A secure and privacy‐preserved delegate‐based blockchain and federated learning for 6G networks

Abstract

Listen

Translate article

SummaryThe 6G networks are envisioned to provide Artificial Intelligence (AI) for distributed devices through deploying machine learning on base stations. However, due to the concern of data privacy, devices are not willing to transmit their raw data to base stations (BSs) for machine learning training. Federated learning (FL) is a promising paradigm that can enable distributed machine learning while protecting data privacy by collaboratively training AI models without sharing raw data. But the traditional FL requires a central node to complete global model update which makes the traditional FL facing single point attack. Further, the traditional FL cannot allow the distributed model sharing among untrusted devices which results in low training efficiency on devices. Blockchain is a promising approach to address the above issues since it can establish a secure decentralized and transaction sharing environment for untrusted devices. In this paper, we integrate blockchain into FL to build a distributed model training and sharing platform, where distributed mobile devices execute local model training and base stations maintain blockchain‐based model sharing. Moreover, in order to accelerate the establishment of blockchain, we design a lightweight consensus strategy based on a delegate committee. To select high‐quality nodes to form delegate committee, we design a deep reinforcement learning (DRL)‐based selection algorithm. Numerical results demonstrate that the proposed framework can achieve a higher precision compared with the traditional FL, and the proposed DRL selection algorithm can reduce consensus latency compared with the benchmarks.