Malicious Models-based Federated Learning in Fog Computing Networks

【Author】 Huang, Xiaoge; Ren, Yang; He, Yong; Chen, Qianbin

【Source】2022 14TH INTERNATIONAL CONFERENCE ON WIRELESS COMMUNICATIONS AND SIGNAL PROCESSING, WCSP

【影响因子】

【Abstract】In the fog computing network (FCN), due to the characteristics of user dispersion and real-time data, the user privacy is an urgent issue to be solved. Federated learning is considered as a potential solution to avoid data leakage between fog nodes and remote clouds. In this paper, a malicious models-based federated learning framework is proposed, in which the dual-layer blockchain contains the main-blockchain and the directed acyclic graph chain is enabled in the network structure to ensure the data security. Moreover, in federated learning, the reliability of local models cannot be guaranteed, when there are malicious models, the global model accuracy will be decreased. An isolation forest-based malicious model detection algorithm is proposed, which could filter malicious local models and perform global aggregation through the Stochastic Gradient Descent algorithm to ensure the security of the global model. Finally, the simulation results show the effectiveness of the proposed algorithm.

【Keywords】Fog computing network; Federated learning; Blockchain; Directed acyclic graph

【发表时间】2022

【收录时间】2023-06-12

【文献类型】实验仿真

【主题类别】

区块链技术-协同技术-雾计算