Optimizing Split Points for Error-Resilient SplitFed Learning

TL;DR

This work examines how the choice of model split point in SplitFed learning affects resilience to packet loss during distributed training, focusing on a Split U-Net for human embryo component segmentation. The authors compare shallow and deep splits across five aggregation strategies while simulating packet loss with probability $P_L$ and affected client count $N_c$, using the Blastocyst dataset. They report a statistically significant advantage for deeper splits under loss, touch on the relative performance of aggregation methods, and discuss mechanisms such as additional CS(BE) layers and client-side skip connections contributing to robustness. The findings inform practical design choices for privacy-preserving, distributed medical imaging applications under realistic network conditions and point to future work on more realistic loss models and recovery methods.

Abstract

Recent advancements in decentralized learning, such as Federated Learning (FL), Split Learning (SL), and Split Federated Learning (SplitFed), have expanded the potentials of machine learning. SplitFed aims to minimize the computational burden on individual clients in FL and parallelize SL while maintaining privacy. This study investigates the resilience of SplitFed to packet loss at model split points. It explores various parameter aggregation strategies of SplitFed by examining the impact of splitting the model at different points-either shallow split or deep split-on the final global model performance. The experiments, conducted on a human embryo image segmentation task, reveal a statistically significant advantage of a deeper split point.

Figures (2)

• Figure 1: Split U-Net architecture
• Figure 2: MJI vs. $P_L$ for shallow split (left) and deep split (right), with variying $N_c$ in the SplitFed U-Net model.