Research map / Adversarial robustness and security / Certified and empirical robustness
ScaleCert: Scalable Certified Defense against Adversarial Patches with Sparse Superficial Layers
Husheng Han, Kaidi Xu, Xing Hu, Xiaobing Chen, Ling Liang, Zidong Du, Qi Guo, Yanzhi Wang, Yunji 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 |
|---|---|---|---|
| use SIN-based sparse pruning and certified detection | ScaleCert · ImageNet, 1% patch | certified accuracy: 36.3 -> 60.4 | Improved |
| use SIN-based sparse pruning and certified detection | ScaleCert · ImageNet, 2% patch | certified accuracy: 33.9 -> 55.4 | Improved |
| use SIN-based sparse pruning and certified detection | ScaleCert · ImageNet, 3% patch | certified accuracy: 31.1 -> 52.8 | Improved |
| use ScaleCert instead of MR | certified defense on CIFAR-10 · CIFAR-10, 0.4% patch | certified accuracy: 82.5 -> 81.0 | Hurt |
| use ScaleCert instead of MR | certified defense on CIFAR-10 · CIFAR-10, 2.4% patch | certified accuracy: 78.1 -> 75.3 | Hurt |
| reduce top-ranking rate from 20% to 10% (ablation) | ScaleCert · ImageNet, 1% patch | certified accuracy: 60.4 -> 56.0 | Hurt |
| reduce top-ranking rate from 20% to 15% (ablation) | ScaleCert · ImageNet, 1% patch | certified accuracy: 60.4 -> 58.4 | Hurt |
| reduce top-ranking rate from 20% to 10% (ablation) | ScaleCert · ImageNet, 2% patch | certified accuracy: 53.3 -> 51.8 | Hurt |
| reduce top-ranking rate from 20% to 15% (ablation) | ScaleCert · ImageNet, 2% patch | certified accuracy: 53.3 -> 54.9 | Improved |
| reduce top-ranking rate from 20% to 10% (ablation) | ScaleCert · ImageNet, 3% patch | certified accuracy: 51.6 -> 48.6 | Hurt |
| reduce top-ranking rate from 20% to 15% (ablation) | ScaleCert · ImageNet, 3% patch | certified accuracy: 51.6 -> 49.6 | Hurt |
| set overlap ratio threshold to 0.3 rather than 0.5 (ablation) | ScaleCert · ImageNet, 2% patch | certified accuracy: 55.2 -> 56.6 | Improved |
Later papers that compared with this method
1 later accepted paper reports ScaleCert as a baseline in their results tables; 1 of them report a better number on most of the results they share with it.
- VIP: Unified Certified Detection and Recovery for Patch Attack with Vision Transformers (ECCV 2022 · better on 2 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
- Certified Defenses for Adversarial Patches (ICLR 2020 · similarity 0.92)
- (De)Randomized Smoothing for Certifiable Defense against Patch Attacks (NeurIPS 2020 · similarity 0.90)
- VIP: Unified Certified Detection and Recovery for Patch Attack with Vision Transformers (ECCV 2022 · similarity 0.90)
- PAD: Patch-Agnostic Defense against Adversarial Patch Attacks (CVPR 2024 · similarity 0.88)
- A Unified, Resilient, and Explainable Adversarial Patch Detector (CVPR 2025 · similarity 0.87)
- On the Scalability of Certified Adversarial Robustness with Generated Data (NeurIPS 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.
