Research map / Adversarial robustness and security / Attacks on text and graphs
Adversarial Attack and Defense for Denoising Diffusion Sampling
Zhao-Rong Lai, Xiwen Yuan, Jian Weng
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 local variation regularization (ADDDS) to ZOD-MC | ZOD-MC · CIFAR-10 BNN posterior sampling, PGD10 | classification accuracy (%): 8.860 -> 9.230 | Improved |
| add local variation regularization (ADDDS) to ZOD-MC | ZOD-MC · CIFAR-10 BNN posterior sampling, clean | classification accuracy (%): 12.030 -> 29.810 | Improved |
| add local variation regularization (ADDDS) to ZOD-MC | ZOD-MC · CIFAR-10 BNN posterior sampling, FGSM | classification accuracy (%): 8.830 -> 14.970 | Improved |
| add local variation regularization (ADDDS) to ZOD-MC | ZOD-MC · CIFAR-100 BNN posterior sampling, PGD10 | classification accuracy (%): 0.530 -> 5.990 | Improved |
| use ADDDS instead of RDMC on CIFAR-10 under PGD10 | RDMC · CIFAR-10 BNN posterior sampling, PGD10 | classification accuracy (%): 8.120 -> 9.230 | No clear effect |
| apply a Gaussian perturbation (sigma=0.1) at sampling time | ZOD-MC · CIFAR-10 BNN posterior sampling | classification accuracy (%): 12.030 -> 11.390 | Hurt |
| apply a Gaussian perturbation (sigma=0.1) at sampling time | proximal sampler · CIFAR-10 BNN posterior sampling | classification accuracy (%): 28.800 -> 21.040 | Hurt |
| apply a Gaussian perturbation (sigma=0.1) at sampling time | ADDDS · CIFAR-10 BNN posterior sampling | classification accuracy (%): 29.810 -> 29.230 | No clear effect |
| apply PGD10 attack at sampling time | ADDDS · CIFAR-10 BNN posterior sampling | classification accuracy (%): 29.810 -> 9.230 | Hurt |
| add LV regularization to the potential minimization (ablation) | potential minimization · Critical point V(x*) vs LV | V(x*) and local variation: V 3.9966 -> 4.0512; LV 5.9605e -08 -> 0.0893 | Mixed |
| replace first-order Monte Carlo samplers by zeroth-order ZOD-MC with LV defense | ADDDS · Gaussian mixtures under PGD10 | MMD and W2 | Improved |
Closest papers by meaning
- Interpreting and Improving Diffusion Models from an Optimization Perspective (ICML 2024 · similarity 0.88)
- Towards Understanding the Robustness of Diffusion-Based Purification: A Stochastic Perspective (ICLR 2025 · similarity 0.88)
- Directly Denoising Diffusion Models (ICML 2024 · similarity 0.87)
- LoRID: Low-Rank Iterative Diffusion for Adversarial Purification (AAAI 2025 · similarity 0.87)
- Sample-specific Noise Injection for Diffusion-based Adversarial Purification (ICML 2025 · similarity 0.87)
- Divide and Conquer: Heterogeneous Noise Integration for Diffusion-based Adversarial Purification (CVPR 2025 · similarity 0.87)
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.
