AI Paper Atlas

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

MultiKD: Backdoor Defense in Federated Graph Learning via Attention-Guided Multi-Teacher Distillation

Jiale Zhang, Yanan Wang, Bosen Rao, Chengcheng Zhu, Xiaobing Sun, Yu Li

AAAI 2026 · Federated learning and privacy › Privacy attacks and secure learning · approach: Defending federated learning against poisoning · 0 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
add inter-layer relation consistency to attention alignment (ablation)MultiKD · NCI1 ASRASR: 25.62 -> 1.85Improved
add inter-layer relation consistency to attention alignment (ablation)MultiKD · PROTEINS ASRASR: 42.62 -> 0.54Improved
add attention map alignment to relation consistency (ablation)MultiKD · NCI1 / PROTEINS ASRASR: 17.70 -> 1.85 (NCI1); 14.12 -> 0.54 (PROTEINS)Improved
add attention map alignment to relation consistency (ablation)MultiKD · DD ASRASR: 15.26 -> 6.84Improved
add attention alignment to relation consistency on clean accuracy (ablation)MultiKD · NCI1 ACCACC: 63.49 -> 62.07Hurt
server-side multi-teacher distillation purification vs FedTGE under UGBAFGL defense · PROTEINS UGBA ASRASR: 72.91 -> 0.12Improved
MultiKD under non-IID splits vs FedTGEFGL defense · NCI1 non-IID UGBA ASRASR: 96.45 -> 1.03Improved
over-weight attention loss relative to relation loss (high alpha, low beta) (ablation)MultiKD · loss weights heatmapASRHurt
increase clean data ratio on server (ablation)MultiKD · NCI1ASR / ACCImproved

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