隨著機器學習與大數據的蓬勃發展，遷移學習已成為高維資料分析中的重要工具，而如何有效結合多組代理資料與目標資料以進行準確的模型選擇，則是高維線性迴歸中的關鍵挑戰。本文將回顧以下重要文獻作為研究基礎：Ing 與 Lai (2011) 提出的 OGA+HDIC+Trim 模型選擇方法，及 Li、Cai 和 Li (2022) 所提出的多輔助資料遷移學習方法 Trans-Lasso。基於這些研究成果，本文提出了兩種創新方法——Trans-OGA 與 Two-Stage OGA，兩種方法均以 OGA+HDIC+Trim 作為核心選模策略，Trans-OGA 透過高維度信息準則（HDIC）對代理資料進行分群並整合；Two-Stage OGA 則根據變數選取結果，逐變數合併代理資料。為驗證本研究方法的有效性，我們設計了多個模擬實驗與真實資料分析，結果顯示，本文提出之方法在多種設定下，顯著提升了估計準確性與模型選擇之正確性。

With the advancement of machine learning and big data, transfer learning has become essential in high-dimensional data analysis. Integrating multiple proxy datasets with target data for accurate model selection remains a key challenge in high-dimensional linear regression. This study builds on two influential works: the OGA+HDIC+Trim method by Ing and Lai (2011) and the Trans-Lasso framework by Li, Cai, and Li (2022). We propose two novel methods—Trans-OGA and Two-Stage OGA—both utilizing OGA+HDIC+Trim for model selection. Trans-OGA clusters and integrates proxy data using the High-Dimensional Information Criterion (HDIC), while Two-Stage OGA merges datasets based on variable selection results. Through simulations and real data experiments, our methods demonstrate substantial improvements in estimation accuracy and model selection under various settings.

1. Ching-Kang Ing and Tze Leung Lai. A Stepwise Regression Method and Consistent Model Selection for High-Dimensional Sparse Linear Models. Statistica Sinica, 21:1473-1513, 2011.

2. Sai Li and T. Tony Cai and Hongzhe Li. Transfer Learning for High-Dimensional Linear Regression: Prediction, Estimation and Minimax Optimality. Journal of the Royal Statistical Society: Series B (Statistical Methodology), 84(1):149-173, 2022.