12 July 2025 in ACM Computing Surveys
The evolution of cybersecurity is closely linked to the development and improvement of artificial intelligence (AI). As a key tool for realizing more cybersecure ecosystems, Intrusion Detection Systems (IDSs) have evolved tremendously in recent years by integrating machine learning (ML) techniques to detect increasingly sophisticated cybersecurity attacks hidden in big data. However, traditional approaches rely on centralized learning, in which data from end nodes are shared with data centers for analysis. Recently, the application of federated learning (FL) in this context has attracted great interest to come up with collaborative intrusion detection approaches where data does not need to be shared. Due to the recent rise of this field, this work presents a complete, contemporary taxonomy for FL-enabled IDS approaches that stems from a comprehensive survey of the literature from 2018 to 2022. Precisely, our discussion includes an analysis of the main ML models, datasets, aggregation functions, as well as implementation libraries employed by the proposed FL-enabled IDS approaches. On top of everything else, we provide a critical view of the current state of the research around this topic, and describe the main challenges and future directions based on the analysis of the literature and our own experience in this area.
15 February 2024 in Internet of Things
JHJosé L. Hernandez-Ramos
10 October 2024 in IEEE Transactions on Emerging Topics in Computing
Federated Learning (FL) represents a promising approach to typical privacy concerns associated with centralized Machine Learning (ML) deployments. Despite its well-known advantages, FL is vulnerable to security attacks such as Byzantine behaviors and poisoning attacks, which can significantly degrade model performance and hinder convergence. The effectiveness of existing approaches to mitigate complex attacks, such as median, trimmed mean, or Krum aggregation functions, has been only partially demonstrated in the case of specific attacks. Our study introduces a novel robust aggregation mechanism utilizing the Fourier Transform (FT), which is able to effectively handle sophisticated attacks without prior knowledge of the number of attackers. Employing this data technique, weights generated by FL clients are projected into the frequency domain to ascertain their density function, selecting the one exhibiting the highest frequency. Consequently, malicious clients' weights are excluded. Our proposed approach was tested against various model poisoning attacks, demonstrating superior performance over state-of-the-art aggregation methods.
12 March 2025 in International Journal of Information Security
Federated Learning (FL) has become an attractive approach to collaboratively train Machine Learning models while data sources’ privacy is still preserved. However, most of existing FL approaches are based on supervised techniques, which could require resource-intensive activities and human intervention to obtain labelled datasets. Furthermore, in the scope of cyberattack detection, such techniques are not able to identify previously unknown threats. In this direction, this work proposes a novel unsupervised FL approach for the identification of potential misbehavior in vehicular environments. This paper presents a cloud-based approach to detect misbehavior in vehicular networks. Our method combines Gaussian Mixture Models and Variational Autoencoders in an FL setting using the VeReMi dataset, allowing each vehicle to train on its own data while sharing insights through a central repository of anomalous events. We employ Restricted Boltzmann Machines to ensure the convergence of the model and Fed+ aggregation function to improve the performance of the model in non-identical and independently distributed scenarios. Experimental results on the VeReMi dataset show that our framework effectively identifies malicious behaviors, enabling robust, collective defense strategies across multiple vehicles. In particular, our approach provides better performance (more than 80%) compared to recent proposals, which are usually based on supervised techniques and artificial divisions of the VeReMi dataset.