Label noise analysis meets adversarial training: A defense against label poisoning in federated learning

Ehsan Hallaji, Roozbeh Razavi-Far, Mehrdad Saif, Enrique Herrera-Viedma

Abstract: Highlights•A novel approach is designed to safeguard deep learners against label poisoning.•We propose formulating label poisoning as a noisy label classification problem.•A GAN-based model is proposed to simulate label poisoning with different mechanisms.•A comparative study is enabled on state-of-the-art noisy deep learners.•The designed approach is tested for federated learning and IoT intrusion detection.