Seminar: “A context-aware localized federated learning approach for cloudlet federation” by Sana Latif

Abstract: This seminar presents a novel approach to federated learning (FL) within a cloudlet federation framework, focusing on addressing the challenges of latency, resource utilization, and data privacy at the edge. Titled "A Context-aware Localized Federated Learning Approach for Cloudlet Federation," the research leverages cloudlet-based edge computing to optimize deep learning model training for healthcare applications, particularly COVID-19 detection. We will discuss using a local centralized broker and edge agents to manage network resources dynamically, enabling efficient model aggregation and training. Key contributions include the implementation of a local Federated Averaging (FedAvg) algorithm, which demonstrates improved accuracy, reduced latency, and faster convergence times compared to traditional FL methods. The results highlight the system's ability to handle non-IID data, minimize network traffic, and enhance resource utilization, making it a viable solution for real-time, healthcare applications.