D2P-Fed: Differentially Private Federated Learning With Efficient Communication

TL;DR

D2P-Fed introduces a discrete Gaussian-based approach to enhance privacy and communication efficiency in federated learning, outperforming previous methods in accuracy and reducing communication costs.

Contribution

It presents a novel discrete Gaussian noise mechanism for federated learning that improves privacy guarantees and communication efficiency simultaneously.

Findings

Outperforms state-of-the-art by 4.7% to 13.0% in accuracy.

Reduces communication cost by one third.

Provides comprehensive analysis of privacy, communication, and convergence.

Abstract

In this paper, we propose the discrete Gaussian based differentially private federated learning (D2P-Fed), a unified scheme to achieve both differential privacy (DP) and communication efficiency in federated learning (FL). In particular, compared with the only prior work taking care of both aspects, D2P-Fed provides stronger privacy guarantee, better composability and smaller communication cost. The key idea is to apply the discrete Gaussian noise to the private data transmission. We provide complete analysis of the privacy guarantee, communication cost and convergence rate of D2P-Fed. We evaluated D2P-Fed on INFIMNIST and CIFAR10. The results show that D2P-Fed outperforms the-state-of-the-art by 4.7% to 13.0% in terms of model accuracy while saving one third of the communication cost.