Adaptive Federated Meta-Learning for Real-Time Personalization in Distributed Cloud-Edge Data Infrastructures

Abstract

With the increasing proliferation of edge devices and context-aware applications, real-time personalization of services is becoming critical in domains such as healthcare, smart homes, and mobile computing. Traditional federated learning (FL) models face challenges in personalization due to data heterogeneity and latency constraints. This study presents an adaptive federated meta-learning (FedMeta) framework that personalizes models in real-time across distributed cloud-edge infrastructures. By leveraging model-agnostic meta-learning (MAML) and dynamically adjusting client-specific learning tasks, the system ensures fast adaptation and minimal communication overhead. Experiments show that FedMeta significantly improves personalization accuracy and system responsiveness compared to centralized and vanilla FL methods.