SlideShare una empresa de Scribd logo
1 de 30
Descargar para leer sin conexión
Distance Metric Learning 
2014-03-06 
전상혁 
2014-03-05 1
Introduction 
•DistanceMetriclearningistolearnadistancemetricfortheinputspaceofdata 
•Dataisfromagivencollectionofpairofsimilar/dissimilarpointsthatpreservesthedistancerelationamongthetrainingdata 
•Learnedmetriccansignificantlyimprovetheperformanceofalgorithmssuchasclassification,clusteringandmanyothertasks 
2014-03-05 2
Distance Metric 
•Adistancemetricordistancefunctionisafunctionthatdefinesadistancebetweenelementsofaset 
•Example1:Euclideanmetric 
•Example2:Discretemetric 
•푖푓푥=푦푡ℎ푒푛푑푥,푦=0,푂푡ℎ푒푟푤푖푠푒,푑푥,푦=1 
•AmetriconasetXisafunction푑:푋×푋→퐑(whereRisrealnumber) 
•푑푥,푦≥0 
•푑푥,푦=0푖푓푎푛푑표푛푙푦푖푓푥=푦 
•푑푥,푦=푑(푦,푥) 
•푑푥,푧≤dx,y+d(y,z) 
2014-03-05 Aurelien Bellet, Tutorial on Metric Learning, 2013 3
Introduction 
2014-03-05 Bellet, A., Habrard, A., and Sebban, M. A Survey on Metric Learning for Feature Vectors and Structured Data, 2013 4 
Bellet, A., Habrard, A., and Sebban, M. A Survey on Metric Learning for Feature Vectors and Structured Data, 2013
Categories of distance metric learning 
•Superviseddistancemetriclearning 
•Constraintsorlabelsgiventothealgorithm 
•Equivalenceconstraintswherepairsofdatapointsthatbelongtothesameclasses(semantically-similar) 
•Inequivalenceconstraintswherepairsofdatapointsthatbelongtothedifferentclasses(semantically-dissimilar) 
•localadaptivesuperviseddistancemetriclearning,NCA,RCA 
•Unsuperviseddistancemetriclearning 
•Dimensionalityreductiontechniques 
•Linearmethods(PCA,MDS)andnon-linearmethods(ISOMAP,LLE,LE) 
•Kernelmethodstowardslearningdistancemetrics 
•Kernelalignment,extensionworkoflearningidealizedkernel 
2014-03-05 Liu Yang, An Overview of Distance Metric Learning, 2007 5 
Liu Yang, Distance Metric Learning: A Comprehensive Survey, 2005 
Liu Yang, An Overview of Distance Metric Learning, 2007
Categories of distance metric learning 
•Maximummarginbaseddistancemetriclearning 
•Formulatedistancemetriclearningasaconstrainedconvexprogrammingproblem,andattempttolearncompletedistancemetricfromtrainingdata 
•Expensivecomputation,overfittingproblemwhenlimitedtrainingsamplesandlargefeaturedimension 
•Largemarginnearestneighborclassification(LMNN),SDPmethodstosolvethekernelizedmarginmaximization 
2014-03-05 Liu Yang, An Overview of Distance Metric Learning, 2007 6 
Liu Yang, Distance Metric Learning: A Comprehensive Survey, 2005 
Liu Yang, An Overview of Distance Metric Learning, 2007
Categories of distance metric learning 
2014-03-05 Liu Yang, An Overview of Distance Metric Learning, 2007 7 
Liu Yang, An Overview of Distance Metric Learning, 2007
Large Margin Nearest Neighbor Classification (LMNN) 
KilianWeinberger,etal.Distancemetriclearningforlargemarginnearestneighborclassification.NIPS,2006. 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 8
Motivation –better KNN 
•KNN:simplenon-parametricclassificationmethod(Alsocanbeusedforclustering,regression,anomalydetection,andetc.) 
•푘isuser-definedconstant,andanunlabeledvectorisclassifiedbyassigningthelabelwhichismostfrequentamongthe푘trainingsamplesnearesttothatunlabeledvector 
•Drawback:whentheclassdistributionisskewed,basic“majorityvoting” classificationmaynotworkswell. 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 9
Introduction 
•Largemarginnearestneighbor(LMNN)method 
•NeighborsareselectedbyusingthelearnedMahanalobisdistancemetric 
•LearnaMahanalobisdistancemetricforkNNclassificationbysemidefiniteprogramming 
•Find 퐿:푅푑→푅푑,which will be used to compute squared distance as 퐷푥푖,푥푗=퐿푥푖−푥푗 2by minimizing some cost function 휖(퐿) 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 10
MahanalobisDistance Metric 
•Relativemeasureofadatapoint’sdistancefromacommonpoint 
•ItdiffersfromEuclideandistanceinthatittakesintoaccountthecorrelationsofthedatasetandisscale-invariant 
•푑 푥, 푦= 푥− 푦⊤훀( 푥− 푦)where훀ispositivesemidefintematrix 
•Example:thedistancebetweentworandomvariable 푥, 푦ofthesamedistributionwithcovariancematrix푆isdefinedas 
•푑 푥, 푦= 푥− 푦⊤푆−1( 푥− 푦) 
•EuclideandistanceisspecialcasewhenSisidentitymatrix 
•Recall:distance퐷푥푖,푥푗=퐿푥푖−푥푗 2 
•Wecanrewriteeq.as퐷푥푖,푥푗=푥푖−푥푗퐌⊤푥푖−푥푗wherethematrix퐌= 퐋⊤퐋,parameterizestheMahalanobisdistancemetricinducedbythelineartransform퐋 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 11
SemidefiniteProgramming 
•Akindofconvexoptimizationproblem 
•푥isavectorofnvariables,푐isaconstantndimensionalvector 
•Linearprogramming(LP) 
•푚푖푛푖푚푖푧푒푐∙푥 푠.푡.푎푖∙푥=푏푖,푖=1,…,푚 푥∈푅+푛 
•Xispositivesemidefinitematrixif푣⊤푋푣≥0푓표푟푎푛푦푣∈푅푛(푋≽0) 
•If퐶(푋)isalinearfunctionofX,then퐶(푋)canbewrittenas퐶∙푋 
•Semidefiniteprogram(SDP) 
•푚푖푛푖푚푖푧푒퐶∙푋 푠.푡.퐴푖∙푋=푏푖,푖=1,…,푚 푋≽0 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 12
Cost Function 
•휖퐿= 푖푗휂푖푗퐋푥푖−푥푗 2+푐 푖푗푙휂푖푗1−푦푖푙1+퐋푥푖−푥푗 2−퐋푥푖−푥푙 2+ 
•Let푥푖,yii=1ndenoteatrainingsetofnlabeledexampleswithinputs푥푖∈푅푑anddiscrete(butnotnecessarilybinary)classlabels푦푖 
•푦푖푗∈{0,1}isbinarymatrixtoindicatewhetherornotthelabels푦푖and푦푗match 
•휂푖푗∈{0,1}toindicatewhetherinput푥푗isatargetneighborofinput푥푖,thisvalueisfixedanddoesnotchangeduringlearning 
•Targetneighbor:eachinstance푥푖hasexactlykdifferenttargetneighborswithindataset,whichallsharethesameclasslabel푦푖 
•Targetneighborsarethedatapointsthatshouldbecomenearestneighborsunderthelearnedmetric 
•C.f.Impostors:sameastargetneighborexceptithasdifferentclasslabel 
•Term푧+=max(푧,0) 
