Decentralized federated learning (DFL) enables peer-to-peer training, but hyperparameter optimization (HPO) in serverless environments remains underexplored. This paper presents a DFL framework featuring a one-shot cooperative HPO mechanism based on Sequential Model-based Algorithm Configuration (SMAC). The architecture employs a Digital Twin (DT) orchestration layer to manage edge nodes and network topologies without centralizing the learning process. Nodes perform independent local SMAC tuning followed by performance-weighted aggregation to derive a global hyperparameter configuration. Evaluated on worker-safety and MNIST datasets using a Raspberry Pi testbed, our approach significantly outperforms fixed settings and Random Search across IID and non-IID conditions (p<0.01p < 0.01 p<0.01). Results demonstrate faster convergence and higher accuracy; specifically, convergence rounds dropped from 42 to 18 for worker safety and from 8 to 3 for MNIST. In representative IID mesh experiments, accuracy rose from 92.4% to 95.3%, and MAE improved from 0.058 to 0.042 compared to fixed hyperparameter settings. These findings validate the efficiency of SMAC-based HPO for decentralized, edge-oriented environments. • A fully decentralized SMAC-based HPO framework integrated into a P2P federated loop. • A one-shot performance-weighted aggregation mechanism to minimize communication overhead. • Integration of Digital Twin (DT) technology as an orchestration and observability plane for edge hardware. • Achieves statistically significant (p<0.01) improvements in convergence speed and accuracy. • Validated on Raspberry Pi devices for industrial worker-safety and image classification tasks.
Moe et al. (Tue,) studied this question.
Synapse has enriched 5 closely related papers on similar clinical questions. Consider them for comparative context: