Distance Metric Learning:

A Comprehensive Survey

Liu Yang
Advisor: Rong Jin
May 8th, 2006
Outline
 Introduction
 Supervised Global Distance Metric Learning
 Supervised Local Distance Metric Learning
 Unsupervised Distance Metric Learning
 Distance Metric Learning based on SVM
 Kernel Methods for Distance Metrics Learning
 Conclusions
 Introduction
 Definition
 Distance Metric learning is to learn a distance metric for the

input space of data from a given collection of pair of

similar/dissimilar points that preserves the distance relation
among the training data pairs.
 Importance
 Many machine learning algorithms, heavily rely on the distance

metric for the input data patterns. e.g. kNN

 A learned metric can significantly improve the performance in

classification, clustering and retrieval tasks:

e.g. KNN classifier, spectral clustering, content-based image
retrieval (CBIR).
Contributions of this Survey
 Review distance metric learning under different learning
conditions
 supervised learning vs. unsupervised learning

 learning in a global sense vs. in a local sense

 distance matrix based on linear kernel vs. nonlinear kernel

 Discuss central techniques of distance metric learning

 K nearest neighbor

 dimension reduction

 semidefinite programming

 kernel learning

 large margin classification

 Global Distance Metric Learning
by Convex Programming
Local Adaptive Distance
Supervised Metric Learning
Distance Metric Learning
Relevant Component Analysis

Local Neighborhood Components Analysis

Linear embedding
Unsupervised PCA, MDS
Distance Metric Learning Nonlinear embedding
LLE, ISOMAP, Laplacian Eigenmaps

Large Margin Nearest Neighbor

Distance Metric Learning Based Distance Metric Learning
based on SVM Cast Kernel Margin
Maximization into a SDP problem

Kernel Alignment with SDP

Kernel Methods for
Distance Metrics Learning Learning with Idealized Kernel
Outline
 Introduction
 Supervised Global Distance Metric Learning
 Supervised Local Distance Metric Learning
 Unsupervised Distance Metric Learning
 Distance Metric Learning based on SVM
 Kernel Methods for Distance Metrics Learning
 Supervised Global Distance Metric
Learning (Xing et al. 2003)
Equivalence constraints: S  {( xi , x j ) | xi and x j belong to the same class}
Inequivalence constraints: D  {( xi , x j ) | xi and x j belong to different classes},
2
d 2A ( x, y )  x  y A
 ( x  y )T A( x  y ), A  S mm is the distance metric

 Goal : keep all the data points within the same classes close,
while separating all the data points from different classes.

 Formulate as a constrained convex programming problem

 minimize the distance between the data pairs in S

 Subject to data pairs in D are well separated

 Global Distance Metric Learning (Cont’d)

 
2 2
min
mm
xi  x j s.t. A  0, xi  x j 1
AR
( xi , x j )S A ( xi , x j )D A

 A is positive semi-definite
 Ensure the negativity and the triangle inequality of the metric

 The number of parameters is quadratic in the number of features

 Difficult to scale to a large number of features

 Simplify the computation

 Global Distance Metric Learning:
Example I

(a) Data Dist. of the original dataset (b) Data scaled by the global metric

 Keep all the data points within the same classes close
 Separate all the data points from different classes
 Global Distance Metric Learning:
Example II

(a) Original data (b) rescaling by learned (c) Rescaling by learned

full A diagonal A

 Diagonalize distance metric A can simplify computation, but

could lead to disastrous results
 Problems with Global Distance
Metric Learning

(a) Data Dist. of the original dataset (b) Data scaled by the global metric

Multimodal data distributions prevent global distance metrics

from simultaneously satisfying constraints on within-class
compactness and between-class separability.
Outline
 Introduction
 Supervised Global Distance Metric Learning
 Supervised Local Distance Metric Learning
 Unsupervised Distance Metric Learning
 Distance Metric Learning based on SVM
 Kernel Methods for Distance Metrics Learning
 Conclusions
 Supervised Local Distance Metric
Learning
 Local Adaptive Distance Metric Learning

 Local Feature Relevance

 Locally Adaptive Feature Relevance Analysis
 Local Linear Discriminative Analysis

 Neighborhood Components Analysis

 Relevant Component Analysis

 Local Adaptive Distance Metric
Learning
 K Nearest Neighbor Classifier

1
Pr(j | x0 ) 
N ( x0 )

xi N ( x0 )
 ( yi  j )

N ( x0 ) : nearest neighbors of x0
  x , y  , ,  x , y   : training examples
1 1 n n

1 yi  j
 ( yi  j )  
0 o.w.
 Local Adaptive Distance Metric
Learning

 Assumption of KNN
 Pr(y|x) in the local NN is constant or smooth
 However, this is not necessarily true!
 Near class boundaries
 Irrelevant dimensions

 Modified local neighborhood by a distance metric

 Elongate the distance along the dimensions where

the class labels change rapidly

 Squeeze the distance along the dimensions that are

almost independent from the class labels

 Local Feature Relevance
[J. Friedman,1994]
 Assume least-squared estimate for predicting f(x) is
Ef   f (x)p(x)dx
 Conditioned at x i  z, then the least-squared estimate of f(x)
p(x) (xi  z )
E[ f | xi  z ]   f (x)p(x|x i =z)dx, p(x|x =z) =
 p(x') (x ' z )
i
i

