Investigated Insider and Outsider Attacks on the Federated Learning Systems
Ibrahim M. Ahmed, Manar Younis Kashmoola · 2022
With the vast usage of smart system devices, such as those used in 6G applications, that rely heavily on distributed machine learning technologies such as Federated Learning. Thus, there is an urgent need to provide a secure federated learning environment. The main challenge that faces federated Learning is thepoising attack. Therefore, this paper proposes a new framework for federated Learning that makes it more secure against GANs (Generative Adversarial Networks) attacks and Sybil attacks. The proposed framework is based on Federated Learning with Microsoft Confidential Consortium Framework (FED_CCF) to create a secure and reliable environment that deceives attackers of the federated learning environment. The performance of the proposed FED_CCF is evaluated using the MNIST dataset in terms of accuracy, where 30% of the devices were malicious devices represented by GAN or Sybil attacks. The experiment results of the proposed FED_CCF system show 96% better accuracy, with no effect of Sybil poisoning attacks, and only 0.18% of GAN poisoning attacks could affect it.