Improved Metric Learning Method based on Distribution Optimization for Data Classification
Advances in Engineering Technology Research · 10.03.2026 · 16 (1) · 434
DOI: 10.56028/aetr.16.1.434.2026 ↗Kurz gefasst
Fachartikel in der Zeitschrift Advances in Engineering Technology Research, online erschienen am 10.03.2026. Verfasst von Yinhao Wang und 1 weiteren Autor:innen.
Abstract (Originalsprache)
The uncertainty of data distribution poses significant challenges to complex data classification tasks. Due to the impact of uncertainty issues, data classification algorithms often suffer from severe overfitting. Taking the person Re-ID task as an example, existing algorithms of similarity measurement have an excellent ability of discrimination on training data. However, these methods perform poorly on the test data. To address this issue, a robust-XQDA metric learning method based on distribution robust optimization is proposed. Our model first learns a set of projection directions based on the existing metric model. Furthermore, the raw metric model is used to find close image pairs and optimize the distribution of all samples. The distribution-robust optimization approach optimizes the distribution by introducing a distribution-constraint item. It introduces uncertainty into the feature matrix, thereby improving the projection subspace kernel function to mitigate the impact of sample distribution and population distribution differences on the robustness of the metric model. This method can effectively alleviate the interference of the significant gap between the distribution of test pairs and training pairs in the classification subspace. Compared to the baseline model, the improved algorithm was tested on three publicly available person re-ID datasets. The results showed that our model had a significant improvement on all three datasets. Especially, it improves the baseline method by 17.47%, 4.74%, and 2.24% matching accuracy on all chosen ranks. It effectively reduces overfitting and ultimately verifies the excellent performance of the method.
Zitieren
APA
Wang, Y., & Wen, D. (2026). Improved Metric Learning Method based on Distribution Optimization for Data Classification. Advances in Engineering Technology Research, 16(1), 434. https://doi.org/10.56028/aetr.16.1.434.2026
Vancouver
Wang Y, Wen D. Improved Metric Learning Method based on Distribution Optimization for Data Classification. Advances in Engineering Technology Research. 2026;16(1):434. doi:10.56028/aetr.16.1.434.2026
BibTeX
@article{Wang2026,
author = {Wang, Yinhao and Wen, Dicheng},
title = {Improved Metric Learning Method based on Distribution Optimization for Data Classification},
journal = {Advances in Engineering Technology Research},
year = {2026},
volume = {16},
number = {1},
pages = {434},
doi = {10.56028/aetr.16.1.434.2026}
}
EUVIMED