| Autor: |
Zheng, Xiaohan, Zhang, Li, Yan, Leilei, Zhao, Lei |
| Předmět: |
|
| Zdroj: |
Applied Intelligence; Nov2024, Vol. 54 Issue 21, p10671-10694, 24p |
| Abstrakt: |
For the semi-supervised regression task, both the similarity of paired samples and the limited label information serve as core indicators. Nevertheless, most traditional semi-supervised regression methods cannot make full use of both simultaneously. To alleviate the above deficiency, this paper proposes a novel semi-supervised regression with label-guided adaptive graph optimization (LGAGO-SSR). Basically, LGAGO-SSR involves two phases: graph representation and label-guided adaptive graph construction. The first phase seeks two low-dimensional manifold spaces based on two similarity matrices. The second phase aims at adaptively learning these similarity matrices by integrating the data structure information in both the low-dimensional manifold spaces and the label spaces. Each phase has its optimization problems, and the final solution is obtained by iteratively solving problems in two phases. Additionally, the idea of decomposition optimization in twin support vector regression (TSVR) is used to accelerate the training of our LGAGO-SSR. Regression results on 12 benchmark datasets with different unlabeled rates demonstrate the effectiveness of LGAGO-SSR in semi-supervised regression tasks. [ABSTRACT FROM AUTHOR] |
| Databáze: |
Complementary Index |
| Externí odkaz: |
|
Full text from SpringerLink