Semi-supervised Regression under Federated Learning Framework Based on Partial Information Estimation

Abstract

Nowadays, data becomes a valuable asset. More people even governments are increasing attention to data sharing and privacy protection. Federated Learning helps to protect privacy and break data island, which is favored by many scholars for its unique advantages. However, less research related to Federated Learning focused on regression problems, not to mention semi-supervised regression. In this paper, we propose the semi-supervised regression under Federated Learning framework based on Partial Information Estimation (Fed-PI), and the effectiveness of models are verified by Monte Carlo simulations. Through the use of Concrete dataset, we experimentally demonstrate that our algorithm performs better compare with the result of other models. These findings would encourage Federated Learning applied to more areas and increase interoperability.