Semi-Supervised Learning with Network Information

• Activity Number: 139 - Recent Advances of Semi-Supervised Learning: Techniques and Applications
• Type: Invited
• Date/Time: Tuesday, August 10, 2021 : 10:00 AM to 11:50 AM
• Sponsor: Section on Statistical Learning and Data Science
• Author(s): Linda Zhao and Junhui Cai* and Haipeng Shen and Dan Yang and Wu Zhu
• Companies: University of Pennsylvania and University of Pennsylvania and University of Hong Kong and University of Hong Kong and University of Pennsylvania
• Keywords: semi-supervised; network; equity holding network

Abstract

Networks are ubiquitous in our lives and play a crucial role in information transmission. The network position, usually captured by centrality, affects individual’s decision making and thus provides information for inference and prediction. Though with a complete network, it is often the case that we might not be able to observe all information of individuals in the network (e.g. Facebook observes the friendship networks, but some users’ information might be missing). We propose a semi-supervised method for a regression problem where a network is fully observed but covariates and outcome are missing at random. By incorporating the network information, the method provides better estimates and prediction, but also a better estimate of centrality. We illustrate our method via a real data example of inferring the performance of Chinese firms with a complete equity holding network.

It is a joint work with Junhui Cai, Haipeng Shen, Dan Yang and Wu Zhu.