•푐>0somepositiveconstant(typicallysetbycrossvalidation) 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 13
Cost Function 
•휖퐿= 푖푗휂푖푗퐋푥푖−푥푗 2+푐 푖푗푙휂푖푗1−푦푖푙1+퐋푥푖−푥푗 2−퐋푥푖−푥푙 2+ 
•Thefirstoptimizationgoalisachievedbyminimizingtheaveragedistancebetweeninstanceandtheirtargetneighbors 
• 
•Thesecondgoalisachievedbyconstrainingimpostors푥푙tobeoneunitfurtherawaythantargetneighbors푥푗 
• 
•푁푖denotethesetoftargetneighborsforadatapoint푥푖 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 14
2014-03-05 15 
Nakul Verma, A tutorial on Metric Learning with some recent advances
Convex Optimization 
•Recall:distance퐷푥푖,푥푗=퐿푥푖−푥푗 2 
•Recall:Wecanrewriteeq.as퐷푥푖,푥푗=푥푖−푥푗퐌⊤푥푖−푥푗wherethematrix퐌=퐋⊤퐋,parameterizestheMahalanobisdistancemetricinducedbythelineartransform퐋 
•Thehingelosscanbe“mimicked”byintroducingslackvariable휁푖푗forallpairsofdifferentlylabeledinputs(i.e.,forall<푖,푗>suchthat푦푖푗=0) 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 16
Results 
•PCAwasusedtoreducethedimensionalityofimage,speech,andtextdata,bothtospeeduptrainingandavoidoverfitting 
•Boththenumberoftargetneighbors(k)andtheweightingparameter(c) inoptimizationproblemweresetbycrossvalidation 
•KNNEuclideandistance 
•KNNMahalanobisdistance 
•Energybasedclassification 
• 
•MulticlassSVM 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 17
Results 
2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 18
GBLMNN: Non-linear method for LMNN 
•Buildingnon-linearmappingformetriclearning 
•GBRT:GradientBoostingRegressionTree 
•Code:http://www.cse.wustl.edu/~kilian/code/code.html 
2014-03-05 Kedem, D., Xu, Z., & Weinberger, K. (n.d.). Gradient Boosted Large Margin Nearest Neighbors, (1), 10–12. 19
Large Margin Component Analysis (LMCA) 
Torresani,L.,&Lee,K.(2007).Largemargincomponentanalysis.AdvancesinNeuralInformationProcessing 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 20
Problem: large number of features 
•Inproblemsinvolvingthousandsoffeatures,distancemetriclearningcannotbeused 
•Highcomputationcomplexity 
•Overffitingproblem 
•Previousworkshavereliedonatwo-stepsolution 
•Applydimensionreductionalgorithm(e.g.PCA) 
•Learnametricintheresultinglow-dimensionalsubspace 
•LMCA:providebettersolutionforhighdimensionalproblem 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 21
Linear Dimensionality Reduction for KNN (1/3) 
•Recall:costfunctionofmetricfunction 
•Inpreviouswork,Lwas퐷×퐷matrix 
•Idea:learnnonsquarematrixsize푑×퐷,푤푖푡ℎ푑≪퐷 
•Problem:costfunction휖퐌,푤ℎ푒푟푒퐌=퐋⊤퐋isnolongerconvex 
•Mhaslowrankd,notD 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 22
Linear Dimensionality Reduction for KNN (2/3) 
•Idea:optimizetheobjectivefunctiondirectlywithrespecttothenonsquarematrixL 
•Although,equationisnon-convex,itisbetterthanminimizingtheobjectivefunctionwithrespecttoLratherthanwithrespecttotherank- deficientD×DmatrixM 
•Sincethisoptimizationinvolvesonly푑∙퐷ratherthan퐷2unknowns,itcanreducetheriskofoverffiting 
•“theoptimalrectangularmatrixLcomputedwithourmethodautomaticallysatisfiestherankconstraintsonMwithoutrequiringthesolutionofdifficultconstrainedminimizationproblems” 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 23
Linear Dimensionality Reduction for KNN (3/3) 
•Minimizecostfunctionusinggradient-basedoptimizers 
•“Wehandlethenon-differentiabilityofh(s)ats=0,byadoptingasmoothhingefunctionasin[8]” 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 24 
[8] J. D. M. Rennieand N. Srebro. Fast maximum margin matrix factorization for collaborative prediction. 
In Proceedings of the 22nd International Conference on Machine Learning (ICML), 2005.
Nonlinear Feature Extraction (1/2) 
•“Ourapproachlearnsalow-rankMahalanobisdistancemetricinahighdimensionalfeaturespaceF,relatedtotheinputsbyanonlinearmap 휙:푅퐷→퐹” 
•kiskernelfunctioncomputedbythefeatureinputproducts 
•LisnowatransformationfromthespaceFintoa푅푑(푑≪퐷) 
• 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 25
Nonlinear Feature Extraction (2/2) 
• 
• 
• 
•UpdateΩ←(Ω−휆Γ)untilconverge 
•Forclassification,weprojectpointsontothelearnedlow-dimspacebyexploitingthekerneltrick:퐿흓=Ω퐤 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 26
Results –High dimensional dataset 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 27
Results –low dimensional dataset 
2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 28
Reference 
•Bellet,A.,Habrard,A.,andSebban,M.ASurveyonMetricLearningforFeatureVectorsandStructuredData,2013 
•LiuYang,DistanceMetricLearning:AComprehensiveSurvey,2005 
•LiuYang,AnOverviewofDistanceMetricLearning,2007 
•MarcoCuturi.KAISTMachineLearningTutorialMetricsandKernelsAfewrecenttopics,2013 
•NakulVerma,AtutorialonMetricLearningwithsomerecentadvances 
•AurelienBellet,TutorialonMetricLearning,2013 
•BrianKulis.TutorialonMetricLearning.InternationalConferenceonMachineLearning(ICML)2010 
2014-03-05 29
Reference 
•K.Q.Weinberger,J.Blitzer,andL.K.Saul(2006).InY.Weiss,B.Schoelkopf, andJ.Platt(eds.),DistanceMetricLearningforLargeMarginNearestNeighborClassification,AdvancesinNeuralInformationProcessingSystems18(NIPS-18).MITPress:Cambridge,MA. 
•K.Q.Weinberger,L.K.Saul.DistanceMetricLearningforLargeMarginNearestNeighborClassification.JournalofMachineLearningResearch(JMLR)2009,10:207-244 
•Torresani,L.,&Lee,K.(2007).Largemargincomponentanalysis. AdvancesinNeuralInformationProcessing 
•Kedem,D.,Xu,Z.,&Weinberger,K.(n.d.).GradientBoostedLargeMarginNearestNeighbors,(1),10–12. 
2014-03-05 30

