Research map / Adversarial robustness and security / Certified and empirical robustness
Robustness via Curvature Regularization, and Vice Versa
Check 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 |
|---|---|---|---|
| Add CURE curvature regularization | ResNet-18 · CIFAR-10 | adversarial accuracy (PGD(20), ε=8): 0.0% -> 36.3% | Helped |
| Add CURE curvature regularization | WideResNet-28×10 · CIFAR-10 | adversarial accuracy (PGD(20), ε=8): 0.0% -> 41.4% | Helped |
| Use CURE instead of adversarial training | ResNet-18 · CIFAR-10 | adversarial accuracy (PGD(20), ε=8): 43.7% -> 36.3% | Hurt |
| Use CURE instead of adversarial training | WideResNet-28×10 · CIFAR-10 | adversarial accuracy (PGD(20), ε=8): 45.8% -> 41.4% | Hurt |
| Add CURE curvature regularization | ResNet-18 · SVHN | adversarial accuracy (PGD(10), ε=12): 0.9% -> 28.4% | Helped |
| Use CURE instead of adversarial training | ResNet-18 · SVHN | adversarial accuracy (PGD(10), ε=12): 33% -> 28.4% | Hurt |
| Fine-tune on adversarial examples | ResNet-18 · CIFAR-10 | adversarial accuracy (PGD(20), ε=4): 0.2% -> 56.9% | Helped |
| Fine-tune on adversarial examples | ResNet-18 · CIFAR-10 | adversarial accuracy (FGSM, ε=4): 38.0% -> 61.0%; 0.2% -> 56.9% | Mixed |
| Increase attack strength from PGD(20) to PGD(40) or PGD(100) | CURE WideResNet-28×10 · CIFAR-10 | adversarial accuracy: 41.4% -> 41.1% | Hurt |
| Use SPSA instead of PGD(20) | CURE WideResNet-28×10 · CIFAR-10 | adversarial accuracy: 41.4% -> 44.5% | Mixed |
| Use the gradient direction instead of the sign of the gradient for the regularizer direction (ablation) | CURE · CIFAR-10 and SVHN | adversarial robustness | No clear effect |
Closest papers by meaning
- When Flatness Does (Not) Guarantee Adversarial Robustness (ICLR 2026 · similarity 0.91)
- Robustness of classifiers: from adversarial to random noise (NeurIPS 2016 · similarity 0.91)
- Efficient Training of Low-Curvature Neural Networks (NeurIPS 2022 · similarity 0.90)
- Compositional Curvature Bounds for Deep Neural Networks (ICML 2024 · similarity 0.90)
- Towards Adversarially Robust Dataset Distillation by Curvature Regularization (AAAI 2025 · similarity 0.88)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · 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.