A Privacy-Preserving Federated Learning Framework for IoT Environment Based on Secure Multi-party Computation
Tieming Geng, Jian Liu, Chin‐Tser Huang
Abstract
Artificial intelligence (AI) techniques, such as federated learning (FL), have been employed in Internet of Things (IoT) networks to harness the enormous amounts of data generated by ubiquitous IoT devices. However, using FL technique in IoT networks brings new security and privacy challenges like model inversion, data poisoning, and central aggregation failure. In this paper, we introduce an innovative framework that integrates blockchain technology and secure multi-party computation (MPC) with FL in IoT networks to address these challenges. Our proposed system enables decentralized and privacy-preserving FL across distributed IoT environments. Utilizing MPC protocols, our framework securely aggregates model updates from clients without revealing individual data, while blockchain technology ensures transparency and immutability of transactions. Security analysis shows that our framework can effectively address the three aforementioned security and privacy challenges and improve the reliability and robustness of FL processes in IoT networks. Our framework can be adapted and integrated with real-world AI-augmented IoT networks, such as smart cities and healthcare, to enhance their protection of security and privacy.