Research map / Adversarial robustness and security / Certified and empirical robustness
Functional Adversarial Attacks
Cassidy Laidlaw, Soheil Feizi
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 CIELUV rather than RGB color space | ReColorAdv · CIFAR-10 | accuracy: 5.9 -> 3.0 | Improved |
| Combine ReColorAdv, StAdv, and delta attacks | Attacks on CIFAR-10 classifiers · CIFAR-10, undefended ResNet-32 | accuracy: 3.0 -> 0.0 | Improved |
| Combine ReColorAdv, StAdv, and delta attacks | Attacks on TRADES · CIFAR-10 | accuracy: 59.2 -> 5.7 | Improved |
| Combine ReColorAdv, StAdv, and delta attacks | Adversarially trained ResNet-32 · CIFAR-10 | accuracy: 3.6 | Improved |
| Convert input images to grayscale (ablation) | CIFAR-10 classification · CIFAR-10 | accuracy: 92.2 -> 88.3 | Hurt |
| Reduce PGD iterations from 300 to 100 (ablation) | ReColorAdv · CIFAR-10, undefended ResNet-32 | accuracy: 3.0 -> 4.4 | Hurt |
| Reduce PGD iterations from 300 to 100 (ablation) | ReColorAdv + StAdv + delta · CIFAR-10, model trained against the same attack combination | accuracy: 3.6 -> 8.8 | Hurt |
| Apply separate perturbation functions to each CIELUV channel (ablation) | ReColorAdv · CIFAR-10, undefended ResNet-32 | accuracy: 4.4% -> 8.4% | Hurt |
| Apply separate perturbation functions to each RGB channel (ablation) | ReColorAdv · CIFAR-10, undefended ResNet-32 | accuracy: 8.2% -> 9.3% | Hurt |
| Use channel-specific RGB perturbation bounds (ablation) | ReColorAdv · CIFAR-10, undefended ResNet-32 | accuracy: 8.2% -> 8.4% | Hurt |
| Use YPbPr rather than CIELUV color space (ablation) | ReColorAdv · CIFAR-10, undefended ResNet-32 | accuracy: 4.4% -> 2.6% | Improved |
| Use HSV' rather than CIELUV color space (ablation) | ReColorAdv · CIFAR-10, undefended ResNet-32 | accuracy: 4.4% -> 2.1% | Improved |
Closest papers by meaning
- PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples (ICLR 2018 · similarity 0.87)
- Sparse and Imperceivable Adversarial Attacks (ICCV 2019 · similarity 0.86)
- ColorFool: Semantic Adversarial Colorization (CVPR 2020 · similarity 0.85)
- Beyond ImageNet Attack: Towards Crafting Adversarial Examples for Black-box Domains (ICLR 2022 · similarity 0.85)
- Advancing Adversarial Robustness in GNeRFs: The IL2-NeRF Attack (CVPR 2025 · similarity 0.85)
- Content-based Unrestricted Adversarial Attack (NeurIPS 2023 · similarity 0.85)
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.
