Research map / Adversarial robustness and security / Black-box and transfer attacks
Square Attack: a query-efficient black-box adversarial attack via random search
Maksym Andriushchenko, Francesco Croce, Nicolas Flammarion, Matthias Hein
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 |
|---|---|---|---|
| use square-shaped updates instead of random pixel positions (c signs) (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 339 -> 73 | Improved |
| use square-shaped updates instead of random pixel positions (c*h^2 signs) (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 401 -> 153 | Improved |
| use one sign per color channel instead of independent signs per pixel (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 153 -> 73 | Improved |
| use square instead of rectangle updates (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 93 -> 73 | Improved |
| use vertical stripes initialization instead of uniform random (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 91 -> 73 | Improved |
| use vertical stripes initialization instead of uniform random (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | median queries: 26 -> 11 | Improved |
| use vertical stripes instead of horizontal stripes initialization (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 83 -> 73 | Improved |
| use one sign per color channel instead of a single sign for all elements (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | avg queries: 129 -> 73 | Improved |
| use one sign per square instead of independent signs per element (ablation) | l_2 Square Attack · ImageNet ResNet-50 l_2 eps=5 | avg queries: 1050 -> 616 | Improved |
| use two-center update eta instead of single-center eta (ablation) | l_2 Square Attack · ImageNet ResNet-50 l_2 eps=5 | avg queries: 650 -> 616 | Improved |
| change the l_2 initialization (gaussian, uniform or vertical stripes instead of eta-grid) (ablation) | l_2 Square Attack · ImageNet ResNet-50 l_2 eps=5 | avg queries: 696 / 660 / 655 vs 616 | No clear effect |
| vary initial square fraction p from 0.0125 to 0.4 (ablation) | l_inf Square Attack · ImageNet ResNet-50 l_inf eps=0.05 | failure rate / avg queries: 0.0% failure, below 140 avg queries | No clear effect |
Later papers that compared with this method
4 later accepted papers report Square Attack as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- Query-efficient Attack for Black-box Image Inpainting Forensics via Reinforcement Learning (AAAI 2025 · better on 0 of 9 shared results)
- Reinforcement Learning Platform for Adversarial Black-box Attacks with Custom Distortion Filters (AAAI 2025 · better on 0 of 1 shared result)
- DiffAttack: Evasion Attacks Against Diffusion-Based Adversarial Purification (NeurIPS 2023 · better on 2 of 2 shared results)
- Diversity can be Transferred: Output Diversification for White- and Black-box Attacks (NeurIPS 2020 · better on 1 of 2 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
- Meta-Learning the Search Distribution of Black-Box Random Search Based Adversarial Attacks (NeurIPS 2021 · similarity 0.92)
- Simple Black-box Adversarial Attacks (ICML 2019 · similarity 0.91)
- BRUSLEATTACK: A QUERY-EFFICIENT SCORE- BASED BLACK-BOX SPARSE ADVERSARIAL ATTACK (ICLR 2024 · similarity 0.89)
- Understanding the Robustness of Randomized Feature Defense Against Query-Based Adversarial Attacks (ICLR 2024 · similarity 0.88)
- Efficient Black-box Adversarial Attacks via Bayesian Optimization Guided by a Function Prior (ICML 2024 · similarity 0.87)
- DifAttack: Query-Efficient Black-Box Adversarial Attack via Disentangled Feature Space (AAAI 2024 · similarity 0.87)
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.
