Demo: Edge Federated Learning over a LoRa Mesh Network

Training machine learning models in resource-constrained edge devices using federated learning is not yet deeply explored by the research community. There are important technical challenges, such as the limited computing capacity of microcontrollers, and the restricted network connectivity and bandwidth of IoT environments to coordinate and execute the federated learning process. In this paper, we demonstrate a prototype implementation of a federated learning model trained over a LoRa mesh network. We leverage the LoRaMesher library to provide the LoRa mesh network service and integrate it into a keyword spotting federated learning application that runs on a microcontroller-based IoT board. The experimentation with the prototype demonstrates the recent vision of providing edge intelligence with low-capacity networked adaptive microcontroller-based nodes.