Stein's identity, Fisher information, and projection pursuit: A triangulation

被引：2

作者：

DasGupta, Anirban ^{[1
]}

机构：

[1] Purdue Univ, W Lafayette, IN 47907 USA

来源：

JOURNAL OF STATISTICAL PLANNING AND INFERENCE | 2007年 / 137卷 / 11期

关键词：

classification; elliptically symmetric; fisher LDF; fisher information; projection pursuit; Stein's identity; unimodality;

D O I：

10.1016/j.jspi.2007.03.019

中图分类号：

O21 [概率论与数理统计]; C8 [统计学];

学科分类号：

020208 ; 070103 ; 0714 ;

摘要：

Two separate structure discovery properties of Fisher's LDF are derived in a mixture multivariate normal setting. One of the properties is related to Fisher information and is proved by using Stein's identity. The other property is on lack of unimodality. The properties are used to give three selection rules for choice of informative projections of high-dimensional data, not necessarily multivariate normal. Their usefulness in the two group-classification problem is studied theoretically and by means of examples. Extensions and various issues about practical implementation are discussed. (C) 2007 Published by Elsevier B.V.

引用

页码：3394 / 3409

页数：16

共 28 条

[1]

[Anonymous], HDB STAT

[2]

BARRON A, 1993, IEEE T INFORMATION T, V3, P930

[3]

Chen D., 1993, Approximation Theory and its Applications, V9, P17, DOI DOI 10.1007/BF02836480

[4]

CHEN ZY, 1994, IMS LNMS, V24

[5]

Cook D., 1993, J COMPUTATIONAL GRAP, V2, P225, DOI [DOI 10.1080/10618600.1993.10474610, DOI 10.2307/1390644]

[6]

DASGUPTA S, 1980, SANKHYA A, V42, P219

[7] NEW ERROR BOUNDS WITH THE NEAREST NEIGHBOR RULE [J].

DEVIJVER, PA .

IEEE TRANSACTIONS ON INFORMATION THEORY, 1979, 25 (06) :749-753

[8]

Devroye L., 1996, A probabilistic theory of pattern recognition

[9] ASYMPTOTICS OF GRAPHICAL PROJECTION PURSUIT [J].

DIACONIS, P ;

FREEDMAN, D .

ANNALS OF STATISTICS, 1984, 12 (03) :793-815

[10] ON NONLINEAR FUNCTIONS OF LINEAR-COMBINATIONS [J].

DIACONIS, P ;

SHAHSHAHANI, M .

SIAM JOURNAL ON SCIENTIFIC AND STATISTICAL COMPUTING, 1984, 5 (01) :175-191

← 1 2 3 →