Federated Learning

Federated learning is valuable for infectious disease surveillance because patient data may be restricted across jurisdictions. Federated learning trains models across decentralized datasets without moving raw data from source institutions. Covariate shift across federated sites can be mitigated through weighting approaches. Heterogeneous non-IID data are a central technical challenge for federated learning. Federated learning is vulnerable to bias propagation and adversarial privacy or security attacks.