data Reconstruction Attack using Guided Diffusion (ICML 2025 · similarity 0.87)
- Score Forgetting Distillation: A Swift, Data-Free Method for Machine Unlearning in Diffusion Models (ICLR 2025 · similarity 0.86)
- An Efficient Membership Inference Attack for the Diffusion Model by Proximal Initialization (ICLR 2024 · similarity 0.86)
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.
