SplitFed resilience to packet loss: Where to split, that is the question
This work addresses the reliability of decentralized machine learning for applications like medical imaging, but it is incremental as it focuses on optimizing an existing hybrid method.
The paper tackled the problem of Split Federated Learning (SFL) robustness against packet loss by investigating how the model split point affects accuracy, finding that a deeper split point provides a statistically significant advantage in experiments on a human embryo image segmentation model.
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.