Towards Serverless Semi-Decentralized Federated Learning with Heterogeneous Optimizers

In three linesNovel serverless semi-decentralized federated learning approach (SSD-FL) eliminating persistent server infrastructure. Lightweight D2D initialization phase followed by fully serverless training with cluster formation. Integrates heterogeneous ML optimizers and network graph-based regularization. Improves convergence speed and communication efficiency.

Summary generated by Claude — human-verified