Research map / Adversarial robustness and security / Certified and empirical robustness
Pre-trained Adversarial Perturbations
Yuanhao Ban, Yinpeng Dong
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 uniform Gaussian sampling (UGS) loss (ablation) | L4A_base · SimCLRv2 ResNet101 | attack success rate AVG (%): 66.95 -> 72.20 | Improved |
| add uniform Gaussian sampling (UGS) loss (ablation) | L4A_base · SimCLRv2 ResNet50 | attack success rate AVG (%): 71.16 -> 77.80 | Improved |
| add uniform Gaussian sampling (UGS) loss (ablation) | L4A_base · MAE ViT16 | attack success rate AVG (%): 94.00 -> 95.30 | Improved |
| use fixed ImageNet mean/std noise instead of uniform sampling (ablation) | L4A_ugs · ResNet101 / ResNet50 SimCLRv2 | attack success rate AVG (%): 68.39 vs 69.50 | Mixed |
| tune fusing weight lambda to 10^-0.5 (ablation) | L4A_ugs · ResNet101 | attack success rate AVG (%): 1.52% | Improved |
| attack lower layers of the network (ablation) | L4A_base · 10 downstream tasks | attack success rate AVG (%) | Improved |
| L4A_ugs versus best baseline UAPEPGD | L4A_ugs · SimCLRv2 ResNet101 | attack success rate AVG (%): 59.34 -> 72.20 | Improved |
| L4A_ugs versus best baseline FFF_no | L4A_ugs · MAE ViT16 | attack success rate AVG (%): 77.69 -> 95.30 | Improved |
| add fused loss on top of base loss (ablation) | L4A_base · CLIP ResNet50 | attack success rate AVG (%): 81.00 -> 91.41 | Improved |
| add UGS to fused loss | L4A_fuse · MOCO ResNet50 | attack success rate AVG (%): 54.72 -> 59.74 | Improved |
| add UGS to L4A_base under adversarial fine-tuning (ablation) | L4A_base · Adversarial fine-tuned ResNet50 | attack success rate AVG (%): 30.46 -> 30.04 | No clear effect |
| use L4A_ugs on adversarial-pretrained standard-finetuned model | L4A_ugs · Adversarial-pretrained ResNet50 | attack success rate AVG (%): 59.45 -> 68.91 | Improved |
Closest papers by meaning
- Improving Adversarial Transferability with Local Perturbation Augmentation (CVPR 2026 · similarity 0.89)
- Adversarial Robustness: From Self-Supervised Pre-Training to Fine-Tuning (CVPR 2020 · similarity 0.88)
- Defense Against Universal Adversarial Perturbations (CVPR 2018 · similarity 0.88)
- Generate Universal Adversarial Perturbations for Few-Shot Learning (NeurIPS 2024 · similarity 0.87)
- Democratic Training Against Universal Adversarial Perturbations (ICLR 2025 · similarity 0.87)
- Pre-trained Model Guided Fine-Tuning for Zero-Shot Adversarial Robustness (CVPR 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.
