Research map / Adversarial robustness and security / Certified and empirical robustness
Defense Against Adversarial Attacks via Controlling Gradient Leaking on Embedded Manifolds
Yueru Li, Shuyu Cheng, Hang Su, Jun Zhu
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 PCA projection to 800 dims with small-scale noise (noise800) over ordinary training | DenseNet · CIFAR10 C&W l2 | mean perturbation norm: 0.090 -> 0.208 | Improved |
| add PCA projection to 300 dims with noise (noise300) over ordinary training | DenseNet · CIFAR10 clean | error rate (%): 5.92 -> 16.53 | Hurt |
| add noise in small-scale PCA directions on top of PCA projection (800 dims) (ablation) | DenseNet · CIFAR10 C&W l2 | mean perturbation norm: 0.160 -> 0.208 | Improved |
| add noise in small-scale PCA directions on top of PCA projection (300 dims) (ablation) | DenseNet · CIFAR10 C&W l2 | mean perturbation norm: 0.256 -> 0.308 | Improved |
| add noise in small-scale directions on top of PCA projection (800 dims) (ablation) | DenseNet · CIFAR10 alpha300 | gradient leaking alpha300: 0.244 -> 0.460 | Improved |
| reduce PCA dimension from 800 to 300 (no noise) (ablation) | DenseNet · CIFAR10 clean | error rate (%): 6.97 -> 11.71 | Hurt |
| reduce PCA dimension from 800 to 300 (no noise) (ablation) | DenseNet · CIFAR10 C&W l2 | mean perturbation norm: 0.160 -> 0.256 | Improved |
| use noise800 instead of TRADES adversarial training | DenseNet · CIFAR10 clean | error rate (%): 21.22 -> 8.82 | Improved |
| use noise800 instead of TRADES adversarial training | DenseNet · CIFAR10 C&W l2 | mean perturbation norm: 0.664 -> 0.208 | Hurt |
| add PCA-projection preprocessing to MMC loss (MMC-P-500) | MMC · CIFAR10 untargeted PGD10 eps=8 | robust accuracy (%): 25.6 -> 42.5 | Improved |
| add PCA-projection preprocessing to MMC loss (MMC-P-300) | MMC · CIFAR10 untargeted PGD50 eps=8 | robust accuracy (%): 9.2 -> 29.2 | Improved |
| add PCA-projection preprocessing to MMC loss (MMC-P-300) | MMC · CIFAR10 clean | accuracy (%): 92.5 -> 87.9 | Hurt |
Closest papers by meaning
- Dual Manifold Adversarial Robustness: Defense against Lp and non-Lp Adversarial Attacks (NeurIPS 2020 · similarity 0.90)
- Adversarial Purification with the Manifold Hypothesis (AAAI 2024 · similarity 0.89)
- On the Need for Topology-Aware Generative Models for Manifold-Based Defenses (ICLR 2020 · similarity 0.89)
- Defense Against Adversarial Images Using Web-Scale Nearest-Neighbor Search (CVPR 2019 · similarity 0.89)
- The Limitations of Adversarial Training and the Blind-Spot Attack (ICLR 2019 · similarity 0.88)
- Enhancing the Adversarial Robustness via Manifold Projection (AAAI 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.
