Conceptual
Browse Instruct Curate

Hardness-Driven Augmentation and Alignment for Multi-Source Domain Adaptation

A multi-source domain adaptation method that quantifies each sample's difficulty via three adaptive hardness measurements and uses those scores to modulate data-augmentation intensity, to reweight Maximum Mean Discrepancy into a clustered inter-domain alignment loss, and to select hard samples for a pseudo-contrastive matrix improving intra-domain alignment and pseudo-label quality.