Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors

Yuto Ogino, Masahiro Yukawa

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

Abstract

Spectral clustering is an empirically successful approach to separating a dataset into some groups with possibly complex shapes based on pairwise affinity. Identifying the number of clusters automatically is still an open issue, although many heuristics have been proposed. In this paper, imposing sparsity on the eigenvectors of graph Laplacian is proposed to attain reasonable approximations of the so-called cluster-indicator-vectors, from which the clusters as well as the cluster number are identified. The proposed algorithm enjoys low computational complexity as it only computes a relevant subset of eigenvectors. It also enjoys better clustering quality than the existing methods, as shown by simulations using nine real datasets.

Original languageEnglish
Title of host publication2018 26th European Signal Processing Conference, EUSIPCO 2018
PublisherEuropean Signal Processing Conference, EUSIPCO
Pages1187-1191
Number of pages5
Volume2018-September
ISBN (Electronic)9789082797015
DOIs
Publication statusPublished - 2018 Nov 29
Event26th European Signal Processing Conference, EUSIPCO 2018 - Rome, Italy
Duration: 2018 Sep 32018 Sep 7

Other

Other26th European Signal Processing Conference, EUSIPCO 2018
CountryItaly
CityRome
Period18/9/318/9/7

Fingerprint

Eigenvalues and eigenfunctions
Computational complexity

ASJC Scopus subject areas

  • Signal Processing
  • Electrical and Electronic Engineering

Cite this

Ogino, Y., & Yukawa, M. (2018). Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors. In 2018 26th European Signal Processing Conference, EUSIPCO 2018 (Vol. 2018-September, pp. 1187-1191). [8553168] European Signal Processing Conference, EUSIPCO. https://doi.org/10.23919/EUSIPCO.2018.8553168

Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors. / Ogino, Yuto; Yukawa, Masahiro.

2018 26th European Signal Processing Conference, EUSIPCO 2018. Vol. 2018-September European Signal Processing Conference, EUSIPCO, 2018. p. 1187-1191 8553168.

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

Ogino, Y & Yukawa, M 2018, Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors. in 2018 26th European Signal Processing Conference, EUSIPCO 2018. vol. 2018-September, 8553168, European Signal Processing Conference, EUSIPCO, pp. 1187-1191, 26th European Signal Processing Conference, EUSIPCO 2018, Rome, Italy, 18/9/3. https://doi.org/10.23919/EUSIPCO.2018.8553168
Ogino Y, Yukawa M. Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors. In 2018 26th European Signal Processing Conference, EUSIPCO 2018. Vol. 2018-September. European Signal Processing Conference, EUSIPCO. 2018. p. 1187-1191. 8553168 https://doi.org/10.23919/EUSIPCO.2018.8553168
Ogino, Yuto ; Yukawa, Masahiro. / Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors. 2018 26th European Signal Processing Conference, EUSIPCO 2018. Vol. 2018-September European Signal Processing Conference, EUSIPCO, 2018. pp. 1187-1191
@inproceedings{3f912e9a67644ee3936ffea89ccaa2aa,
title = "Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors",
abstract = "Spectral clustering is an empirically successful approach to separating a dataset into some groups with possibly complex shapes based on pairwise affinity. Identifying the number of clusters automatically is still an open issue, although many heuristics have been proposed. In this paper, imposing sparsity on the eigenvectors of graph Laplacian is proposed to attain reasonable approximations of the so-called cluster-indicator-vectors, from which the clusters as well as the cluster number are identified. The proposed algorithm enjoys low computational complexity as it only computes a relevant subset of eigenvectors. It also enjoys better clustering quality than the existing methods, as shown by simulations using nine real datasets.",
author = "Yuto Ogino and Masahiro Yukawa",
year = "2018",
month = "11",
day = "29",
doi = "10.23919/EUSIPCO.2018.8553168",
language = "English",
volume = "2018-September",
pages = "1187--1191",
booktitle = "2018 26th European Signal Processing Conference, EUSIPCO 2018",
publisher = "European Signal Processing Conference, EUSIPCO",

}

TY - GEN

T1 - Spectral clustering with automatic cluster-number identification via finding sparse eigenvectors

AU - Ogino, Yuto

AU - Yukawa, Masahiro

PY - 2018/11/29

Y1 - 2018/11/29

N2 - Spectral clustering is an empirically successful approach to separating a dataset into some groups with possibly complex shapes based on pairwise affinity. Identifying the number of clusters automatically is still an open issue, although many heuristics have been proposed. In this paper, imposing sparsity on the eigenvectors of graph Laplacian is proposed to attain reasonable approximations of the so-called cluster-indicator-vectors, from which the clusters as well as the cluster number are identified. The proposed algorithm enjoys low computational complexity as it only computes a relevant subset of eigenvectors. It also enjoys better clustering quality than the existing methods, as shown by simulations using nine real datasets.

AB - Spectral clustering is an empirically successful approach to separating a dataset into some groups with possibly complex shapes based on pairwise affinity. Identifying the number of clusters automatically is still an open issue, although many heuristics have been proposed. In this paper, imposing sparsity on the eigenvectors of graph Laplacian is proposed to attain reasonable approximations of the so-called cluster-indicator-vectors, from which the clusters as well as the cluster number are identified. The proposed algorithm enjoys low computational complexity as it only computes a relevant subset of eigenvectors. It also enjoys better clustering quality than the existing methods, as shown by simulations using nine real datasets.

UR - http://www.scopus.com/inward/record.url?scp=85059806403&partnerID=8YFLogxK

UR - http://www.scopus.com/inward/citedby.url?scp=85059806403&partnerID=8YFLogxK

U2 - 10.23919/EUSIPCO.2018.8553168

DO - 10.23919/EUSIPCO.2018.8553168

M3 - Conference contribution

VL - 2018-September

SP - 1187

EP - 1191

BT - 2018 26th European Signal Processing Conference, EUSIPCO 2018

PB - European Signal Processing Conference, EUSIPCO

ER -