Unsupervised Domain Adaptation Based on Pseudo-Label Confidence

Unsupervised domain adaptation aims to align the distributions of data in source and target domains, as well assign labels domain. In this paper, we propose a new method named Domain Adaptation based on Pseudo-Label Confidence (UDA-PLC). Concretely, UDA-PLC first learns feature representation by projecting domains into latent subspace. subspace, distribution two are aligned discriminability features both is improved. Then, applies Structured Prediction (SP) Nearest Class Prototype (NCP) predicting pseudo-labels domain, it takes fraction samples with high confidence rather than all pseudo-labeled next iterative learning. Finally, experimental results validate that proposed outperforms several state-of-the-art methods three benchmark sets.