Research map / Adversarial robustness and security / Attacks on text and graphs
GreedyFool: Distortion-Aware Sparse Adversarial Attack
Xiaoyi Dong, Dongdong Chen, Jianmin Bao, Chuan Qin, Lu Yuan, Weiming Zhang, Nenghai Yu, Dong Chen
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 the Reduce stage that drops redundant pixels (ablation) | GreedyFool · ImageNet eps=10 | mean perturbed pixels: 537.10 -> 426.77 | Improved |
| add GAN-based distortion map guidance (ablation) | GreedyFool · ImageNet eps=10 | binary classifier accuracy (lower is more invisible): 75.65 -> 68.50 | Improved |
| add distortion map guidance (ablation) | GreedyFool · ImageNet eps=10 | mean perturbed pixels: 537.10 -> 607.22 | Hurt |
| use gradient magnitude and sign instead of only gradient sign (ablation) | GreedyFool · ImageNet eps=10 | mean perturbed pixels: 1149.39 -> 537.10 | Improved |
| use GreedyFool instead of SparseFool for non-target sparse attack | GreedyFool · ImageNet eps=255 | median perturbed pixels: 80.50 -> 27.00 | Improved |
| use GreedyFool instead of PGD0 for target attack with 200 pixel budget on ImageNet | GreedyFool · ImageNet target attack eps=255 | success rate (%): 5.56 -> 15.52 | Improved |
| use GreedyFool instead of PGD0 for single-pixel non-target attack on CIFAR10 | GreedyFool · CIFAR10 eps=255 m=1 | fooling rate (%): 15.22 -> 12.41 | Hurt |
| increase confidence factor kappa from 0 to 6 in the Increasing stage only (ablation) | GreedyFool · ImageNet black-box transfer DenseNet161->ResNet50 | fooling rate (%): 17.00 -> 32.33 | Improved |
| use GreedyFool instead of PGD0 on ImageNet runtime | GreedyFool · ImageNet | seconds per image: 3.232 -> 5.253 | Hurt |
| use GreedyFool instead of SparseFool for invisibility against SRM detector | GreedyFool · ImageNet eps=255 | SRM detection rate (%): 74.36 -> 61.54 | Improved |
Later papers that compared with this method
2 later accepted papers report GreedyFool (κ=15) as a baseline in their results tables; 2 of them report a better number on most of the results they share with it.
- Transferable Sparse Adversarial Attack (CVPR 2022 · better on 12 of 14 shared results)
- Sparse and Imperceptible Adversarial Attack via a Homotopy Algorithm (ICML 2021 · better on 6 of 6 shared results)
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.
- SparseFool — SparseFool: A Few Pixels Make a Big Difference (CVPR 2019) · better on 9 of 9 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
- SuperDeepFool: a new fast and accurate minimal adversarial attack (NeurIPS 2024 · similarity 0.88)
- Towards Efficient Training and Evaluation of Robust Models against $l_0$ Bounded Adversarial Perturbations (ICML 2024 · similarity 0.87)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.87)
- GSE: Group-wise Sparse and Explainable Adversarial Attacks (ICLR 2025 · similarity 0.86)
- $\sigma$-zero: Gradient-based Optimization of $\ell_0$-norm Adversarial Examples (ICLR 2025 · similarity 0.86)
- Transferable Structural Sparse Adversarial Attack Via Exact Group Sparsity Training (CVPR 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.
