Research map / Adversarial robustness and security / Black-box and transfer attacks
The LogBarrier Adversarial Attack: Making Effective Use of Decision Boundary Information
Chris Finlay, Aram-Alexandre Pooladian, Adam Oberman
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 |
|---|---|---|---|
| use LogBarrier instead of PGD (l2) | l2 attack · MNIST | percent misclassified: 52.58 -> 99.10 | Improved |
| use LogBarrier instead of CW (l2) | l2 attack · CIFAR10 ResNeXt34 | percent misclassified: 90.40 -> 99.90 | Improved |
| use LogBarrier instead of CW (l2) | l2 attack · ImageNet-1K | percent misclassified: 74.86 -> 98.40 | Improved |
| use LogBarrier instead of PGD (l2) | l2 attack · ImageNet-1K | percent misclassified: 90.00 -> 98.40 | Hurt |
| use LogBarrier instead of IFGSM (linf) | linf attack · MNIST | percent misclassified: 73.40 -> 94.80 | Improved |
| use LogBarrier instead of IFGSM (linf) | linf attack · CIFAR10 ResNeXt34 | percent misclassified: 75.80 -> 98.70 | Improved |
| use LogBarrier instead of IFGSM (linf) | linf attack · ImageNet-1K | percent misclassified: 99.60 -> 95.20 | Hurt |
| use LogBarrier instead of IFGSM against softmax-temperature obfuscation (MNIST) | linf attack · MNIST obfuscated (T = 2) | percent misclassified at linf 0.3: 32.90 -> 99.80 | Improved |
| use LogBarrier instead of IFGSM against softmax-temperature obfuscation (CIFAR10) | linf attack · CIFAR10 obfuscated (T = 20) | percent misclassified at linf 8/255: 43.90 -> 54.40 | Improved |
| use LogBarrier instead of IFGSM against adversarial training (MNIST) | linf attack · MNIST adversarially trained | linf distance to perturb 90% of images: 6.51E-1 -> 2.85E-1 | Improved |
| add multi-step adversarial training to the MNIST model (ablation) | MNIST net · MNIST, LogBarrier attack | linf distance to perturb 90% of images: 2.27E-1 -> 2.85E-1 | No clear effect |
| use LogBarrier instead of IFGSM on adversarially trained model (CIFAR10) | linf attack · CIFAR10 adversarial training | percent misclassified at linf 4/255: 31.60 -> 23.40 | Hurt |
Closest papers by meaning
- With Friends Like These, Who Needs Adversaries? (NeurIPS 2018 · similarity 0.88)
- Guessing Smart: Biased Sampling for Efficient Black-Box Adversarial Attacks (ICCV 2019 · similarity 0.88)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.87)
- PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples (ICLR 2018 · similarity 0.87)
- Robust Feature-Level Adversaries are Interpretability Tools (NeurIPS 2022 · similarity 0.87)
- ADBA: Approximation Decision Boundary Approach for Black-Box Adversarial Attacks (AAAI 2025 · similarity 0.86)
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.
