Inferring feature relevances from metric learning

Alexander Schulz; Bassam Mokbel; Michael Biehl; Barbara Hammer

doi:10.1109/SSCI.2015.225

Inferring feature relevances from metric learning

Alexander Schulz, Bassam Mokbel, Michael Biehl, Barbara Hammer

Section for Transport Biology

4 Citations (Scopus)

Abstract

Powerful metric learning algorithms have been proposed in the last years which do not only greatly enhance the accuracy of distance-based classifiers and nearest neighbor database retrieval, but which also enable the interpretability of these operations by assigning explicit relevance weights to the single data components. Starting with the work SSCI13Stretal, it has been noticed, however, that this procedure has very limited validity in the important case of high data dimensionality or high feature correlations: The resulting relevance profiles are random to a large extend, leading to invalid interpretation and fluctuations of its accuracy for novel data. While the work SSCI13Stretal proposes a first cure by means of L2-regularisation, it only preserves strongly relevant features, leaving weakly relevant and not necessarily unique features undetected. In this contribution, we enhance the technique by an efficient linear programming scheme which enables the unique identification of a relevance interval for every observed feature, this way identifying both, strongly and weakly relevant features for a given metric.

Original language	Danish
Title of host publication	Proceedings - 2015 IEEE Symposium Series on Computational Intelligence : SSCI 2015
Number of pages	8
Publisher	IEEE
Publication date	2015
Pages	1599-1606
ISBN (Print)	9781479975600
DOIs	https://doi.org/10.1109/SSCI.2015.225
Publication status	Published - 2015

Series	Proceedings - 2015 IEEE Symposium Series on Computational Intelligence, SSCI 2015

Access to Document

10.1109/SSCI.2015.225

Cite this

Inferring feature relevances from metric learning. / Schulz, Alexander; Mokbel, Bassam; Biehl, Michael et al.
Proceedings - 2015 IEEE Symposium Series on Computational Intelligence: SSCI 2015. IEEE, 2015. p. 1599-1606 (Proceedings - 2015 IEEE Symposium Series on Computational Intelligence, SSCI 2015).

Research output: Chapter in Book/Report/Conference proceeding › Article in proceedings › Research › peer-review

@inproceedings{d601f52bf45548d9a8f08ff85b748eeb,

title = "Inferring feature relevances from metric learning",

abstract = "Powerful metric learning algorithms have been proposed in the last years which do not only greatly enhance the accuracy of distance-based classifiers and nearest neighbor database retrieval, but which also enable the interpretability of these operations by assigning explicit relevance weights to the single data components. Starting with the work [1], it has been noticed, however, that this procedure has very limited validity in the important case of high data dimensionality or high feature correlations: the resulting relevance profiles are random to a large extend, leading to invalid interpretation and fluctuations of its accuracy for novel data. While the work [1] proposes a first cure by means of L2-regularisation, it only preserves strongly relevant features, leaving weakly relevant and not necessarily unique features undetected. In this contribution, we enhance the technique by an efficient linear programming scheme which enables the unique identification of a relevance interval for every observed feature, this way identifying both, strongly and weakly relevant features for a given metric.",

author = "Alexander Schulz and Bassam Mokbel and Michael Biehl and Barbara Hammer",

year = "2015",

doi = "10.1109/SSCI.2015.225",

language = "Dansk",

isbn = "9781479975600",

series = "Proceedings - 2015 IEEE Symposium Series on Computational Intelligence, SSCI 2015",

publisher = "IEEE",

pages = "1599--1606",

booktitle = "Proceedings - 2015 IEEE Symposium Series on Computational Intelligence",

}

TY - GEN

T1 - Inferring feature relevances from metric learning

AU - Schulz, Alexander

AU - Mokbel, Bassam

AU - Biehl, Michael

AU - Hammer, Barbara

PY - 2015

Y1 - 2015

N2 - Powerful metric learning algorithms have been proposed in the last years which do not only greatly enhance the accuracy of distance-based classifiers and nearest neighbor database retrieval, but which also enable the interpretability of these operations by assigning explicit relevance weights to the single data components. Starting with the work [1], it has been noticed, however, that this procedure has very limited validity in the important case of high data dimensionality or high feature correlations: the resulting relevance profiles are random to a large extend, leading to invalid interpretation and fluctuations of its accuracy for novel data. While the work [1] proposes a first cure by means of L2-regularisation, it only preserves strongly relevant features, leaving weakly relevant and not necessarily unique features undetected. In this contribution, we enhance the technique by an efficient linear programming scheme which enables the unique identification of a relevance interval for every observed feature, this way identifying both, strongly and weakly relevant features for a given metric.

AB - Powerful metric learning algorithms have been proposed in the last years which do not only greatly enhance the accuracy of distance-based classifiers and nearest neighbor database retrieval, but which also enable the interpretability of these operations by assigning explicit relevance weights to the single data components. Starting with the work [1], it has been noticed, however, that this procedure has very limited validity in the important case of high data dimensionality or high feature correlations: the resulting relevance profiles are random to a large extend, leading to invalid interpretation and fluctuations of its accuracy for novel data. While the work [1] proposes a first cure by means of L2-regularisation, it only preserves strongly relevant features, leaving weakly relevant and not necessarily unique features undetected. In this contribution, we enhance the technique by an efficient linear programming scheme which enables the unique identification of a relevance interval for every observed feature, this way identifying both, strongly and weakly relevant features for a given metric.

U2 - 10.1109/SSCI.2015.225

DO - 10.1109/SSCI.2015.225

M3 - Konferencebidrag i proceedings

SN - 9781479975600

T3 - Proceedings - 2015 IEEE Symposium Series on Computational Intelligence, SSCI 2015

SP - 1599

EP - 1606

BT - Proceedings - 2015 IEEE Symposium Series on Computational Intelligence

PB - IEEE

ER -