Client Selection and Resource Allocation via Graph Neural Networks for Efficient Federated Learning in Healthcare Environments

Two of the most significant challenges in decentralized federated learning are resource allocation and client selection. In order to address certain aspects of these challenges, in this paper we introduce a novel approach based on graph neural networks (GNNs). In the present work, we aim to ensure differential privacy guarantees in optimal client selection and resource allocation. Our comparative analysis against two baseline schemes reveals that our solution maintains a relatively low total delay, even as the number of clients increases. Furthermore, our preliminary results indicate that GNNs contribute to differentially private client selection and resource allocation in federated learning, especially in healthcare environments.