FedEHR-Gen: Federated Synthetic Time-Series EHR Generation via Latent Space Alignment and Distribution-Aware Aggregation
Researchers have developed FedEHR-Gen, a novel federated learning framework for generating synthetic Electronic Health Records (EHRs). This approach addresses the challenge of data privacy by enabling cross-hospital modeling without pooling sensitive patient data. FedEHR-Gen utilizes a two-stage process involving a federated autoencoder for latent space alignment and a federated temporal conditional variational autoencoder for stable time-series generation, outperforming standard federated baselines in fidelity and utility. AI
IMPACT Enables privacy-preserving synthetic EHR generation for research and development across institutions.