The use of mobile edge computing is increasingly prevalent, especially in catering to user devices that come with a multitude of sensors. These sensors produce vast amounts of data, like images recording human activities or the real-time locations of vehicles, as seen in smart city scenarios [22, 55]. However, transferring this training data from the user’s device to a server can pose a threat to data privacy. FL is an emerging distributed ML approach that gains traction as a solution to mitigate data privacy concerns [20]. With FL, user devices can jointly train an ML model without having to disclose their private data to a server. The user devices, acting as clients, iteratively train their local models on their private data and send the local model updates to a server.