摘要： 联邦学习通过让用户使用本地数据训练模型来保护隐私，但现有的工作普遍忽略了真实场景下参与方的异质性。针对传统联邦学习难以避免与不恰当参与方共享模型的问题，提出一种联邦学习参与方选择方案。设计新的可撤销的密文策略属性基加密算法，在不泄露隐私的前提下对参与方实现高效的访问控制。对所提出的方案进行安全性的深入讨论，同时使用公开的数据集进行模拟实验，结果表明所提方案在提供可靠参与方选择应用的同时极大提升了模型的性能，能促进联邦学习在智慧城市等场景中的广泛落地。

中图分类号：TP181
文献标识码：A
DOI:10.19358/j.issn.2097-1788.2023.04.011
引用格式：邱浩宸,张信明.基于属性加密的联邦学习参与方选择方案［J］.网络安全与数据治理，2023，42（4）：62-69.

Client selection scheme for federated learning with attributebased encryption

Abstract： Federated learning protects privacy by enabling users to train models with local data, but existing works has generally ignored applicability in realworld scenarios. In view of the problem that traditional federated learning frameworks cannot avoid sharing models with inappropriate participants, an efficient federated learning participant selection scheme is proposed. A new ciphertextpolicy attribute based encryption with attribute revocation is designed to achieve efficient access control for participants without disclosing privacy. The security of the proposed scheme is discussed in depth, and the simulation experiment is carried out with open datesets. The results show that the scheme greatly improves the performance of the model while providing reliable participant selection applications, which can promote the widespread implementation of federated learning in smart cities and other scenarios.

Key words : attribute-based encryption; federated learning; attribute revocation