Cloud Edge Patterns for Federated Learning Applied to the Architectural Design of a Privacy-Preserving Monitoring System for Diabetic Patients

This study aims to explore the current state of the art of patterns used in combination with federated learning (FL) in a cloud continuum (CC) scenario, where computational and storage resources range from the edge, through the use of IoT devices, to a concentrated environment through the use of cloud infrastructure. Using a case study on the monitoring system for diabetic patients, we identified key federated learning challenges, including data heterogeneity, malicious clients, and reconstructing data from the model. These challenges were categorized into macro areas, and corresponding patterns were associated with mitigating or resolving these issues. A three-layer architecture is further proposed to support such application patterns, ensuring robustness, scalability, and data analysis with privacy preservation. This work classifies federated learning challenges and proposes patterns to mitigate these problems while providing the baseline for a secure system architecture for real-world deployment in the medical field.