Outlier-resisting graph embedding

Yanwei Pang, Yuan Yuan*

*Corresponding author for this work

Research output: Contribution to journalArticle

Abstract

Graph embedding is a general framework for subspace learning. However, because of the well-known outlier-sensitiveness disadvantage of the L2-norm, conventional graph embedding is not robust to outliers which occur in many practical applications. In this paper, an improved graph embedding algorithm (termed LPP-L1) is proposed by replacing L2-norm with L1-norm. In addition to its robustness property, LPP-L1 avoids small sample size problem. Experimental results on both synthetic and real-world data demonstrate these advantages.

Original languageEnglish
Pages (from-to)968-974
Number of pages7
JournalNeurocomputing
Volume73
Issue number4-6
Early online date9 Oct 2009
DOIs
Publication statusPublished - Jan 2010

Fingerprint

Sample Size
Learning

Bibliographical note

Bayesian Networks / Design and Application of Neural Networks and Intelligent Learning Systems (KES 2008 / Bio-inspired Computing: Theories and Applications (BIC-TA 2007)

Keywords

  • dimensionality reduction
  • graph embedding
  • L1-norm
  • locality preserving projection
  • outlier

Cite this

Pang, Yanwei ; Yuan, Yuan. / Outlier-resisting graph embedding. In: Neurocomputing. 2010 ; Vol. 73, No. 4-6. pp. 968-974.
@article{3572ae2bff764207a988a4048fb6b626,
title = "Outlier-resisting graph embedding",
abstract = "Graph embedding is a general framework for subspace learning. However, because of the well-known outlier-sensitiveness disadvantage of the L2-norm, conventional graph embedding is not robust to outliers which occur in many practical applications. In this paper, an improved graph embedding algorithm (termed LPP-L1) is proposed by replacing L2-norm with L1-norm. In addition to its robustness property, LPP-L1 avoids small sample size problem. Experimental results on both synthetic and real-world data demonstrate these advantages.",
keywords = "dimensionality reduction, graph embedding, L1-norm, locality preserving projection, outlier",
author = "Yanwei Pang and Yuan Yuan",
note = "Bayesian Networks / Design and Application of Neural Networks and Intelligent Learning Systems (KES 2008 / Bio-inspired Computing: Theories and Applications (BIC-TA 2007)",
year = "2010",
month = "1",
doi = "10.1016/j.neucom.2009.08.020",
language = "English",
volume = "73",
pages = "968--974",
journal = "Neurocomputing",
issn = "0925-2312",
publisher = "Elsevier",
number = "4-6",

}

Pang, Y & Yuan, Y 2010, 'Outlier-resisting graph embedding', Neurocomputing, vol. 73, no. 4-6, pp. 968-974. https://doi.org/10.1016/j.neucom.2009.08.020

Outlier-resisting graph embedding. / Pang, Yanwei; Yuan, Yuan.

In: Neurocomputing, Vol. 73, No. 4-6, 01.2010, p. 968-974.

Research output: Contribution to journalArticle

TY - JOUR

T1 - Outlier-resisting graph embedding

AU - Pang, Yanwei

AU - Yuan, Yuan

N1 - Bayesian Networks / Design and Application of Neural Networks and Intelligent Learning Systems (KES 2008 / Bio-inspired Computing: Theories and Applications (BIC-TA 2007)

PY - 2010/1

Y1 - 2010/1

N2 - Graph embedding is a general framework for subspace learning. However, because of the well-known outlier-sensitiveness disadvantage of the L2-norm, conventional graph embedding is not robust to outliers which occur in many practical applications. In this paper, an improved graph embedding algorithm (termed LPP-L1) is proposed by replacing L2-norm with L1-norm. In addition to its robustness property, LPP-L1 avoids small sample size problem. Experimental results on both synthetic and real-world data demonstrate these advantages.

AB - Graph embedding is a general framework for subspace learning. However, because of the well-known outlier-sensitiveness disadvantage of the L2-norm, conventional graph embedding is not robust to outliers which occur in many practical applications. In this paper, an improved graph embedding algorithm (termed LPP-L1) is proposed by replacing L2-norm with L1-norm. In addition to its robustness property, LPP-L1 avoids small sample size problem. Experimental results on both synthetic and real-world data demonstrate these advantages.

KW - dimensionality reduction

KW - graph embedding

KW - L1-norm

KW - locality preserving projection

KW - outlier

UR - http://www.scopus.com/inward/record.url?scp=75749094310&partnerID=8YFLogxK

U2 - 10.1016/j.neucom.2009.08.020

DO - 10.1016/j.neucom.2009.08.020

M3 - Article

AN - SCOPUS:75749094310

VL - 73

SP - 968

EP - 974

JO - Neurocomputing

JF - Neurocomputing

SN - 0925-2312

IS - 4-6

ER -