Wireless Federated Learning with Local Differential Privacy

TL;DR

This paper explores wireless federated learning with local differential privacy, leveraging the superposition property of wireless channels to enhance privacy and bandwidth efficiency, and analyzes the convergence and tradeoffs involved.

Contribution

It introduces a private wireless gradient aggregation scheme that improves privacy scaling with the number of users and analyzes the convergence and resource-privacy tradeoffs.

Findings

Privacy leakage per user scales as 1/√K with K users.

Wireless superposition enables bandwidth-efficient gradient aggregation.

The proposed scheme offers better privacy guarantees compared to orthogonal transmission.

Abstract

In this paper, we study the problem of federated learning (FL) over a wireless channel, modeled by a Gaussian multiple access channel (MAC), subject to local differential privacy (LDP) constraints. We show that the superposition nature of the wireless channel provides a dual benefit of bandwidth efficient gradient aggregation, in conjunction with strong LDP guarantees for the users. We propose a private wireless gradient aggregation scheme, which shows that when aggregating gradients from $K$ users, the privacy leakage per user scales as $\mathcal{O}\big(\frac{1}{\sqrt{K}} \big)$ compared to orthogonal transmission in which the privacy leakage scales as a constant. We also present analysis for the convergence rate of the proposed private FL aggregation algorithm and study the tradeoffs between wireless resources, convergence, and privacy.