Fortify the Guardian, Not the Treasure: Resilient Adversarial Detectors
Fuente:
arXiv
Gespeichert in:
| Hauptverfasser: | , , |
|---|---|
| Format: | Preprint |
| Veröffentlicht: |
2024
|
| Schlagworte: | |
| Online-Zugang: | |
| Tags: |
Tag hinzufügen
Keine Tags, Fügen Sie den ersten Tag hinzu!
|
| _version_ | 1866914851869163520 |
|---|---|
| author | Lapid, Raz Dubin, Almog Sipper, Moshe |
| author_facet | Lapid, Raz Dubin, Almog Sipper, Moshe |
| contents | This paper presents RADAR-Robust Adversarial Detection via Adversarial Retraining-an approach designed to enhance the robustness of adversarial detectors against adaptive attacks, while maintaining classifier performance. An adaptive attack is one where the attacker is aware of the defenses and adapts their strategy accordingly. Our proposed method leverages adversarial training to reinforce the ability to detect attacks, without compromising clean accuracy. During the training phase, we integrate into the dataset adversarial examples, which were optimized to fool both the classifier and the adversarial detector, enabling the adversarial detector to learn and adapt to potential attack scenarios. Experimental evaluations on the CIFAR-10 and SVHN datasets demonstrate that our proposed algorithm significantly improves a detector's ability to accurately identify adaptive adversarial attacks -- without sacrificing clean accuracy. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2404_12120 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Fortify the Guardian, Not the Treasure: Resilient Adversarial Detectors Lapid, Raz Dubin, Almog Sipper, Moshe Computer Vision and Pattern Recognition Artificial Intelligence This paper presents RADAR-Robust Adversarial Detection via Adversarial Retraining-an approach designed to enhance the robustness of adversarial detectors against adaptive attacks, while maintaining classifier performance. An adaptive attack is one where the attacker is aware of the defenses and adapts their strategy accordingly. Our proposed method leverages adversarial training to reinforce the ability to detect attacks, without compromising clean accuracy. During the training phase, we integrate into the dataset adversarial examples, which were optimized to fool both the classifier and the adversarial detector, enabling the adversarial detector to learn and adapt to potential attack scenarios. Experimental evaluations on the CIFAR-10 and SVHN datasets demonstrate that our proposed algorithm significantly improves a detector's ability to accurately identify adaptive adversarial attacks -- without sacrificing clean accuracy. |
| title | Fortify the Guardian, Not the Treasure: Resilient Adversarial Detectors |
| topic | Computer Vision and Pattern Recognition Artificial Intelligence |
| url | https://arxiv.org/abs/2404.12120 |