BriefGPT.xyz
Sep, 2023
针对来自投毒的联邦学习的客户端梯度反转
Client-side Gradient Inversion Against Federated Learning from Poisoning
HTML
PDF
Jiaheng Wei, Yanjun Zhang, Leo Yu Zhang, Chao Chen, Shirui Pan...
TL;DR
该研究提出了一种名为Client-side poisoning Gradient Inversion (CGI)的新攻击方法,旨在从聚合的全局模型中恢复训练样本,展示了在有限知识的客户端对聚合全局模型进行恶意操纵的可行性,并且具备对拜占庭-鲁棒聚合规则的隐蔽性。
Abstract
federated learning
(FL) enables distributed participants (e.g., mobile devices) to train a global model without sharing data directly to a central server. Recent studies have revealed that FL is vulnerable to
gradient i
→