Computational complexity of kernel-based density-ratio estimation: a condition number analysis

被引:8
作者
Kanamori, Takafumi [1 ]
Suzuki, Taiji [2 ]
Sugiyama, Masashi [3 ]
机构
[1] Nagoya Univ, Dept Comp Sci & Math Informat, Chikusa Ku, Nagoya, Aichi 4648601, Japan
[2] Univ Tokyo, Dept Math Informat, Bunkyo Ku, Tokyo 1138656, Japan
[3] Tokyo Inst Technol, Dept Comp Sci, Meguro Ku, Tokyo 1528552, Japan
关键词
Kernel; Density-ratio; Condition number; Smoothed analysis; SMOOTHED ANALYSIS; COVARIATE SHIFT; PROBABILISTIC ANALYSIS; DIVERGENCE; ALGORITHM; THEOREM;
D O I
10.1007/s10994-012-5323-6
中图分类号
TP18 [人工智能理论];
学科分类号
081104 ; 0812 ; 0835 ; 1405 ;
摘要
In this study, the computational properties of a kernel-based least-squares density-ratio estimator are investigated from the viewpoint of condition numbers. The condition number of the Hessian matrix of the loss function is closely related to the convergence rate of optimization and the numerical stability. We use smoothed analysis techniques and theoretically demonstrate that the kernel least-squares method has a smaller condition number than other M-estimators. This implies that the kernel least-squares method has desirable computational properties. In addition, an alternate formulation of the kernel least-squares estimator that possesses an even smaller condition number is presented. The validity of the theoretical analysis is verified through numerical experiments.
引用
收藏
页码:431 / 460
页数:30
相关论文
共 90 条
[81]   Mutual Information Approximation via Maximum Likelihood Estimation of Density Ratio [J].
Suzuki, Taiji ;
Sugiyama, Masashi ;
Tanaka, Toshiyuki .
2009 IEEE INTERNATIONAL SYMPOSIUM ON INFORMATION THEORY, VOLS 1- 4, 2009, :463-+
[82]   Mutual information estimation reveals global associations between stimuli and biological processes [J].
Suzuki, Taiji ;
Sugiyama, Masashi ;
Kanamori, Takafumi ;
Sese, Jun .
BMC BIOINFORMATICS, 2009, 10
[83]  
Tao T, 2007, ACM S THEORY COMPUT, P248
[84]   Characterizations, bounds, and probabilistic analysis of two complexity measures for linear programming problems [J].
Todd, MJ ;
Tunçel, L ;
Ye, YY .
MATHEMATICAL PROGRAMMING, 2001, 90 (01) :59-69
[85]   ROUNDING-OFF ERRORS IN MATRIX PROCESSES [J].
TURING, AM .
QUARTERLY JOURNAL OF MECHANICS AND APPLIED MATHEMATICS, 1948, 1 (03) :287-308
[86]  
Vershynin R, 2006, ANN IEEE SYMP FOUND, P133
[87]   NUMERICAL INVERTING OF MATRICES OF HIGH ORDER [J].
VONNEUMANN, J ;
GOLDSTINE, HH .
BULLETIN OF THE AMERICAN MATHEMATICAL SOCIETY, 1947, 53 (11) :1021-1099
[88]  
Yamada M., 2011, Journal of Machine Learning Research - Proceedings Track, V15, P807
[89]  
Yamada M, 2010, AAAI CONF ARTIF INTE, P643
[90]  
Zadrozny Bianca., 2004, LEARNING EVALUATING