Más contenido relacionado

La actualidad más candente

Pca(principal components analysis)
Pca(principal components analysis)Pca(principal components analysis)
Pca(principal components analysis)kalung0313
 
Linear models and multiclass classification
Linear models and multiclass classificationLinear models and multiclass classification
Linear models and multiclass classificationNdSv94
 
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...Christian Robert
 
Introduction to ggplot2
Introduction to ggplot2Introduction to ggplot2
Introduction to ggplot2maikroeder
 
Introduction to Principle Component Analysis
Introduction to Principle Component AnalysisIntroduction to Principle Component Analysis
Introduction to Principle Component AnalysisSunjeet Jena
 
Point and Interval Estimation
Point and Interval EstimationPoint and Interval Estimation
Point and Interval EstimationShubham Mehta
 
Introduction to calculus
Introduction to calculusIntroduction to calculus
Introduction to calculussheetslibrary
 
Introduction to Generalized Linear Models
Introduction to Generalized Linear ModelsIntroduction to Generalized Linear Models
Introduction to Generalized Linear Modelsrichardchandler
 
Correspondence analysis(step by step)
Correspondence analysis(step by step)Correspondence analysis(step by step)
Correspondence analysis(step by step)Nguyen Van Chuc
 
Chi square Test Using SPSS
Chi square Test Using SPSSChi square Test Using SPSS
Chi square Test Using SPSSDr Athar Khan
 

La actualidad más candente (20)

Pca(principal components analysis)
Pca(principal components analysis)Pca(principal components analysis)
Pca(principal components analysis)
 
Linear models and multiclass classification
Linear models and multiclass classificationLinear models and multiclass classification
Linear models and multiclass classification
 
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
Statistics (1): estimation Chapter 3: likelihood function and likelihood esti...
 
Decision tree
Decision treeDecision tree
Decision tree
 
Linear and Logistics Regression
Linear and Logistics RegressionLinear and Logistics Regression
Linear and Logistics Regression
 
Simple Linear Regression
Simple Linear RegressionSimple Linear Regression
Simple Linear Regression
 
Principal Component Analysis
Principal Component AnalysisPrincipal Component Analysis
Principal Component Analysis
 
Pca ppt
Pca pptPca ppt
Pca ppt
 
Logistic regression
Logistic regressionLogistic regression
Logistic regression
 
Sampling Distributions and Estimators
Sampling Distributions and Estimators Sampling Distributions and Estimators
Sampling Distributions and Estimators
 
Introduction to ggplot2
Introduction to ggplot2Introduction to ggplot2
Introduction to ggplot2
 
Introduction to Principle Component Analysis
Introduction to Principle Component AnalysisIntroduction to Principle Component Analysis
Introduction to Principle Component Analysis
 
Correlation analysis
Correlation analysisCorrelation analysis
Correlation analysis
 
Point and Interval Estimation
Point and Interval EstimationPoint and Interval Estimation
Point and Interval Estimation
 
Introduction to calculus
Introduction to calculusIntroduction to calculus
Introduction to calculus
 
Introduction to Generalized Linear Models
Introduction to Generalized Linear ModelsIntroduction to Generalized Linear Models
Introduction to Generalized Linear Models
 
Correspondence analysis(step by step)
Correspondence analysis(step by step)Correspondence analysis(step by step)
Correspondence analysis(step by step)
 