 The improvement in prediction error with knowing x i  z

I i2 ( z )  E[( f (x)-Ef ) 2 | xi  z ]  E[( f (x)-E(f (x) | xi  z ) 2 | xi  z ]  ( Ef  E[ f | xi  z ]) 2


 Consider z  ( z1 , , zm ) , a measure of relative influence of the

ith input variable to the variation ofp f(x) at x = z is given by
ri 2 ( z )  I i2 ( zi ) /  I k2 ( zk )
k 1
 Locally Adaptive Feature Relevance
Analysis [C. Domeniconi, 2002]
 Use a Chi-squared distance analysis to compute metric for
producing a neighborhood, in which

 The posterior probabilities are approximately constant

 Highly adaptive to query locations

 Chi-squared distance between the true and estimated posterior

at the test point x 0 J
[ p ( j | X)  p ( j | x 0 )]
2
r (X, x 0 )  
j 1 p( j | x 0 )

 Use the Chi-squared distance for feature relevance:

---- to tell to which extent the ith dimension can be relied on for
predicting p(j|x 0 )
 Local Relevance Measure
in ith Dimension

 ri (z) measures the distance between Pr(j|z) and the conditional

expectation of Pr(j|x) at location z

 Calculate ri (z) for each point z in the neighborhood of x 0


J
[Pr( j | z )  Pr( j | xi  zi )]2
ri (z) =  


j 1 Pr( j | xi  zi )
Pr( j | xi  zi )  E (Pr(j|x) | xi  zi ) is a conditional expectation of p(j|x)


 The closer Pr( j | xi  zi ) is to p(j|z), the more information the ith
dimension provides for predicting p(j|z)
 Locally Adaptive Feature
Relevance Analysis
 A local relevance measure in dimension i
 1
ri (x 0 ) 
K

zN (x 0 )
ri ( z ) N (x 0 ) is the neighborhood of x 0

 Relative relevance
Ri ( x0 ) t  
w i ( x0 )  q
, where Ri ( x0 )  (max q
j 1 rj ( x0 ))  ri ( x0 )
 l 0
R (
l 1
x ) t

t= 1 or 2, corresponds to linear and quadratic weighting.

q
 Weighted distance D(x,y) =  w (x  y )
i=1
i i i
2
 Local Linear Discriminative Analysis
[T. Hastie et al. 1996]
 LDA finds principle eigenvectors of matrix T = Sw -1Sb.

 to keep patterns from the same class close

 separate patterns from different classes apart

Sb : the between-class covariance matrix

Sw : the within-class covariance matrix

 LDA metric : stacking principle eigenvectors of T together

Local Linear
Discriminative Analysis
 Need local adaptation of the nearest neighbor metric

 Initialize  as identical matrix

 Given a testing point x 0 , iterate below two steps:

 Estimate Sb and Sw based on the local neighbor

of x 0 measured by 
 Form a local metric behaving like LDA metric
1 1 1 1
   
  S w [ S w Sb S w   I]S w
2 2 2 2

 is a small tuning parameter to prevent neighborhoods

extending to infinity
 Local Linear Discriminative Analysis
 Local Sb shows the inconsistency of the class centriods
 The estimated metric
 shrinks the neighborhood in directions in which the local class
centroids differ to produce a neighborhood in which the class
centriod coincide
 shrinks neighborhoods in directions orthogonal to these local
decision boundaries, and elongates them parallel to the boundaries.
 Neighborhood Components Analysis
[J. Goldberger et al. 2005]
 NCA learns a Mahalanobis distance metric for the KNN
classifier by maximizing the leave-one-out cross validation.

 The probability of classifying x i correctly, pi   pij

jC
weighted counting involving pairwise distance i

2
exp( Ax i  Ax j )
Here Ci  { j | ci  c j }, pij 
 exp( Ax i  Ax k )
2

k i
 The expected number of correctly classification points:
n
f(A) =  pi ,
i=1

Overfitting, Scalability problem, # parameters is quadratic in #features.

 RCA [N. Shen et al. 2002]
 Constructs a Mahalanobis distance metric based on a sum of
 in-chunklet covariance matrices

 Chunklet : data have same but unknown class labels

unlabeled data chuklet data labeled data

 Sum of in-chunklet covariance matrices for p points in k
chunklets:
^ 1 k
n j ^ ^ ^

n
C (x ji  m j )(x ji  m j ) ,
T
chunklet j : {x ji }i=1 , with mean m j
j

pj 1 i 1
1
^ 2
yC x
Information maximization
under chunklet constraints
[A. Bar-Hillel etal, 2003]
 Maximizes the mutual information I(X,Y)
 Constraints: within-chunklet compactness
nj 2
k
1
max I(X,Y) s.t.
f F
 ji j
p j 1 i 1
y  m y
 K . (*)

m yj is the transformed mean in the jth chunklet.

K is threshold constant.

