Research map / Adversarial robustness and security / Black-box and transfer attacks
Admix: Enhancing the Transferability of Adversarial Attacks
Xiaosen Wang, Xuanran He, Jingdong Wang, Kun He
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 |
|---|---|---|---|
| admix a small portion of other-category images into the input gradient (Admix instead of SIM) | MI-FGSM with SIM · Inc-v3 to Inc-v4 | attack success rate (%): 69.4 -> 82.6 | Improved |
| admix instead of SIM | MI-FGSM with SIM · Inc-v3 to Inc-v3_ens3 | attack success rate (%): 32.5 -> 39.0 | Improved |
| replace SIM by Admix in SI-TIM | SI-TIM · Inc-v3 to Inc-v3_ens3 | attack success rate (%): 48.2 -> 59.1 | Improved |
| replace SIM by Admix in SI-DIM | SI-DIM · Inc-v3 to Inc-v3_ens3 | attack success rate (%): 48.0 -> 52.2 | Improved |
| replace SIM by Admix in SI-TI-DIM, ensemble attack | SI-TI-DIM · IncRes-v2_ens | attack success rate (%): 86.8 -> 90.0 | Improved |
| replace SI-TI-DIM by Admix-TI-DIM against advanced defenses | SI-TI-DIM · Nine defense models average | attack success rate (%): 79.9 -> 83.3 | Improved |
| use Admix instead of SIM in ensemble attack | SIM · IncRes-v2_ens | attack success rate (%): 59.5 -> 67.8 | Improved |
| increase number of sampled images m2 from 0 to 3 (ablation) | Admix · Inc-v3 source, six other models | attack success rate (%) | Improved |
| increase m2 beyond 3 (ablation) | Admix · Inc-v3 source, normally trained models | attack success rate (%) | Hurt |
| increase m2 beyond 3 (ablation) | Admix · Inc-v3 source, adversarially trained models | attack success rate (%) | Improved |
| increase admix strength eta from 0 to 0.2 (ablation) | Admix · Inc-v3 source | attack success rate (%) | Improved |
| mix labels in mixup (Mixup vs Mixup without label mixing) (ablation) | Mixup attack · Inc-v3 source | attack success rate (%) | Hurt |
Later papers that compared with this method
10 later accepted papers report Admix as a baseline in their results tables; 10 of them report a better number on most of the results they share with it.
- Improving Adversarial Transferability with Local Perturbation Augmentation (CVPR 2026 · better on 1 of 1 shared result)
- Prompting Adversarial Transferability via Path Flatness Attack (AAAI 2026 · better on 1 of 1 shared result)
- Enhancing Transferability of Targeted Adversarial Examples via Inverse Target Gradient Competition and Spatial Distance Stretching (ICCV 2025 · better on 4 of 4 shared results)
- Boosting Adversarial Transferability by Block Shuffle and Rotation (CVPR 2024 · better on 3 of 3 shared results)
- LRS: Enhancing Adversarial Transferability through Lipschitz Regularized Surrogate (AAAI 2024 · better on 1 of 1 shared result)
- Improving the Transferability of Adversarial Samples by Path-Augmented Method (CVPR 2023 · better on 12 of 12 shared results)
- Boosting Adversarial Transferability via Gradient Relevance Attack (ICCV 2023 · better on 9 of 9 shared results)
- Improving Adversarial Transferability via Intermediate-level Perturbation Decay (NeurIPS 2023 · better on 3 of 3 shared results)
2 more not listed.
Matched by method name in results tables read by AI. Authors' own tables tend to favour the newer method, and a comparison can be missed when a method is renamed.
Closest papers by meaning
- Improving the Transferability of Adversarial Samples With Adversarial Transformations (CVPR 2021 · similarity 0.90)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
- Learning to Transform Dynamically for Better Adversarial Transferability (CVPR 2024 · similarity 0.88)
- Boosting Adversarial Transferability by Block Shuffle and Rotation (CVPR 2024 · similarity 0.88)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.88)
- AIM: Additional Image Guided Generation of Transferable Adversarial Attacks (AAAI 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.
