Decentralized Federated Learning with Differential Privacy

Abstract

Federated learning (FL) offers a promising paradigm for training machine learning models on decentralized data sources without directly exchanging data. However, traditional FL approaches often struggle with data heterogeneity, communication bottlenecks, and, crucially, privacy concerns. This paper proposes a novel decentralized federated learning framework that integrates differential privacy (DP) mechanisms to mitigate privacy risks while maintaining model convergence in a heterogeneous environment. The core of the framework lies in a distributed aggregation strategy and a tailored DP mechanism designed to minimize information leakage from individual client updates. We demonstrate the effectiveness of this approach through a theoretical analysis and outline the key components of the system. The primary goal is to provide a robust and scalable solution for collaborative model training in scenarios where data privacy is paramount and system heterogeneity is significant.