Metric learning by directly minimizing the k-NN training error

Konstantin Chernoff; Marco Loog; Mads Nielsen

Metric learning by directly minimizing the k-NN training error

Konstantin Chernoff, Marco Loog, Mads Nielsen

Department of Computer Science

1 Citation (Scopus)

Abstract

This paper presents an approach for computing global distance metrics that minimize the k-NN leave-one-out (LOO) error. The approach optimizes an energy function that corresponds to a smoothened version of the k-NN LOO error. The generalization of the proposed approach is further improved by controlling the k parameter through a heuristic. Evaluation of the proposed approach on several public datasets showed that it was able to compete with an established state-of-the art approach.

Original language	English
Title of host publication	2012 21st International Conference on Pattern Recognition (ICPR)
Number of pages	4
Publisher	IEEE
Publication date	2012
Pages	1265-1268
ISBN (Print)	978-1-4673-2216-4
Publication status	Published - 2012
Event	21st International Conference on Pattern Recognition - Tsukuba Science City , Japan Duration: 11 Nov 2012 → 15 Nov 2012 Conference number: 21

Conference

Conference	21st International Conference on Pattern Recognition
Number	21
Country/Territory	Japan
City	Tsukuba Science City
Period	11/11/2012 → 15/11/2012

Access to Document

http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=6460369&queryText%3DMetric+learning+by+directly+minimizing+the+k-NN+training+error

Cite this

Chernoff, K, Loog, M & Nielsen, M 2012, Metric learning by directly minimizing the k-NN training error. in 2012 21st International Conference on Pattern Recognition (ICPR). IEEE, pp. 1265-1268, 21st International Conference on Pattern Recognition , Tsukuba Science City , Japan, 11/11/2012. <http://ieeexplore.ieee.org/xpl/articleDetails.jsp?tp=&arnumber=6460369&queryText%3DMetric+learning+by+directly+minimizing+the+k-NN+training+error>

@inproceedings{e5090c5e40194e038ad5cac9f01920b0,

title = "Metric learning by directly minimizing the k-NN training error",

abstract = "This paper presents an approach for computing global distance metrics that minimize the k-NN leave-one-out (LOO) error. The approach optimizes an energy function that corresponds to a smoothened version of the k-NN LOO error. The generalization of the proposed approach is further improved by controlling the k parameter through a heuristic. Evaluation of the proposed approach on several public datasets showed that it was able to compete with an established state-of-the art approach.",

keywords = "learning (artificial intelligence), LOO error minimization, energy function, global distance metrics, k-NN training error minimization, leave-one-out error minimization, metric learning, public datasets, Art, Iris, Machine learning, Measurement, Optimization, Training, Vectors",

author = "Konstantin Chernoff and Marco Loog and Mads Nielsen",

year = "2012",

language = "English",

isbn = "978-1-4673-2216-4 ",

pages = "1265--1268",

booktitle = "2012 21st International Conference on Pattern Recognition (ICPR)",

publisher = "IEEE",

note = "21st International Conference on Pattern Recognition ; Conference date: 11-11-2012 Through 15-11-2012",

}

TY - GEN

T1 - Metric learning by directly minimizing the k-NN training error

AU - Chernoff, Konstantin

AU - Loog, Marco

AU - Nielsen, Mads

N1 - Conference code: 21

PY - 2012

Y1 - 2012

N2 - This paper presents an approach for computing global distance metrics that minimize the k-NN leave-one-out (LOO) error. The approach optimizes an energy function that corresponds to a smoothened version of the k-NN LOO error. The generalization of the proposed approach is further improved by controlling the k parameter through a heuristic. Evaluation of the proposed approach on several public datasets showed that it was able to compete with an established state-of-the art approach.

AB - This paper presents an approach for computing global distance metrics that minimize the k-NN leave-one-out (LOO) error. The approach optimizes an energy function that corresponds to a smoothened version of the k-NN LOO error. The generalization of the proposed approach is further improved by controlling the k parameter through a heuristic. Evaluation of the proposed approach on several public datasets showed that it was able to compete with an established state-of-the art approach.

KW - learning (artificial intelligence)

KW - LOO error minimization

KW - energy function

KW - global distance metrics

KW - k-NN training error minimization

KW - leave-one-out error minimization

KW - metric learning

KW - public datasets

KW - Art

KW - Iris

KW - Machine learning

KW - Measurement

KW - Optimization

KW - Training

KW - Vectors

M3 - Article in proceedings

SN - 978-1-4673-2216-4

SP - 1265

EP - 1268

BT - 2012 21st International Conference on Pattern Recognition (ICPR)

PB - IEEE

T2 - 21st International Conference on Pattern Recognition

Y2 - 11 November 2012 through 15 November 2012

ER -