Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Purification by Consistency-Aware Latent Space Optimization on Data Manifolds
Shuhai Zhang, Jiahao Yang, Hui Luo, Jie Chen, Ke Wang, Feng Liu, Bo Han, Mingkui Tan
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 |
|---|---|---|---|
| replace consistency model with GAN latent space (ablation) | CMAP · CIFAR-10 PGD+EOT | robust accuracy (%): 74.60 -> 40.80 | Hurt |
| replace consistency model with VAE (ablation) | CMAP · CIFAR-10 PGD+EOT | robust accuracy (%): 74.60 -> 58.80 | Hurt |
| replace consistency model with VAE (ablation) | CMAP · ImageNet-100 PGD+EOT | robust accuracy (%): 39.87 -> 43.60; standard 61.93 -> 68.20 | Mixed |
| add DiffPure on top of CMAP | CMAP · CIFAR-10 BPDA+EOT | robust accuracy (%): 83.53 -> 89.47 | Improved |
| add DiffPure on top of CMAP | CMAP · CIFAR-10 standard accuracy | accuracy (%): 88.73 -> 91.33 | Improved |
| add DiffPure on top of CMAP | CMAP · CIFAR-10 PGD-l2+EOT | robust accuracy (%): 84.00 -> 83.60 | No clear effect |
| use CMAP instead of best purification baseline | CMAP · CIFAR-10 PGD+EOT | robust accuracy (%): 55.87 -> 74.60 | Improved |
| use CMAP instead of FreqPure | CMAP · ImageNet-100 BPDA+EOT | robust accuracy (%): 19.58 -> 49.47 | Improved |
| use CMAP instead of GNSP | CMAP · CIFAR-10 BPDA+EOT | robust accuracy (%): 88.53 -> 83.53 | Hurt |
| use CMAP instead of GNSP at larger epsilon 16/255 | CMAP · CIFAR-10 AutoAttack | robust accuracy (%): 33.80 -> 80.73 | Improved |
| use CMAP instead of FreqPure | CMAP · CIFAR-10 AutoAttack-l2 | robust accuracy (%): 88.26 -> 83.40 | Hurt |
| use CMAP instead of FreqPure | CMAP · CIFAR-10 clean accuracy | accuracy (%): 93.96 -> 88.73 | Hurt |
Closest papers by meaning
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · similarity 0.89)
- Instant Adversarial Purification with Adversarial Consistency Distillation (CVPR 2025 · similarity 0.89)
- Adversarial Training on Purification (AToP): Advancing Both Robustness and Generalization (ICLR 2024 · similarity 0.88)
- Robust Overfitting Does Matter: Test-Time Adversarial Purification With FGSM (CVPR 2024 · similarity 0.87)
- Sample-specific Noise Injection for Diffusion-based Adversarial Purification (ICML 2025 · similarity 0.87)
- DiffBreak: Is Diffusion-Based Purification Robust? (NeurIPS 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.
