FedHUG: Federated Heterogeneous Unsupervised Generalization for Remote Physiological Measurements
The FedHUG framework proposes a new protocol to improve model generalization for remote physiological measurements using unsupervised federated learning. It addresses data biases from different devices and users, and label distribution issues by introducing Minimal Bias Aggregation (MBA) and the Global Distribution-aware Learning Controller (GDLC). This results in superior performance across various estimation techniques.