Research map / Adversarial robustness and security / Black-box and transfer attacks
Boosting Adversarial Transferability via Residual Perturbation Attack
Jinjia Peng, Zeze Tao, Huibing Wang, Meng Wang, Yang Wang
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 |
|---|---|---|---|
| use residual gradient (current minus EMA reference) as perturbation direction | flatness-based transfer attack · ImageNet, Inc-v3 surrogate, 8 targets | average attack success rate (%): 63.6 -> 64.6 (vs PGN) | Improved |
| combine ResPA with DIM | DIM · ImageNet, Res-50 surrogate | average attack success rate (%): 79.6 -> 91.0 | Improved |
| combine ResPA with TIM | TIM · ImageNet, Res-50 surrogate | average attack success rate (%): 70.6 -> 90.9 | Improved |
| combine ResPA with SSA | SSA · ImageNet, Res-50 surrogate | average attack success rate (%): 82.4 -> 86.6 | Improved |
| combine ResPA with Admix | Admix · ImageNet, Res-50 surrogate | average attack success rate (%): 70.2 -> 81.8 | Improved |
| ResPA in ensemble-surrogate setting vs PGN | ResPA · ImageNet, ensemble of Res-50/Vgg-19/Den-121 | average attack success rate (%): 95.8 -> 96.0 | Improved |
| ResPA against defense models vs PGN | ResPA · ImageNet, 7 defense models, Inc-v3 surrogate | average attack success rate (%): 58.8 -> 59.8 | Improved |
| ResPA on Re-ID attack vs TPA | ResPA · Market-1501 Re-ID, average mAP | mAP (%) lower is better: 19.20 -> 18.77 | Improved |
| increase sampling number N (ablation) | ResPA · ImageNet, Res-50 surrogate | attack success rate: stabilizes after N>15 (CNNs), N>25 (ViT/adv-trained) | Improved |
| increase sampling boundary beta beyond 2.5 epsilon (ablation) | ResPA · ImageNet, Res-50 surrogate | attack success rate | Hurt |
| increase penalty coefficient gamma above 0.9 (ablation) | ResPA · ImageNet, Res-50 surrogate | attack success rate | Hurt |
| vary EMA decay theta within [0.2, 1] (ablation) | ResPA · ImageNet, Res-50 surrogate | attack success rate | No clear effect |
Closest papers by meaning
- Boosting the Transferability of Adversarial Attacks with Reverse Adversarial Perturbation (NeurIPS 2022 · similarity 0.93)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (AAAI 2024 · similarity 0.90)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.89)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · similarity 0.88)
- Transferability Bound Theory: Exploring Relationship between Adversarial Transferability and Flatness (NeurIPS 2024 · similarity 0.88)
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.
