Enregistré dans:
| Auteurs principaux: | , , |
|---|---|
| Format: | Preprint |
| Publié: |
2024
|
| Sujets: | |
| Accès en ligne: | https://arxiv.org/abs/2402.04010 |
| Tags: |
Ajouter un tag
Pas de tags, Soyez le premier à ajouter un tag!
|
| _version_ | 1866916117092499456 |
|---|---|
| author | Wang, Yihan Zhu, Yifan Gao, Xiao-Shan |
| author_facet | Wang, Yihan Zhu, Yifan Gao, Xiao-Shan |
| contents | Availability attacks can prevent the unauthorized use of private data and commercial datasets by generating imperceptible noise and making unlearnable examples before release. Ideally, the obtained unlearnability prevents algorithms from training usable models. When supervised learning (SL) algorithms have failed, a malicious data collector possibly resorts to contrastive learning (CL) algorithms to bypass the protection. Through evaluation, we have found that most of the existing methods are unable to achieve both supervised and contrastive unlearnability, which poses risks to data protection. Different from recent methods based on contrastive error minimization, we employ contrastive-like data augmentations in supervised error minimization or maximization frameworks to obtain attacks effective for both SL and CL. Our proposed AUE and AAP attacks achieve state-of-the-art worst-case unlearnability across SL and CL algorithms with less computation consumption, showcasing prospects in real-world applications. |
| format | Preprint |
| id |
arxiv_https___arxiv_org_abs_2402_04010 |
| institution | arXiv |
| publishDate | 2024 |
| record_format | arxiv |
| spellingShingle | Efficient Availability Attacks against Supervised and Contrastive Learning Simultaneously Wang, Yihan Zhu, Yifan Gao, Xiao-Shan Machine Learning Availability attacks can prevent the unauthorized use of private data and commercial datasets by generating imperceptible noise and making unlearnable examples before release. Ideally, the obtained unlearnability prevents algorithms from training usable models. When supervised learning (SL) algorithms have failed, a malicious data collector possibly resorts to contrastive learning (CL) algorithms to bypass the protection. Through evaluation, we have found that most of the existing methods are unable to achieve both supervised and contrastive unlearnability, which poses risks to data protection. Different from recent methods based on contrastive error minimization, we employ contrastive-like data augmentations in supervised error minimization or maximization frameworks to obtain attacks effective for both SL and CL. Our proposed AUE and AAP attacks achieve state-of-the-art worst-case unlearnability across SL and CL algorithms with less computation consumption, showcasing prospects in real-world applications. |
| title | Efficient Availability Attacks against Supervised and Contrastive Learning Simultaneously |
| topic | Machine Learning |
| url | https://arxiv.org/abs/2402.04010 |