SplitFed resilience to packet loss: Where to split, that is the question

TL;DR

This paper examines how the placement of the split point in Split Federated Learning affects its robustness to packet loss, finding that deeper splits offer better accuracy under communication issues.

Contribution

It introduces an analysis of split point placement in SFL and demonstrates the impact of split depth on model accuracy during packet loss scenarios.

Findings

Deeper split points improve robustness to packet loss.

Statistically significant accuracy difference based on split location.

Experiments on human embryo image segmentation validate results.

Abstract

Decentralized machine learning has broadened its scope recently with the invention of Federated Learning (FL), Split Learning (SL), and their hybrids like Split Federated Learning (SplitFed or SFL). The goal of SFL is to reduce the computational power required by each client in FL and parallelize SL while maintaining privacy. This paper investigates the robustness of SFL against packet loss on communication links. The performance of various SFL aggregation strategies is examined by splitting the model at two points -- shallow split and deep split -- and testing whether the split point makes a statistically significant difference to the accuracy of the final model. Experiments are carried out on a segmentation model for human embryo images and indicate the statistically significant advantage of a deeper split point.