Blockchain-Assisted Privacy-Preserving and Robust Federated Learning in Edge Computing

Abstract

Federated learning (FL) enables distributed clients to collaboratively train machine learning models without sharing raw data, making it well-suited for edge computing environments. However, deploying FL at the edge introduces critical challenges, including potential model poisoning by malicious clients or edge servers, increased privacy risks, and communication bottlenecks. To address these issues, we propose a two-layer blockchain-assisted FL framework that ensures privacy, robustness, and efficiency. In our design, each client uploads an encrypted local update to an edge server for privacy-preserving validation, while a lightweight, sparse upload strategy is adopted to reduce communication overhead. We design a two-layer defense strategy that employs the secure cosine similarity technique and a truth discovery algorithm to resist Byzantine attacks from both malicious clients and edge servers, while a consortium blockchain maintained by edge servers guarantees decentralized and tamper-resistant global model updates. Theoretical analysis and extensive experiments under both IID and non-IID data partitions demonstrate that the proposed scheme effectively defends against targeted and untargeted poisoning attacks and preserves the privacy of local model updates. It also substantially reduces communication overhead. These results indicate that our framework provides a practical and scalable solution for privacy-preserving and robust FL in edge environments.