Research map / Adversarial robustness and security / Black-box and transfer attacks
Leveraging Spatial Invariance to Boost Adversarial Transferability
Zihan Zhou, Li Li, Yanli Ren, Chuan Qin, Guorui Feng
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 |
|---|---|---|---|
| use SID instead of BSR on a Nesterov gradient base | SID-NI · IncV3 surrogate, average black-box | attack success rate: 66.7 -> 73.8 | Improved |
| add SID transformation (without DIM) to TIM | SID-TIM · IncV3 surrogate, average black-box | attack success rate: SIA-TI-DIM 82.9 -> SID-TIM 83.2 | Improved |
| add SIM to SID-TIM | SID-SI-TIM · IncV3 surrogate, average black-box | attack success rate: 83.2 -> 89.2 | Improved |
| use SID against adversarially trained models vs SIA and BSR | SID · IncV3 surrogate | attack success rate: average improvement of 16.8% over SIA and 13.7% over BSR | Improved |
| use SID against normally trained models vs SIA and BSR | SID · IncV3 surrogate | attack success rate: average improvement of 3.5% and 4.0% | Mixed |
| use SID in ensemble-model attack vs SIA and BSR | SID · ensemble-model, nine models | attack success rate: average improvement of 5.2% and 4.0% | Improved |
| use SID-SI-TIM against defenses vs SSA-TI-DIM | SID-SI-TIM · eight defenses average | attack success rate: 79.5 -> 84.3 | Improved |
| add multi-scale downsampling instead of fixed-scale DIM-n (ablation) | SID without local fusion · IncV3 surrogate, six black-box models | attack success rate | Improved |
| add linear fusion rather than only frequency fusion (ablation) | SID local fusion · IncV3 surrogate | attack success rate | Improved |
| use both linear and frequency fusion together (ablation) | SID local fusion · IncV3 surrogate | attack success rate | Improved |
| add local image fusion to multi-scale transformation (ablation) | SID · IncV3 surrogate | attack success rate | Improved |
| add multi-scale transformation to local image fusion (ablation) | SID · IncV3 surrogate, adversarially trained models | attack success rate | Improved |
Closest papers by meaning
- Spatially Transformed Adversarial Examples (ICLR 2018 · similarity 0.89)
- Excessive Invariance Causes Adversarial Vulnerability (ICLR 2019 · similarity 0.88)
- Boosting Adversarial Transferability with Spatial Adversarial Alignment (NeurIPS 2025 · similarity 0.88)
- Regional Homogeneity: Towards Learning Transferable Universal Adversarial Perturbations Against Defenses (ECCV 2020 · similarity 0.87)
- Breaking Barriers in Physical-World Adversarial Examples: Improving Robustness and Transferability via Robust Feature (AAAI 2025 · similarity 0.86)
- Pixel2Feature Attack (P2FA): Rethinking the Perturbed Space to Enhance Adversarial Transferability (ICML 2025 · similarity 0.86)
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.
