Edge-cloud collaborative learning for large-scale physical education data processing

With the development of digital education, physical education is generating increasing amounts of multi-source data, including wearable sensing records, training logs, classroom activity data, and performance evaluation results. These data are useful for intelligent teaching support, but their distributed and time-sensitive nature also makes large-scale processing difficult. Cloud-centered processing may introduce high communication cost and delayed response, while purely local edge processing lacks sufficient global coordination across different sites. To address this issue, this paper proposes ECL-PE, an edge-cloud collaborative learning framework for large-scale physical education data processing. In ECL-PE, edge nodes perform local preprocessing and preliminary updating, while the cloud conducts global aggregation and model synchronization. Experiments on a representative activity-recognition dataset show that ECL-PE achieves better accuracy, macro-F1, latency, convergence behavior, robustness under compressed updates, and cumulative communication cost than several comparison methods. The study provides a benchmark-based attempt to connect edge-cloud collaborative learning with distributed PE-related activity data processing.