Bayesian Federated Cause-of-Death Classification and Quantification Under Distribution Shift
The episode discusses a paper on Bayesian Federated Cause-of-Death Classification and Quantification Under Distribution Shift. The hosts explain how this method uses local models trained on private data from different regions to make accurate cause-of-death estimates without sharing raw records. They detail how the approach handles data shift and improves upon existing rigid or data-sharing trade-off methods.