基于差分隐私的联邦学习数据隐私安全技术

Privacy-preserving Method of Federated Learning Based on Differential Privacy

联邦学习方法在大数据时代有效解决了“数据孤岛”问题,也在一定程度上保障了数据隐私安全。然而,联邦学习的许多方面仍面临隐私风险。首先归纳总结了联邦学习面临的常见隐私威胁,并针对不同类型的隐私威胁归纳出对应的隐私保护措施;其次重点针对差分隐私方法进行了探讨,归纳总结了一些差分隐私的实现方法;最后基于差分隐私设计了一种适用于联邦学习系统的隐私保护手段。

Federated learning method effectively solves the problem of data isolation in the era of big data,and also ensures the security of data privacy to a certain extent. However, there are still many privacy risks associated with federated learning. First,this paper summarizes the common privacy threats in federated learning system, and summarizes the corresponding privacy protection measures for different types of privacy threats. Secondly,this paper mainly focuses on differential privacy method, and summarizes the concrete mechanism of differential privacy. Finally, this paper comes up with a privacy protection method based on the differential privacy, which can fit in federated learning system.