Participant Selection for Efficient and Trusted Federated Learning in Blockchain-Assisted Hierarchical Federated Learning Architectures

Federated learning has attracted widespread attention due to its strong capabilities of privacy protection, making it a powerful supporting technology for addressing data silos in the future. However, federated learning still lags significantly behind traditional centralized learning in terms of learning efficiency and system security. In this paper, we first construct a hierarchical federated learning architecture integrated with blockchain based on the cooperation of the cloud, edge, and terminal, which has the ability to enhance the security of federated learning while reducing the introduction costs of blockchain. Under this architecture, we propose a semi-asynchronous aggregation scheme at the edge layer and introduce a hierarchical aggregation scheme that combines it with synchronous aggregation at the cloud end to improve system efficiency. Furthermore, we present a multi-objective node selection scheme that considers various influencing factors such as security and efficiency. We formulate the node selection problem as a Markov Decision Process (MDP) and propose a solution based on deep reinforcement learning to address it more efficiently. The experimental results show that the proposed scheme can effectively improve system efficiency and enhance system security. In addition, the proposed DQN-based node selection algorithm can efficiently realize the selection of the optimal policy.