PCA
PCAPCA
PCA
 
Chi square Test Using SPSS
Chi square Test Using SPSSChi square Test Using SPSS
Chi square Test Using SPSS
 
6. point and interval estimation
6. point and interval estimation6. point and interval estimation
6. point and interval estimation
 

Destacado

Metric learning ICML2010 tutorial
Metric learning  ICML2010 tutorialMetric learning  ICML2010 tutorial
Metric learning ICML2010 tutorialzukun
 
Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...
Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...
Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...Koichi Hamada
 
論文輪読: Deep neural networks are easily fooled: High confidence predictions for...
論文輪読: Deep neural networks are easily fooled: High confidence predictions for...論文輪読: Deep neural networks are easily fooled: High confidence predictions for...
論文輪読: Deep neural networks are easily fooled: High confidence predictions for...mmisono
 
Improving neural networks by preventing co adaptation of feature detectors
Improving neural networks by preventing co adaptation of feature detectorsImproving neural networks by preventing co adaptation of feature detectors
Improving neural networks by preventing co adaptation of feature detectorsJunya Saito
 
Information-Theoretic Metric Learning
Information-Theoretic Metric LearningInformation-Theoretic Metric Learning
Information-Theoretic Metric LearningKoji Matsuda
 
Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3
Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3
Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3Daiki Shimada
 
Image net classification with Deep Convolutional Neural Networks
Image net classification with Deep Convolutional Neural NetworksImage net classification with Deep Convolutional Neural Networks
Image net classification with Deep Convolutional Neural NetworksShingo Horiuchi
 
Deep Residual Learning (ILSVRC2015 winner)
Deep Residual Learning (ILSVRC2015 winner)Deep Residual Learning (ILSVRC2015 winner)
Deep Residual Learning (ILSVRC2015 winner)Hirokatsu Kataoka
 
Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料
Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料
Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料tm_2648
 

Destacado (12)

Metric learning ICML2010 tutorial
Metric learning  ICML2010 tutorialMetric learning  ICML2010 tutorial
Metric learning ICML2010 tutorial
 
Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...
Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...
Laplacian Pyramid of Generative Adversarial Networks (LAPGAN) - NIPS2015読み会 #...
 
論文輪読: Deep neural networks are easily fooled: High confidence predictions for...
論文輪読: Deep neural networks are easily fooled: High confidence predictions for...論文輪読: Deep neural networks are easily fooled: High confidence predictions for...
論文輪読: Deep neural networks are easily fooled: High confidence predictions for...
 
An Introduction to Metric Learning for Clustering
An Introduction to Metric Learning for ClusteringAn Introduction to Metric Learning for Clustering
An Introduction to Metric Learning for Clustering
 
Improving neural networks by preventing co adaptation of feature detectors
Improving neural networks by preventing co adaptation of feature detectorsImproving neural networks by preventing co adaptation of feature detectors
Improving neural networks by preventing co adaptation of feature detectors
 
Information-Theoretic Metric Learning
Information-Theoretic Metric LearningInformation-Theoretic Metric Learning
Information-Theoretic Metric Learning
 
Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3
Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3
Adversarial Networks の画像生成に迫る @WBAFLカジュアルトーク#3
 
Image net classification with Deep Convolutional Neural Networks
Image net classification with Deep Convolutional Neural NetworksImage net classification with Deep Convolutional Neural Networks
Image net classification with Deep Convolutional Neural Networks
 
Deep Residual Learning (ILSVRC2015 winner)
Deep Residual Learning (ILSVRC2015 winner)Deep Residual Learning (ILSVRC2015 winner)
Deep Residual Learning (ILSVRC2015 winner)
 
20150930
2015093020150930
20150930
 
MIRU2014 tutorial deeplearning
MIRU2014 tutorial deeplearningMIRU2014 tutorial deeplearning
MIRU2014 tutorial deeplearning
 
Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料
Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料
Deep Convolutional Generative Adversarial Networks - Nextremer勉強会資料
 

Similar a Distance Metric Learning

240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx
240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx
240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptxthanhdowork
 
Unsupervised Distance Based Detection of Outliers by using Anti-hubs
Unsupervised Distance Based Detection of Outliers by using Anti-hubsUnsupervised Distance Based Detection of Outliers by using Anti-hubs
Unsupervised Distance Based Detection of Outliers by using Anti-hubsIRJET Journal
 
Enhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wildEnhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wildPrerana Mukherjee
 
Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...
Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...
Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...Maninda Edirisooriya
 
Ability Study of Proximity Measure for Big Data Mining Context on Clustering
Ability Study of Proximity Measure for Big Data Mining Context on ClusteringAbility Study of Proximity Measure for Big Data Mining Context on Clustering
Ability Study of Proximity Measure for Big Data Mining Context on ClusteringKamleshKumar394
 
Sparse inverse covariance estimation using skggm
Sparse inverse covariance estimation using skggmSparse inverse covariance estimation using skggm
Sparse inverse covariance estimation using skggmManjari Narayan
 
k-Nearest Neighbors with brief explanation.pptx
k-Nearest Neighbors with brief explanation.pptxk-Nearest Neighbors with brief explanation.pptx
k-Nearest Neighbors with brief explanation.pptxgamingzonedead880
 
Machine Learning Foundations for Professional Managers
Machine Learning Foundations for Professional ManagersMachine Learning Foundations for Professional Managers
Machine Learning Foundations for Professional ManagersAlbert Y. C. Chen
 
An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...
An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...
An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...IDES Editor
 
Robust inference via generative classifiers for handling noisy labels
Robust inference via generative classifiers for handling noisy labelsRobust inference via generative classifiers for handling noisy labels
Robust inference via generative classifiers for handling noisy labelsKimin Lee
 
P.maria sheeba 15 mco010
P.maria sheeba 15 mco010P.maria sheeba 15 mco010
P.maria sheeba 15 mco010W3Edify
 
