SplitFed: When Federated Learning Meets Split Learning

Federated learning (FL) and split (SL) are two popular distributed machine approaches. Both follow a model-to-data scenario; clients train test models without sharing raw data. SL provides better model privacy than FL due to the architecture between server. Moreover, makes option for resource-constrained environments. However, performs slower relay-based training across multiple clients. In this regard, paper presents novel approach, named splitfed (SFL), that amalgamates approaches eliminating their inherent drawbacks, along with refined architectural configuration incorporating differential PixelDP enhance data robustness. Our analysis empirical results demonstrate (pure) SFL similar accuracy communication efficiency as while significantly decreasing its computation time per global epoch in Furthermore, SL, over improves number of Besides, performance robustness measures is further evaluated under extended experimental settings.