Research map / Adversarial robustness and security / Black-box and transfer attacks
Boosting Adversarial Transferability via Gradient Relevance Attack
Hegui Zhu, Yuchen Ren, Xiaoyan Sui, Lianping Yang, Wuming Jiang
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 gradient relevance framework and decay indicator to MI-FGSM (GRA) versus Admix | Inc-v3 source · ILSVRC 2012 val 1k, 7 targets average | attack success rate (%): 62.3 -> 72.6 | Improved |
| use average gradient relevance (GRA) instead of individual relevance (I-GRA) (ablation) | GRA · Inc-v3 source, 7 targets average | attack success rate (%): 71.7 -> 72.6 | Improved |
| combine GRA with combined transformation (DI, TI, SI) | GRA · Inc-v3 source, 7 targets average | attack success rate (%): 72.6 -> 90.4 | Improved |
| apply GRA in an ensemble attack versus Admix on defended models | GRA · nine defended models | attack success rate (%): at least +12.3% over other attacks; 83.0% average | Improved |
| apply GRA on online models | GRA ensemble · Tencent Cloud and Baidu AI Cloud | attack success rate (%): +11.1% over VTNI, +11.4% over Admix | Improved |
| increase sample quantity m (ablation) | GRA · Inc-v3 source | attack success rate (%) | Improved |
| use GRA with m=5 versus VTNI with m=20 (ablation) | GRA · Inc-v4 and IncRes-v2ens targets | attack success rate (%): 78.1 vs 76.5 (Inc-v4); 24.8 vs 18.8 (IncRes-v2ens) | Improved |
| enlarge sample range factor beta to 3.5 (ablation) | GRA · Inc-v3 source, 7 targets | attack success rate (%) | Improved |
| set attenuation factor eta near 0.94 (ablation) | GRA decay indicator · Inc-v3 source, normal and adversarial groups | average attack success rate (%) | Improved |
| GRA versus Admix on IncRes-v2 white-box and IncRes-v2 source normal targets | GRA · IncRes-v2 source, normally trained targets | attack success rate (%): Inc-v3 89.9 (Admix) vs 86.0 (GRA) | Mixed |
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 9 of 9 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
- Boosting the Transferability of Adversarial Attacks with Reverse Adversarial Perturbation (NeurIPS 2022 · similarity 0.90)
- Enhancing Adversarial Transferability by Balancing Exploration and Exploitation with Gradient-Guided Sampling (ICCV 2025 · similarity 0.89)
- Boosting Adversarial Transferability by Achieving Flat Local Maxima (NeurIPS 2023 · 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)
- Towards Transferable Adversarial Attacks with Centralized Perturbation (AAAI 2024 · 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.
