摘要
针对高维线性回归模型参数估计问题,基于迁移学习提出一种新算法,即联合Trans-Lasso算法。该算法使用高维统计技术有效结合大量与目标样本相似的辅助样本及少量的目标样本,在充分考虑辅助样本信息性的前提下对目标模型进行参数估计,有效降低了估计误差。通过数值模拟将迁移弹性网算法、OrcaleTrans-Lasso算法、联合Trans-Lasso算法与传统的Lasso算法估计性能进行比较,结果表明,联合Trans-Lasso算法的估计误差最小,提高率最高。
Aiming at the parameter estimation problem of high-dimensional linear regression model,the study proposes a new algorithm based on transfer learning-Joint Trans-Lasso algorithm.This algorithm uses high-dimensional statistical techniques to effectively combine a large number of auxiliary samples similar to the target sample,and a small number of target samples.Under the premise of fully considering the information of auxiliary samples,the parameters of the target model are estimated,which effectively reduces the estimation error.At the same time,the estimated performance of the migration elastic net algorithm,Orcale Trans-Lasso algorithm,the Joint Trans-Lasso algorithm and the traditional Lasso algorithm are compared through numerical simulation.The results show that the Joint Trans-Lasso algorithm has the smallest estimation error and the highest improvement rate.
作者
刘毅
Liu Yi(School of Mathematics and Statistics,North China University of Water Resources and Electric Power,Zhengzhou 450046,China)
出处
《黑龙江科学》
2023年第20期22-24,28,共4页
Heilongjiang Science
关键词
高维
Lasso
迁移学习
参数估计
High-dimensional
Lasso
Transfer learning
Parameter estimation