Federated learning in wireless networks

Abstract

Wireless devices, including mobile phones, laptops, autonomous driving, UAT, etc., generate an enormous amount of local data, which could be employed to train machine learning models for decision-making and improve prediction accuracy. The major problem is that these data sets are localized among the users, and using data sets for training ML Models could threaten Privacy, Security, and bandwidth constraints. Federated Learning encounters these problems by replacing the centralized training of the ML Model on a server with decentralized training on localized multi-user data sets. Our work is based upon employing a real-life server-client hardware setup using USRP’s to test the model’s accuracy, in which the clients can transmit the generated ML Parameters(Weights and Bias) to the Server.