Detection of GAN Generated Images Related Poisoning Attacks in Federated Learning Systems

The rise of Generative Adversarial Networks has raised significant concerns about poisoning attacks in Federated Learning systems. Attackers often inject synthetic images created by these networks to harm model performance and compromise user privacy. While many existing solutions focus on specific threats, our study introduces a new machine-learning algorithm that effectively detects models using synthetic data and defends against label-flipping attacks. This dual detection method not only enhances the reliability of the system but also ensures that it can accurately tell apart real models and synthetic models. By addressing these critical challenges, we aim to improve the overall resilience of Federated Learning systems against emerging threats. To validate our method, we will conduct thorough tests across various attack scenarios, evaluating its performance in different settings. Our goal is to strengthen the security of the system and promote its wider use in a variety of applications, ultimately fostering greater trust in these advanced technologies.