Research map / Adversarial robustness and security / Black-box and transfer attacks
Aha! Adaptive History-Driven Attack for Decision-Based Black-Box Models
Jie Li, Rongrong Ji, Peixian Chen, Baochang Zhang, Xiaopeng Hong, Ruixin Zhang, Shaoxin Li, Jilin Li, Feiyue Huang, Yongjian 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 |
|---|---|---|---|
| add dimension reduction (subspace sampling) (ablation) | AHA · ImageNet ResNet50 | mean L2: 10.823 -> 6.203 | Improved |
| add dimension reduction (subspace sampling) (ablation) | AHA · ImageNet Inception-V3 | mean L2: 22.196 -> 10.976 | Improved |
| add DCT low-pass on top of dimension reduction (ablation) | AHA · ImageNet ResNet50 | mean L2: 6.203 -> 16.200 | Hurt |
| use all historical queries rather than only failed ones (ablation) | AHA · ImageNet ResNet50 | mean L2: 6.950 -> 6.203 | Improved |
| use all historical queries rather than only successful ones (ablation) | AHA · ImageNet ResNet50 | mean L2: 7.465 -> 6.203 | Improved |
| replace 1/5th success rule with ratio-based adaptive coefficient h=(r+0.8)^2 (ablation) | AHA · ImageNet ResNet50 | mean L2: 6.638 -> 6.203 | Improved |
| replace 1/5th success rule with ratio-based adaptive coefficient h=(r+0.8)^2 (ablation) | AHA · ImageNet Inception-V3 | mean L2: 11.507 -> 10.976 | Improved |
| change h from (r+0.8)^2 to (r+0.6)^2 (ablation) | AHA · ImageNet ResNet50 | mean L2: 6.203 -> 20.813 | Hurt |
| change h from (r+0.8)^2 to (r+0.9)^2 (ablation) | AHA · ImageNet ResNet50 | mean L2: 6.203 -> 9.643 | Hurt |
| change h from (r+0.8)^2 to r+0.8 (ablation) | AHA · ImageNet ResNet50 | mean L2: 6.203 -> 6.168 | No clear effect |
| use AHA instead of Evolutionary attack | Evolutionary · ImageNet Inception-V3 | attack success rate: 64% -> 91% | Improved |
| use AHA instead of Evolutionary attack | Evolutionary · Face++ API | mean L2: 14.544 -> 10.910 | Improved |
Closest papers by meaning
- Learning Black-Box Attackers with Transferable Priors and Query Feedback (NeurIPS 2020 · similarity 0.88)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.88)
- Query-Efficient and Scalable Black-Box Adversarial Attacks on Discrete Sequential Data via Bayesian Optimization (ICML 2022 · similarity 0.87)
- Query-Efficient Black-Box Adversarial Attack With Customized Iteration and Sampling (TPAMI 2022 · similarity 0.87)
- Query-Efficient Black-Box Adversarial Attacks Guided by a Transfer-Based Prior (TPAMI 2021 · similarity 0.87)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · 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.