Uncertainty aware multidimensional ensemble data visualization and exploration
Uncertainty aware multidimensional ensemble data visualization and explorationUncertainty aware multidimensional ensemble data visualization and exploration
Uncertainty aware multidimensional ensemble data visualization and explorationSubhashis Hazarika
 
MLSEV Virtual. Searching for Anomalies
MLSEV Virtual. Searching for AnomaliesMLSEV Virtual. Searching for Anomalies
MLSEV Virtual. Searching for AnomaliesBigML, Inc
 
Irrera gold2010
Irrera gold2010Irrera gold2010
Irrera gold2010grssieee
 
SVM Based Identification of Psychological Personality Using Handwritten Text
SVM Based Identification of Psychological Personality Using Handwritten Text SVM Based Identification of Psychological Personality Using Handwritten Text
SVM Based Identification of Psychological Personality Using Handwritten Text IJERA Editor
 
Satellite Image Classification using Decision Tree, SVM and k-Nearest Neighbor
Satellite Image Classification using Decision Tree, SVM and k-Nearest NeighborSatellite Image Classification using Decision Tree, SVM and k-Nearest Neighbor
Satellite Image Classification using Decision Tree, SVM and k-Nearest NeighborNational Cheng Kung University
 

Similar a Distance Metric Learning (20)

240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx
240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx
240408_JW_labseminar[Asymmetric Transitivity Preserving Graph Embedding].pptx
 
Unsupervised Distance Based Detection of Outliers by using Anti-hubs
Unsupervised Distance Based Detection of Outliers by using Anti-hubsUnsupervised Distance Based Detection of Outliers by using Anti-hubs
Unsupervised Distance Based Detection of Outliers by using Anti-hubs
 
Enhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wildEnhanced characterness for text detection in the wild
Enhanced characterness for text detection in the wild
 
Weighting NN
Weighting NNWeighting NN
Weighting NN
 
Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...
Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...
Lecture 11 - KNN and Clustering, a lecture in subject module Statistical & Ma...
 
Ability Study of Proximity Measure for Big Data Mining Context on Clustering
Ability Study of Proximity Measure for Big Data Mining Context on ClusteringAbility Study of Proximity Measure for Big Data Mining Context on Clustering
Ability Study of Proximity Measure for Big Data Mining Context on Clustering
 
Sparse inverse covariance estimation using skggm
Sparse inverse covariance estimation using skggmSparse inverse covariance estimation using skggm
Sparse inverse covariance estimation using skggm
 
Thesis presentation
Thesis presentationThesis presentation
Thesis presentation
 
k-Nearest Neighbors with brief explanation.pptx
k-Nearest Neighbors with brief explanation.pptxk-Nearest Neighbors with brief explanation.pptx
k-Nearest Neighbors with brief explanation.pptx
 
PNN and inversion-B
PNN and inversion-BPNN and inversion-B
PNN and inversion-B
 
Machine Learning Foundations for Professional Managers
Machine Learning Foundations for Professional ManagersMachine Learning Foundations for Professional Managers
Machine Learning Foundations for Professional Managers
 
An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...
An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...
An Efficient K-Nearest Neighbors Based Approach for Classifying Land Cover Re...
 
Robust inference via generative classifiers for handling noisy labels
Robust inference via generative classifiers for handling noisy labelsRobust inference via generative classifiers for handling noisy labels
Robust inference via generative classifiers for handling noisy labels
 
P.maria sheeba 15 mco010
P.maria sheeba 15 mco010P.maria sheeba 15 mco010
P.maria sheeba 15 mco010
 
Uncertainty aware multidimensional ensemble data visualization and exploration
Uncertainty aware multidimensional ensemble data visualization and explorationUncertainty aware multidimensional ensemble data visualization and exploration
Uncertainty aware multidimensional ensemble data visualization and exploration
 
MLSEV Virtual. Searching for Anomalies
MLSEV Virtual. Searching for AnomaliesMLSEV Virtual. Searching for Anomalies
MLSEV Virtual. Searching for Anomalies
 
Irrera gold2010
Irrera gold2010Irrera gold2010
Irrera gold2010
 
DBSCAN (1) (4).pptx
DBSCAN (1) (4).pptxDBSCAN (1) (4).pptx
DBSCAN (1) (4).pptx
 
SVM Based Identification of Psychological Personality Using Handwritten Text
SVM Based Identification of Psychological Personality Using Handwritten Text SVM Based Identification of Psychological Personality Using Handwritten Text
SVM Based Identification of Psychological Personality Using Handwritten Text
 
Satellite Image Classification using Decision Tree, SVM and k-Nearest Neighbor
Satellite Image Classification using Decision Tree, SVM and k-Nearest NeighborSatellite Image Classification using Decision Tree, SVM and k-Nearest Neighbor
Satellite Image Classification using Decision Tree, SVM and k-Nearest Neighbor
 

Más de Sanghyuk Chun

Más de Sanghyuk Chun (6)

Introduction to E-book
Introduction to E-bookIntroduction to E-book
Introduction to E-book
 
Markov Chain Basic
Markov Chain BasicMarkov Chain Basic
Markov Chain Basic
 
Coordinate Descent method
Coordinate Descent methodCoordinate Descent method
Coordinate Descent method
 
Gradient descent method
Gradient descent methodGradient descent method
Gradient descent method
 
K-means and GMM
K-means and GMMK-means and GMM
K-means and GMM
 
PAC Learning
PAC LearningPAC Learning
PAC Learning
 

Último

Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxRustici Software
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerThousandEyes
 
Vector Search -An Introduction in Oracle Database 23ai.pptx
Vector Search -An Introduction in Oracle Database 23ai.pptxVector Search -An Introduction in Oracle Database 23ai.pptx
Vector Search -An Introduction in Oracle Database 23ai.pptxRemote DBA Services
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Jeffrey Haguewood
 
