A Blockchain-Based Dynamic and Fair Federated Learning for IoT Data Sharing of Inferior Networking Conditions

Data sharing among edge devices belonging to different stakeholders can open new possibilities for emerging Internet -of -Things (IoT) edge intelligence and collaboration. However, traditional data sharing methods enabled through a centralized mechanism were no longer suitable for edge scenarios due to security, privacy and scalability issues. Decentralized device -to -device (D2D) data sharing and fusion is a promising paradigm in edge networks. But cost -inefficient trust management, low efficiency, vulnerable privacy protection and vague assessment of contribution issues are new obstacles. In this paper, we first proposed an IoT data sharing system that supports blockchain and federated learning (FL) to facilitate secure, efficient, and privacyenhanced decentralized data sharing and fusion. To improve the quality of FL -enabled data sharing and promote fairness of contribution evaluation under inferior networking conditions, we proposed a blockchainbased dynamic and fair FL scheme with an adaptive dynamic learning mechanism and a gradient entropy -based evaluation method. In addition, the learning operations automatically performed by the smart contract reach a consensus through the proposed proof -of -ability algorithm, which realized data sharing with high efficiency and autonomy. Simulation results show that the proposed system and mechanism can improve the data sharing efficiency and learning quality, and guarantee fair evaluation and security.