Given the exponential emergence of connected objects, intelligent systems and massive demands for connection to the Internet network is increasingly evolving, which generates security problems of work traffic that is routed in quantity via the WAN network precisely called the Internet, others such as bandwidth pressure and bottlenecks and latency that must be performed and improved in order to do a cyclical training of learning model very quickly. For these reasons that are mentioned in the first paragraph, it is necessary to implement various resilient security strategies and relevant access control management to the various resources, whether hardware or software that are part of the IOT infrastructure and even the smart city to deal with malicious programs in terms of data loss, theft, falsification of parameters that are essential for the training phase of federated learning model. In this article, we will discuss the different research that has been carried out by researchers with the aim of developing secure, resilient and sustainable federated learning architectures against any anomaly or intruder coming from either inside the IoT network or from outside.