Subtype-aware Unsupervised Domain Adaptation for Medical Diagnosis

Recent advances in unsupervised domain adaptation (UDA) show that transferable prototypical learning presents a powerful means for class conditional alignment, which encourages the closeness of cross-domain centroids. However, inner-class compactness and underlying fine-grained subtype structure remained largely underexplored. In this work, we propose to adaptively carry out subtype-aware alignment by explicitly enforcing class-wise separation subtype-wise with intermediate pseudo labels. Our key insight is unlabeled subtypes can be divergent one another different label shifts, while inheriting local proximity within subtype. The cases or without prior information on numbers are investigated discover an online fashion. proposed dynamic UDA achieves promising results medical diagnosis task.