Kernel Distribution Consistency Based Local Domain Adaptation Learning: Kernel Distribution Consistency Based Local Domain Adaptation Learning

Abstract

In allusion to domain adaptation learning (DAL) problems, this paper proposes a novel so-called kernel distribution consistency based local domain adaptation classifier (KDC-LDAC). Firstly, in some universally reproduced kernel Hilbert space (URKHS), the KDC-LDAC trains a kernel distribution consistency regularized domain adaptation support vector machine (SVM) based on the structure risk minimization model, which extends the formulation of classical SVMs to the domain adaptation learning schema. And secondly, according to the idea of local learning, the proposed method predicts the label of each data point in target domain based on its neighbors and their labels in the URKHS. The last but not least, the KDC-LDACs learning a discriminant function to classify the unseen data in target domain with training data well predicted in the local learning procedure. Experimental results on artificial and real world problems show the advantages or comparable effectiveness of the proposed approach compared to related approaches.