Research map / Adversarial robustness and security / Certified and empirical robustness
Minimally distorted Adversarial Examples with a Fast Adaptive Boundary Attack
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 FAB restarts from 10 to 100 (ablation) | FAB · MNIST + CIFAR-10, l∞ | avg. robust accuracy: 46.18 -> 45.47 | Helped |
| Increase FAB restarts from 10 to 100 (ablation) | FAB · MNIST + CIFAR-10, l2 | avg. robust accuracy: 36.41 -> 35.57 | Helped |
| Increase FAB restarts from 10 to 100 (ablation) | FAB · MNIST + CIFAR-10, l1 | avg. robust accuracy: 33.26 -> 29.46 | Helped |
| Add bias toward the original input in the gradient step (ablation) | FAB · Ablation comparison with DeepFool | robust accuracy | Helped |
| Add a backward step toward the original input after finding an adversarial point (ablation) | FAB · Ablation comparison with DeepFool | robust accuracy | Helped |
| Add random restarts (ablation) | FAB · Ablation comparison with DeepFool | robust accuracy | Helped |
| Rescale the classifier by 10^6 while using PGD-CE with step size ε/10 (ablation) | PGD-CE · CIFAR-10, ResNet-110, ε=8/255 | robust accuracy: 90.2% -> 12.9% | Helped |
| Rescale the classifier by 10^6 while using PGD-CE with step size ε/2 (ablation) | PGD-CE · CIFAR-10, ResNet-110, ε=8/255 | robust accuracy: 90.2% -> 2.5% | Helped |
| Use FAB, which is invariant to classifier rescaling | FAB · CIFAR-10, ResNet-110, ε=8/255 | robust accuracy: 90.2% -> 0.3% | Helped |
| Use targeted FAB with top-ranked target classes instead of normal FAB | FAB · CIFAR-10, CIFAR-100, and ImageNet | robust accuracy | Helped |
| Use targeted FAB with top-ranked target classes instead of normal FAB | FAB · MNIST | robust accuracy | Hurt |
Closest papers by meaning
- Fast Minimum-norm Adversarial Attacks through Adaptive Norm Constraints (NeurIPS 2021 · similarity 0.91)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.89)
- Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations (ICML 2024 · similarity 0.89)
- Error Correcting Output Codes Improve Probability Estimation and Adversarial Robustness of Deep Neural Networks (NeurIPS 2019 · similarity 0.88)
- Mind the Box: $l_1$-APGD for Sparse Adversarial Attacks on Image Classifiers (ICML 2021 · similarity 0.88)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 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.