【Author】
Sun, Jin; Wu, Ying; Wang, Shangping; Fu, Yixue; Chang, Xiao
【Source】IEEE COMMUNICATIONS LETTERS
【Abstract】Federated learning is an emerging technology that solves the privacy problem of training data in multi-party machine learning. However, this technology is vulnerable to a series of system security problems. In this letter, we leverage Hyperledger Fabric permissioned blockchain architecture to build a secure and reliable federated learning platform across multiple data owners, where individual local updates are encrypted based on threshold homomorphic encryption and then recorded on a distributed ledger. The security analysis shows that our solution can effectively deal with the existing privacy and security issues in the federated learning system. The numerical results show that the scheme is feasible and efficient.
【Keywords】Blockchains; Security; Fabrics; Collaborative work; Peer-to-peer computing; Data models; Privacy; Federated learning; blockchain; homomorphic encryption; security and privacy
【摘要】联邦学习是解决多方机器学习中训练数据隐私问题的新兴技术。然而，该技术容易受到一系列系统安全问题的影响。在这封信中，我们利用 Hyperledger Fabric 许可的区块链架构来构建跨多个数据所有者的安全可靠的联邦学习平台，其中基于阈值同态加密对单个本地更新进行加密，然后记录在分布式账本上。安全性分析表明，我们的解决方案可以有效处理联邦学习系统中存在的隐私和安全问题。数值结果表明该方案是可行且有效的。
【关键词】区块链；安全;面料；协作工作；对等计算；数据模型；隐私;联邦学习；区块链；同态加密；安全和隐私
评论