Research map / Adversarial robustness and security / Certified and empirical robustness
Enhancing Adversarial Training With Second-Order Statistics of Weights
Gaojie Jin, Xinping Yi, Wei Huang, Sven Schewe, Xiaowei Huang
Check 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 S²O to vanilla adversarial training | PreAct ResNet18 · CIFAR-10/100 and SVHN, ℓ∞ and ℓ₂ | PGD-20 accuracy: 52.77 -> 55.11; 28.02 -> 30.58; 60.91 -> 64.83; 68.83 -> 69.42; 42.20 -> 44.07; 66.76 -> 76.19 | Improved |
| Add S²O to TRADES | ResNet18 · CIFAR-10, ℓ∞ | PGD-20 accuracy: 53.81 -> 55.20 | Improved |
| Add S²O to TRADES+AWP | ResNet18 · CIFAR-10, ℓ∞ | PGD-20 accuracy: 56.18 -> 57.29 | Improved |
| Add S²O to TRADES+AWP | WideResNet · CIFAR-10, ℓ∞ | PGD-20 accuracy: 59.67 -> 61.12 | Improved |
| Add S²O to TRADES+AWP | WideResNet · CIFAR-10, ℓ∞ | Auto Attack accuracy: 56.2 -> 55.9 | Hurt |
| Add S²O to AVMixup | WideResNet · CIFAR-10, ℓ∞ | PGD-20 accuracy: 59.75 -> 60.43 | Improved |
| Add S²O to AVMixup | WideResNet · CIFAR-10, ℓ∞ | Auto Attack accuracy: 39.7 -> 39.3 | Hurt |
| Add S²O to MART | WideResNet · CIFAR-10, ℓ∞ | PGD-20 accuracy: 58.31 -> 59.29 | Improved |
| Add S²O to TRADES+AWP | WideResNet · CIFAR-100, ℓ∞ | PGD-20 accuracy: 34.04 -> 35.95 | Improved |
| Add S²O to vanilla adversarial training | VGG16 · CIFAR-10, ℓ∞ | PGD-20 accuracy: 49.21 -> 50.53 | Improved |
| Add S²O to vanilla adversarial training | MobileNetV2 · CIFAR-10, ℓ∞ | PGD-20 accuracy: 50.76 -> 52.93 | Improved |
| Add S²O to vanilla adversarial training | PreAct ResNet18 · CIFAR-10, ℓ∞ black-box attack | PGD-20 accuracy: 62.63 -> 63.87 | Improved |
Closest papers by meaning
- S$^{2}$ 2O: Enhancing Adversarial Training With Second-Order Statistics of Weights (TPAMI 2025 · similarity 0.98)
- Randomized Adversarial Training via Taylor Expansion (CVPR 2023 · similarity 0.88)
- Improving Adversarial Robustness by Putting More Regularizations on Less Robust Samples (ICML 2023 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.87)
- Doubly Robust Instance-Reweighted Adversarial Training (ICLR 2024 · similarity 0.87)
- Stability and Generalization of Adversarial Training for Shallow Neural Networks with Smooth Activation (NeurIPS 2024 · 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.
