Group Transfer Learning for High‐Dimensional Linear Regression

ABSTRACT Transfer learning is a powerful technique for enhancing parameter estimation in target datasets by incorporating information from related source datasets. However, traditional transfer learning methods often assess the overall transferability of source data without fully considering partially transferable components. To address this, we propose a novel group transfer learning (GTL) algorithm that efficiently integrates both fully and partially transferable data by grouping parameters based on their transferability. Our approach includes an adaptive mechanism to identify transferable parameter groups when their transferability is unknown. We provide theoretical guarantees for the effectiveness of the GTL algorithm in parameter estimation and its ability to accurately detect transferable segments. Extensive simulations and real‐world case studies demonstrate that our method surpasses existing approaches in estimation accuracy and robustness.