Research map / Adversarial robustness and security / Black-box and transfer attacks
Introducing Competition To Boost the Transferability of Targeted Adversarial Examples Through Clean Feature Mixup
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 Clean Feature Mixup to RDI | Targeted transfer attacks from RN-50 · ImageNet-Compatible, 10 target models | average targeted attack success rate (%): 69.2 -> 74.6 | Helped |
| Add Clean Feature Mixup to RDI | Targeted transfer attacks from Inc-v3 · ImageNet-Compatible, 10 target models | average targeted attack success rate (%): 29.1 -> 39.2 | Helped |
| Add Clean Feature Mixup to RDI | Targeted transfer attacks from RN-50 · ImageNet-Compatible, robust and Transformer targets | average targeted attack success rate (%): 27.8 -> 30.8 | Helped |
| Add Clean Feature Mixup to RDI | Targeted transfer attacks from adv-RN-50 · ImageNet-Compatible, robust and Transformer targets | average targeted attack success rate (%): 51.8 -> 58.6 | Helped |
| Add Clean Feature Mixup to RDI | Targeted transfer attacks from RN-50 · CIFAR-10, nine target models including defensive ensembles | average targeted attack success rate (%): 76.5 -> 89.3 | Helped |
| Add image-wise clean-feature shuffling (ablation) | CFM-RDI · ImageNet-Compatible, adv-RN-50 source, 10 selected targets | average targeted attack success rate (%): 55.7 -> 61.9 | Helped |
| Use per-channel rather than scalar mixing ratios (ablation) | CFM-RDI · ImageNet-Compatible, adv-RN-50 source, 10 selected targets | average targeted attack success rate (%): 61.0 -> 61.9 | Helped |
| Add clean-feature mixing (ablation) | CFM-RDI · ImageNet-Compatible, adv-RN-50 source, 10 selected targets | average targeted attack success rate (%): 59.2 -> 61.9 | Helped |
| Vary batch size among 5, 10, 20, and 30 (ablation) | CFM-RDI · ImageNet-Compatible, adv-RN-50 source, 10 selected targets | average targeted attack success rate (%): 61.4%, 61.6%, 61.9%, and 61.0% | No clear effect |
| Vary mixing probability and maximum mixing ratio (ablation) | CFM-RDI · ImageNet-Compatible, adv-RN-50 source, 10 selected targets | targeted attack success rate | No clear effect |
Closest papers by meaning
- Improving Transferable Targeted Attacks with Feature Tuning Mixup (CVPR 2025 · similarity 0.91)
- Improving Transferability of Adversarial Examples With Input Diversity (CVPR 2019 · similarity 0.90)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.89)
- Breaking Barriers in Physical-World Adversarial Examples: Improving Robustness and Transferability via Robust Feature (AAAI 2025 · similarity 0.88)
- Wide Two-Layer Networks can Learn from Adversarial Perturbations (NeurIPS 2024 · similarity 0.87)
- Adversarial Attacks against Closed-Source MLLMs via Feature Optimal Alignment (NeurIPS 2025 · similarity 0.87)
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.