Energy Efficient Aspects of Federated Learning – Mechanisms and Opportunities

The role of machine learning in IoT-enabled applications, most predominantly in societal applications, is instrumental. That the active researchers take a series of steps utilizing innovative technologies to override emerging challenges such as data privacy, latency, scalability, energy consumption, and so forth. Federated Learning (FL), a subset of machine learning paradigm, has shifted the mindset of researchers, including system architects, in recent years while solving the existing challenges of the inclusion of machine learning in applications; it has taken a more distinct shape in terms of handling learning mechanisms in a decentralized fashion. In fact, energy-efficient mechanisms need to be incorporated into the FL frameworks/architectures. There is still no work that expresses the energy reduction opportunities of FL algorithms or architectures. This paper has primarily focused on revealing the energy-efficient aspects of FL. Besides, it is reinforced with the discussions on the recent developments of FL and prominent future research contributions in various application domains with an emphasis on energy efficiency. The article would benefit researchers, more specifically the system engineers or tool developers, who deal with the inclusion of FL for applications.