Label-Free Algorithm Selection for Medical Imaging UDA

A new method selects optimal unsupervised domain adaptation algorithms and hyperparameters without requiring target domain labels. It uses an agreement reference constructed from multiple label-free signals to score candidate models against a consensus prediction.