Optimizing Privacy and Latency Tradeoffs in Split Federated Learning Over Wireless Networks

In this letter, a novel cut layer selection scheme is designed to minimize the overall latency in split federated learning (SFL) over wireless networks, while maintaining an acceptable privacy level. Considering a tradeoff between overall latency and privacy level in terms of the cut layer selection, we establish a theoretical framework for managing cut layer selection in SFL to optimize the cut layer point. Furthermore, we discuss the impact of a differential privacy technique designed to enhance privacy by effectively concealing individual information. We evaluate the performance of the proposed scheme and provide insights on optimizing the overall latency of SFL while maintaining the desired privacy level through cut layer selection.

I. Introduction

Split Federated Learning (SFL) is a promising technology for distributed learning that combines the strengths of federated and split learning paradigms. In SFL, clients train only a part of the full model, called the client-side model, reducing their computational load. These client-side models are then synchronized to improve convergence speed. Correspondingly, this approach has obtained significant attention, particularly in wireless networks where mobile devices (MDs) are known to have limited battery and computational resources. However, while offering notable advantages, SFL introduces additional communication overhead when communicating with servers and raises privacy concerns due to the frequent exchange of client-side model outputs and model updates, which are correlated to the raw data. Therefore, the choice of the cut layer in SFL, which divides the model into client- and server-side models, greatly affects the overall latency and privacy of SFL. Optimizing SFL management to address these challenges remains a complicated issue [1], [2].