Research map / Adversarial robustness and security / Certified and empirical robustness
Adversarial Attacks Are Reversible With Natural Supervision
Chengzhi Mao, Mia Chiquier, Hao Wang, Junfeng Yang, Carl Vondrick
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 |
|---|---|---|---|
| add self-supervised reverse attack at inference | Semi-SL · CIFAR-10 AutoAttack | robust accuracy: 60.22% -> 67.79% | Improved |
| add self-supervised reverse attack at inference | RO PreRes-18 · CIFAR-10 AutoAttack | robust accuracy: 48.95% -> 58.20% | Improved |
| add self-supervised reverse attack at inference | TRADES · CIFAR-10 PGD 50 | robust accuracy: 55.05% -> 57.00% | Improved |
| add self-supervised reverse attack at inference | AWP · CIFAR-10 BIM 200 | robust accuracy: 63.64% -> 63.69% | Improved |
| add self-supervised reverse attack at inference | RO ResNet18 · CIFAR-100 AutoAttack | robust accuracy: 19.23% -> 25.45% | Improved |
| add self-supervised reverse attack at inference | Semi-SL · SVHN AutoAttack | robust accuracy: 53.61% -> 65.50% | Improved |
| add self-supervised reverse attack at inference | FBF ResNet50 · ImageNet PGD10 | robust accuracy: 28.68% -> 31.36% | Improved |
| add self-supervised reverse attack at inference | MART · CIFAR-10 L2 C&W 200 | robust accuracy: 43.23% -> 45.85% | Improved |
| increase defense-aware attacker's lambda_s from 0 to 10 (ablation) | Ours vs Semi-SL/RO · CIFAR-10 PGD 50 adaptive attack | robust accuracy: lambda_s 0 to 10 | Mixed |
| use a random reverse perturbation instead of natural supervision (ablation) | reverse defense · CIFAR-10 clean vs robust accuracy | robust accuracy | Hurt |
| increase number of reverse-attack iterations (ablation) | Ours · CIFAR-10 accuracy vs time budget | robust accuracy | Improved |
| remove Gaussian blur from contrastive augmentations (ablation) | contrastive branch G · CIFAR | robust accuracy | Improved |
Closest papers by meaning
- A New Defense Against Adversarial Images: Turning a Weakness into a Strength (NeurIPS 2019 · similarity 0.90)
- One Man's Trash Is Another Man's Treasure: Resisting Adversarial Examples by Adversarial Examples (CVPR 2020 · similarity 0.90)
- PixelDefend: Leveraging Generative Models to Understand and Defend against Adversarial Examples (ICLR 2018 · similarity 0.89)
- Towards Deep Learning Models Resistant to Adversarial Attacks (ICLR 2018 · similarity 0.89)
- Countering Adversarial Images using Input Transformations (ICLR 2018 · similarity 0.89)
- Defense Against Adversarial Images Using Web-Scale Nearest-Neighbor Search (CVPR 2019 · similarity 0.89)
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.
