Global Metric Learning by Gradient Descent

Jens Hocke; Thomas Martinetz

doi:10.1007/978-3-319-11179-7_17

Global Metric Learning by Gradient Descent

Jens Hocke, Thomas Martinetz

Institut für Neuro- und Bioinformatik

6 Zitate (Scopus)

Abstract

The k-NN classifier can be very competitive if an appropriate distance measure is used. It is often used in applications because the classification decisions are easy to interpret. Here, we demonstrate how to find a good Mahalanobis distance for k-NN classification by a simple gradient descent without any constraints. The cost term uses global distances and unlike other methods there is a soft transition in the influence of data points. It is evaluated and compared to other metric learning and feature weighting methods on datasets from the UCI repository, where the described gradient method also shows a high robustness. In the comparison the advantages of global approaches are demonstrated.

Originalsprache	Englisch
Titel	Artificial Neural Networks and Machine Learning - ICANN 2014
Redakteure/-innen	Stefan Wermter, Cornelius Weber, Wlodzislaw Duch, Timo Honkela, Petia Koprinkova-Hristova, Sven Magg, Günther Palm, Allessandro E.P. Villa
Seitenumfang	7
Herausgeber (Verlag)	Springer Verlag
Erscheinungsdatum	01.01.2014
Auflage	1
Seiten	129-135
ISBN (Print)	978-3-319-11178-0
ISBN (elektronisch)	978-3-319-11179-7
DOIs	https://doi.org/10.1007/978-3-319-11179-7_17
Publikationsstatus	Veröffentlicht - 01.01.2014
Veranstaltung	24th International Conference on Artificial Neural Networks - Depat. of Informatics, Knowledge Technology, University of hamburg, Hamburg, Deutschland Dauer: 15.09.2014 → 19.09.2014 http://icann2014.org

UN SDGs

Dieser Output leistet einen Beitrag zu folgendem(n) Ziel(en) für nachhaltige Entwicklung

SDG 3 – Gesundheit und Wohlergehen
SDG 9 – Industrie, Innovation und Infrastruktur

Dokumente

10.1007/978-3-319-11179-7_17

https://link.springer.com/chapter/10.1007/978-3-319-11179-7_17

Zitieren

Hocke, J., & Martinetz, T. (2014). Global Metric Learning by Gradient Descent. in S. Wermter, C. Weber, W. Duch, T. Honkela, P. Koprinkova-Hristova, S. Magg, G. Palm, & A. E. P. Villa (Hrsg.), Artificial Neural Networks and Machine Learning - ICANN 2014 (1 Aufl., S. 129-135). (Theoretical Computer Science and General Issues; Band 8681). Springer Verlag. https://doi.org/10.1007/978-3-319-11179-7_17

Hocke, Jens ; Martinetz, Thomas. / Global Metric Learning by Gradient Descent. Artificial Neural Networks and Machine Learning - ICANN 2014. Hrsg. / Stefan Wermter ; Cornelius Weber ; Wlodzislaw Duch ; Timo Honkela ; Petia Koprinkova-Hristova ; Sven Magg ; Günther Palm ; Allessandro E.P. Villa. 1. Aufl. Springer Verlag, 2014. S. 129-135 (Theoretical Computer Science and General Issues).

@inproceedings{2613824b8f9640b8b4fb0f89a6d5f8f7,

title = "Global Metric Learning by Gradient Descent",

abstract = "The k-NN classifier can be very competitive if an appropriate distance measure is used. It is often used in applications because the classification decisions are easy to interpret. Here, we demonstrate how to find a good Mahalanobis distance for k-NN classification by a simple gradient descent without any constraints. The cost term uses global distances and unlike other methods there is a soft transition in the influence of data points. It is evaluated and compared to other metric learning and feature weighting methods on datasets from the UCI repository, where the described gradient method also shows a high robustness. In the comparison the advantages of global approaches are demonstrated.",

