Federated Learning of User Mobility Anomaly Based on Graph Attention Networks
Yingjie Tang, Jia Ruijuan, Xiaoming Zhou, Li Zhao, Hao Jin, Chenglin Zhao · 2021 7th International Conference on Computer and Communications (ICCC) · 2021
In recent years, more and more applications require low latency and high processing capacity. The computing capacity of mobile devices limits their performance to support compute- intensive applications. Opportunistic computation offloading reduces computing burden and prolong battery life of mobile devices. However, the computation resource usage on edge servers becomes unbalanced due to user mobility. Therefore, it is important to detect user mobility anomaly in order to improve computation resource usage efficiency on edge servers.In this paper, user mobility anomaly detection is investigated in the scenario of radio access network. Considering the impact of user mobility privacy and communication overhead on learning data, an algorithm is proposed named Graph Attention Network based User Mobility Anomaly Representation and Detection based on Federated Learning (GUMARD-FL). The simulation results show that the accuracy of GUMARD-FL is close to that of GUMARD-CL, and the best neighborhood value ξ is less than 5.