Presentation 2012-12-12
Learning of Mahalanobis discriminant functions by a neural network : For non-normally distributed signals
Hiroyuki Izumi, Yoshifusa Ito, Cidambi Srinivasan,
PDF Download Page PDF download Page Link
Abstract(in Japanese) (See Japanese page)
Abstract(in English) Generally known is that a posterior probability and its monotone transform can be used as Bayesian discriminant functions and a neural network can learn the posterior probability. In the two-category normal-distribution case, a shift of the logit transform of the posterior probability can be used as a Mahalanobis discriminant function. The size of the shift can be estimated by the network having learned the posterior probability. In this paper we show, with simulations, that the Mahalanobis discriminant function can be obtained in such a way in the two-category non-normal-distribution case.
Keyword(in Japanese) (See Japanese page)
Keyword(in English) Neural network / Mahalanobis / Bayesian / discriminant function / non-normal distribution
Paper # NC2012-90
Date of Issue

Conference Information
Committee NC
Conference Date 2012/12/5(1days)
Place (in Japanese) (See Japanese page)
Place (in English)
Topics (in Japanese) (See Japanese page)
Topics (in English)
Chair
Vice Chair
Secretary
Assistant

Paper Information
Registration To Neurocomputing (NC)
Language JPN
Title (in Japanese) (See Japanese page)
Sub Title (in Japanese) (See Japanese page)
Title (in English) Learning of Mahalanobis discriminant functions by a neural network : For non-normally distributed signals
Sub Title (in English)
Keyword(1) Neural network
Keyword(2) Mahalanobis
Keyword(3) Bayesian
Keyword(4) discriminant function
Keyword(5) non-normal distribution
1st Author's Name Hiroyuki Izumi
1st Author's Affiliation Aichi Gakuin University()
2nd Author's Name Yoshifusa Ito
2nd Author's Affiliation Aichi Medical University
3rd Author's Name Cidambi Srinivasan
3rd Author's Affiliation University of Kentucky
Date 2012-12-12
Paper # NC2012-90
Volume (vol) vol.112
Number (no) 345
Page pp.pp.-
#Pages 6
Date of Issue