机构:
Stanford Univ, Elect Engn Dept, Stanford, CA 94305 USAStanford Univ, Elect Engn Dept, Stanford, CA 94305 USA
Sweetkind-Singer, JA
[1
]
机构:
[1] Stanford Univ, Elect Engn Dept, Stanford, CA 94305 USA
来源:
2003 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY - PROCEEDINGS
|
2003年
关键词:
D O I:
暂无
中图分类号:
TP [自动化技术、计算机技术];
学科分类号:
0812 ;
摘要:
Regularization penalties are commonly used in linear regression to reduce overfitting [1]. We introduce a log regularization penalty, motivated by a minimum-description-length (MDL) perspective [2] and from ideas in algorithmic complexity [3], and compare it to the more commonly used penalties known as ridge regression and the lasso [1].
机构:
School of Economics and Management, China University of Geosciences,Wuhan
Center for Resources and Environmental Economic Research, China University of Geosciences,WuhanSchool of Economics and Management, China University of Geosciences,Wuhan
SHI Yue-yong
CAO Yong-xiu
论文数: 0引用数: 0
h-index: 0
机构:
School of Statistics and Mathematics, Zhongnan University of Economics and LawSchool of Economics and Management, China University of Geosciences,Wuhan
CAO Yong-xiu
YU Ji-chang
论文数: 0引用数: 0
h-index: 0
机构:
School of Statistics and Mathematics, Zhongnan University of Economics and LawSchool of Economics and Management, China University of Geosciences,Wuhan
YU Ji-chang
JIAO Yu-ling
论文数: 0引用数: 0
h-index: 0
机构:
School of Statistics and Mathematics, Zhongnan University of Economics and LawSchool of Economics and Management, China University of Geosciences,Wuhan
机构:
Hubei Normal Univ, Sch Math & Stat, Huangshi 435002, Hubei, Peoples R ChinaHubei Normal Univ, Sch Math & Stat, Huangshi 435002, Hubei, Peoples R China
Hu, Hongchang
Zeng, Zhen
论文数: 0引用数: 0
h-index: 0
机构:
Nanjin Univ Finana & Econ, Sch Appl Math, Nanjing, Peoples R ChinaHubei Normal Univ, Sch Math & Stat, Huangshi 435002, Hubei, Peoples R China