Learning Rate Optimization for Federated Learning Exploiting Over-the-Air Computation

Federated learning (FL) as a promising edge-learning framework can effectively address the latency and privacy issues by featuring distributed at devices model aggregation in central server. In order to enable efficient wireless data aggregation, over-the-air computation (AirComp) has recently attracted great attention. However, fading of channels produce aggregate distortions an AirComp-based FL scheme. this paper, we propose modified federated averaging (FedAvg) algorithm introducing local rates present convergence analysis. To combat distortion, rate is optimized adapt channel, which termed dynamic (DLR). We begin our discussion considering multiple-input-single-output (MISO) scenario, since underlying optimization problem convex closed-form solution. Our studies are extended more general multiple-input-multiple-output (MIMO) case iterative method derived. also asymptotic analysis give near-optimal receive beamforming solution when number antennas approaches infinity. Extensive simulation results demonstrate effectiveness proposed DLR scheme reducing distortion guaranteeing testing accuracy on MNIST CIFAR10 datasets. addition, close-form verified through numerical simulations.