Realization for Implementing Federated Learning Based Intrusion Detection with Non-IID IoT Datasets
Ping-Yen Lin, Lin-Huang Chang, Tsung-Han Lee · 2023
With the advance of technology, there is an increasing variety of Internet of Things (IoT) devices, which could be vulnerable to hacker attacks. Therefore, intrusion detection for these IoT devices is crucial. Federated learning, a novel distributed learning approach, which trains multiple clients together and aggregates their weights through a server to achieve collaborative learning. These weights do not disclose the clients' individual dataset information, thereby preserve client privacy while benefiting from the knowledge of other clients. As the intrusion detection datasets for IoT devices come from different domains, the attack categories in these datasets are often highly dissimilar. Therefore, it is necessary to address the Non-Independent and Identically Distributed (non-IID) nature of the data in federated learning. In this paper, we propose three methods to handle the issues for different numbers of class outputs: maximum value, union, and federated transfer learning.