CLASS PROTOTYPE ALIGNMENT THROUGH OPTIMAL TRANSPORT OF BARYCENTERS FOR UNSUPERVISED DOMAIN ADAPTATION
Bulletin of Toraighyrov University. Physics & Mathematics series · 30.06.2026 · (2, 2026) · 42-53
DOI: 10.48081/bgqf2032 ↗Kurz gefasst
Fachartikel in der Zeitschrift Bulletin of Toraighyrov University. Physics & Mathematics series, online erschienen am 30.06.2026. Verfasst von S. V. Khan und 1 weiteren Autor:innen.
Abstract (Originalsprache)
In this paper, the problem of domain adaptation for image classification using the Optimal Transport approach is studied. A proposed method considers the class structure of the data, in which Wasserstein barycenter’s are calculated for each class of the training sample, acting as semantic prototypes. These prototypes are used to align the distribution of the target data sample by clustering and building a transport plan between barycenter’s and target samples. Additionally, pseudo-labels are generated for the unlabeled data, which allows them to be included in the model training process. The experimental evaluation is carried out on datasets of handwritten digits showing differences in feature distributions. Linear regression is used as the basic method. The results show that the proposed approach provides higher classification accuracy and better consistency between the two datasets. Thus, the method confirms the effectiveness of using structured transfer of distributions to improve the quality of models in conditions of a distribution shift. Keywords: Unsupervised Domain Adaptation, Optimal Transport, Wasserstein Barycenter’s, Class Prototype Alignment, Domain Shift, Image Classification, Pseudo-Labels, K-Means Clustering.
Zitieren
APA
Khan, S. V., & Aliyev, Y. N. (2026). CLASS PROTOTYPE ALIGNMENT THROUGH OPTIMAL TRANSPORT OF BARYCENTERS FOR UNSUPERVISED DOMAIN ADAPTATION. Bulletin of Toraighyrov University. Physics & Mathematics series, 42-53. https://doi.org/10.48081/bgqf2032
Vancouver
Khan SV, Aliyev YN. CLASS PROTOTYPE ALIGNMENT THROUGH OPTIMAL TRANSPORT OF BARYCENTERS FOR UNSUPERVISED DOMAIN ADAPTATION. Bulletin of Toraighyrov University. Physics & Mathematics series. 2026:42-53. doi:10.48081/bgqf2032
BibTeX
@article{Khan2026,
author = {Khan, S. V. and Aliyev, Y. N.},
title = {CLASS PROTOTYPE ALIGNMENT THROUGH OPTIMAL TRANSPORT OF BARYCENTERS FOR UNSUPERVISED DOMAIN ADAPTATION},
journal = {Bulletin of Toraighyrov University. Physics & Mathematics series},
year = {2026},
number = {2, 2026},
pages = {42--53},
doi = {10.48081/bgqf2032}
}
EUVIMED