Research map / Adversarial robustness and security / Certified and empirical robustness
Can Adversarial Training Be Manipulated By Non-Robust Features?
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 |
|---|---|---|---|
| Apply hypocritical rather than adversarial perturbations to the training data | PGD-AT with defense budget 8/255 · CIFAR-10 | AutoAttack accuracy: 46.13 -> 35.44 | Hurt |
| Apply hypocritical rather than adversarial perturbations to the training data | PGD-AT with defense budget 8/255 · SVHN | AutoAttack accuracy: 42.48 -> 35.43 | Hurt |
| Apply hypocritical rather than adversarial perturbations to the training data | PGD-AT with defense budget 8/255 · CIFAR-100 | AutoAttack accuracy: 22.71 -> 18.74 | Hurt |
| Apply hypocritical rather than adversarial perturbations to the training data | PGD-AT with defense budget 8/255 · Tiny-ImageNet | AutoAttack accuracy: 16.83 -> 12.53 | Hurt |
| Increase the defense budget from 8/255 to 14/255 (ablation) | PGD-AT on hypocritically perturbed CIFAR-10 · CIFAR-10 | PGD-20 accuracy: 37.61 -> 52.92 | Helped |
| Use MART with defense budget 14/255 instead of PGD-AT with 8/255 | Adversarial training on hypocritically perturbed CIFAR-10 · CIFAR-10 | PGD-20 accuracy: 37.61 -> 53.82 | Helped |
| Use TRADES with defense budget 12/255 instead of PGD-AT with 8/255 | Adversarial training on hypocritically perturbed CIFAR-10 · CIFAR-10 | PGD-20 accuracy: 37.61 -> 51.77 | Helped |
| Add Gaussian Smoothing to PGD-AT (ablation) | PGD-AT with defense budget 8/255 on hypocritically perturbed data · CIFAR-10 | PGD-20 accuracy: 37.61 -> 42.80 | Helped |
| Add Cutout to PGD-AT (ablation) | PGD-AT with defense budget 8/255 on hypocritically perturbed data · CIFAR-10 | PGD-20 accuracy: 37.61 -> 39.77 | Helped |
| Add Random Noise to PGD-AT (ablation) | PGD-AT with defense budget 8/255 on hypocritically perturbed data · CIFAR-10 | PGD-20 accuracy: 37.61 -> 38.35 | Helped |
| Add AutoAugment to PGD-AT (ablation) | PGD-AT with defense budget 8/255 on hypocritically perturbed data · CIFAR-10 | PGD-20 accuracy: 37.61 -> 40.19 | Helped |
| Use hypocritical rather than adversarial perturbations (ablation) | PGD-AT trained with a clean/perturbed data mix · CIFAR-10, clean proportion 0.4 | PGD-20 accuracy: 49.68 -> 46.98 | Hurt |
Closest papers by meaning
- Are Adversarial Examples Created Equal? A Learnable Weighted Minimax Risk for Robustness under Non-uniform Attacks (AAAI 2021 · similarity 0.91)
- Adversarial Training and Robustness for Multiple Perturbations (NeurIPS 2019 · similarity 0.91)
- On the Vulnerability of Adversarially Trained Models Against Two-faced Attacks (ICLR 2024 · similarity 0.91)
- Better Safe Than Sorry: Preventing Delusive Adversaries with Adversarial Training (NeurIPS 2021 · similarity 0.90)
- Certified Adversarial Robustness with Additive Noise (NeurIPS 2019 · similarity 0.90)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · similarity 0.90)
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.