Research map / Adversarial robustness and security / Black-box and transfer attacks
On the Adversarial Transferability of Generalized “Skip Connections”
Yisen Wang, Yichuan Mo, Dongxian Wu, Mingjie Li, Xingjun Ma, Zhouchen Lin
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 SGM to PGD | PGD ResNet-50 source · VGG16 target | attack success rate: 45.00 -> 58.68 | Improved |
| add SGM to SIA | SIA ResNet-50 source · VGG16 target | attack success rate: 90.88 -> 95.93 | Improved |
| add SGM to PGD with ViT-B source | PGD ViT-B source · VGG16 target | attack success rate: 16.34% to 21.04% | Improved |
| add SGM to PGD with Inception-V3 source (no skip connections) | PGD Inception-V3 source · VGG16 target | attack success rate: 29.74 -> 45.43 | Improved |
| add SGM to MFAA in model ensemble | MFAA ensemble source · SE154 target | attack success rate: 61.70% to 90.32% | Improved |
| add SGM to Logit targeted attack | Logit ResNet-50 source · VGG16 target | attack success rate: 37.90 -> 63.49 | Improved |
| add SGM to RFA against CFA defense | RFA · CFA defense | attack success rate: 9.40% to 13.13% | Improved |
| add MUP baseline to PGD with Inception-V3 source | PGD Inception-V3 source · VGG16 target | attack success rate: 29.74 -> 21.62 | Hurt |
| add MUP baseline to PGD with ResNet-50 source | PGD ResNet-50 source · VGG16 target | attack success rate: 45.00 -> 50.84 | Improved |
| decay gradient of residual modules (lower gamma) for ResNet-50 source (ablation) | SGM PGD ResNet-50 · ImageNet black-box | attack success rate: peak around gamma = 0.2 | Improved |
| decay gradients for Inception-V3 source (no skip connections) (ablation) | SGM PGD Inception-V3 · ImageNet black-box | attack success rate: optimal around gamma = 0.6 | Improved |
| lower gamma from 1.0 to 0.5 on adversarial confidence on true class (ablation) | PGD+SGM image (a) · ImageNet samples | ground-truth confidence (lower is better): 16.84 -> 0.56 | Improved |
Closest papers by meaning
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.87)
- Improving the Adversarial Transferability of Vision Transformers with Virtual Dense Connection (AAAI 2024 · similarity 0.87)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · similarity 0.86)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.86)
- Improving Transferable Targeted Adversarial Attacks with Model Self-Enhancement (CVPR 2024 · similarity 0.86)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 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.
