I. Introduction
Recently, federated learning (FL) as a way to train analytical models in distributed intrusion detection systems has attracted a lot of research attention [1]–[4]. This computational paradigm may benefit the privacy of the data used to train the model, and provides a natural way to extend the training data set in a privacy-preserving manner. The experiments showed that attack detection models trained in federated mode on multiple datasets have a higher detection rate of previously unknown attacks when compared to models trained on one dataset.