author = "Jens Hocke and Thomas Martinetz",

year = "2014",

month = jan,

day = "1",

doi = "10.1007/978-3-319-11179-7\_17",

language = "English",

isbn = "978-3-319-11178-0",

series = "Theoretical Computer Science and General Issues",

publisher = "Springer Verlag",

pages = "129--135",

editor = "Stefan Wermter and Cornelius Weber and Wlodzislaw Duch and Timo Honkela and Petia Koprinkova-Hristova and Sven Magg and G{\"u}nther Palm and Villa, \{Allessandro E.P.\}",

booktitle = "Artificial Neural Networks and Machine Learning - ICANN 2014",

edition = "1",

note = "24th International Conference on Artificial Neural Networks ; Conference date: 15-09-2014 Through 19-09-2014",

url = "http://icann2014.org",

}

Hocke, J & Martinetz, T 2014, Global Metric Learning by Gradient Descent. in S Wermter, C Weber, W Duch, T Honkela, P Koprinkova-Hristova, S Magg, G Palm & AEP Villa (Hrsg.), Artificial Neural Networks and Machine Learning - ICANN 2014. 1 Aufl., Theoretical Computer Science and General Issues, Bd. 8681, Springer Verlag, S. 129-135, 24th International Conference on Artificial Neural Networks, Hamburg, Deutschland, 15.09.14. https://doi.org/10.1007/978-3-319-11179-7_17

Global Metric Learning by Gradient Descent. / Hocke, Jens; Martinetz, Thomas.
Artificial Neural Networks and Machine Learning - ICANN 2014. Hrsg. / Stefan Wermter; Cornelius Weber; Wlodzislaw Duch; Timo Honkela; Petia Koprinkova-Hristova; Sven Magg; Günther Palm; Allessandro E.P. Villa. 1. Aufl. Springer Verlag, 2014. S. 129-135 (Theoretical Computer Science and General Issues; Band 8681).

Publikation: Kapitel in Büchern/Berichten/Konferenzbänden › Konferenzbeitrag › Begutachtung

TY - GEN

T1 - Global Metric Learning by Gradient Descent

AU - Hocke, Jens

AU - Martinetz, Thomas

PY - 2014/1/1

Y1 - 2014/1/1

N2 - The k-NN classifier can be very competitive if an appropriate distance measure is used. It is often used in applications because the classification decisions are easy to interpret. Here, we demonstrate how to find a good Mahalanobis distance for k-NN classification by a simple gradient descent without any constraints. The cost term uses global distances and unlike other methods there is a soft transition in the influence of data points. It is evaluated and compared to other metric learning and feature weighting methods on datasets from the UCI repository, where the described gradient method also shows a high robustness. In the comparison the advantages of global approaches are demonstrated.

AB - The k-NN classifier can be very competitive if an appropriate distance measure is used. It is often used in applications because the classification decisions are easy to interpret. Here, we demonstrate how to find a good Mahalanobis distance for k-NN classification by a simple gradient descent without any constraints. The cost term uses global distances and unlike other methods there is a soft transition in the influence of data points. It is evaluated and compared to other metric learning and feature weighting methods on datasets from the UCI repository, where the described gradient method also shows a high robustness. In the comparison the advantages of global approaches are demonstrated.

U2 - 10.1007/978-3-319-11179-7_17

DO - 10.1007/978-3-319-11179-7_17

M3 - Conference contribution

SN - 978-3-319-11178-0

T3 - Theoretical Computer Science and General Issues

SP - 129

EP - 135

BT - Artificial Neural Networks and Machine Learning - ICANN 2014

A2 - Wermter, Stefan

A2 - Weber, Cornelius

A2 - Duch, Wlodzislaw

A2 - Honkela, Timo

A2 - Koprinkova-Hristova, Petia

A2 - Magg, Sven

A2 - Palm, Günther

A2 - Villa, Allessandro E.P.

PB - Springer Verlag

T2 - 24th International Conference on Artificial Neural Networks

Y2 - 15 September 2014 through 19 September 2014

ER -

Global Metric Learning by Gradient Descent

Abstract

UN SDGs

Dokumente

Fingerprint

Zitieren