Quantization-Aware Federated Learning with Coarsely Quantized Measurements

In this work, we present an energy-efficient federated learning framework using coarsely quantized measured data for Internet of Things (IoT) networks. In particular, we develop a quantization-aware Federated Averaging Least Mean Square (QA-FedAvg-LMS) algorithm that can learn parameters in an energy-efficient fashion using measurements quantized with few bits and develop a bias compensation strategy to further improve the performance of the QA-FedAvg-LMS algorithm. We carry out a statistical analysis of the proposed QA-FedAvg-LMS algorithm. Numerical results assess the QA-FedAvg-LMS algorithm against existing techniques for a parameter estimation task in a scenario where IoT devices operate in federated learning mode.