Introduction to Multilingual Retrieval Augmented Generation (RAG)
Introduction to Multilingual Retrieval Augmented Generation (RAG)Introduction to Multilingual Retrieval Augmented Generation (RAG)
Introduction to Multilingual Retrieval Augmented Generation (RAG)Zilliz
 
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamDEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamUiPathCommunity
 
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...apidays
 
ICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesrafiqahmad00786416
 
CNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In PakistanCNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In Pakistandanishmna97
 
[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdfSandro Moreira
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MIND CTI
 
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdfRising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdfOrbitshub
 
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Orbitshub
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWERMadyBayot
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfsudhanshuwaghmare1
 
MS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectorsMS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectorsNanddeep Nachan
 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...DianaGray10
 
WSO2's API Vision: Unifying Control, Empowering Developers
WSO2's API Vision: Unifying Control, Empowering DevelopersWSO2's API Vision: Unifying Control, Empowering Developers
WSO2's API Vision: Unifying Control, Empowering DevelopersWSO2
 
Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...
Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...
Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...Angeliki Cooney
 

Último (20)

Understanding the FAA Part 107 License ..
Understanding the FAA Part 107 License ..Understanding the FAA Part 107 License ..
Understanding the FAA Part 107 License ..
 
Corporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptxCorporate and higher education May webinar.pptx
Corporate and higher education May webinar.pptx
 
How to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected WorkerHow to Troubleshoot Apps for the Modern Connected Worker
How to Troubleshoot Apps for the Modern Connected Worker
 
Vector Search -An Introduction in Oracle Database 23ai.pptx
Vector Search -An Introduction in Oracle Database 23ai.pptxVector Search -An Introduction in Oracle Database 23ai.pptx
Vector Search -An Introduction in Oracle Database 23ai.pptx
 
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
Web Form Automation for Bonterra Impact Management (fka Social Solutions Apri...
 
Introduction to Multilingual Retrieval Augmented Generation (RAG)
Introduction to Multilingual Retrieval Augmented Generation (RAG)Introduction to Multilingual Retrieval Augmented Generation (RAG)
Introduction to Multilingual Retrieval Augmented Generation (RAG)
 
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 AmsterdamDEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
DEV meet-up UiPath Document Understanding May 7 2024 Amsterdam
 
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
Apidays New York 2024 - Passkeys: Developing APIs to enable passwordless auth...
 
ICT role in 21st century education and its challenges
ICT role in 21st century education and its challengesICT role in 21st century education and its challenges
ICT role in 21st century education and its challenges
 
CNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In PakistanCNIC Information System with Pakdata Cf In Pakistan
CNIC Information System with Pakdata Cf In Pakistan
 
[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf[BuildWithAI] Introduction to Gemini.pdf
[BuildWithAI] Introduction to Gemini.pdf
 
MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024MINDCTI Revenue Release Quarter One 2024
MINDCTI Revenue Release Quarter One 2024
 
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdfRising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
Rising Above_ Dubai Floods and the Fortitude of Dubai International Airport.pdf
 
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
Navigating the Deluge_ Dubai Floods and the Resilience of Dubai International...
 
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWEREMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
EMPOWERMENT TECHNOLOGY GRADE 11 QUARTER 2 REVIEWER
 
Boost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdfBoost Fertility New Invention Ups Success Rates.pdf
Boost Fertility New Invention Ups Success Rates.pdf
 
MS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectorsMS Copilot expands with MS Graph connectors
MS Copilot expands with MS Graph connectors
 
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
Connector Corner: Accelerate revenue generation using UiPath API-centric busi...
 
WSO2's API Vision: Unifying Control, Empowering Developers
WSO2's API Vision: Unifying Control, Empowering DevelopersWSO2's API Vision: Unifying Control, Empowering Developers
WSO2's API Vision: Unifying Control, Empowering Developers
 
Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...
Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...
Biography Of Angeliki Cooney | Senior Vice President Life Sciences | Albany, ...
 

