Research map / Adversarial robustness and security / Black-box and transfer attacks
An Adaptive Model Ensemble Adversarial Attack for Boosting Adversarial Transferability
Bin Chen, Jiali Yin, Shukai Chen, Bohao Chen, Ximeng Liu
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 adaptive gradient modulation (AGM) to naive ensemble attack (ablation) | Ens · CIFAR-10 black-box | average ASR (%): 27.29 -> 39.79 | Improved |
| add disparity-reduced filter (DRF) to naive ensemble attack (ablation) | Ens · CIFAR-10 ViT targets | average ASR on ViTs (%): 23.94 -> 47.57 | Improved |
| combine AGM and DRF (full AdaEA) (ablation) | Ens · CIFAR-10 black-box | average ASR (%): 27.29 -> 44.78 | Improved |
| use AdaEA instead of SVRE ensemble on ImageNet | ensemble attack · ImageNet black-box | average ASR (%): 46.39 -> 49.36 | Improved |
| use SVRE variance reduction with momentum or input-diversity base attacks | ensemble attack on CIFAR-10 · CIFAR-10 black-box (DI2-FGSM) | average ASR (%): 72.44 -> 33.49 | Hurt |
| increase number of surrogate models from two to four (ablation) | ensemble attack · CIFAR-10 black-box | average ASR (%): around 20% on both CNNs and ViTs | Improved |
| use ViT-only versus CNN-only surrogate ensemble (ablation) | AdaEA · CIFAR-10 black-box | average ASR (%): CNN-only to ViT targets 13.58; ViT-only to CNN targets 36.53 | Mixed |
| increase weighting scale beta in AGM | AdaEA · CIFAR-10 black-box | average ASR | Mixed |
| increase DRF binarization threshold eta | AdaEA · CIFAR-10 black-box | average ASR | Mixed |
| AdaEA with ViT-only surrogates (0 CNNs, 3 ViTs) versus plain Ens on CNN targets (ablation) | AdaEA · CIFAR-10 CNN targets | average ASR on CNNs (%): 36.67 -> 36.53 | No clear effect |
Closest papers by meaning
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 2024 · similarity 0.94)
- Understanding Model Ensemble in Transferable Adversarial Attack (ICML 2025 · similarity 0.94)
- Ensemble Diversity Facilitates Adversarial Transferability (CVPR 2024 · similarity 0.91)
- Enhancing Adversarial Transferability with Checkpoints of a Single Model's Training (CVPR 2025 · similarity 0.90)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.90)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · 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.
