AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

WaNet - Imperceptible Warping-based Backdoor Attack

Tuan Anh Nguyen, Anh Tuan Tran

ICLR 2021 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 125 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 elastic image warping as the trigger instead of added patch or noise perturbationsWaNet · GTSRB human inspectionsuccess fooling rate: up to 7.7% -> 28%Improved
Add the noise training mode around the fixed warping field (ablation)WaNet · CIFAR-10, Neural Cleansedefense detectionImproved
Remove the noise training mode (ablation)WaNet · CIFAR-10, Neural Cleansedefense detectionHurt
Use small warping strength or grid size (ablation)WaNet · CIFAR-10backdoor accuracyHurt
Increase warping strength or grid size from small values (ablation)WaNet · CIFAR-10backdoor accuracyImproved
Vary the randomly generated warping field (ablation)WaNet · CIFAR-10clean accuracy: 94.42 ± 0.08No clear effect
Vary the randomly generated warping field (ablation)WaNet · CIFAR-10backdoor accuracy: 99.40 ± 0.21No clear effect
Vary the randomly generated warping field (ablation)WaNet · CIFAR-10noise accuracy: 93.16 ± 0.43No clear effect
Switch from all-to-one to all-to-all target labels (ablation)WaNet · MNIST, CIFAR-10, GTSRB, and CelebAattack accuracyHurt
Use the fixed warp for physical screen recapture attacksWaNet · Physical screen recaptureattack success rate: 96%Improved

Later papers that compared with this method

14 later accepted papers report WaNet as a baseline in their results tables; 12 of them report a better number on most of the results they share with it.

6 more not listed.

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

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