AI Paper Atlas

Research map / Adversarial robustness and security / Certified and empirical robustness

CertMask: Certifiable Defense Against Adversarial Patches via Theoretically Optimal Mask Coverage

Xuntao Lyu, Ching-Chi Lin, Abdullah Al Arafat, Georg von der Brüggen, Jian-Jia Chen, Zhishan Guo

AAAI 2026 · Adversarial robustness and security › Certified and empirical robustness · approach: Provable defenses and their limits · 0 citations (OpenAlex)

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 changedWhereMeasuredEffect
use CertMask instead of PatchCleanserViT · ImageNet 2% patchcertified robust accuracy: 62.1 -> 75.5Improved
use CertMask instead of PatchCleanserViT · CIFAR-10 2.4% patchcertified robust accuracy: 89.1 -> 96.4Improved
use CertMask instead of PatchCleanserViT · ImageNette 2% patchcertified robust accuracy: 96.4 -> 99.0Improved
use CertMask instead of PatchCleanserResNet · ImageNette 2% patchcertified robust accuracy: 94.4 -> 98.9Improved
use CertMask instead of PG-DSDS · ImageNet 2% patchcertified robust accuracy: 15.7 -> 75.5Improved
use CertMask instead of PatchCleanserViT · ImageNet 2% patchclean accuracy: 83.9 -> 84.5No clear effect
use Offset Tiling instead of Replicated Tiling (ablation)ViT · ImageNet 3% patchcertified robust accuracy: 78.6 -> 74.0Hurt
use Offset Tiling instead of Replicated Tiling (ablation)ResNet · ImageNet 3% patchcertified robust accuracy: 73.7 -> 68.5Hurt
use Offset Tiling instead of Replicated Tiling (ablation)MLP · CIFAR-10 2.4% patchclean accuracy: 95.7 -> 98.0Improved
increase coverage k from 1 to 6 (ablation)ViT · ImageNet 3% patchcertified robust accuracy: 78.6 -> 74.0Hurt
increase coverage k from 1 to 6 (ablation)ViT · ImageNet 3% patchclean accuracy: 82.0 -> 84.5Improved
increase mask size from 40px to 120px (ablation)ViT · ImageNet (k=6)certified robust accuracy: 76 -> 62Hurt

Closest papers by meaning

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.

Open scime draft

Get the next scime Dispatch and a monthly field update

Free. We send a confirmation link first; every email has a one-click unsubscribe. Privacy