Non-Euclidean or non-metric measures can be informative

EM Pekalska, A Harol, RPW Duin, B Spillmann, H Bunke

Research output: Chapter in Book/Conference proceedings/Edited volumeConference contributionScientificpeer-review

41 Citations (Scopus)

Abstract

StatisticallearningalgorithmsoftenrelyontheEuclideandistance.Inpractice,non-Euclideanornon-metricdissimilaritymeasuresmayarisewhencontours,spectraorshapesarecomparedbyeditdistancesorasaconsequenceofrobustobjectmatching[1,2].Itisanopenissuewhethersuchmeasuresareadvantageousforstatisticallearningorwhethertheyshouldbeconstrainedtoobeythemetricaxioms. Thek-nearestneighbor(NN)ruleiswidelyappliedtogeneraldissimilaritydataasthemostnaturalapproach.Alternativemethodsexistthatembedsuchdataintosuitablerepresentationspacesinwhichstatisticalclassi¿ersareconstructed[3].Inthispaper,weinvestigatetherelationbetweennon-Euclideanaspectsofdissimilaritydataandtheclassi¿cationperformanceofthedirectNNruleandsomeclassi¿erstrainedinrepresentationspaces.Thisisevaluatedonaparameterizedfamilyofeditdistances,inwhichparametervaluescontrolthestrengthofnon-Euclideanbehavior.Our¿ndingisthatthediscriminativepowerofthismeasureincreaseswithincreasingnon-Euclideanandnon-metricaspectsuntilacertainoptimumisreached.Theconclusionisthatstatisticalclassi¿ersperformwellandtheoptimalvaluesoftheparameterscharacterizeanon-Euclideanandsomewhatnon-metricmeasure
Original languageUndefined/Unknown
Title of host publicationStructural, syntactic and statistical pattern recognition
EditorsDY Yeung, JT Kwok, A Fred, F Roli, D de Ridder
Place of PublicationBerlin-Heidelberg
PublisherSpringer
Pages871-880
Number of pages10
ISBN (Print)3-540-37236-9
Publication statusPublished - 2006
EventJoint IAPR International Workshops SSPR 2006 and SPR 2006, Hong Kong, China - Heidelberg
Duration: 17 Aug 200619 Aug 2006

Publication series

Name
PublisherSpringer
NameLecture Notes in Computer Science
Volume4109
ISSN (Print)0302-9743

Conference

ConferenceJoint IAPR International Workshops SSPR 2006 and SPR 2006, Hong Kong, China
Period17/08/0619/08/06

Keywords

  • conference contrib. refereed
  • CWTS 0.75 <= JFIS < 2.00

Cite this

Pekalska, EM., Harol, A., Duin, RPW., Spillmann, B., & Bunke, H. (2006). Non-Euclidean or non-metric measures can be informative. In DY. Yeung, JT. Kwok, A. Fred, F. Roli, & D. de Ridder (Eds.), Structural, syntactic and statistical pattern recognition (pp. 871-880). (Lecture Notes in Computer Science; Vol. 4109). Springer.