Research map / Adversarial robustness and security / Certified and empirical robustness
Boundary thickness and robustness in learning models
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 |
|---|---|---|---|
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR10 | PGD-20 8-pixel accuracy (%): 2.0 -> 11.7 | Helped |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR10 | black-box attack accuracy (%): 46.3 -> 78.0 | Helped |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR10 | OOD accuracy (%): 78.5 -> 83.6 | Helped |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR10 | clean accuracy (%): 96.0 -> 94.4 | Hurt |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR100 | clean accuracy (%): 78.3 -> 72.2 | Hurt |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR100 | OOD accuracy (%): 51.3 -> 52.5 | Helped |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR100 | black-box attack accuracy (%): 37.3 -> 60.1 | Helped |
| mix random noise labeled NONE class into mixup (noisy mixup) | mixup, ResNet-18 · CIFAR100 | PGD-20 8-pixel accuracy (%): 0.0 -> 1.5 | Helped |
| use mixup instead of standard training (thicker boundary) | ResNet/VGG/DenseNet · CIFAR10-C | OOD robustness vs boundary thickness | Helped |
| add L2 regularization, L1 regularization, large learning rate, early stopping or cutout (ablation) | adversarial training ResNet-18 · CIFAR10 | boundary thickness and robust generalization gap | Helped |
| learning rate decay (ablation) | standard training · CIFAR10 | boundary thickness | Hurt |
| measure thickness instead of margin as predictor of robustness level | adversarially trained models · CIFAR10 | differentiating robustness levels | Helped |
Closest papers by meaning
- Hold me tight! Influence of discriminative features on deep network boundaries (NeurIPS 2020 · similarity 0.89)
- Exploring and Exploiting Decision Boundary Dynamics for Adversarial Robustness (ICLR 2023 · similarity 0.88)
- Fantastic Robustness Measures: The Secrets of Robust Generalization (NeurIPS 2023 · similarity 0.88)
- Robust Models Are More Interpretable Because Attributions Look Normal (ICML 2022 · similarity 0.88)
- Reducing Excessive Margin to Achieve a Better Accuracy vs. Robustness Trade-off (ICLR 2022 · similarity 0.88)
- Robustness via Curvature Regularization, and Vice Versa (CVPR 2019 · 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.