BriefGPT.xyz
Nov, 2023
FedTruth: 拜占庭容错和后门抵制的联邦学习框架
FedTruth: Byzantine-Robust and Backdoor-Resilient Federated Learning Framework
HTML
PDF
Sheldon C. Ebron Jr., Kan Yang
TL;DR
FedTruth是一种针对FL中的模型污染问题的鲁棒防御方法，通过动态聚合权重估计全局模型更新，考虑了所有良性客户的贡献，并在实证研究中证明了其对拜占庭攻击和后门攻击的毒化更新的影响有很好的缓解效果。
Abstract
federated learning
(FL) enables collaborative machine learning
model training
across multiple parties without sharing raw data. However, FL's distributed nature allows
→