Energy Efficiency in Federated Learning: A Survey on Models, Strategies and Perspectives
Selman Sezgin, Kahina Mokrani, Julien Jacques, Sylvain Allio · HAL (Le Centre pour la Communication Scientifique Directe) · 2025
Federated Learning (FL) is an emerging distributed learning technique in which multiple devices optimize a model without transmitting raw data to a third-party server, thus facilitating the protection of user data privacy. In FL, devices train the model locally and share only the model weights for an aggregation step. Despite promising application possibilities in various sectors such as healthcare or smart mobility, optimizing the energy efficiency of FL remains an open issue, driven on the one hand by resource constraints in certain environments such as IoT, and on the other hand by the current increase in the size of learning models, whose training requires significant processing capabilities. In this survey, we analyze the energy challenges of FL that have been identified in the literature. We provide a comprehensive review of energy efficiency strategies for FL and discuss the underlying energy models. A portion of this work is a case study focused on FL in vehicular networks and that conducts an in-depth analysis of the impact of vehicle mobility on the energy cost of FL. Finally, we provide some research directions for the development of sustainable FL solutions.