Research map / Adversarial robustness and security / Black-box and transfer attacks
Enhancing Transferability of Targeted Adversarial Examples via Inverse Target Gradient Competition and Spatial Distance Stretching
Zhankai Li, Weiping Wang, Jie Li, Shigeng Zhang, Yunan Hu, Song Guo
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 SDS (top-k spatial distance stretching) to ITC (ablation) | ITDS · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 27.9 -> 33.7 | Improved |
| add SDS to ITC with DIM (ablation) | ITDS-DIM · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 53.1 -> 61.7 | Improved |
| add ITC competition to a non-competition baseline (ITC only vs CFM) | ITDS · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 24.6 -> 27.9 | Improved |
| drop norm normalization in momentum gradient (ablation) | ITDS-DIM · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 55.3 -> 61.7 | Improved |
| do not use momentum gradient as competitive gradient (ablation) | ITDS-DIM · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 47.9 -> 61.7 | Improved |
| use momentum gradient as competitor without normalization (ablation) | ITDS-DIM · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 61.7 -> 0.4 | Hurt |
| replace CFM competition with ITDS (ITC+SDS) on RN50 surrogate, DIM-combined | ITDS-DIM vs CFM-DIM · ImageNet targeted attack | BAvg TASR (%): 46.0 -> 61.7 | Improved |
| replace CFM competition with ITDS (ITC+SDS) on RegN surrogate, DIM-combined | ITDS-DIM vs CFM-DIM · ImageNet targeted attack | BAvg TASR (%): 47.9 -> 63.2 | Improved |
| use ITDS-DIM instead of CFM-DIM against Score-Opt defense | ITDS-DIM · ImageNet targeted attack vs Score-Opt | TASR (%): 51.5 -> 91.3 | Improved |
| use ITDS-DIM instead of CFM-DIM against GDMP defense | ITDS-DIM · ImageNet targeted attack vs GDMP | TASR (%): 4.1 -> 31.2 | Improved |
| use S2 (geometric space) instead of S1 (feature space) in SDS (ablation) | ITDS · ImageNet targeted attack (RN50 surrogate) | BAvg TASR (%): 33.7 -> 33.6 | No clear effect |
| increase number of target class samples n beyond 3 (ablation) | ITDS-DIM · ImageNet targeted attack (RN50 surrogate) | TASR | No clear effect |
Earlier methods this paper compares with
Named baselines from earlier accepted papers in this paper's results tables: better than 1 of 1 on most shared results.
- Admix — Admix: Enhancing the Transferability of Adversarial Attacks (ICCV 2021) · better on 4 of 4 shared results
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
- Introducing Competition To Boost the Transferability of Targeted Adversarial Examples Through Clean Feature Mixup (CVPR 2023 · similarity 0.89)
- Towards Transferable Targeted Attack (CVPR 2020 · similarity 0.89)
- Enhancing Adversarial Transferability with Adversarial Weight Tuning (AAAI 2025 · similarity 0.89)
- Boosting Adversarial Transferability through Augmentation in Hypothesis Space (CVPR 2025 · similarity 0.88)
- Improving Transferable Targeted Adversarial Attacks with Model Self-Enhancement (CVPR 2024 · similarity 0.87)
- Everywhere Attack: Attacking Locally and Globally to Boost Targeted Transferability (AAAI 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.
