Current Result Document : 0 / 0
| 한글제목(Korean Title) |
Multilevel Semi-Supervised Regression with Hierarchically Structured Networks |
| 영문제목(English Title) |
Multilevel Semi-Supervised Regression with Hierarchically Structured Networks |
| 저자(Author) |
김명준
김재현
신현정
Myungjun Kim
Jaehyun Kim
Hyunjung Shin
|
| 원문수록처(Citation) |
VOL 47 NO. 02 PP. 0748 ~ 0750 (2020. 12) |
한글내용
(Korean Abstract) |
|
영문내용
(English Abstract) |
In many situations, data have a hierarchical or clustered structure in diverse domains including biological sciences, parentoffspring relations, designed experiments, etc. In such cases conventional approaches exhibit difficulties of handling large scale data with limited number of labeled data. As a result, this inspires new approaches for utilizing hierarchically structured data. In this paper, we propose a network-based approach to circumvent the problems by extending ideas from semi-supervised learning for hierarchically structured networks to a regression framework. The proposed method conceptualizes the given data into a multilayer network setting, in which interlayer connections allow information in one layer can be propagated to others. Then, to impose a regression framework, we develop a Gaussian process approach, which provides uncertainties in prediction along with advantages in model selection. The experimental results show that the proposed algorithms perform well with data of hierarchical structure and exhibit its strength in situations of scarce amount of labeled data compared to related approaches.
|
| 키워드(Keyword) |
|
| 원문 |
PDF 다운로드
|
