With the rapid evolution of Internet-of-Things (IoT) devices and the emergence of Autonomous Vehicles (AVs), machine learning processes pose a growing privacy issue. Federated learning (FL) and current cryptography can mitigate this problem; however, these solutions might not be efficient enough during the decades-long lifespans of such gadgets.
In this paper, a generalization of FL schemes, incorporating sharing a part of raw data, is presented with a proof-of-concept experiment. Besides anonymization, the exchanged data portion can also directly support real-time decision-making. In contrast with cryptographical approaches, the proposed FL scheme can guarantee a certain level of privacy during the whole lifetime of IoT devices or AVs.