.png){kind=icon}
返回
Transfer learning for high-dimensional expectile regression
.png){kind=icon}
PRE
AI
DOI：10.1080/03610918.2025.2578277.png)
.png){kind=icon}
原文求助.png){kind=icon}
分享.png){kind=icon}
收藏
摘要
En 中文
Transfer learning has emerged as a pivotal way for enhancing the performance of target models through the transfer of source data, especially when there is a relative scarcity of data available for the target domain. High-dimensional data often show variability resulting from non-uniform covariate influences or heteroscedasticity, causing heterogeneity. Nevertheless, contemporary transfer learning methodologies frequently disregard the heterogeneity and heavy-tailed attributes of high-dimensional data, which may ultimately undermine their efficacy. Penalized expectile regression method offers effective means for detecting heteroscedasticity within high-dimensional datasets. In this study, a robust transfer learning approach using expectile regression is proposed. With identified transferable sources, this paper puts forward a two-stage transfer learning approach. To lessen the detrimental impact of using non-informative sources, we provide an approach to selecting transferable sources that is based on cross-validation. Our suggested method's efficacy is demonstrated by a practical study and numerical simulations.
Keyword:
Expectile regression
Lasso
transfer learning
期刊
C
IF:
0.8
论文数:
226
被引数:
4.7K
机构
引用论文
分享收藏Estimation of high dimensional mean regression in the absence of symmetry and light tail assumptions
OAAI
OAAI分享收藏分享收藏One-step sparse estimates in nonconcave penalized likelihood models非凹惩罚似然模型中的一步稀疏估计.png)
ANNALS OF STATISTICS
OAAI
ANNALS OF STATISTICSIF3.7
OAAI分享收藏
查看更多内容