Research map / Adversarial robustness and security / Certified and empirical robustness
Beating Attackers At Their Own Games: Adversarial Example Detection Using Adversarial Gradient Directions
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 AGD similarities rather than transformation-based feature-distance scores | Adversarial example detection · CIFAR-10 and ImageNet, five attacks on average | AUC-ROC (%): 97.9% and 98.6% | Helped |
| Use the top 4 predicted classes instead of only the top 1 (ablation) | AGD detector · CIFAR-10 | AUC-ROC (%): 94.1 -> 99.8 | Helped |
| Use the top 4 predicted classes instead of only the top 1 (ablation) | AGD detector · ImageNet | AUC-ROC (%): 98.2 -> 99.5 | Helped |
| Increase random transformations from one to multiple (ablation) | AGD detector · CIFAR-10, FGSM detection | AUC-ROC (%): 99.6 -> 99.7 | No clear effect |
| Increase random transformations from one to four (ablation) | AGD detector · ImageNet, FGSM detection | AUC-ROC (%): 99.5 -> 99.7 | No clear effect |
| Combine all three AGD similarity scores alpha, beta, and gamma rather than using only one or two (ablation) | AGD detector · CIFAR-10 and ImageNet | Detection performance | Helped |
| Use AGD-based scores instead of the Rand-1 L1 score under an adaptive attack | ImageNet white-box detection · Adaptive PGD, LR=0.1 | Detection AUC: 53.0 -> 64.1 | Helped |
| Use AGD detector with K=4 rather than Rand-1 | ImageNet white-box defense · Adaptive PGD, LR=0.1 | Attack success rate: 79.4 -> 53.5 | Helped |
Closest papers by meaning
- Evading Adversarial Example Detection Defenses with Orthogonal Projected Gradient Descent (ICLR 2022 · similarity 0.91)
- GAT: Generative Adversarial Training for Adversarial Example Detection and Robust Classification (ICLR 2020 · similarity 0.90)
- Detecting Adversarial Examples Through Image Transformation (AAAI 2018 · similarity 0.90)
- ML-LOO: Detecting Adversarial Examples with Feature Attribution (AAAI 2020 · similarity 0.90)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.90)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · 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.