Secure and Efficient Federated Learning in Edge Computing

Abstract

Federated Learning (FL) has emerged as a promising paradigm for privacy-preserving Machine Learning (ML). It enables distributed end devices (clients) to collaboratively train a shared global model without exposing their local data. However, FL typically assumes that all clients are benign and trust the coordinating central server, which is unrealistic for many real-world scenarios. In practice, clients can harm the FL process by sharing poisonous model updates (known as poisoning attack) or sending counterfeit yet harmless parameters to the central server to obtain the trained global model without actual contribution (known as free-riding attack), while the central server could malfunction or misbehave. Moreover, the deployment of FL for real-world applications is hindered by the high communication overhead between the server and clients that are often at the network edge with limited bandwidth.

This thesis aims to develop novel FL approaches toward secure and efficient FL in edge computing. First, a novel lightweight blockchain-based FL framework is devised to mitigate the single point of failure of traditional FL. This is achieved by removing the centralized model aggregation to the distributed blockchain nodes. Incorporating the Inter-Planetary File System and Verifiable Random Function, the proposed framework is energy-efficient and scalable with the blockchain network size. Next, a secure and efficient federated edge learning system is proposed, based on the developed blockchain-based FL framework, with a communication-efficient training scheme to reduce the communication cost of clients and a secure model aggregation protocol to build defense against poisoning attacks. Then, an original Shapley value-based defense mechanism is designed to further enhance the robustness of FL, not only against adversarial poisoning attack but also the stealthy free-riding attack. Extensive experiments show that the proposed approach can detect typical free-riding attacks with high precision and is resistant to poisoning attacks launched by adversarial clients.