Federated Learning via Intelligent Reflecting Surface

Over-the-air computation (AirComp) based federated learning (FL) is capable of achieving fast model aggregation by exploiting the waveform superposition property multiple access channels. However, performance severely limited unfavorable wireless propagation In this paper, we propose to leverage intelligent reflecting surface (IRS) achieve yet reliable for AirComp-based FL. To optimize performance, formulate an optimization problem that jointly optimizes device selection, beamformer at base station (BS), and phase shifts IRS maximize number devices participating in each communication round under certain mean-squared-error (MSE) requirements. tackle formulated highly-intractable problem, a two-step framework. Specifically, induce sparsity selection first step, followed solving series MSE minimization problems find maximum feasible set second step. We then alternating framework, supported difference-of-convex-functions programming algorithm low-rank optimization, efficiently design beamformers BS IRS. Simulation results will demonstrate our proposed deployment can lower training loss higher FL prediction accuracy than baseline algorithms.