Tutorial on Federated Learning at the Networks of the Future conference 2023 in Izmir
On October 4, our PhD member, Léo Lavaur, had the wonderful opportunity to host a tutorial session on “Federated Learning × Security in Network Management” together with one of his advisors, professor Yann Busnel, at the 14th international conference on Networks of the Future (NoF 2023) in Izmir, Turkey. Feedbacks were great, and the session brought very interesting discussions! All materials are in open access on GitHub!
Abstract
Federated learning (FL) is a machine learning (ML) paradigm that enables distributed agents to learn collaborative models without sharing data. In the context of network security, FL promises to improve the detection and mitigation of attacks, notably by virtually extending the local dataset of each participant. However, one of the major challenges of this recent technology is the heterogeneity of the data used by the participants. Indeed, some participants with very different monitoring contexts could penalize the global model. Furthermore, identifying malicious contributions is made more difficult in heterogeneous environments.
In this tutorial, we will first present the fundamentals of federated learning, then focus on its use in network monitoring, and more specifically, in collaborative intrusion detection (Federated Learning-based Intrusion Detection System—FIDS). Secondly, we will address some of the open research questions in this context, before focusing on the problem of training data heterogeneity. Finally, we will discuss the security of FL architectures, and more specifically, the problem of poisoning attacks. All these parts will be illustrated by hands-on exercises, guided step by step throughout the tutorial.
Léo Lavaur
Léo Lavaur received the engineering degree in information security from the National Engineering School, South Brittany (ENSIBS), Vannes, France, in 2020. He is currently pursuing the Ph.D. degree in cybersecurity with the Engineering School, IMT Atlantique and the Chair on Cybersecurity in Critical Networked Infrastructures (Cyber CNI), Rennes, France. During his studies, he also worked in industry with Orange Cyberdefense as a part-time Employee for three years, where he worked on application security, and Wi-Fi rogue access-point detection and location.
He now studies the collaboration in security systems, and how to share data without compromising security. His current research focuses on the challenges of using federated learning as a framework for collaborative intrusion detection systems. In particular, he works on the detections of malicious contributions in heterogeneous environments, as well as on the creation of datasets for evaluating FIDS in heterogeneous settings.
Léo’s research focuses on federated architectures, intrusion detection, and machine learning. He studies the collaboration in security systems, and how to share data without compromising security. With the chair and its test bed, he works on building a distributed observatory for attack detection, characterization, and mitigation.
- We welcomed the CEO of EDF - October 30, 2023
- Tutorial on Federated Learning at the Networks of the Future conference 2023 in Izmir - October 27, 2023
- IMT Symposium: Léo Lavaur, “Federated Approaches to Defend Cyber-Attacks” - April 13, 2022
