AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

LIRA: Learnable, Imperceptible and Robust Backdoor Attacks

Khoa Doan, Yingjie Lao, Weijie Zhao, Ping Li

ICCV 2021 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 233 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 LIRA learned trigger instead of WaNet warping triggerbackdoor attack · GTSRB human inspection (backdoor images)fooling rate (%): 38.6 -> 60.8Improved
use LIRA instead of WaNetbackdoor attack · GTSRB human inspection (both)fooling rate (%): 28.0 -> 50.4Improved
use LIRA instead of WaNetbackdoor attack · CIFAR10 (all-to-one)attack success rate: 0.99 -> 1.00Improved
use LIRA instead of WaNetbackdoor attack · MNIST (all-to-all)attack success rate: 0.95 -> 0.99Improved
use LIRA instead of WaNetbackdoor attack · CIFAR10 (all-to-all)attack success rate: 0.93 -> 0.94Improved
use LIRA instead of WaNetbackdoor attack · T-ImageNet (all-to-all)attack success rate: 0.58 -> 0.59Improved
use LIRA instead of WaNetbackdoor attack · GTSRB (all-to-one)clean accuracy: 0.99 -> 0.99No clear effect
use autoencoder versus U-Net trigger generator (ablation)LIRA · trigger generatorattack performanceNo clear effect
vary epsilon in range 0.001 to 0.1 (ablation)LIRA · all datasetsattack performance: 0.001 to 0.1No clear effect
weight alpha larger than beta (clean over backdoor loss) (ablation)LIRA classifier · trainingconvergenceMixed

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.

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