Mutual-Supervised Federated Learning and Blockchain-Based IoT Data Sharing

Abstract

Due to the decentralized, tamper-proof, and auditable properties of blockchain, more and more scholars and researchers are studying the application of blockchain technology in IoT data sharing. Federated learning is an effective way to enable data sharing, but can be compromised by dishonest data owners who may provide malicious models. In addition, dishonest data requesters may also infer private information from model parameters. To solve the above problems, a secure data sharing mechanism based on mutual-supervised federated learning and blockchain, BPCV-FL, is proposed. This mechanism ensures data privacy by adopting gradient descent algorithm with differential privacy protection in local model training and ensures the reliability of shared data through mutual supervision on the blockchain. Experimental results show that the proposed BPCV-FL has high accuracy and security in IoT data sharing.