Research map / Adversarial robustness and security / Certified and empirical robustness
Towards Defending Multiple $$\ell _p$$-Norm Bounded Adversarial Perturbations via Gated Batch Normalization
Aishan Liu, Shiyu Tang, Xinyun Chen, Lei Huang, Haotong Qin, Xianglong Liu, Dacheng Tao
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 |
|---|---|---|---|
| use Conv gate in first GBN and FC gates in other GBN blocks instead of Conv gate everywhere (ablation) | GBN VGG-16 · CIFAR-10 | PGD-l1 robust accuracy (%): 30.4 -> 59.6 | Improved |
| use Conv gate in first GBN and FC gates elsewhere instead of FC gate everywhere (ablation) | GBN VGG-16 · CIFAR-10 | PGD-l1 robust accuracy (%): 35.1 -> 59.6 | Improved |
| use Conv gate in first GBN and FC gates elsewhere instead of Conv gate everywhere (ablation) | GBN VGG-16 · CIFAR-10 | clean accuracy (%): 41.3 -> 83.6 | Improved |
| use Conv gate in first GBN and FC gates elsewhere instead of Conv gate everywhere (ablation) | GBN VGG-16 · CIFAR-10 | PGD-linf robust accuracy (%): 20.2 -> 58.1 | Improved |
| use soft-gated weighting instead of hard top-1 gate (ablation) | GBN ResNet-20 · CIFAR-10 | PGD-l1 robust accuracy (%): 57.8 -> 58.1 | Improved |
| use soft-gated weighting instead of hard top-1 gate (ablation) | GBN ResNet-20 · CIFAR-10 | clean accuracy (%): 79.8 -> 80.2 | Improved |
| replace vanilla BN with GBN | ResNet-20 · CIFAR-10 | PGD-l1 robust accuracy (%): 0.1 -> 58.1 | Improved |
| replace vanilla BN with GBN | ResNet-20 · CIFAR-10 | clean accuracy (%): 89.4 -> 80.2 | Hurt |
| replace vanilla BN with GBN | ResNet-34 · Tiny-ImageNet | clean accuracy (%): 54.0 -> 43.2 | Hurt |
| replace MBN (2-way gated multiple BN) with GBN | ResNet-34 · Tiny-ImageNet | all-attacks robust accuracy (%): 18.3 -> 37.7 | Improved |
| replace MBN with GBN | ResNet-20 · CIFAR-10 | all-attacks robust accuracy (%): 20.0 -> 48.2 | Improved |
| replace MSD with GBN | LeNet · MNIST | all-attacks robust accuracy (%): 38.1 -> 68.5 | Improved |
Closest papers by meaning
- Fast Minimum-norm Adversarial Attacks through Adaptive Norm Constraints (NeurIPS 2021 · similarity 0.86)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.85)
- Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations (ICML 2024 · similarity 0.83)
- Kernel Learning with Adversarial Features: Numerical Efficiency and Adaptive Regularization (NeurIPS 2025 · similarity 0.83)
- The Cost of Compression: Tight Quadratic Black-Box Attacks on Sketches for $\ell_2$ Norm Estimation (NeurIPS 2025 · similarity 0.83)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.82)
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.
