【Author】
Peng, Lei; Yang, Zhixiang; Guo, Shaoyong; Qiu, Xuesong; Li, Wenjing; Yu, Peng
【Abstract】Federated learning (FL) is seen as a road towards privacy-preserving distributed artificial intelligence (AI) while keeping the raw training data on the local device. By leveraging blockchain, this paper puts forward a blockchain and FL fusioned framework to manage the security and trust issues when applying FL over mobile edge networks. First, a two-layered architecture is proposed that consists of two types of blockchains: local model update chain (LMUC) assisted by device-to-device (D2D) communication and global model update chain (GMUC) supporting task sharding. D2D-assisted LMUC is designed to chronologically and efficiently record all of the local model training results, which can help to form a long-term reputation of local devices. The GMUC is proposed to provide both security and efficiency by preventing mobile edge computing (MEC) nodes from malfunctioning and dividing it into logically-isolated FL task-specific chains. Then, a reputation-learning based incentive mechanism is introduced to make the participant local devices more trustful with the reward implemented by a smart contract. Finally, a case study is given to show that the proposed framework performs well in terms of FL learning accuracy and blockchain time delay.
【Keywords】blockchain; federated learning; security; trustful mobile edge network
【摘要】联邦学习 (FL) 被视为通向保护隐私的分布式人工智能 (AI) 的道路，同时将原始训练数据保留在本地设备上。通过利用区块链，本文提出了一种区块链和 FL 融合框架来管理在移动边缘网络上应用 FL 时的安全和信任问题。首先，提出了一种由两种类型的区块链组成的两层架构：由设备到设备（D2D）通信辅助的本地模型更新链（LMUC）和支持任务分片的全局模型更新链（GMUC）。 D2D 辅助的 LMUC 旨在按时间顺序高效记录所有本地模型训练结果，有助于形成本地设备的长期声誉。 GMUC 被提议通过防止移动边缘计算 (MEC) 节点发生故障并将其划分为逻辑隔离的 FL 任务特定链来提供安全性和效率。然后，引入基于信誉学习的激励机制，使参与者本地设备对智能合约实现的奖励更加信任。最后，给出了一个案例研究，表明所提出的框架在 FL 学习准确性和区块链时间延迟方面表现良好。
【关键词】区块链；联邦学习；安全;可信赖的移动边缘网络
评论