Towards Secure and Robust Federated Learning
Federated Learning (FL) is a new way to train machine learning models across many decentralized devices while keeping data private. In Intrusion Detection Systems (IDS), FL allows different networks to share security intelligence without sharing sensitive raw data. However, this decentralized structure makes the system vulnerable to security risks, especially poisoning attacks. In these attacks, malicious clients provide fake updates to damage the global model's accuracy. This paper studies how to defend FL-based IDS against these threats. We evaluate two main defenses: robust aggregation techniques and client selection strategies. Our experiments show that these methods successfully maintain high IDS performance even when many clients are compromised. The findings offer a practical guide for building secure and privacy-preserving intrusion detection systems