Research map / Adversarial robustness and security / Black-box and transfer attacks
Boosting the Transferability of Adversarial Attacks with Reverse Adversarial Perturbation
Read the paperCheck your idea against 108,967 papers
AI-extracted summary. Written by AI from the paper's title only (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 reverse adversarial perturbation (RAP) to MI | MI attack · Targeted ResNet-50 -> Dense-121 | success rate (%): 6.3 -> 17.5 | Helped |
| add RAP to untargeted baselines (I, MI, TI, DI, SI, Admix) | baseline transfer attacks · Untargeted ImageNet, avg over targets | success rate (%): +9.6% (I), +16.3% (MI) | Helped |
| add RAP to I on VGG-16 surrogate | I attack · Untargeted VGG-16 -> Res-50 | success rate (%): 53.7 -> 53.0 | No clear effect |
| add late-start (start RAP at iteration 100) to RAP (ablation) | RAP · Targeted ResNet-50 -> Dense-121, MI | success rate (%): 17.5 -> 29.6 | Helped |
| add RAP-LS to combinational attacks | MTDI/MTDSI/MTDAI · Targeted, avg over targets | success rate (%): +14.2%, +11.8%, +9.3% | Helped |
| add RAP-LS to MTDAI against ViT-B/16 target | MTDAI · Targeted ResNet-50 -> ViT-B/16 | success rate (%): 3.3 -> 7.4 | Helped |
| add RAP-LS to ensemble attacks on ensemble-AT defense models | MTDI/MTDSI/MTDAI · Targeted ens-adv-Inc-Res-v2 | success rate (%): 12.2 -> 26.0 | Helped |
| RAP vs generative TTP | MTDSI+RAP-LS · Targeted Inc-v3 | success rate (%): 40.3 -> 59.3 | Helped |
| Increase inner iterations T for fixed epsilon_n (ablation) | RAP · Targeted ResNet-50 | success rate | Helped |
| Increase RAP neighborhood size epsilon_n (ablation) | RAP · Targeted ResNet-50 | success rate | Helped |
| Increase late-start iteration K_LS beyond 100 (ablation) | RAP-LS · Targeted ResNet-50 | success rate | No clear effect |
| add RAP-LS to MTDAI against Google Cloud Vision API | MTDAI · Google Cloud Vision targeted | images attacked of 500: 232 -> 342 | Helped |
Closest papers by meaning
- Rethinking the Backward Propagation for Adversarial Transferability (NeurIPS 2023 · similarity 0.91)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.91)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.91)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.90)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (AAAI 2024 · similarity 0.90)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · 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.