Reducing hubness for kernel regression

Kazuo Hara, Ikumi Suzuki, Kei Kobayashi, Kenji Fukumizu, Miloš Radovanović

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Abstract

In this paper, we point out that hubness—some samples in a high-dimensional dataset emerge as hubs that are similar to many other samples—influences the performance of kernel regression. Because the dimension of feature spaces induced by kernels is usually very high, hubness occurs, giving rise to the problem of multicollinearity, which is known as a cause of instability of regression results. We propose hubnessreduced kernels for kernel regression as an extension of a previous approach for kNN classification that reduces spatial centrality to eliminate hubness.

Original languageEnglish
Title of host publicationSimilarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings
PublisherSpringer Verlag
Pages339-344
Number of pages6
Volume9371
ISBN (Print)9783319250861
DOIs
Publication statusPublished - 2015
Externally publishedYes
Event8th International Conference on Similarity Search and Applications, SISAP 2015 - Glasgow, United Kingdom
Duration: 2015 Oct 122015 Oct 14

Publication series

NameLecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)
Volume9371
ISSN (Print)03029743
ISSN (Electronic)16113349

Other

Other8th International Conference on Similarity Search and Applications, SISAP 2015
CountryUnited Kingdom
CityGlasgow
Period15/10/1215/10/14

Fingerprint

Kernel Regression
kernel
Multicollinearity
Centrality
Feature Space
High-dimensional
Eliminate
Regression

ASJC Scopus subject areas

  • Computer Science(all)
  • Theoretical Computer Science

Cite this

Hara, K., Suzuki, I., Kobayashi, K., Fukumizu, K., & Radovanović, M. (2015). Reducing hubness for kernel regression. In Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings (Vol. 9371, pp. 339-344). (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9371). Springer Verlag. https://doi.org/10.1007/978-3-319-25087-8_33

Reducing hubness for kernel regression. / Hara, Kazuo; Suzuki, Ikumi; Kobayashi, Kei; Fukumizu, Kenji; Radovanović, Miloš.

Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings. Vol. 9371 Springer Verlag, 2015. p. 339-344 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics); Vol. 9371).

Research output: Chapter in Book/Report/Conference proceedingConference contribution

Hara, K, Suzuki, I, Kobayashi, K, Fukumizu, K & Radovanović, M 2015, Reducing hubness for kernel regression. in Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings. vol. 9371, Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics), vol. 9371, Springer Verlag, pp. 339-344, 8th International Conference on Similarity Search and Applications, SISAP 2015, Glasgow, United Kingdom, 15/10/12. https://doi.org/10.1007/978-3-319-25087-8_33
Hara K, Suzuki I, Kobayashi K, Fukumizu K, Radovanović M. Reducing hubness for kernel regression. In Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings. Vol. 9371. Springer Verlag. 2015. p. 339-344. (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)). https://doi.org/10.1007/978-3-319-25087-8_33
Hara, Kazuo ; Suzuki, Ikumi ; Kobayashi, Kei ; Fukumizu, Kenji ; Radovanović, Miloš. / Reducing hubness for kernel regression. Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings. Vol. 9371 Springer Verlag, 2015. pp. 339-344 (Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)).
@inproceedings{dcab97eb918e40369e269e7cae15b923,
title = "Reducing hubness for kernel regression",
abstract = "In this paper, we point out that hubness—some samples in a high-dimensional dataset emerge as hubs that are similar to many other samples—influences the performance of kernel regression. Because the dimension of feature spaces induced by kernels is usually very high, hubness occurs, giving rise to the problem of multicollinearity, which is known as a cause of instability of regression results. We propose hubnessreduced kernels for kernel regression as an extension of a previous approach for kNN classification that reduces spatial centrality to eliminate hubness.",
author = "Kazuo Hara and Ikumi Suzuki and Kei Kobayashi and Kenji Fukumizu and Miloš Radovanović",
year = "2015",
doi = "10.1007/978-3-319-25087-8_33",
language = "English",
isbn = "9783319250861",
volume = "9371",
series = "Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)",
publisher = "Springer Verlag",
pages = "339--344",
booktitle = "Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings",
address = "Germany",

}

TY - GEN

T1 - Reducing hubness for kernel regression

AU - Hara, Kazuo

AU - Suzuki, Ikumi

AU - Kobayashi, Kei

AU - Fukumizu, Kenji

AU - Radovanović, Miloš

PY - 2015

Y1 - 2015

N2 - In this paper, we point out that hubness—some samples in a high-dimensional dataset emerge as hubs that are similar to many other samples—influences the performance of kernel regression. Because the dimension of feature spaces induced by kernels is usually very high, hubness occurs, giving rise to the problem of multicollinearity, which is known as a cause of instability of regression results. We propose hubnessreduced kernels for kernel regression as an extension of a previous approach for kNN classification that reduces spatial centrality to eliminate hubness.

AB - In this paper, we point out that hubness—some samples in a high-dimensional dataset emerge as hubs that are similar to many other samples—influences the performance of kernel regression. Because the dimension of feature spaces induced by kernels is usually very high, hubness occurs, giving rise to the problem of multicollinearity, which is known as a cause of instability of regression results. We propose hubnessreduced kernels for kernel regression as an extension of a previous approach for kNN classification that reduces spatial centrality to eliminate hubness.

UR - http://www.scopus.com/inward/record.url?scp=84951787726&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=84951787726&partnerID=8YFLogxK

U2 - 10.1007/978-3-319-25087-8_33

DO - 10.1007/978-3-319-25087-8_33

M3 - Conference contribution

AN - SCOPUS:84951787726

SN - 9783319250861

VL - 9371

T3 - Lecture Notes in Computer Science (including subseries Lecture Notes in Artificial Intelligence and Lecture Notes in Bioinformatics)

SP - 339

EP - 344

BT - Similarity Search and Applications - 8th International Conference, SISAP 2015, Proceedings

PB - Springer Verlag

ER -