Learning Distance Metrics for Multi-Label Classification

Henry Gouk; Bernhard Pfahringer; Michael Cree

Learning Distance Metrics for Multi-Label Classification

Henry Gouk
, Bernhard Pfahringer
, Michael Cree

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

Abstract

Distance metric learning is a well studied problem in the field of machine learning, where it is typically used to improve the accuracy of instance based learning techniques. In this paper we propose a distance metric learning algorithm that is specialised for multi-label classification tasks, rather than the multiclass setting considered by most work in this area. The method trains an embedder that can transform instances into a feature space where Euclidean distance provides an estimate of the Jaccard distance between the corresponding label vectors. In addition to a linear Mahalanobis style metric, we also present a nonlinear extension that provides a substantial boost in performance. We show that this technique significantly improves upon current approaches for instance based multi-label classification, and also enables interesting data visualisations.

Original language	English
Title of host publication	Proceedings of The 8th Asian Conference on Machine Learning
Editors	Robert J. Durrant, Kee-Eung Kim
Place of Publication	The University of Waikato, Hamilton, New Zealand
Publisher	PMLR
Pages	318-333
Number of pages	16
Volume	63
Publication status	E-pub ahead of print - 18 Nov 2016
Event	8th Asian Conference on Machine Learning - Hamilton, New Zealand Duration: 16 Nov 2016 → 18 Nov 2016 http://www.acml-conf.org/2016/

Publication series

Name	Proceedings of Machine Learning Research
Publisher	PMLR
Volume	63
ISSN (Electronic)	2640-3498

Conference

Conference	8th Asian Conference on Machine Learning
Abbreviated title	ACML 2016
Country/Territory	New Zealand
City	Hamilton
Period	16/11/16 → 18/11/16
Internet address	http://www.acml-conf.org/2016/

Keywords / Materials (for Non-textual outputs)

Distance Metric Learning
Multi-Label Classification
Instance Based Learning

Access to Document

http://proceedings.mlr.press/v63/Gouk8.html

Cite this

@inproceedings{4c1e27411a7941738ba7bc002806ccd9,

title = "Learning Distance Metrics for Multi-Label Classification",

abstract = "Distance metric learning is a well studied problem in the field of machine learning, where it is typically used to improve the accuracy of instance based learning techniques. In this paper we propose a distance metric learning algorithm that is specialised for multi-label classification tasks, rather than the multiclass setting considered by most work in this area. The method trains an embedder that can transform instances into a feature space where Euclidean distance provides an estimate of the Jaccard distance between the corresponding label vectors. In addition to a linear Mahalanobis style metric, we also present a nonlinear extension that provides a substantial boost in performance. We show that this technique significantly improves upon current approaches for instance based multi-label classification, and also enables interesting data visualisations.",

keywords = "Distance Metric Learning, Multi-Label Classification, Instance Based Learning",

author = "Henry Gouk and Bernhard Pfahringer and Michael Cree",

year = "2016",

month = nov,

day = "18",

language = "English",

volume = "63",

series = "Proceedings of Machine Learning Research",

publisher = "PMLR",

pages = "318--333",

editor = "Durrant, \{Robert J.\} and Kee-Eung Kim",

booktitle = "Proceedings of The 8th Asian Conference on Machine Learning",

note = "8th Asian Conference on Machine Learning, ACML 2016 ; Conference date: 16-11-2016 Through 18-11-2016",

url = "http://www.acml-conf.org/2016/",

}

Gouk, H, Pfahringer, B & Cree, M 2016, Learning Distance Metrics for Multi-Label Classification. in RJ Durrant & K-E Kim (eds), Proceedings of The 8th Asian Conference on Machine Learning. vol. 63, Proceedings of Machine Learning Research, vol. 63, PMLR, The University of Waikato, Hamilton, New Zealand, pp. 318-333, 8th Asian Conference on Machine Learning, Hamilton, New Zealand, 16/11/16. <http://proceedings.mlr.press/v63/Gouk8.html>

Learning Distance Metrics for Multi-Label Classification. / Gouk, Henry; Pfahringer, Bernhard; Cree, Michael.
Proceedings of The 8th Asian Conference on Machine Learning. ed. / Robert J. Durrant; Kee-Eung Kim. Vol. 63 The University of Waikato, Hamilton, New Zealand: PMLR, 2016. p. 318-333 (Proceedings of Machine Learning Research; Vol. 63).

Research output: Chapter in Book/Report/Conference proceeding › Conference contribution

TY - GEN

T1 - Learning Distance Metrics for Multi-Label Classification

AU - Gouk, Henry

AU - Pfahringer, Bernhard

AU - Cree, Michael

PY - 2016/11/18

Y1 - 2016/11/18

N2 - Distance metric learning is a well studied problem in the field of machine learning, where it is typically used to improve the accuracy of instance based learning techniques. In this paper we propose a distance metric learning algorithm that is specialised for multi-label classification tasks, rather than the multiclass setting considered by most work in this area. The method trains an embedder that can transform instances into a feature space where Euclidean distance provides an estimate of the Jaccard distance between the corresponding label vectors. In addition to a linear Mahalanobis style metric, we also present a nonlinear extension that provides a substantial boost in performance. We show that this technique significantly improves upon current approaches for instance based multi-label classification, and also enables interesting data visualisations.

AB - Distance metric learning is a well studied problem in the field of machine learning, where it is typically used to improve the accuracy of instance based learning techniques. In this paper we propose a distance metric learning algorithm that is specialised for multi-label classification tasks, rather than the multiclass setting considered by most work in this area. The method trains an embedder that can transform instances into a feature space where Euclidean distance provides an estimate of the Jaccard distance between the corresponding label vectors. In addition to a linear Mahalanobis style metric, we also present a nonlinear extension that provides a substantial boost in performance. We show that this technique significantly improves upon current approaches for instance based multi-label classification, and also enables interesting data visualisations.

KW - Distance Metric Learning

KW - Multi-Label Classification

KW - Instance Based Learning

M3 - Conference contribution

VL - 63

T3 - Proceedings of Machine Learning Research

SP - 318

EP - 333

BT - Proceedings of The 8th Asian Conference on Machine Learning

A2 - Durrant, Robert J.

A2 - Kim, Kee-Eung

PB - PMLR

CY - The University of Waikato, Hamilton, New Zealand

T2 - 8th Asian Conference on Machine Learning

Y2 - 16 November 2016 through 18 November 2016

ER -

Learning Distance Metrics for Multi-Label Classification

Abstract

Publication series

Conference

Keywords / Materials (for Non-textual outputs)

Access to Document

Fingerprint

Cite this