xG Security: Zero-Trust and Moving Target Defense in Decentralized Learning Environment
Zeyad Abdelhay, Ahmed Refaey · 2024
The divergence of Artificial Intelligence (AI) with Next-Generation (xG) mobile networks is inevitable as it is driven by the demand for more intelligent mobile networks that can optimize the data collected from users’ devices and utilize the distributed nature of Federated Learning (FL). This poses a multitude of security challenges, including authentication, data integrity, and secure communication channels between participating network nodes. Traditional deployments of FL have not proven resilience against a range of attacks, like port scanning, man-in-the-middle, and network mapping. This paper proposes the SDP-FedStellar framework as a possible solution, aiming to address the security gap at the intersection of xG and FL. We establish a zero-trust security model using SDP’s dynamic controller-based authentication and authorization to ensure the privacy of user and model data privacy throughout the federated learning process, to enhance the overall security of xG networks running centralized or decentralized FL. This framework strengthens the network and each node’s ability to dynamically defend itself against attackers targeting malicious nodes.