Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Robustness Against the Union of Multiple Perturbation 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 |
|---|---|---|---|
| Use MSD instead of Max training | 4-layer CNN · MNIST, union of ℓ∞, ℓ2, and ℓ1 attacks | adversarial accuracy: 42.1% -> 58.4% | Helped |
| Use MSD instead of Avg training | 4-layer CNN · MNIST, union of ℓ∞, ℓ2, and ℓ1 attacks | adversarial accuracy: 34.9% -> 58.4% | Helped |
| Use MSD instead of Avg training | pre-activation ResNet18 · CIFAR10, union of ℓ∞, ℓ2, and ℓ1 attacks | adversarial accuracy: 40.6% -> 47.0% | Helped |
| Use MSD instead of Max training | pre-activation ResNet18 · CIFAR10, union of ℓ∞, ℓ2, and ℓ1 attacks | adversarial accuracy: 34.9% -> 47.0% | Helped |
| Remove the ℓ2 adversary from MSD training (ablation) | MSD · MNIST | PGD adversarial accuracy: drops by 1% on ℓ1 and ℓ∞; drops by 2% on ℓ2 | Hurt |
| Randomly vary the number of ℓ1 coordinates updated (ablation) | MSD ℓ1 steepest-descent attack · Adversarial attack experiments | attack success rate | Helped |
Closest papers by meaning
- Adversarial Training and Robustness for Multiple Perturbations (NeurIPS 2019 · similarity 0.94)
- Toward Efficient Robust Training against Union of $\ell_p$ Threat Models (NeurIPS 2022 · similarity 0.91)
- Accurate, reliable and fast robustness evaluation (NeurIPS 2019 · similarity 0.91)
- RAMP: Boosting Adversarial Robustness Against Multiple $l_p$ Perturbations for Universal Robustness (NeurIPS 2024 · similarity 0.90)
- Towards Million-Scale Adversarial Robustness Evaluation With Stronger Individual Attacks (CVPR 2025 · similarity 0.89)
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 2024 · similarity 0.89)
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.