Research map / Adversarial robustness and security / Black-box and transfer attacks
BlackboxBench: A Comprehensive Benchmark of Black-Box Adversarial Attacks
Meixi Zheng, Xuanchen Yan, Zihao Zhu, Hongrui Chen, Baoyuan Wu
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 |
|---|---|---|---|
| combine SIA with model-fusion method Enslogit | SIA transfer attack · ImageNet untargeted l_inf | ASR (%): 75.1 -> 78.5 | Improved |
| combine SIA with CWA | SIA transfer attack · ImageNet untargeted l_inf | ASR (%): 75.1 -> 77.1 | Improved |
| combine individual transfer attacks into pairwise composites | transfer-based attacks · ImageNet untargeted l_inf, 98 composites | ASR | Improved |
| combine SIA with model-tuning methods (LGV, Bayesian attack) | SIA transfer attack · ImageNet untargeted l_inf | ASR | No clear effect |
| increase perturbation budget from 1/255 to 32/255 (ablation) | Parsimonious Attack · ImageNet score-based untargeted | ASR: 0.27 -> 1.00 | Improved |
| increase perturbation budget from smallest to 32/255 (ablation) | Square Attack · ImageNet score-based untargeted | ASR: 0.31 -> 1.00 | Improved |
| increase perturbation budget to 32/255 (ablation) | NES · ImageNet score-based untargeted | ASR: -> 0.79 | Improved |
| use surrogate models for gradient estimation (combination methods) | score-based attacks (MCG, PRGF, BASES, CG-attack) · ImageNet score-based | ASR across budgets | Improved |
| use semantic-perturbation transfer attacks (FIA, NAA, DRA, IAA) with low norm budget (ablation) | transfer-based attacks · ImageNet transfer, varying epsilon | ASR | Hurt |
| add more iterations to momentum-based attacks (MI, NI, PI, CWA) (ablation) | transfer-based attacks · ImageNet transfer | ASR | Hurt |
| use complex target architectures (ConvNeXt-T, ViT-B/16) instead of ResNet-50 or VGG-19 | query-based attacks · ImageNet untargeted l_inf | AQN/ASR ratio | Improved |
Closest papers by meaning
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.92)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.91)
- Black-Box Adversarial Attack with Transferable Model-based Embedding (ICLR 2020 · similarity 0.90)
- Improving Query Efficiency of Black-box Adversarial Attack (ECCV 2020 · similarity 0.90)
- TransferBench: Benchmarking Ensemble-based Black-box Transfer Attacks (NeurIPS 2025 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.89)
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.
