Research map / Adversarial robustness and security / Black-box and transfer attacks
Meta Gradient Adversarial Attack
Zheng Yuan, Jie Zhang, Yunpei Jia, Chuanqi Tan, Tao Xue, Shiguang Shan
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 MGAA to DIM (targeted) | DIM · ImageNet targeted, black-box avg | targeted attack success rate (%): 19.42 -> 47.90 | Improved |
| add MGAA to MIM (targeted) | MIM · ImageNet targeted, black-box avg | targeted attack success rate (%): 2.84 -> 8.56 | Improved |
| add MGAA to TI-DIM (targeted) | TI-DIM · ImageNet targeted, black-box avg | targeted attack success rate (%): 23.86 -> 38.18 | Improved |
| add MGAA to TI-DIM (targeted), white-box | TI-DIM · ImageNet targeted, white-box avg | targeted attack success rate (%): 42.85 -> 85.52 | Improved |
| add MGAA to DIM (untargeted) | DIM · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 79.4 -> 88.0 | Improved |
| add MGAA to MIM (untargeted) | MIM · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 44.4 -> 52.0 | Improved |
| add MGAA to TI-DIM on CIFAR10 | TI-DIM · CIFAR10 untargeted, Adv_ResNet-18_ll | untargeted attack success rate (%): 46.69 -> 68.77 | Improved |
| add meta-train ensemble step (ablation) | MGAA · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 81.5 -> 99.1 | Improved |
| add meta-test step (ablation) | MGAA · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 96.3 -> 99.1 | Improved |
| increase meta-train iterations K from 1 to 8 (ablation) | MGAA w/ TI-DIM · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 96.3 -> 99.1 | Improved |
| increase sampled tasks T from 10 to 120 (ablation) | MGAA w/ TI-DIM · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 94.5 -> 98.2 | Improved |
| increase ensembled models n beyond 5 (ablation) | MGAA w/ TI-DIM · ImageNet untargeted, Adv_Inceptionv3 | untargeted attack success rate (%): 98.6 -> 98.6 | No clear effect |
Closest papers by meaning
- Generalizable Black-Box Adversarial Attack With Meta Learning (TPAMI 2023 · similarity 0.91)
- Training Meta-Surrogate Model for Transferable Adversarial Attack (AAAI 2023 · similarity 0.90)
- Improving Black-box Adversarial Attacks with a Transfer-based Prior (NeurIPS 2019 · similarity 0.89)
- Query-Efficient Black-Box Adversarial Attacks Guided by a Transfer-Based Prior (TPAMI 2021 · similarity 0.89)
- Improving Integrated Gradient-based Transferable Adversarial Examples by Refining the Integration Path (AAAI 2025 · similarity 0.87)
- Rethinking Model Ensemble in Transfer-based Adversarial Attacks (ICLR 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.
