The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. ex. Some numerals are expressed as "XNUMX".
Copyrights notice
The original paper is in English. Non-English content has been machine-translated and may contain typographical errors or mistranslations. Copyrights notice
La régression logistique du noyau (KLR) est un algorithme de classification puissant et flexible, qui possède la capacité de fournir la confiance de la prédiction de classe. Cependant, sa formation, généralement réalisée par des méthodes (quasi-)Newton, prend beaucoup de temps. Dans cet article, nous proposons un algorithme alternatif de classification probabiliste appelé Classificateur probabiliste des moindres carrés (LSPC). KLR modélise la probabilité a posteriori de classe par la combinaison log-linéaire des fonctions du noyau et ses paramètres sont appris par maximum de vraisemblance (régularisé). En revanche, LSPC utilise la combinaison linéaire de fonctions de noyau et ses paramètres sont appris par ajustement des moindres carrés régularisés de la véritable probabilité a posteriori de classe. Grâce à cette formulation des moindres carrés régularisés linéaires, la solution de LSPC peut être calculée analytiquement simplement en résolvant un système régularisé d'équations linéaires d'une manière par classe. Ainsi, LSPC est très efficace sur le plan informatique et numériquement stable. Par des expériences, nous montrons que le temps de calcul du LSPC est deux fois plus rapide que celui du KLR, avec une précision de classification comparable.
The copyright of the original papers published on this site belongs to IEICE. Unauthorized use of the original or translated papers is prohibited. See IEICE Provisions on Copyright for details.
Copier
Masashi SUGIYAMA, "Superfast-Trainable Multi-Class Probabilistic Classifier by Least-Squares Posterior Fitting" in IEICE TRANSACTIONS on Information,
vol. E93-D, no. 10, pp. 2690-2701, October 2010, doi: 10.1587/transinf.E93.D.2690.
Abstract: Kernel logistic regression (KLR) is a powerful and flexible classification algorithm, which possesses an ability to provide the confidence of class prediction. However, its training--typically carried out by (quasi-)Newton methods--is rather time-consuming. In this paper, we propose an alternative probabilistic classification algorithm called Least-Squares Probabilistic Classifier (LSPC). KLR models the class-posterior probability by the log-linear combination of kernel functions and its parameters are learned by (regularized) maximum likelihood. In contrast, LSPC employs the linear combination of kernel functions and its parameters are learned by regularized least-squares fitting of the true class-posterior probability. Thanks to this linear regularized least-squares formulation, the solution of LSPC can be computed analytically just by solving a regularized system of linear equations in a class-wise manner. Thus LSPC is computationally very efficient and numerically stable. Through experiments, we show that the computation time of LSPC is faster than that of KLR by two orders of magnitude, with comparable classification accuracy.
URL: https://global.ieice.org/en_transactions/information/10.1587/transinf.E93.D.2690/_p
Copier
@ARTICLE{e93-d_10_2690,
author={Masashi SUGIYAMA, },
journal={IEICE TRANSACTIONS on Information},
title={Superfast-Trainable Multi-Class Probabilistic Classifier by Least-Squares Posterior Fitting},
year={2010},
volume={E93-D},
number={10},
pages={2690-2701},
abstract={Kernel logistic regression (KLR) is a powerful and flexible classification algorithm, which possesses an ability to provide the confidence of class prediction. However, its training--typically carried out by (quasi-)Newton methods--is rather time-consuming. In this paper, we propose an alternative probabilistic classification algorithm called Least-Squares Probabilistic Classifier (LSPC). KLR models the class-posterior probability by the log-linear combination of kernel functions and its parameters are learned by (regularized) maximum likelihood. In contrast, LSPC employs the linear combination of kernel functions and its parameters are learned by regularized least-squares fitting of the true class-posterior probability. Thanks to this linear regularized least-squares formulation, the solution of LSPC can be computed analytically just by solving a regularized system of linear equations in a class-wise manner. Thus LSPC is computationally very efficient and numerically stable. Through experiments, we show that the computation time of LSPC is faster than that of KLR by two orders of magnitude, with comparable classification accuracy.},
keywords={},
doi={10.1587/transinf.E93.D.2690},
ISSN={1745-1361},
month={October},}
Copier
TY - JOUR
TI - Superfast-Trainable Multi-Class Probabilistic Classifier by Least-Squares Posterior Fitting
T2 - IEICE TRANSACTIONS on Information
SP - 2690
EP - 2701
AU - Masashi SUGIYAMA
PY - 2010
DO - 10.1587/transinf.E93.D.2690
JO - IEICE TRANSACTIONS on Information
SN - 1745-1361
VL - E93-D
IS - 10
JA - IEICE TRANSACTIONS on Information
Y1 - October 2010
AB - Kernel logistic regression (KLR) is a powerful and flexible classification algorithm, which possesses an ability to provide the confidence of class prediction. However, its training--typically carried out by (quasi-)Newton methods--is rather time-consuming. In this paper, we propose an alternative probabilistic classification algorithm called Least-Squares Probabilistic Classifier (LSPC). KLR models the class-posterior probability by the log-linear combination of kernel functions and its parameters are learned by (regularized) maximum likelihood. In contrast, LSPC employs the linear combination of kernel functions and its parameters are learned by regularized least-squares fitting of the true class-posterior probability. Thanks to this linear regularized least-squares formulation, the solution of LSPC can be computed analytically just by solving a regularized system of linear equations in a class-wise manner. Thus LSPC is computationally very efficient and numerically stable. Through experiments, we show that the computation time of LSPC is faster than that of KLR by two orders of magnitude, with comparable classification accuracy.
ER -