Research map / Adversarial robustness and security / Black-box and transfer attacks
Practical Relative Order Attack in Deep Ranking
Mo Zhou, Le Wang, Zhenxing Niu, Qilin Zhang, Yinghui Xu, Nanning Zheng, 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 the semantics-preserving L_QA+ term with xi=10 (ablation) | white-box OA · Fashion-MNIST (k=5, eps=4/255) | SRC tau_S: 0.561 -> 0.412 | Hurt |
| add the semantics-preserving L_QA+ term with xi=10 (ablation) | white-box OA · Fashion-MNIST (k=5, eps=4/255) | mean rank mR: 27.2 -> 9.1 | Improved |
| add L_QA+ with xi=10^3 on SOP (ablation) | white-box OA · Stanford Online Products (k=5, eps=4/255) | mean rank mR: 973.9 -> 4.9 | Improved |
| add L_QA+ with xi=10^3 on SOP (ablation) | white-box OA · Stanford Online Products (k=5, eps=4/255) | SRC tau_S: 0.932 -> 0.448 | Hurt |
| reduce perturbation search space to 3x32x32 (ablation) | black-box OA · Stanford Online Products and JD SnapShop | SRC tau_S | Improved |
| use NES instead of random search | black-box OA · Fashion-MNIST (k=5, N=inf, eps=4/255) | SRC tau_S: 0.309 -> 0.416 | Improved |
| use Beta distribution search instead of uniform random search | black-box OA · Fashion-MNIST (k=5, N=inf, eps=4/255) | SRC tau_S: 0.309 -> 0.360 | Improved |
| use SPSA instead of NES with N=k and large eps | black-box OA · Stanford Online Products (k=5, N=k, eps=16/255) | SRC tau_S: 0.030 -> 0.144 | Improved |
| increase perturbation budget eps from 2/255 to 16/255 | white-box OA · Fashion-MNIST (k=5, N=inf) | SRC tau_S: 0.286 -> 0.599 | Improved |
| increase perturbation budget eps beyond 4/255 under truncated list | black-box OA with Rand · Stanford Online Products (k=5, N=50) | SRC tau_S: 0.216 -> 0.126 | Hurt |
| increase number of selected candidates k from 5 to 25 | white-box OA · Fashion-MNIST (eps=4/255) | SRC tau_S: 0.412 -> 0.108 | Hurt |
Closest papers by meaning
- Adversarial Attack and Defense in Deep Ranking (TPAMI 2024 · similarity 0.90)
- Adversarial Ranking Attack and Defense (ECCV 2020 · similarity 0.89)
- Perturbation-Invariant Adversarial Training for Neural Ranking Models: Improving the Effectiveness-Robustness Trade-Off (AAAI 2024 · similarity 0.86)
- Attack-in-the-Chain: Bootstrapping Large Language Models for Attacks Against Black-Box Neural Ranking Models (AAAI 2025 · similarity 0.85)
- Adversarial Perturbations Are Formed by Iteratively Learning Linear Combinations of the Right Singular Vectors of the Adversarial Jacobian (ICML 2025 · similarity 0.84)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.83)
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.
