ABSTRACT
Individual and integrated uses of the federated learning (FL) and internet of things (IoT) have lately been applied in numerous network-related situations, and as a result, they have been experiencing a rising interest in the academic research community. In a decentralized fashion, FL solves concerns about the security and privacy of the data generated by the IoT. Additionally, it may be possible to train different learning algorithms using local material, with the exception of transferring data using intelligent algorithms based on artificial intelligence (AI). However, the fact that the models and devices used in complex IoT networks are heterogeneous has severely hampered the capacity of the FL training process to function effectively. As a result, it is practically unsuited for direct distribution via IoT networks, despite the fact that attempts are now being made to address this problem and overcome this difficult challenge. As a consequence of this, the primary features of FL in the internet of things (FL-IoT) are outlined in this paper from the perspectives of both security and privacy. We widen the scope of our research to include the investigation and analysis of algorithms, models, and protocols that are at the cutting edge of FL technology. The primary emphasis is on how effective these algorithms, models, and protocols are when used in a practical setting across IoT-based systems and networks. After this, a comparative study of newly available security solutions for FL-IoT is discussed. These protection solutions may be based on non-cryptographic or cryptographic solutions, and they can be implemented across heterogeneous, dynamic IoT networks.
