Go to ICLR 2025 Conference homepageA Provably Robust Algorithm for Differentially Private Clustered Federated Learning
Federated Learning (FL), which is a decentralized machine learning (ML) approach, often incorporates differential privacy (DP) to enhance data privacy guarantees. However, differentially private federated learning (DPFL) introduces performance disparities across clients, particularly affecting minority groups. Some recent works have attempted to address large data heterogeneity in vanilla FL settings through clustering clients, but these methods remain sensitive and prone to errors further exacerbated by the DP noise, making them inappropriate for DPFL settings. We propose an algorithm for differentially private clustered FL, which is robust to the DP noise in the system and identifies clients’ clusters correctly. To this end, we propose to cluster clients based on both their model updates and training loss values. Furthermore, when clustering clients’ model updates, our proposed approach addresses the server’s uncertainties by employing large batch sizes as well as Gaussian Mixture Models (GMM) to reduce the impact of DP and stochastic noise and avoid potential clustering errors. This idea is efficient especially in privacy-sensitive scenarios with more DP noise. We provide theoretical analysis justifying our approach, and evaluate it extensively across diverse data distributions and privacy budgets. Our experimental results show its effectiveness in addressing large data heterogeneity in DPFL systems with a small computational cost.