User-Empowered Federated Learning in the Automotive Domain

The proliferation of data generated through everyday device usage has prompted privacy concerns among users. In the automotive sector, this issue is particularly acute, given the substantial volumes of data collected in accordance with manufacturers' privacy policies. Privacy-Enhancing Technologies (PETs), such as Federated Learning (FL), offer a solution by safeguarding the confidentiality of car data while enabling decentralised machine learning model training, thus preventing the need for centralised data aggregation. These FL-based models stand to benefit significantly from the diverse data distributions inherent in training across various features extracted from different cars. However, it remains imperative to ensure user awareness regarding their data processing, despite FL's privacy-preserving mechanisms. To address this, we propose a User-Empowered FL approach, built upon the Flower Framework, empowering users to decide their participation in model training or merely inference without impacting the global model. We demonstrate this approach through an automotive case study utilising the EngineFaultDB dataset. Finally, we outline future directions, particularly focusing on handling unlabelled data through self-supervised learning methodologies.