Research map / Adversarial robustness and security / Attacks on text and graphs
Effective Targeted Attacks for Adversarial Self-Supervised Learning
Minseon Kim, Hyeonjeong Ha, Sooel Son, Sung Ju Hwang
Read the paperCheck your idea against 137,217 papers
AI-extracted. The measured effects below were read by AI from the paper's full text (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 |
|---|---|---|---|
| replace untargeted attack with TARO targeted attack | BYOL · CIFAR-10 linear eval | PGD robust accuracy: 16.20 -> 31.20 | Improved |
| replace untargeted attack with TARO targeted attack | SimSiam · CIFAR-10 linear eval | PGD robust accuracy: 32.28 -> 44.71 | Improved |
| replace untargeted attack with TARO targeted attack | SimSiam · CIFAR-10 linear eval | clean accuracy: 71.78 -> 74.87 | Improved |
| replace untargeted attack with TARO targeted attack | BYOL · CIFAR-10 linear eval | AutoAttack robust accuracy: 0.01 -> 22.01 | Improved |
| add TARO to contrastive adversarial SSL | RoCL · CIFAR-10 linear eval | PGD robust accuracy: 42.89 -> 45.37 | Improved |
| add TARO to contrastive adversarial SSL | ACL · CIFAR-10 linear eval | PGD / AutoAttack accuracy: 39.37 -> 39.71 PGD; 35.97 -> 35.81 AutoAttack | No clear effect |
| add TARO on larger dataset | SimSiam* · CIFAR-100 linear eval | PGD robust accuracy: 13.34 -> 22.18 | Improved |
| add TARO on larger dataset | RoCL · CIFAR-100 linear eval | PGD robust accuracy: 17.17 -> 18.91 | Improved |
| add TARO for transfer from CIFAR-100 to CIFAR-10 | RoCL · CIFAR-10 transfer | clean accuracy: 73.93 -> 65.21 | Hurt |
| use random in-batch target instead of untargeted attack (ablation) | SimSiam · CIFAR-10 | PGD robust accuracy: 32.28 -> 42.85 | Improved |
| use score-selected target instead of random target (ablation) | SimSiam · CIFAR-10 | PGD robust accuracy: 42.85 -> 44.71 | Improved |
| use score-selected target instead of random target (ablation) | RoCL · CIFAR-10 | PGD robust accuracy: 43.45 -> 45.37 | Improved |
Closest papers by meaning
- Adversarial Self-Supervised Contrastive Learning (NeurIPS 2020 · similarity 0.89)
- Unlabeled Data Improves Adversarial Robustness (NeurIPS 2019 · similarity 0.89)
- 3SAT: A Simple Self-Supervised Adversarial Training Framework (AAAI 2025 · similarity 0.88)
- Are Adversarial Examples Created Equal? A Learnable Weighted Minimax Risk for Robustness under Non-uniform Attacks (AAAI 2021 · similarity 0.88)
- Self-Supervised Adversarial Training via Diverse Augmented Queries and Self-Supervised Double Perturbation (NeurIPS 2024 · similarity 0.87)
- Adversarial Training Can Provably Improve Robustness: Theoretical Analysis of Feature Learning Process Under Structured Data (ICLR 2025 · 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.
