Privacy protection federated learning system based on blockchain and edge computing in mobile crowdsourcing
Weilong Wang,Yingjie Wang,Yan Huang,Chunxiao Mu,Zice Sun,Xiangrong Tong,Zhipeng Cai,Weilong Wang,Yingjie Wang,Yan Huang,Chunxiao Mu,Zice Sun,Xiangrong Tong,Zhipeng Cai
DOI: https://doi.org/10.1016/j.comnet.2022.109206
IF: 5.493
2022-10-09
Computer Networks
Abstract:With the rapid popularization and development of the Internet of Things (IoT) and 5G networks, mobile crowdsourcing (MCS) has become an indispensable part in today’s society. However, when task participants submit tasks, they are likely to expose their data privacy and location privacy. These privacy will be maliciously attacked and exploited by attackers (external attackers and untrusted third party). With the rapid increase of MCS data throughput, traditional cloud platforms can no longer meet the huge data processing needs. To solve these problems, this paper proposes an MCS federated learning system based on Blockchain and edge computing. This paper uses federated learning as the framework of the MCS system. The system protects data privacy and location privacy by using the Double local disturbance Localized Differential Privacy (DLD-LDP) proposed in this paper. Because the sensed data exists in multiple modalities (text, video, audio, etc.), this paper uses the Multi-modal Transformer (MulT) method to merge the multi-modal data before subsequent operations. To solve the problem that the third party is untrusted, we utilize Blockchain to distribute tasks and collect models in a distributed way. A reputation calculation method (Sig-RCU) is proposed to calculate the real-time reputation of task participants. Through conducting experiments on real data sets, the effectiveness and adaptation of the proposed DLD-LDP algorithm and Sig-RCU algorithm are verified.
computer science, information systems,telecommunications,engineering, electrical & electronic, hardware & architecture