Decentralized Federated Learning with Byzantine Fault Tolerance using Graph Neural Networks

Abstract

Federated learning (FL) offers a promising approach to training machine learning models on decentralized data without directly exchanging the data itself. However, FL systems are vulnerable to Byzantine attacks, where malicious participants introduce corrupted data or models to compromise the learning process. This paper proposes a novel decentralized federated learning framework incorporating Graph Neural Networks (GNNs) for Byzantine fault tolerance. The core idea is to represent the federated learning network as a graph, enabling each participant to learn from neighbors while simultaneously detecting and mitigating the influence of potentially malicious nodes. Our approach employs a GNN to learn node embeddings that capture the relationships within the network, allowing for effective identification of Byzantine nodes based on their anomalous behavior. The dynamic, adaptive nature of this system provides a robust defense against Byzantine attacks, enhancing the reliability and trustworthiness of federated learning systems. We demonstrate the effectiveness of our framework through a theoretical analysis and outline a potential implementation strategy.