Domain Adaptation via Bregman divergence minimization

Abstract

In recent years, the Fisher linear discriminant analysis (FLDA) based classification models are among the most successful approaches and have shown effective performance in different classification tasks. However, when the learning data (source domain) have a different distribution compared with the testing data (target domain), the FLDA-based models may not work well, and the performance degrades, dramatically. To face this issue, we offer an optimal domain adaptation via Bregman divergence minimization (DAB) approach, in which the discriminative features of source and target domains are simultaneously learned via domain invariant representation. DAB is designed based on the constraints of FLDA, with the aim of the coupled marginal and conditional distribution shifts adaptation through Bregman divergence minimization. Thus, the resulting representation can show well functionality like FLDA and simultaneously discriminate across various classes, as well. Moreover, our proposed approach can be easily kernelized to deal with nonlinear tasks. Different experiments on various benchmark datasets demonstrate that our DAB can constructively face with the cross domain divergence and outperforms other novel state-of-the-art domain adaptation approaches in crossdistribution domains.

Visual Transductive learning via Iterative Label Correction

“…Subsequently, some UDA methods take advantage of label correction to mitigate the disruptive impact of unreliable pseudo-labeled samples. As a result, pseudo-labeling methods can be categorized into two paradigms: with label correction [27,28] and without label correction [29]. In pseudo labeling without label correction, we infer asymmetric tri-training for unsupervised domain adaptation [30], in which three classifiers are trained asymmetrically, while two of them are trained with labeled source samples and used for pseudo labeling, and the third one is trained with pseudo-labeled samples.…”