Peer-to-Peer Federated Learning for COVID-19 Detection Using Transformers

The simultaneous advances in deep learning and the Internet of Things (IoT) have benefited distributed paradigms. Federated is one most promising frameworks, where a server works with local learners to train global model. intrinsic heterogeneity IoT devices, or non-independent identically (Non-I.I.D.) data, combined unstable communication network environment, causes bottleneck that slows convergence degrades efficiency. Additionally, majority weight averaging-based model aggregation approaches raise questions about fairness. In this paper, we propose peer-to-peer federated (P2PFL) framework based on Vision Transformers (ViT) models help solve some above issues classify COVID-19 vs. normal cases Chest-X-Ray (CXR) images. Particularly, clients jointly iterate aggregate order build robust experimental results demonstrate proposed approach capable significantly improving performance an Area Under Curve (AUC) 0.92 0.99 for hospital-1 hospital-2, respectively.