Toward Multiple Federated Learning Services Resource Sharing in Mobile Edge Networks

Federated Learning is a new learning scheme for collaborative training a\nshared prediction model while keeping data locally on participating devices. In\nthis paper, we study a new model of multiple federated learning services at the\nmulti-access edge computing server. Accordingly, the sharing of CPU resources\namong learning services at each mobile device for the local training process\nand allocating communication resources among mobile devices for exchanging\nlearning information must be considered. Furthermore, the convergence\nperformance of different learning services depends on the hyper-learning rate\nparameter that needs to be precisely decided. Towards this end, we propose a\njoint resource optimization and hyper-learning rate control problem, namely\nMS-FEDL, regarding the energy consumption of mobile devices and overall\nlearning time. We design a centralized algorithm based on the block coordinate\ndescent method and a decentralized JP-miADMM algorithm for solving the MS-FEDL\nproblem. Different from the centralized approach, the decentralized approach\nrequires many iterations to obtain but it allows each learning service to\nindependently manage the local resource and learning process without revealing\nthe learning service information. Our simulation results demonstrate the\nconvergence performance of our proposed algorithms and the superior performance\nof our proposed algorithms compared to the heuristic strategy.\n