ABSTRACT
Machine Learning (ML) models are being deployed in a wide range of domains owing to their capacity to deliver high performance across a range of challenging tasks including safety-critical and privacy-sensitive applications. Moreover, the computing requirements of increasingly complex ML models presents a significant challenge to the hardware industry.
Against this backdrop, Federated Learning (FL) has emerged as a promising technique that enables privacy-preserving development of ML models on low-energy Edge devices. FL is a distributed approach that enables learning from data belonging to multiple participants, without compromising privacy since user data are never directly shared. Instead, FL relies on training a global model by aggregating knowledge from local models. Despite its reputation as a privacy-enhancing strategy, recent studies reveal its susceptibility to sophisticated attacks that can undermine integrity and, as well as disrupt their operations. Notably, the constraints posed by the limited hardware resources in edge devices compound these challenges. Gaining insight into these potential risks and exploring hardware-friendly solutions is vital for effectively implementing trustworthy and power-efficient FL systems in edge environments.
This chapter contributes a review and perspective of the triad of privacy, security, and hardware optimization in FL settings.
