Research map / Adversarial robustness and security / Black-box and transfer attacks
Triangle Attack: A Query-Efficient Decision-Based Adversarial Attack
Xiaosen Wang, Zeliang Zhang, Kangheng Tong, Dihong Gong, Kun He, Zhifeng Li, Wei Liu
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 |
|---|---|---|---|
| optimize in the low-frequency space with a mask instead of the input space (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%): 3.0 -> 51.5 | Improved |
| sample the line in frequency space rather than input space (perturbation still in input space) (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%): 3.0 -> 22.5 | Improved |
| optimize the perturbation in frequency space rather than input space (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%): 22.5 -> 25.5 | Improved |
| restrict to the low-frequency subspace (top 10%) rather than the full frequency space (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%): 25.5 -> 51.5 | Improved |
| increase the low-frequency ratio (ablation) | Triangle Attack · ImageNet ResNet-18, 1000 queries | attack success rate (%) | Hurt |
| restrict each adversarial example to the decision boundary with binary search (increase N_bs) (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%) | Hurt |
| set gamma to 0.01 rather than 0.05 or 0.005 (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%) | Improved |
| increase lambda with gamma = 0.01 (ablation) | Triangle Attack · ImageNet ResNet-18, RMSE 0.01 | attack success rate (%) | Mixed |
| increase the dimension d of the directional line from 1 to 10 (ablation) | Triangle Attack · ImageNet ResNet-18, 1000 queries | attack success rate (%) | Hurt |
| increase the angle bound tau (ablation) | Triangle Attack · ImageNet ResNet-18, 1000 queries | attack success rate (%) | Hurt |
| Triangle Attack versus runner-up attack on ResNet-101 | Triangle Attack · ImageNet ResNet-101, RMSE 0.1/0.05/0.01 | attack success rate (%): 1.0%, 7.5% and 13.0% | Improved |
| Triangle Attack versus Surfree on 1000 images of VGG-16 | Triangle Attack · ImageNet VGG-16 (1000 images), RMSE 0.01 | attack success rate (%): 36.5 -> 39.7 | Improved |
Closest papers by meaning
- A Geometry-Inspired Decision-Based Attack (ICCV 2019 · similarity 0.87)
- TtBA: Two-third Bridge Approach for Decision-Based Adversarial Attack (ICML 2025 · similarity 0.86)
- DifAttack: Query-Efficient Black-Box Adversarial Attack via Disentangled Feature Space (AAAI 2024 · similarity 0.85)
- Seeing is Not Believing: Adversarial Natural Object Optimization for Hard-Label 3D Scene Attacks (CVPR 2025 · similarity 0.85)
- AttackBench: Evaluating Gradient-based Attacks for Adversarial Examples (AAAI 2025 · similarity 0.85)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · 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.
