Mobile Edge Collaborative Computing Technology Based On Federated Learning

With the rapid development of the Internet of Things in recent years,the number of connected smart devices has also increased exponentially,and the resulting applications have increasingly stringent requirements for latency.The emerging Mobile Edge Computing technology can help mobile devices offload tasks that require high precision and low latency to nearby edge servers,speeding up the computing process and alleviating the resource shortage of Io T devices.At the same time,data security and privacy protection issues have gradually become the focus of attention,which also makes federated learning gradually evolve into an important technology in the field of artificial intelligence.In the final analysis,federated learning is also a distributed learning method,so how to reasonably apply federated learning in mobile edge computing scenarios to give full play to its distributed synergy advantages and improve overall efficiency has become a problem that needs to be solved.With the development of mobile edge computing technology,ultra-dense networks and UAV swarms assisted edge computing not only occupy the mainstream position because of their stronger computing resources and processing capabilities,but more importantly,they can group and collaboratively process computing tasks to achieve edge intelligence,which is the most compatible with future technological development.However,as an emerging technology,although federated learning has a high degree of matching with them in terms of scenarios,it still faces many challenges in practical applications,for example,it cannot meet the resource scheduling in task offloading,and it cannot reasonably utilize the collaboration characteristic of edge computing devices.The most prominent one is the barrel effect problem faced by federated learning in complex scenarios,which is the overall efficiency will be affected by the device with the slowest training speed,and this issue is still unresolved in the academic community.This paper studies the application of federated learning in ultra-dense networks and UAV swarm-assisted edge computing scenarios.The main work and innovations are as follows:1.A collaborative federated learning algorithm between servers is proposed for the federated learning optimization problem in the ultra-dense edge computing scenario.The optimization problem is first formally defined as a mixed-integer linear programming problem with two optimization variables,and then heuristics are used to find feasible solutions.At the same time,a series of rigorous theoretical analysises are carried out,which verifies that our algorithm has a significant optimization effect,which provides a strong theoretical guarantee for our proposed algorithm.Extensive numerical results show that our proposed algorithm can greatly reduce the client’s waiting time without affecting the training accuracy,thereby improving the overall training efficiency.2.An intra-UAV grouping algorithm is proposed for the federated learning optimization problem in the scenario of UAV swarm-assisted edge computing.First of all,in order to improve the efficiency of the grouped UAV swarms,the roles of UAVs in each group are divided.Then,in order to make full use of communication and computing resources,the transmission tasks are assigned to some UAVs in each group as relay nodes.The final experimental results verify that our proposed algorithm can greatly reduce the grouping time of UAVs.Moreover,compared with traditional multi-UAV and UAV swarm-based edge computing,the energy consumption of UAVs after grouping is significantly reduced.3.In the scenario of UAV swarm-assisted edge computing,based on the above-mentioned intra-UAV grouping algorithm,a collaborative federated learning algorithm between UAV groups is proposed.Firstly,based on the grouping of UAV swarms,we study the problem of minimizing the overall training delay of UAVs under the constraints of federated learning global accuracy guarantee and energy consumption optimization.Then to solve the proposed problem,a low-complexity joint training optimization algorithm is proposed.The final simulation results verify the effectiveness of the proposed joint training optimization algorithm,and at the same time demonstrate the feasibility of federated learning in the above scenarios.