Research map / Adversarial robustness and security / Certified and empirical robustness
Bias in Zeroth-Order Normal Estimation for Decision-Based Attacks
Feiyang Wang, Hangwei Qian, Xingquan Zuo, Gang Chen, Ivor Tsang
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 SAR post-refinement | HSJA · ImageNet ViT (10k queries) | average L2: 2.9681 -> 2.5712 | Improved |
| add SAR post-refinement | CGBA · ImageNet ViT (10k queries) | average L2: 1.5685 -> 1.3542 | Improved |
| add SAR post-refinement | TtBA · ImageNet ViT (10k queries) | average L2: 1.5019 -> 1.3613 | Improved |
| add SAR post-refinement | HSJA · ImageNet ViT (10k queries) | ASR: 62.6% -> 71.8% | Improved |
| add SAR post-refinement | CGBA · ImageNet VGG (10k queries) | average L2: 1.1743 -> 0.9596 | Improved |
| add SAR post-refinement | CGBA · ImageNet Engstrom adversarially trained | ASR: 30.6% -> 38.0% | Improved |
| add SAR post-refinement | HSJA · ImageNet Inception (10k queries) | average L2: 4.1391 -> 4.2420 | Hurt |
| replace binary region PAR with continuous SAR rescaling (ablation) | HSJA · ImageNet ViT (10k queries) | average L2: 2.7396 -> 2.5712 | Improved |
| replace binary region PAR with continuous SAR rescaling (ablation) | CGBA · ImageNet ViT (10k queries) | average L2: 1.4584 -> 1.3542 | Improved |
| replace binary region PAR with continuous SAR rescaling (ablation) | TtBA · ImageNet VGG (10k queries) | average L2: 1.0685 -> 0.9829 | Improved |
| add SAR post-refinement under a 5,000 query budget | TtBA · MNIST CNN (5k queries) | average L2: 2.272 -> 1.356 | Improved |
| add SAR post-refinement under a 5,000 query budget | HSJA · ImageNet VGG (5k queries) | average L2: 4.207 -> 4.169 | Improved |
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.
- TtBA — TtBA: Two-third Bridge Approach for Decision-Based Adversarial Attack (ICML 2025) · better on 8 of 8 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
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.86)
- Adversarial Perturbations Are Formed by Iteratively Learning Linear Combinations of the Right Singular Vectors of the Adversarial Jacobian (ICML 2025 · similarity 0.85)
- Backdoor Attacks Against No-Reference Image Quality Assessment Models via a Scalable Trigger (AAAI 2025 · similarity 0.84)
- Closing the Gap: Achieving Better Accuracy-Robustness Tradeoffs against Query-Based Attacks (AAAI 2024 · similarity 0.84)
- Strong Transferable Adversarial Attacks via Ensembled Asymptotically Normal Distribution Learning (CVPR 2024 · similarity 0.84)
- Revisiting Zeroth-Order Optimization: Minimum-Variance Two-Point Estimators and Directionally Aligned Perturbations (ICLR 2025 · similarity 0.84)
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.
