Asynchronous Decentralized Federated Anomaly Detection for 6G Networks

The long-term vision for 6G security is to implement AI-assisted frameworks that achieve security automation without disrupting normal usage. Deep learning-based anomaly detection is one of the essential components in the envisioned 6G network security paradigm. However, the future 6G architectures, such as the space-air-ground-sea integrated networks (SAGSINs), suffer from limited bandwidth resources beyond the ground. In such scenarios, traditional centralized or centralized-distributed anomaly detection algorithms may encounter high communication latency. Furthermore, due to the high heterogeneity of 6G networks, the training speed of conventional synchronous distributed algorithms may be seriously affected by slow devices. This paper proposes an end-to-end asynchronous decentralized federated unsupervised anomaly detection approach named DEFEND for 6G networks, which enables multiple devices to collaborate asynchronously through peer-to-peer communication for proactive and automatic threat detection while preserving data privacy. The proposed DEFEND employs meta-learning to facilitate better adaptation to the evolving data distribution. To address the complex nested optimization problem of meta-learning, we utilize a computationally efficient cutting plane approximation method. We theoretically prove that the convergence rate of DEFEND is <inline-formula xmlns:mml="http://www.w3.org/1998/Math/MathML" xmlns:xlink="http://www.w3.org/1999/xlink"> <tex-math notation="LaTeX">$\mathcal {O}(\epsilon ^{-1})$ </tex-math></inline-formula> and demonstrate the effectiveness of DEFEND through comprehensive experiments in public datasets.