Distance Metric Learning

  • 1. Distance Metric Learning 2014-03-06 전상혁 2014-03-05 1
  • 3. Distance Metric •Adistancemetricordistancefunctionisafunctionthatdefinesadistancebetweenelementsofaset •Example1:Euclideanmetric •Example2:Discretemetric •푖푓푥=푦푡ℎ푒푛푑푥,푦=0,푂푡ℎ푒푟푤푖푠푒,푑푥,푦=1 •AmetriconasetXisafunction푑:푋×푋→퐑(whereRisrealnumber) •푑푥,푦≥0 •푑푥,푦=0푖푓푎푛푑표푛푙푦푖푓푥=푦 •푑푥,푦=푑(푦,푥) •푑푥,푧≤dx,y+d(y,z) 2014-03-05 Aurelien Bellet, Tutorial on Metric Learning, 2013 3
  • 4. Introduction 2014-03-05 Bellet, A., Habrard, A., and Sebban, M. A Survey on Metric Learning for Feature Vectors and Structured Data, 2013 4 Bellet, A., Habrard, A., and Sebban, M. A Survey on Metric Learning for Feature Vectors and Structured Data, 2013
  • 5. Categories of distance metric learning •Superviseddistancemetriclearning •Constraintsorlabelsgiventothealgorithm •Equivalenceconstraintswherepairsofdatapointsthatbelongtothesameclasses(semantically-similar) •Inequivalenceconstraintswherepairsofdatapointsthatbelongtothedifferentclasses(semantically-dissimilar) •localadaptivesuperviseddistancemetriclearning,NCA,RCA •Unsuperviseddistancemetriclearning •Dimensionalityreductiontechniques •Linearmethods(PCA,MDS)andnon-linearmethods(ISOMAP,LLE,LE) •Kernelmethodstowardslearningdistancemetrics •Kernelalignment,extensionworkoflearningidealizedkernel 2014-03-05 Liu Yang, An Overview of Distance Metric Learning, 2007 5 Liu Yang, Distance Metric Learning: A Comprehensive Survey, 2005 Liu Yang, An Overview of Distance Metric Learning, 2007
  • 6. Categories of distance metric learning •Maximummarginbaseddistancemetriclearning •Formulatedistancemetriclearningasaconstrainedconvexprogrammingproblem,andattempttolearncompletedistancemetricfromtrainingdata •Expensivecomputation,overfittingproblemwhenlimitedtrainingsamplesandlargefeaturedimension •Largemarginnearestneighborclassification(LMNN),SDPmethodstosolvethekernelizedmarginmaximization 2014-03-05 Liu Yang, An Overview of Distance Metric Learning, 2007 6 Liu Yang, Distance Metric Learning: A Comprehensive Survey, 2005 Liu Yang, An Overview of Distance Metric Learning, 2007
  • 7. Categories of distance metric learning 2014-03-05 Liu Yang, An Overview of Distance Metric Learning, 2007 7 Liu Yang, An Overview of Distance Metric Learning, 2007
  • 8. Large Margin Nearest Neighbor Classification (LMNN) KilianWeinberger,etal.Distancemetriclearningforlargemarginnearestneighborclassification.NIPS,2006. 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 8
  • 9. Motivation –better KNN •KNN:simplenon-parametricclassificationmethod(Alsocanbeusedforclustering,regression,anomalydetection,andetc.) •푘isuser-definedconstant,andanunlabeledvectorisclassifiedbyassigningthelabelwhichismostfrequentamongthe푘trainingsamplesnearesttothatunlabeledvector •Drawback:whentheclassdistributionisskewed,basic“majorityvoting” classificationmaynotworkswell. 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 9
  • 10. Introduction •Largemarginnearestneighbor(LMNN)method •NeighborsareselectedbyusingthelearnedMahanalobisdistancemetric •LearnaMahanalobisdistancemetricforkNNclassificationbysemidefiniteprogramming •Find 퐿:푅푑→푅푑,which will be used to compute squared distance as 퐷푥푖,푥푗=퐿푥푖−푥푗 2by minimizing some cost function 휖(퐿) 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 10
  • 11. MahanalobisDistance Metric •Relativemeasureofadatapoint’sdistancefromacommonpoint •ItdiffersfromEuclideandistanceinthatittakesintoaccountthecorrelationsofthedatasetandisscale-invariant •푑 푥, 푦= 푥− 푦⊤훀( 푥− 푦)where훀ispositivesemidefintematrix •Example:thedistancebetweentworandomvariable 푥, 푦ofthesamedistributionwithcovariancematrix푆isdefinedas •푑 푥, 푦= 푥− 푦⊤푆−1( 푥− 푦) •EuclideandistanceisspecialcasewhenSisidentitymatrix •Recall:distance퐷푥푖,푥푗=퐿푥푖−푥푗 2 •Wecanrewriteeq.as퐷푥푖,푥푗=푥푖−푥푗퐌⊤푥푖−푥푗wherethematrix퐌= 퐋⊤퐋,parameterizestheMahalanobisdistancemetricinducedbythelineartransform퐋 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 11
  • 12. SemidefiniteProgramming •Akindofconvexoptimizationproblem •푥isavectorofnvariables,푐isaconstantndimensionalvector •Linearprogramming(LP) •푚푖푛푖푚푖푧푒푐∙푥 푠.푡.푎푖∙푥=푏푖,푖=1,…,푚 푥∈푅+푛 •Xispositivesemidefinitematrixif푣⊤푋푣≥0푓표푟푎푛푦푣∈푅푛(푋≽0) •If퐶(푋)isalinearfunctionofX,then퐶(푋)canbewrittenas퐶∙푋 •Semidefiniteprogram(SDP) •푚푖푛푖푚푖푧푒퐶∙푋 푠.푡.퐴푖∙푋=푏푖,푖=1,…,푚 푋≽0 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 12
  • 13. Cost Function •휖퐿= 푖푗휂푖푗퐋푥푖−푥푗 2+푐 푖푗푙휂푖푗1−푦푖푙1+퐋푥푖−푥푗 2−퐋푥푖−푥푙 2+ •Let푥푖,yii=1ndenoteatrainingsetofnlabeledexampleswithinputs푥푖∈푅푑anddiscrete(butnotnecessarilybinary)classlabels푦푖 •푦푖푗∈{0,1}isbinarymatrixtoindicatewhetherornotthelabels푦푖and푦푗match •휂푖푗∈{0,1}toindicatewhetherinput푥푗isatargetneighborofinput푥푖,thisvalueisfixedanddoesnotchangeduringlearning •Targetneighbor:eachinstance푥푖hasexactlykdifferenttargetneighborswithindataset,whichallsharethesameclasslabel푦푖 •Targetneighborsarethedatapointsthatshouldbecomenearestneighborsunderthelearnedmetric •C.f.Impostors:sameastargetneighborexceptithasdifferentclasslabel •Term푧+=max(푧,0) •푐>0somepositiveconstant(typicallysetbycrossvalidation) 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 13
