1（College of Computer Science and Technology， Huaqiao University， Xiamen 361021， China）2（Xiamen Ke Laboratory of Data Security and Blockchain Technology,Huaqiao University, Xiamen 361021,China）3（Institute of Artificial Intelligence and Future Networks， Beijing Normal University， Zhuhai 519000， China）4（Guangdong Key Lab of AI and Multi-Modal Data Processing， BNU-HKBU United International College， Zhuhai 519000， China）
Abstract：A significant number of Internet of Things （IoT） devices upload their data to the cloud， which lead to some serious problem such as network delay and waste of resource. Edge computing， as a new computation paradigm， can solve the above problems. However， edge computing also has some severe problems to been solved， such as user privacy and data security. Federated learning is a new and hot technology of artificial intelligence that can solve the problem of privacy data and data silos. Applying federated learning to the scenarios of edge computing can solve the problem of user privacy. After lots of research， in this paper， the overview of federated learning and federated learning training model based on edge computing is introduced. Then， we compare methods of federated learning application on edge aggregation， edge cache， and computation offloading， point out the problems of the above existing methods， and give the idea to solve the problems. Finally， we give some future researches directions and challenges of federated learning applications on edge computing.