Research map / Adversarial robustness and security / Certified and empirical robustness
S$^{2}$ 2O: Enhancing Adversarial Training With Second-Order Statistics of Weights
Gaojie Jin, Xinping Yi, Wei Huang, Sven Schewe, Xiaowei Huang
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 S2O to adversarial training | AT PreAct ResNet-18 · CIFAR-10 l_inf | PGD-20 accuracy: 52.77 -> 55.11 | Improved |
| add S2O to adversarial training | AT PreAct ResNet-18 · CIFAR-100 l_inf | PGD-20 accuracy: 28.02 -> 30.58 | Improved |
| add S2O to adversarial training | AT PreAct ResNet-18 · SVHN l_inf | PGD-20 accuracy: 60.91 -> 64.83 | Improved |
| add S2O to adversarial training under l2 | AT PreAct ResNet-18 · SVHN l_2 | PGD accuracy: 66.76 -> 76.19 | Improved |
| add S2O to adversarial training | AT PreAct ResNet-18 · CIFAR-10 | training time per epoch: 309s -> 368s | Hurt |
| compute correlation matrices from adversarial data only (ablation) | AT+S2O · CIFAR-10 l_inf ResNet-18 | clean accuracy: 83.65 -> 82.89 | Hurt |
| compute correlation matrices from adversarial data only (ablation) | AT+S2O · CIFAR-10 l_inf ResNet-18 | training time per epoch: 368s -> 340s | Improved |
| add S2O to TRADES | TRADES · CIFAR-10 ResNet-18 | AutoAttack accuracy: 48.6 -> 49.5 | Improved |
| add S2O to TRADES+AWP | TRADES+AWP · CIFAR-10 ResNet-18 | AutoAttack accuracy: 51.7 -> 52.4 | Improved |
| add S2O to TRADES+AWP | TRADES+AWP · CIFAR-10 WRN | AutoAttack accuracy: 56.2 -> 55.9 | Hurt |
| add S2O to TRADES | TRADES · CIFAR-100 WRN | clean accuracy: 60.38 -> 63.40 | Improved |
| add S2O to DDPM-data adversarial training | Pang et al. (2022) · CIFAR-10 WRN-28-10 | AutoAttack accuracy: 61.04 -> 61.45 | Improved |
Closest papers by meaning
- Enhancing Adversarial Training With Second-Order Statistics of Weights (CVPR 2022 · similarity 0.98)
- Weight-covariance alignment for adversarially robust neural networks (ICML 2021 · 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)
- Exploring Adversarial Robustness of Deep State Space Models (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.
