Introducing Federated Learning into Internet of Things Ecosystems – Maintaining Cooperation Between Competing Parties

In practical realizations of a Federated Learning ecosystems, the parties cooperating during the training process, and that later use the trained/global model may consist of competing institutions. This can result in incentives for malicious behavior, which can infringe on the safety and data privacy of other participants. Additionally, even in cases devoid of foul play, the format of the data stored locally, and the equipment available for training, may differ between participating institutions. This necessitates creation of a flexible and adaptable preprocessing pipeline, including a comprehensive registration and data preparation process. Among others, it should identify the affiliation of the joining device(s), maintain appropriate data privacy mechanisms, and compensate for the heterogeneity of the devices that are to participate in model training. In this context, the practical aspects of deploying federated learning solutions, in real-life production environments, are discussed.