AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Rethinking the Backdoor Attacks' Triggers: A Frequency Perspective

Yi Zeng, Won Park, Z. Morley Mao, Ruoxi Jia

ICCV 2021 · Adversarial robustness and security › Backdoor and poisoning · approach: Trigger-based backdoor attacks · 207 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
detect in DCT frequency domain instead of image domainbackdoor detector · CIFAR-10, BadNetsBDR (%): 1.38 -> 90.50Improved
detect in DCT frequency domain instead of image domainbackdoor detector · GTSRB, Troj-WMBDR (%): 31.51 -> 100Improved
detect in DCT frequency domain instead of image domainbackdoor detector · CIFAR-10, l0 invBDR (%): 0.00 -> 100Improved
fine-tune CIFAR-10 detector with 200 GTSRB-distribution samplesDCT detector · GTSRB, transfer from CIFAR-10ACC (%): 68.23 -> 89.44Improved
fine-tune CIFAR-10 detector with 200 GTSRB-distribution samplesDCT detector · GTSRB, Blend triggerBDR (%): 96.18 -> 45.67Hurt
fine-tune CIFAR-10 detector with 100 TSRD clean samples (size 200)DCT detector · TSRD, BadNetsACC (%): 61.17 -> 82.53Improved
train detector on combined CIFAR-10 and GTSRB instead of GTSRB aloneDCT detector · TSRD, Blend triggerBDR (%): 88.94 -> 89.68Improved
add combined training data then tune on TSRD, versus no tuningDCT detector · TSRD, Blend triggerACC (%): 64.18 -> 84.61Improved
use bilevel smooth trigger instead of naive low-pass filtered triggersbackdoor attack · CIFAR-10ASR (%): 75.54 -> 97.25Improved
use bilevel smooth trigger instead of naive low-pass filtered triggersbackdoor attack · CIFAR-10clean accuracy (%): 76.29 -> 84.54Improved
upgrade Meta Neural Analysis to consider smooth trigger generationMNA defense · smooth-trigger poisoned classifiersAUC: 0.0776 -> 0.694Improved
fine-tune detector with the smooth trigger instead of simple filtered patternsDCT detector · unseen low-pass filtered triggersdetection rate (%): 89.37Improved

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