Optimizing Split Points for Error-Resilient SplitFed Learning

TL;DR

This paper investigates how different split points in SplitFed learning affect model performance under packet loss, finding that deeper splits offer significant resilience benefits in a human embryo image segmentation task.

Contribution

It introduces an analysis of split point selection in SplitFed learning, demonstrating the impact on error resilience and model accuracy under packet loss conditions.

Findings

Deeper split points improve error resilience.

Packet loss impacts model performance based on split location.

Deeper splits yield statistically significant advantages.

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.