Scholarlink

<p>This paper investigates transfer learning in heterogeneous multi-source environments characterized by distributional divergence between the target and auxiliary domains. To mitigate challenges of statistical bias and computational efficiency, we propose a Sparse Optimization for Transfer Learning (SOTL) framework based on <InlineEquation ID="IEq4"> <EquationSource Format="TEX">\(L_0\)</EquationSource> </InlineEquation>-regularization. The method extends the Joint Estimation Transferred from Strata (JETS) paradigm with two key innovations: (1) <InlineEquation ID="IEq5"> <EquationSource Format="TEX">\(L_0\)</EquationSource> </InlineEquation>-constrained exact sparsity for parameter space compression and complexity reduction, and (2) refined optimization that prioritizes target parameters over redundant ones. Simulation studies demonstrate that SOTL substantially enhances both estimation accuracy and computational speed, particularly under adversarial auxiliary domain settings. Empirical analyses on the publicly available Communities and Crime dataset and Beijing Multi-site Air Quality dataset further confirm the statistical robustness of SOTL in cross-domain transfer.</p>

错误：搜索内容不能为空，请输入英文关键词
错误：关键词超出字数限制，请精简
高级检索

Sparse optimization for transfer learning: a \(L_0\)-regularized framework for multi-source domain adaptation

  • 期刊论文

  • 发表日期：

    2026年2月24日

查看全文
  • Chenqi Gong,
  • Hu Yang

摘要

This paper investigates transfer learning in heterogeneous multi-source environments characterized by distributional divergence between the target and auxiliary domains. To mitigate challenges of statistical bias and computational efficiency, we propose a Sparse Optimization for Transfer Learning (SOTL) framework based on \(L_0\) -regularization. The method extends the Joint Estimation Transferred from Strata (JETS) paradigm with two key innovations: (1) \(L_0\) -constrained exact sparsity for parameter space compression and complexity reduction, and (2) refined optimization that prioritizes target parameters over redundant ones. Simulation studies demonstrate that SOTL substantially enhances both estimation accuracy and computational speed, particularly under adversarial auxiliary domain settings. Empirical analyses on the publicly available Communities and Crime dataset and Beijing Multi-site Air Quality dataset further confirm the statistical robustness of SOTL in cross-domain transfer.