Data Privacy Attacks on Federated Learning

Federated Learning, which enables cooperative model training over distributed edge devices, has emerged as a paradigm shift in machine learning. However, there are serious security issues with this dispersed strategy. This study thoroughly analyses security concerns in federated learning, including data poisoning, sybil attacks, model inversion attacks, and privacy violations. We carefully examine these dangers and their potential effects using a vast body of research. We also offer a taxonomy of defense tactics put forth by researchers worldwide. These techniques address flaws at data aggregation and model aggregation phases, ranging from differential privacy mechanisms to safe multiparty computations. This paper seeks to direct future research efforts in reinforcing federated learning against increasing attacks by thoroughly reviewing security problems and cutting-edge defenses.