Topology-Independent Robustness of the Weighted Mean under Label Poisoning Attacks in Heterogeneous Decentralized Learning
拓扑无关的加权平均在异构去中心化学习中对抗标签污染攻击的鲁棒性
机构 * School of Computer Science and Engineering, Sun Yat-Sen University(中山大学计算机科学与工程学院) ; School of Engineering and Applied Science, Harvard University(哈佛大学工程与应用科学学院) ; Department of Electrical and Electronic Engineering, Imperial College London(伦敦帝国理工学院电子与电气工程系)
AI总结 本文研究了在异构去中心化学习中,加权平均聚合器在标签污染攻击下的鲁棒性,发现其在特定网络拓扑条件下可优于鲁棒聚合器。