Research map / Adversarial robustness and security / Black-box and transfer attacks
Structure Invariant Transformation for better Adversarial Transferability
Xiaosen Wang, Zeliang Zhang, Jianping Zhang
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 |
|---|---|---|---|
| apply different transformations locally to image blocks (SIA) instead of global transformations | MI-FGSM · ImageNet, ResNet-18 to 7 models | average attack success rate (%): 59.7 -> 92.1 | Improved |
| add any single block transformation (weakest, DCT) (ablation) | SIA · ImageNet, ResNet-18 to 7 models | average attack success rate (%): 59.7 -> 67.3 | Improved |
| include each of the ten transformations (adding back any removed one) (ablation) | SIA · ImageNet, ResNet-18 to 7 models | average attack success rate (%): 88.3-90.7 -> 92.1 | Improved |
| combine two transformations instead of one (ablation) | SIA · ImageNet, ResNet-18 to 7 models | average attack success rate (%) | Improved |
| apply transformation per image block rather than to the whole image s x s times (ablation) | SIA · ImageNet, ResNet-18 to 7 models | average attack success rate (%) | Improved |
| increase number of blocks s beyond 3 (ablation) | SIA · ImageNet, ResNet-18 to 7 models | attack success rate | Hurt |
| increase number of transformed images N from 1 to 20 (ablation) | SIA · ImageNet, ResNet-18 to 7 models | attack success rate | Improved |
| increase N beyond 20 (ablation) | SIA · ImageNet, ResNet-18 to 7 models | attack success rate | No clear effect |
| integrate SIA into SGM and LinBP | input-transformation attacks · ImageNet, ResNet-18 source | average attack success rate (%): +3.3% (SGM), +3.6% (LinBP) over best baselines | Improved |
| use SIA instead of best baseline on ensemble models | ensemble attack · ImageNet, 8 models | attack success rate (%): +6.5% over winner-up | Improved |
Closest papers by meaning
- Boosting the Transferability of Adversarial Attack on Vision Transformer with Adaptive Token Tuning (NeurIPS 2024 · similarity 0.90)
- Boosting Adversarial Transferability by Block Shuffle and Rotation (CVPR 2024 · similarity 0.89)
- Learning to Transform Dynamically for Better Adversarial Transferability (CVPR 2024 · similarity 0.89)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
- Improving the Adversarial Transferability of Vision Transformers with Virtual Dense Connection (AAAI 2024 · similarity 0.89)
- Harnessing the Computation Redundancy in ViTs to Boost Adversarial Transferability (NeurIPS 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.
