Optimizing federated learning efficiency: static and adaptive data selection strategies for heterogeneous environment

Federated learning (FL) enables collaborative model training while preserving data privacy. However, traditional approaches such as FedAvg struggle in heterogeneous environments due to variability in data distribution and client resources. This paper introduces static and adaptive data selection algorithms to optimize convergence efficiency. The static method ensures balanced participation with predefined parameters, while the adaptive approach dynamically adjusts the selection to address client variability. Experiments show that the static algorithm excels in stable environments, whereas the adaptive method is more robust in dynamic scenarios.