Decentralized Privacy-Preserving Intelligence in 6G Networks via Federated Learning

Abstract

The 6G wireless network is set to become the first network that can interconnect billions of intelligent devices and provide ultra-low latency services, including holographic telepresence and autonomous mobility. The existing cloud-based machine learning architecture is unable to address these needs, as it has too much overhead on transmission of data and the growing concern of privacy. A decentralized privacy-preserving learning framework based on federated learning (FL) is proposed for beamforming prediction in 6G networks. According to the suggested solution, edge devices will train their models locally and send the encrypted model update only to a central aggregation system, thus avoiding the exposure of raw data. The framework features several machine learning models, such as Random Forest (RF), Feed Forward Neural Network (FNN), Convolutional Neural Network (CNN), Multilayer Perceptron (MLP), XGBoost, and Decision Tree, to measure predictive performance. Empirical experimentation has shown that the ensemble tree-based models are very effective in prediction accuracy with efficient distributed training. Challenges such as non-IID data distribution, straggler clients, and model poisoning are also identified. Additional improvements, including differential privacy, secure aggregation, and blockchain-based auditing, can also be made in the future to enhance secure and scalable federated learning deployment in next-generation 6G ecosystems.