A Trusted and Decentralized Federated Learning Framework for IoT devices in Smart City

Abstract

Smart cities, through investment in human and social capital as well as traditional and modern communications infrastructure, drive sustainable economic growth and high quality of life. IoT devices, as crucial components of smart cities, leverage the Federated Learning (FL) paradigm to contribute significantly to the advancement of smart cities by providing the value of data while safeguarding local data. However, existing federated learning frameworks for smart cities face several issues, including excessive centralization of decision-making and data, unbalanced resource allocation, and security and privacy concerns. To address these challenges, This paper aims to propose a blockchain-based federated learning framework for IoT devices in smart cities. We combine multiple security and privacy-preserving techniques such as differential privacy and Trusted Execution Environments (TEE) with federated learning to establish a framework that facilitates secure and trustworthy data exchange and FL requirements for IoT devices in smart cities. Moreover, we offload the local training computation from the limited IoT devices to the edge server and execute trust aggregation on the blockchain smart contracts. We build a prototype of our designed system and conduct experiments on diverse federated learning datasets. Experimental results demonstrate that our scheme achieves high efficiency while ensuring security and privacy. Through this work, we provide a viable solution for federated learning in smart cities, thereby advancing the sustainable development and intelligence of smart cities.