Research map / Adversarial robustness and security / Certified and empirical robustness
Large Norms of CNN Layers Do Not Hurt Adversarial Robustness
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 |
|---|---|---|---|
| Apply ℓ1 norm decay with β=10^-3 | VGG-11 · CIFAR-10 test | clean accuracy: 90.4 -> 91.1 | Helped |
| Apply ℓ1 norm decay with β=10^-3 | VGG-11 · CIFAR-10, Auto Attack ℓ∞ ε=1/255 | robust accuracy: 60.2 -> 62.5 | Helped |
| Apply ℓ1 norm decay with β=10^-3 | ResNet-18 · CIFAR-10 test | clean accuracy: 93.2 -> 93.4 | Helped |
| Apply ℓ1 norm decay with β=10^-3 | ResNet-18 · CIFAR-10, Auto Attack ℓ∞ ε=1/255 | robust accuracy: 37.0 -> 37.7 | Helped |
| Apply strong norm regularization, such as ℓ1 norm decay with β=10^-2 | ResNet-18 · CIFAR-10, Auto Attack ℓ∞ ε=1/255 | robust accuracy: 37.0 -> 24.5 | Hurt |
| Use stronger norm regularization | CNNs across VGG-11, ResNet-18, SENet-18, and RegNetX-200MF · CIFAR-10, Auto Attack ℓ∞ ε=1/255 | robust accuracy | Mixed |
| Use norm regularization (weight decay, singular value clipping, or norm decay) | CNNs across four architectures · CIFAR-10 | robust accuracy | Mixed |
| Use norm regularization | CNNs across four architectures · CIFAR-10 test | clean accuracy | Helped |
| Adversarially train the network | VGG-11, ResNet-18, SENet-18, and RegNetX-200MF · CIFAR-10 | layer ℓ∞ norms | Mixed |
Closest papers by meaning
- Improved techniques for deterministic l2 robustness (NeurIPS 2022 · similarity 0.91)
- Improved deterministic l2 robustness on CIFAR-10 and CIFAR-100 (ICLR 2022 · similarity 0.91)
- Adversarial Robustness against Multiple and Single $l_p$-Threat Models via Quick Fine-Tuning of Robust Classifiers (ICML 2022 · similarity 0.90)
- Robust Design of Deep Neural Networks Against Adversarial Attacks Based on Lyapunov Theory (CVPR 2020 · similarity 0.89)
- Adversarial Robustness through Local Linearization (NeurIPS 2019 · similarity 0.89)
- Theoretical Analysis of Robust Overfitting for Wide DNNs: An NTK Approach (ICLR 2024 · similarity 0.88)
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.