AI Paper Atlas

Research map / Adversarial robustness and security / Backdoor and poisoning

Progressive Backdoor Erasing via Connecting Backdoor and Adversarial Attacks

Bingxu Mu, Zhenxing Niu, Le Wang, Xue Wang, Qiguang Miao, Rong Jin, Gang Hua

CVPR 2023 · Adversarial robustness and security › Backdoor and poisoning · approach: Training-time backdoor defense · 18 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 progressive adversarial fine-tuning and clean-image filtering to erase the backdoorPBE · CIFAR-10, Badnet, all-to-oneASR: 100.00 -> 1.09Improved
Use PBE with a clean extra dataset rather than PBE without onePBE · CIFAR-10, Badnet, all-to-oneASR: 11.30 -> 1.09Improved
Use PBE without a clean extra datasetPBE · CIFAR-10, Badnet, all-to-oneASR: 100.00 -> 11.30Improved
Use PBE without a clean extra datasetPBE · CIFAR-10, Blend, all-to-oneASR: 100.00 -> 1.16Improved
Use PBE without a clean extra datasetPBE · CIFAR-10, Badnet, all-to-oneASR: 100.00 -> 11.30Mixed
Use PBE with a clean extra datasetPBE · CIFAR-10, WaNet, all-to-oneASR: 99.50 -> 0.46Improved
Use PBE with a clean extra datasetPBE · GTSRB, Badnet, all-to-oneASR: 100.00 -> 0.47Improved
Use PBE with a clean extra datasetPBE · Tiny-ImageNet, WaNet, all-to-oneASR: 99.92 -> 5.32Improved
Use PBE with a clean extra datasetPBE · CIFAR-10, Badnet, all-to-allASR: 94.41 -> 0.62Improved
Fine-tune on adversarial examples generated by untargeted attacksInfected model · Backdoor attacks across tested datasetsASRImproved
Progressively update the extra dataset by ranking training images against the infected and purified modelsPBE · CIFAR-10 Blend attackAverage precision: 0.25 -> 0.93Improved
Plant a backdoor in the modelClassifier adversarial examples · CIFAR-10, five backdoor attacksFeature L2 distance to triggered-image features: 102.58 -> 85.11; 135.91 -> 78.18; 124.22 -> 75.09; 40.42 -> 28.66; 48.13 -> 15.85Improved

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