Understanding Clipping for Federated Learning: Convergence and Client-Level Differential Privacy

TL;DR

This paper investigates the effects of clipping in federated learning with differential privacy, providing both empirical insights and convergence analysis to understand its impact on model performance and privacy guarantees.

Contribution

It offers the first rigorous theoretical and empirical analysis of clipping in federated learning with differential privacy, revealing its effects on convergence and update similarity.

Findings

Clipped FedAvg performs well despite data heterogeneity.

Clients' updates tend to become similar across deep architectures.

The convergence analysis links clipping bias to update distribution.

Abstract

Providing privacy protection has been one of the primary motivations of Federated Learning (FL). Recently, there has been a line of work on incorporating the formal privacy notion of differential privacy with FL. To guarantee the client-level differential privacy in FL algorithms, the clients' transmitted model updates have to be clipped before adding privacy noise. Such clipping operation is substantially different from its counterpart of gradient clipping in the centralized differentially private SGD and has not been well-understood. In this paper, we first empirically demonstrate that the clipped FedAvg can perform surprisingly well even with substantial data heterogeneity when training neural networks, which is partly because the clients' updates become similar for several popular deep architectures. Based on this key observation, we provide the convergence analysis of a differential private (DP) FedAvg algorithm and highlight the relationship between clipping bias and the distribution of the clients' updates. To the best of our knowledge, this is the first work that rigorously investigates theoretical and empirical issues regarding the clipping operation in FL algorithms.