Joint Resource Allocation for Efficient Federated Learning in Internet of Things Supported by Edge Computing
Jianyang Ren, Junshuai Sun, Hui Tian, Wanli Ni, Gaofeng Nie, Yingying Wang
Abstract
Federated learning (FL) and edge computing are both important technologies to support the future Internet of Things (IoT). Despite that the network supported by edge computing has great potential to promote FL, it is more challenging to achieve efficient FL due to more complex resource coupling in it. Focus on this problem, we formulate a problem which minimizes the weighted sum of system cost and learning cost by jointly optimizing bandwidth, computation frequency, transmission power allocation and subcarrier assignment. In order to solve this mixed-integer non-linear problem, we first decouple the bandwidth allocation subproblem from the original problem and obtain a closed-form solution. Further considering the remaining joint optimization problem of computation frequency, transmission power and subcarrier, an iterative algorithm with polynomial time complexity is designed. In an iteration, the latency and computation frequency optimization subproblem and transmission power and subcarrier optimization subproblem are solved using the proposed algorithms in turn. The iterative algorithm is repeated until convergence. Finally, to verify the performance of the algorithm, we compare the proposed algorithm with five baselines. Numerical results show the significant performance gain and the robustness of the proposed algorithm.