A value-added IoT service for cellular networks using federated learning

Abstract

The number of Internet-of-Things (IoT) devices is expected to reach 64 billion by 2025. These IoT devices will mostly use cellular networks for transferring a huge amount of IoT data to the cloud for machine learning (ML) based forecasting. Keeping in view a large number of application scenarios for highly resource constraint IoT devices connected with the cellular networks, we propose a value-added IoT service (VAIS) for the cellular network operators based on the federated learning (FL) paradigm. Through simulation experiments, we show, for real air quality data and specific ML models, the proposed VAIS reduces the backhaul data by 70× and requires less energy than its equivalent cloud-based conventional approach, however, with a slight increase in communication time. From the insights we gained, we believe that a properly designed VAIS would efficiently utilize network resources, not only reduce management for IoT users but also the operating costs for cellular operators, and encourage IoT applications on limited backhaul cellular networks.