Presentation 2015-11-26
[Poster Presentation] Approximate Distribution Followed by A Principal Component Term Corresponding with the Population Eigenvalue of zero value in the Q Statistic
Yasuyuki Kobayashi,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) The Mahalanobis distance requires all the population eigenvalues of the population covariance matrix larger than zero. However, some population eigenvalues of the population covariance matrix from the probability distribution followed by learning samples may contain zero value. Including this case, when the eigenvalues of the sample covariance are extremely small, it is said that the Q statistic is adequate to analyze with the Mahalanobis distance. However, the existing approximate distribution models followed by the Q statistic cannot handle the case including the population eigenvalues of zero value. Therefore, this report has proposed the approximate probability distribution models followed by the sample eigenvalue and the principal component of the Q statistic corresponding with the population eigenvalue of zero value, considering the effect of numerical errors theoretically. The numerical experiment has shown that the empirical distributions obtained by Monte Carlo simulation coincides with the proposed approximate distribution models on the whole.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Mahalanobis Distance / Q Statistic / Principal Component / Probability Distribution / Approximate Model
Paper # IBISML2015-52
Date of Issue 2015-11-19 (IBISML)

Conference Information
Committee IBISML
Conference Date 2015/11/25(3days)
Place (in Japanese) (See Japanese page)
Place (in English) Epochal Tsukuba
Topics (in Japanese) (See Japanese page)
Topics (in English) Information-Based Induction Science Workshop (IBIS2015)
Chair Takashi Washio(Osaka Univ.)
Vice Chair Kenji Fukumizu(ISM) / Masashi Sugiyama(Tokyo Inst. of Tech.)
Secretary Kenji Fukumizu(ISM) / Masashi Sugiyama(Kyoto Univ.)
Assistant Koji Tsuda(Univ. of Tokyo) / Hisashi Kashima(Kyoto Univ.)

Paper Information
Registration To Technical Committee on Infomation-Based Induction Sciences and Machine Learning
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) [Poster Presentation] Approximate Distribution Followed by A Principal Component Term Corresponding with the Population Eigenvalue of zero value in the Q Statistic
Sub Title (in English)
Keyword(1) Mahalanobis Distance
Keyword(2) Q Statistic
Keyword(3) Principal Component
Keyword(4) Probability Distribution
Keyword(5) Approximate Model
1st Author's Name Yasuyuki Kobayashi
1st Author's Affiliation Teikyo University(Teikyo Univ.)
Date 2015-11-26
Paper # IBISML2015-52
Volume (vol) vol.115
Number (no) IBISML-323
Page pp.pp.1-7(IBISML),
#Pages 7
Date of Issue 2015-11-19 (IBISML)