Research map / Adversarial robustness and security / Certified and empirical robustness
Relating Adversarially Robust Generalization to Flat Minima
David Stutz, Matthias Hein, Bernt Schiele
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 adversarial weight perturbations (AT-AWP) | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 54.3 | Improved |
| add additional unlabeled examples | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 48.9 | Improved |
| use TRADES instead of standard AT | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 56.7 | Improved |
| add AutoAugment | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 54.0 | Improved |
| add label noise | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 56.2 | Improved |
| add weight clipping | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 56.5 | Improved |
| add Entropy-SGD | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 58.6 | Improved |
| add label smoothing | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 61.2 | No clear effect |
| replace ReLU with SiLU | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 60.0 | No clear effect |
| increase epsilon to 9/255 | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 60.9 | No clear effect |
| add early stopping | AT baseline · CIFAR10 (ResNet-18) | robust test error (%): 62.8 -> 54.6 | Improved |
| add early stopping | TRADES · CIFAR10 (ResNet-18) | robust test error (%): 56.7 -> 53.4 | Improved |
Closest papers by meaning
- Fantastic Robustness Measures: The Secrets of Robust Generalization (NeurIPS 2023 · similarity 0.94)
- When Flatness Does (Not) Guarantee Adversarial Robustness (ICLR 2026 · similarity 0.92)
- Adversarial Weight Perturbation Helps Robust Generalization (NeurIPS 2020 · similarity 0.92)
- Disentangling Adversarial Robustness and Generalization (CVPR 2019 · similarity 0.90)
- A Flat Minima Perspective on Understanding Augmentations and Model Robustness (AAAI 2026 · similarity 0.90)
- Explaining the role of Intrinsic Dimensionality in Adversarial Training (ICML 2025 · similarity 0.89)
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.