  • 14. Cost Function •휖퐿= 푖푗휂푖푗퐋푥푖−푥푗 2+푐 푖푗푙휂푖푗1−푦푖푙1+퐋푥푖−푥푗 2−퐋푥푖−푥푙 2+ •Thefirstoptimizationgoalisachievedbyminimizingtheaveragedistancebetweeninstanceandtheirtargetneighbors • •Thesecondgoalisachievedbyconstrainingimpostors푥푙tobeoneunitfurtherawaythantargetneighbors푥푗 • •푁푖denotethesetoftargetneighborsforadatapoint푥푖 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 14
  • 15. 2014-03-05 15 Nakul Verma, A tutorial on Metric Learning with some recent advances
  • 16. Convex Optimization •Recall:distance퐷푥푖,푥푗=퐿푥푖−푥푗 2 •Recall:Wecanrewriteeq.as퐷푥푖,푥푗=푥푖−푥푗퐌⊤푥푖−푥푗wherethematrix퐌=퐋⊤퐋,parameterizestheMahalanobisdistancemetricinducedbythelineartransform퐋 •Thehingelosscanbe“mimicked”byintroducingslackvariable휁푖푗forallpairsofdifferentlylabeledinputs(i.e.,forall<푖,푗>suchthat푦푖푗=0) 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 16
  • 17. Results •PCAwasusedtoreducethedimensionalityofimage,speech,andtextdata,bothtospeeduptrainingandavoidoverfitting •Boththenumberoftargetneighbors(k)andtheweightingparameter(c) inoptimizationproblemweresetbycrossvalidation •KNNEuclideandistance •KNNMahalanobisdistance •Energybasedclassification • •MulticlassSVM 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 17
  • 18. Results 2014-03-05 Kilian Weinberger, et al. Distance metric learning for large margin nearest neighbor classification. NIPS, 2006. 18
  • 19. GBLMNN: Non-linear method for LMNN •Buildingnon-linearmappingformetriclearning •GBRT:GradientBoostingRegressionTree •Code:http://www.cse.wustl.edu/~kilian/code/code.html 2014-03-05 Kedem, D., Xu, Z., & Weinberger, K. (n.d.). Gradient Boosted Large Margin Nearest Neighbors, (1), 10–12. 19
  • 20. Large Margin Component Analysis (LMCA) Torresani,L.,&Lee,K.(2007).Largemargincomponentanalysis.AdvancesinNeuralInformationProcessing 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 20
  • 21. Problem: large number of features •Inproblemsinvolvingthousandsoffeatures,distancemetriclearningcannotbeused •Highcomputationcomplexity •Overffitingproblem •Previousworkshavereliedonatwo-stepsolution •Applydimensionreductionalgorithm(e.g.PCA) •Learnametricintheresultinglow-dimensionalsubspace •LMCA:providebettersolutionforhighdimensionalproblem 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 21
  • 22. Linear Dimensionality Reduction for KNN (1/3) •Recall:costfunctionofmetricfunction •Inpreviouswork,Lwas퐷×퐷matrix •Idea:learnnonsquarematrixsize푑×퐷,푤푖푡ℎ푑≪퐷 •Problem:costfunction휖퐌,푤ℎ푒푟푒퐌=퐋⊤퐋isnolongerconvex •Mhaslowrankd,notD 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 22
  • 23. Linear Dimensionality Reduction for KNN (2/3) •Idea:optimizetheobjectivefunctiondirectlywithrespecttothenonsquarematrixL •Although,equationisnon-convex,itisbetterthanminimizingtheobjectivefunctionwithrespecttoLratherthanwithrespecttotherank- deficientD×DmatrixM •Sincethisoptimizationinvolvesonly푑∙퐷ratherthan퐷2unknowns,itcanreducetheriskofoverffiting •“theoptimalrectangularmatrixLcomputedwithourmethodautomaticallysatisfiestherankconstraintsonMwithoutrequiringthesolutionofdifficultconstrainedminimizationproblems” 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 23
  • 24. Linear Dimensionality Reduction for KNN (3/3) •Minimizecostfunctionusinggradient-basedoptimizers •“Wehandlethenon-differentiabilityofh(s)ats=0,byadoptingasmoothhingefunctionasin[8]” 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 24 [8] J. D. M. Rennieand N. Srebro. Fast maximum margin matrix factorization for collaborative prediction. In Proceedings of the 22nd International Conference on Machine Learning (ICML), 2005.
  • 25. Nonlinear Feature Extraction (1/2) •“Ourapproachlearnsalow-rankMahalanobisdistancemetricinahighdimensionalfeaturespaceF,relatedtotheinputsbyanonlinearmap 휙:푅퐷→퐹” •kiskernelfunctioncomputedbythefeatureinputproducts •LisnowatransformationfromthespaceFintoa푅푑(푑≪퐷) • 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 25
  • 26. Nonlinear Feature Extraction (2/2) • • • •UpdateΩ←(Ω−휆Γ)untilconverge •Forclassification,weprojectpointsontothelearnedlow-dimspacebyexploitingthekerneltrick:퐿흓=Ω퐤 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 26
  • 27. Results –High dimensional dataset 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 27
  • 28. Results –low dimensional dataset 2014-03-05 Torresani, L., & Lee, K. (2007). Large margin component analysis. Advances in Neural Information Processing 28
  • 29. Reference •Bellet,A.,Habrard,A.,andSebban,M.ASurveyonMetricLearningforFeatureVectorsandStructuredData,2013 •LiuYang,DistanceMetricLearning:AComprehensiveSurvey,2005 •LiuYang,AnOverviewofDistanceMetricLearning,2007 •MarcoCuturi.KAISTMachineLearningTutorialMetricsandKernelsAfewrecenttopics,2013 •NakulVerma,AtutorialonMetricLearningwithsomerecentadvances •AurelienBellet,TutorialonMetricLearning,2013 •BrianKulis.TutorialonMetricLearning.InternationalConferenceonMachineLearning(ICML)2010 2014-03-05 29
  • 30. Reference •K.Q.Weinberger,J.Blitzer,andL.K.Saul(2006).InY.Weiss,B.Schoelkopf, andJ.Platt(eds.),DistanceMetricLearningforLargeMarginNearestNeighborClassification,AdvancesinNeuralInformationProcessingSystems18(NIPS-18).MITPress:Cambridge,MA. •K.Q.Weinberger,L.K.Saul.DistanceMetricLearningforLargeMarginNearestNeighborClassification.JournalofMachineLearningResearch(JMLR)2009,10:207-244 •Torresani,L.,&Lee,K.(2007).Largemargincomponentanalysis. AdvancesinNeuralInformationProcessing •Kedem,D.,Xu,Z.,&Weinberger,K.(n.d.).GradientBoostedLargeMarginNearestNeighbors,(1),10–12. 2014-03-05 30