Over-the-Air Federated Learning In Broadband Communication
This addresses privacy and efficiency issues in federated learning for wireless edge applications, but appears incremental as it builds on existing FL and MIMO concepts.
The paper tackles the limitations of current federated learning approaches, such as vulnerability to inference attacks and decreased test accuracy with differential privacy, by proposing a novel integration of federated learning into MIMO systems, though no concrete results or numbers are provided.
Federated learning (FL) is a privacy-preserving distributed machine learning paradigm that operates at the wireless edge. It enables clients to collaborate on model training while keeping their data private from adversaries and the central server. However, current FL approaches have limitations. Some rely on secure multiparty computation, which can be vulnerable to inference attacks. Others employ differential privacy, but this may lead to decreased test accuracy when dealing with a large number of parties contributing small amounts of data. To address these issues, this paper proposes a novel approach that integrates federated learning seamlessly into the inner workings of MIMO (Multiple-Input Multiple-Output) systems.