AI Paper Atlas

Research map / Federated learning and privacy / Privacy attacks and secure learning

Eliminate Distance Differences Induced by Backdoor Attacks: Layer-Selective Training and Clipping to Mask Backdoor Models

Xuzeng Li, Tao Zhang, Xiangyun Tang, Jiacheng Wang, Jian Wang, Jiawen Kang, Jiqiang Liu, Zhen Han, Dusit Niyato, Dong In Kim

CVPR 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Defending federated learning against poisoning

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
add layer-selective training with constraint loss and clipping (LaySelFL) to BadnetBadnet · CIFAR10, Multi-Krum defense, ResNet18ASR (%): 3.0 -> 53.0Improved
add LaySelFL to A3FLA3FL · CIFAR10, Multi-Krum defense, ResNet18ASR (%): 1.0 -> 99.8Improved
add LaySelFL to A3FLA3FL · CIFAR10, Multimetric defense, ResNet18ASR (%): 1.4 -> 76.2Improved
add LaySelFL to Badnet against undefended and similarity/direction defensesBadnet · CIFAR10, AlignIns/Foolsgold/FedAvg, ResNet18ASR (%): 97.7 -> 34.2 (AlignIns)Hurt
add LaySelFL to A3FL under RFAA3FL · SVHN, RFA defense, ResNet18ASR (%): 100 -> 0.7Hurt
add LaySelFL to attacks overallBadnet and A3FL · three datasets, five defensesshare of attacks prevented by defenses: 26.6% -> 4%Improved
use EfficientNet instead of ResNet18LaySelFL · A3FL, CIFAR10, AverageASR (%): 99.5 -> 61.3Hurt
add targeted clipping to sensitive-layer constrained training (ablation)SLI+CT · CIFAR10, Badnet, Multi-Krum, ResNet18ASR (%): 34.9 -> 54.2Improved
add targeted clipping to sensitive-layer constrained training (ablation)SLI+CT · SVHN, A3FL, AlignIns, ResNet18ASR (%): 39.5 -> 22.1Hurt
add constrained training to SLI+TC (ablation)SLI+TC · FMNIST, A3FL, Multi-Krum, ResNet18ASR (%): 1.4 -> 100Improved
add constrained training to SLI+TC (ablation)SLI+TC · CIFAR10, Badnet, FedAvg, ResNet18ASR (%): 97.3 -> 36.4Hurt
increase number of selected sensitive layers (ablation)LaySelFL · A3FL, ResNet18ASR (%)Improved

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