Semi-supervised regression using diffusion on graphs

• DOI: 10.1016/j.asoc.2021.107188
• Publication Date: 2021-02-21T15:10:53Z
• Authors: Mohan Timilsina; Alejandro Figueroa; Mathieu d’Aquin; Haixuan Yang
• Keywords: Supervised Learning

In real-world machine learning applications, unlabeled training data are readily available, but labeled expensive and hard to obtain. Therefore, semi-supervised algorithms have gathered much attention. Previous studies in this area mainly focused on a classification problem, whereas regression has received less paper, we proposed novel algorithm using heat diffusion with boundary-condition that guarantees closed-form solution. Experiments from artificial real datasets business, biomedical, physical, social domain show the boundary-based method can effectively outperform top state of art methods.