Unsupervised domain adaptation with progressive adaptation of subspaces

Unsupervised Domain Adaptation (UDA) aims to classify unlabeled target domain by transferring knowledge from labeled source with shift. Most of the existing UDA methods try mitigate adverse impact induced shift via reducing discrepancy. However, such approaches easily suffer a notorious mode collapse issue due lack labels in domain. Naturally, one effective ways this is reliably estimate pseudo for domain, which itself hard. To overcome this, we propose novel method named Progressive Subspaces approach (PAS) utilize an intuition that appears much reasonable gradually obtain reliable labels. Speci fically, progressively and steadily refine shared subspaces as bridge transfer adaptively anchoring/selecting leveraging those samples Subsequently, refined can turn provide more pseudo-labels making highly mitigated. Our thorough evaluation demonstrates PAS not only common UDA, but also outperforms state-of-the arts challenging Partial (PDA) situation, where label set subsumes one.