Unsupervised domain adaptation through transferring both the source-knowledge and target-relatedness simultaneously

<abstract><p>Unsupervised domain adaptation (UDA) is an emerging research topic in the field of machine learning and pattern recognition, which aims to help unlabeled target by transferring knowledge from source domain. To perform UDA, a variety methods have been proposed, most concentrate on scenario single (1S1T). However, real applications, usually with multiple domains are involved (1SmT), cannot be handled directly those 1S1T models. Unfortunately, although few related works 1SmT UDA nearly none them model leverage target-relatedness jointly. overcome these shortcomings, we herein propose more general through both source-knowledge target-relatedness, UDA-SKTR for short. In this way, not only supervision but also potential relatedness among simultaneously modeled exploitation process UDA. addition, construct alternating optimization algorithm solve variables proposed convergence guarantee. Finally, extensive experiments benchmark datasets, validate effectiveness superiority method.</p></abstract>