Research map / Adversarial robustness and security / Black-box and transfer attacks
Adversarial Ranking Attack and Defense
Mo Zhou, Zhenxing Niu, Le Wang, Qilin Zhang, Gang Hua
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 maximum-shift-distance adversarial training defense (ablation) | ranking model · MNIST, CA+ eps=0.3 w=1 | mean rank after attack (%): 2.1 -> 30.7 | Improved |
| add maximum-shift-distance adversarial training defense (ablation) | ranking model · MNIST clean retrieval | Recall@1 (%): 99.1 -> 98.3 | Hurt |
| add defense (ablation) | ranking model · Fashion-MNIST, CA+ eps=0.3 w=1 | mean rank after attack (%): 1.3 -> 32.5 | Improved |
| add defense (ablation) | ranking model · Fashion-MNIST clean retrieval | Recall@1 (%): 88.8 -> 79.6 | Hurt |
| add defense (ablation) | ranking model · SOP, CA- eps=0.06 w=1 | mean rank after attack (%): 100.0 -> 97.3 | Mixed |
| add defense (ablation) | ranking model · SOP clean retrieval | Recall@1 (%): 63.1 -> 46.4 | Hurt |
| add defense (ablation) | ranking model · SOP, QA- eps=0.06 m=1 | mean rank after attack (%): 73.8 -> 51.4 | Improved |
| raise SP balancing parameter xi from 0 to 100 for QA+ (ablation) | SP-QA+ · MNIST m=1 | mean rank of chosen candidate (%): 0.2 -> 14.1 | Hurt |
| raise xi to 10^4 (ablation) | SP-QA+ · MNIST m=1 | mean rank of chosen candidate (%): 14.1 -> 37.9 | Hurt |
| use ResNet-18 instead of C2F1 as the ranking model | ranking model · MNIST CA- transfer/white-box | mean rank after attack (%): 93.9 -> 66.7 | Improved |
| evaluate universal perturbation on unseen images | I-CA+ · MNIST | mean rank (%): 18.1 -> 18.5 | No clear effect |
| transfer adversarial candidates from C2F1 to Res18 | CA+ · MNIST black-box | mean rank (%): 50 -> 31.3 | Improved |
Closest papers by meaning
- Adversarial Attack and Defense in Deep Ranking (TPAMI 2024 · similarity 0.98)
- Perturbation-Invariant Adversarial Training for Neural Ranking Models: Improving the Effectiveness-Robustness Trade-Off (AAAI 2024 · similarity 0.90)
- Practical Relative Order Attack in Deep Ranking (ICCV 2021 · similarity 0.89)
- Benchmarking Adversarial Robustness on Image Classification (CVPR 2020 · similarity 0.88)
- Deep Defense: Training DNNs with Improved Adversarial Robustness (NeurIPS 2018 · similarity 0.87)
- Adversarial Perturbations Are Formed by Iteratively Learning Linear Combinations of the Right Singular Vectors of the Adversarial Jacobian (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.
