Conformal Risk Minimization for Semi-Supervised Domain Adaptation via Optimal Transport

TL;DR - This paper integrates conformal risk minimization into semi-supervised domain adaptation, using optimal-transport pseudolabels to compensate for scarce labeled target-domain data. The approach aims to produce clinically useful prediction sets that remain coverage-valid and compact under patient-population shifts.

• Jointly optimizes domain invariance, predictive performance, and conformal-set efficiency rather than applying conformal prediction post hoc.
• Uses optimal transport to pseudolabel unlabeled target examples, supplying the training signal needed to estimate nonconformity thresholds with few target labels.
• Targets compact, distribution-free prediction sets while preserving formal coverage guarantees.
• Supports domain-specific constraints, such as excluding mutually contradictory diagnoses in skin-lesion classification.