Differentially Private Federated Learning on Heterogeneous Data

TL;DR

This paper introduces DP-SCAFFOLD, a differentially private federated learning algorithm designed to handle highly heterogeneous data while ensuring privacy against both third-party and server observers, with proven convergence and superior performance.

Contribution

The work presents a novel federated learning method that integrates differential privacy into SCAFFOLD, addressing data heterogeneity and privacy protection simultaneously.

Findings

DP-SCAFFOLD outperforms DP-FedAvg with increased local updates and heterogeneity.

The algorithm guarantees convergence for convex and non-convex objectives.

Significant practical gains demonstrated through numerical experiments.

Abstract

Federated Learning (FL) is a paradigm for large-scale distributed learning which faces two key challenges: (i) efficient training from highly heterogeneous user data, and (ii) protecting the privacy of participating users. In this work, we propose a novel FL approach (DP-SCAFFOLD) to tackle these two challenges together by incorporating Differential Privacy (DP) constraints into the popular SCAFFOLD algorithm. We focus on the challenging setting where users communicate with a "honest-but-curious" server without any trusted intermediary, which requires to ensure privacy not only towards a third-party with access to the final model but also towards the server who observes all user communications. Using advanced results from DP theory, we establish the convergence of our algorithm for convex and non-convex objectives. Our analysis clearly highlights the privacy-utility trade-off under data heterogeneity, and demonstrates the superiority of DP-SCAFFOLD over the state-of-the-art algorithm DP-FedAvg when the number of local updates and the level of heterogeneity grow. Our numerical results confirm our analysis and show that DP-SCAFFOLD provides significant gains in practice.