Improving Transfer Learning via Uniform Boundedness Prior

In this paper, we study the task of transfer learning and aim to improve its prediction accuracy by leveraging auxiliary data from different sources. Specifically, we introduce uniform bounded constraints associated with source data to the target Lasso model. To optimize such a structural statistical model, alternating direction method of multipliers (ADMM) is employed. We conduct experiments on both synthetic and real-world datasets. Experimental results demonstrate superior reliability and accuracy of the proposed model as compared to classical regression models including Lasso and Trans-Lasso.