Federated Cubic Regularized Newton Learning with Sparsification-amplified Differential Privacy

TL;DR

This paper proposes DP-FCRN, a federated learning algorithm using cubic regularization and second-order methods, which enhances privacy and reduces communication costs through sparsification and noise perturbation, with proven convergence and privacy guarantees.

Contribution

It introduces a novel federated cubic regularized Newton method that combines second-order optimization, sparsification, and differential privacy for improved efficiency and privacy.

Findings

Lower iteration complexity compared to first-order methods.

Effective privacy amplification via sparsification.

Successful validation on benchmark datasets.

Abstract

This paper investigates the use of the cubic-regularized Newton method within a federated learning framework while addressing two major concerns that commonly arise in federated learning: privacy leakage and communication bottleneck. We introduce a federated learning algorithm called Differentially Private Federated Cubic Regularized Newton (DP-FCRN). By leveraging second-order techniques, our algorithm achieves lower iteration complexity compared to first-order methods. We also incorporate noise perturbation during local computations to ensure privacy. Furthermore, we employ sparsification in uplink transmission, which not only reduces the communication costs but also amplifies the privacy guarantee. Specifically, this approach reduces the necessary noise intensity without compromising privacy protection. We analyze the convergence properties of our algorithm and establish the privacy guarantee. Finally, we validate the effectiveness of the proposed algorithm through experiments on a benchmark dataset.