Blockchain-enabled Edge Computing Framework for Hierarchic Cluster-based Federated Learning

Abstract

Federated learning implements decentralized machine learning tasks without exposing users' private data. However, in practical scenarios, intelligent devices data pertain to different fields are non-independent and identically distributed (non-IID), which leads to a decrease in the accuracy of the global model. In addition, if there are untrusted devices participated in federated learning, the global model accuracy will be decreased. To address the above-mentioned issues, in this paper, we propose a blockchain-enabled hierarchic cluster-based federated learning in edge computing framework to improve the accuracy of the global model and ensure the local model credibility. Firstly, we propose the hierarchic cluster-based federated learning (HCFL) algorithm, which realizes hierarchically aggregation based on user cosine similarity to improve global model accuracy. Moreover, blockchain technology is enabled in the proposed HCFL algorithm to verify the local model gradient from IDs before global aggregation. Moreover, incentive mechanism is proposed to dynamically adjust reward of IDs for promote IDs train trusted models. Finally, simulation results demonstrate the efficiency and performance of the blockchain-enabled hierarchic cluster-based federated learning framework.