Research map / Adversarial robustness and security / Black-box and transfer attacks
Finding Optimal Tangent Points for Reducing Distortions of Hard-label Attacks
Chen Ma, Xiangyu Guo, Li Chen, Jun-Hai Yong, Yisen Wang
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 |
|---|---|---|---|
| replace HSJA jump direction with tangent-point direction (TA) | HSJA · ImageNet targeted, Inception-v3 @10K | mean l2 distortion: 30.425 -> 25.566 | Improved |
| replace HSJA jump direction with tangent-point direction (TA) | HSJA · ImageNet untargeted, Inception-v3 @1K | mean l2 distortion: 57.295 -> 52.523 | Improved |
| replace HSJA jump direction with tangent-point direction (TA) | HSJA · CIFAR-10 targeted, GDAS @10K | mean l2 distortion: 0.301 -> 0.302 | No clear effect |
| replace hemisphere with semi-ellipsoid (G-TA vs TA) | TA · CIFAR-10 targeted, GDAS @10K | mean l2 distortion: 0.302 -> 0.298 | Improved |
| replace hemisphere with semi-ellipsoid (G-TA vs TA) | TA · ImageNet targeted, ResNet-101 @10K | mean l2 distortion: 15.414 -> 15.633 | Hurt |
| replace hemisphere with semi-ellipsoid (G-TA vs TA) | TA · CIFAR-10 untargeted, WRN-28 @10K | mean l2 distortion: 0.233 -> 0.232 | No clear effect |
| add TA to QEBA-S | QEBA-S · ImageNet targeted, Inception-v3 @1K | mean l2 distortion: 79.604 -> 75.622 | Improved |
| add TA to QEBA-S | QEBA-S · ImageNet targeted, Inception-v3 @10K | mean l2 distortion: 18.414 -> 17.799 | Improved |
| use TA/G-TA instead of HSJA under l_inf norm | HSJA · ImageNet targeted, Inception-v3 @10K | mean l_inf distortion: 0.147 -> 0.149 | Hurt |
| choose initial target-class image closest to original rather than random or farthest (ablation) | Tangent Attack · ImageNet targeted, ResNet-50 | mean l2 distortion | Improved |
| vary radius ratio r in G-TA (ablation) | G-TA · ImageNet targeted, ResNet-50 | mean l2 distortion | No clear effect |
| jump to optimal tangent point instead of random jump direction (RandomHSJA) (ablation) | HSJA · ImageNet targeted, ResNet-50 | mean l2 distortion | 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.
- Sign-OPT — Sign-OPT: A Query-Efficient Hard-label Adversarial Attack (ICLR 2020) · better on 4 of 4 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
- Boosting Ray Search Procedure of Hard-label Attacks with Transfer-based Priors (ICLR 2025 · similarity 0.85)
- TtBA: Two-third Bridge Approach for Decision-Based Adversarial Attack (ICML 2025 · similarity 0.85)
- Manifold Constraints for Imperceptible Adversarial Attacks on Point Clouds (AAAI 2024 · similarity 0.85)
- Curvature-Invariant Adversarial Attacks for 3D Point Clouds (AAAI 2024 · similarity 0.85)
- ADBA: Approximation Decision Boundary Approach for Black-Box Adversarial Attacks (AAAI 2025 · similarity 0.85)
- GSBA$^K$: $top$-$K$ Geometric Score-based Black-box Attack (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.
