Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarially Robust Models may not Transfer Better: Sufficient Conditions for Domain Transferability from the View of Regularization
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 |
|---|---|---|---|
| Increase feature-extractor regularization with Jacobian regularization or weight decay | ResNet-18 · CIFAR-10 to SVHN and ImageNet to CIFAR-10 | relative domain transfer accuracy | Helped |
| Increase feature-extractor regularization with Jacobian regularization or weight decay | ResNet-18 · CIFAR-10 and ImageNet source domains | adversarial accuracy | Hurt |
| Increase the last-layer norm using last-layer regularization or orthogonal training | ResNet-18 · CIFAR-10 to SVHN and ImageNet to CIFAR-10 | relative domain transfer accuracy | Helped |
| Increase the last-layer norm using last-layer regularization or orthogonal training | ResNet-18 · CIFAR-10 and ImageNet source domains | adversarial accuracy: around $-0.9$ | Hurt |
| Add Gaussian-noise augmentation | ResNet-18 · CIFAR-10 to SVHN and ImageNet to CIFAR-10 | relative domain transfer accuracy | Helped |
| Add Gaussian-noise augmentation | ResNet-18 · CIFAR-10 and ImageNet source domains | adversarial accuracy | Mixed |
| Add posterize augmentation | ResNet-18 · CIFAR-10 to SVHN and ImageNet to CIFAR-10 | relative domain transfer accuracy | Helped |
| Add rescaling or Gaussian-blur augmentation | ResNet-18 and WideResNet50 · ImageNet to CIFAR-10 | relative domain transfer accuracy | Helped |
| Add rotation augmentation | ResNet-18 · CIFAR-10 to SVHN and ImageNet to CIFAR-10 | relative domain transfer accuracy | No clear effect |
| Add translation augmentation | ResNet-18 · CIFAR-10 to SVHN and ImageNet to CIFAR-10 | relative domain transfer accuracy | No clear effect |
| Improve adversarial robustness without imposing additional conditions on the feature class | Models in the paper's theoretical analysis · Domain transferability | relative domain transfer loss | No clear effect |
| Shrink the feature-extractor function class through regularization | Training algorithms in the paper's theoretical analysis · Source-to-target domain transfer | relative domain transfer loss upper bound | Helped |
Closest papers by meaning
- Adversarially robust transfer learning (ICLR 2020 · similarity 0.90)
- Do Adversarially Robust ImageNet Models Transfer Better? (NeurIPS 2020 · similarity 0.90)
- Adversarially Robust Hypothesis Transfer Learning (ICML 2024 · similarity 0.88)
- Sharpness-Aware Model-Agnostic Long-Tailed Domain Generalization (AAAI 2024 · similarity 0.88)
- Initialization Matters for Adversarial Transfer Learning (CVPR 2024 · similarity 0.88)
- Transferability Bound Theory: Exploring Relationship between Adversarial Transferability and Flatness (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.