Learning Target-Domain-Specific Classifier for Partial Domain Adaptation

Unsupervised domain adaptation (UDA) aims at reducing the distribution discrepancy when transferring knowledge from a labeled source to an unlabeled target domain. Previous UDA methods assume that and domains share identical label space, which is unrealistic in practice since information of agnostic. This article focuses on more realistic scenario, i.e., partial (PDA), where space subsumed space. In PDA outliers are absent may be wrongly matched (technically named negative transfer), leading performance degradation methods. proposes novel target-domain-specific classifier learning-based (TSCDA) method. TSCDA presents soft-weighed maximum mean criterion partially align feature distributions alleviate transfer. Also, it learns target-specific for with pseudolabels multiple auxiliary classifiers further address shift. A module peers-assisted learning used minimize prediction difference between classifiers, makes discriminant Extensive experiments conducted three benchmark data sets show outperforms other state-of-the-art large margin, e.g., 4% 5.6% averagely Office-31 Office-Home, respectively.