Let B =A T A, (*) can be further written into

nj 2
k
1
max | B | s.t.  x ji  m j  K , B 0
B p j 1 i 1 B
 RCA algorithm applied to
synthetic Gaussian data

 (a) The fully labeled data set with 3 classes.

 (b) Same data unlabeled; classes' structure is less evident.
 (c) The set of chunklets
 (d) The centered chunklets, and their empirical covariance.
 (e) The RCA transformation applied to the chunklets. (centered)
 (f) The original data after applying the RCA transformation.
Outline
 Introduction
 Supervised Global Distance Metric Learning
 Supervised Local Distance Metric Learning
 Unsupervised Distance Metric Learning
 Distance Metric Learning based on SVM
 Kernel Methods for Distance Metrics Learning
 Conclusions
 Unsupervised Distance Metric Learning

 Most dimension reduction approaches are to learn a distance

metric without label information. e.g. PCA

 I will present five methods for dimensionality reduction.

linear nonlinear
Global PCA, MDS ISOMAP
Local LLE, Laplacian Eigenmap

 A Unified Framework for Dimension Reduction

 Solution 1

 Solution 2
 Dimensionality Reduction Algorithms
 PCA finds the subspace that best preserves the variance of the data.

 MDS finds the subspace that best preserves the interpoint distances.

 Isomap finds the subspace that best preserves the geodesic interpoint
distances. [Tenenbaum et al, 2000].

 LLE finds the subspace that best preserves the local linear structure of the
data [Roweis and Saul, 2000].

 Laplacian Eigenmap finds the subspace that best preserves local

neighborhood information in the adjacency graph [M. Belkin and P.
Niyogi,2003].
 Multidimensional Scaling (MDS)
 MDS finds the rank m projection that best preserves the
inter-point distance given by matrix D
 Converts distances to inner products B= (D)= X T X
 Calculate X [VMDS, MDS ] =eig(B)
1
X = V MDS ( )
MDS 2

 Rank m projections Y closet to X

1
Y= V ( )
MDS
m
MDS 2
m
 Given the distance matrix among
cities, MDS produces this map:
 PCA (Principal Component Analysis)
 PCA finds the subspace that best preserves the data
variance.  =Var(X)
[V PCA ,  PCA ]=eig(  )

 PCA projection
Y = Vof XX
PCA with rank m
m

1
 PCAVvs.
PCA MDS MDS
 XV ,  PCA   MDS , Y PCA  ( PCA 2
) Y MDS

 In the Euclidean case, MDS only differs from PCA by

starting with D and calculating X.
 Isometric Feature Mapping (ISOMAP)
[Tenenbaum et al, 2000]
 Geodesic :the shortest curve on a manifold
that connects two points on the manifold
A B
e.g. on a sphere, geodesics are great circles

 Geodesic distance: length of the geodesic

 Points far apart measured by geodesic dist.

appear close measured by Euclidean dist.
 ISOMAP
 Take a distance matrix as input

 Construct a weighted graph G based on neighborhood relations

 Estimate pairwise geodesic distance by

“a sequence of short hops” on G

 Apply MDS to the geodesic distance matrix

 Locally Linear Embedding (LLE)
[Roweis and Saul, 2000]
 LLE finds the subspace that best preserves the local
linear structure of the data

 Assumption: manifold is locally “linear”

Each sample in the input space is a linearly weighted

average of its neighbors.
 A good projection should best preserve this geometric
locality property
 LLE
 W: a linear representation of every data point by its neighbors
 Choose W by minimized the reconstruction
2
error
n K
minimizing  x W x
i=1
i
j 1
ij ij

n
s.t. W
j 1
ij  1, x i ; Wij  0 if x j is not a neighbor of x i

 Calculate a neighborhood preserving mapping Y, by minimizing

the reconstruction error
K
 (Y)= yi   Wij* yij , where W*  arg min  (W)
i 1 W

 Y is given by the eigenvectors of the m lowest nonzero

T
eigenvalues of matrix (I-W) (I-W)
 Laplacian Eigenmap
[M. Belkin and P. Niyogi,2003]
 Laplacian Eigenmap finds the subspace that best preserves local
neighborhood information in adjacency graph

 Graph Laplacian: Given a graph G with weight matrix W

D is a diagonal matrix with Dii   Wji
L =D –W is the graph Laplacian j

 Detailed steps:
 Construct adjacency graph G.
 Weight the edges: Wij  1, if nodes i and j are connected, and 0 otw.
 Generalized eigen-decomposition of Lf= Df
 Embedding : eigenvectors with top m nonzero eigenvalues
 A Unified Framework for
Dimension Reduction Algorithms
 All use an eigendecomposition to obtain a lower-dimensional embedding
of data lying on a non-linear manifold.
 Normalize affinity matrix  Hij ^
H  j
H
^
eig (H)  the m largest positive eigenvalues t and eigenvectors v t

 The embedding of xhas two alternative solutions

 Solution 1 : (MDS i& Isomap)
ei with eit = t vit
is the best approximation of ^ in the squared error sense.
ei , e j H ij
 Solution 2 : (LLE & Laplacian Eigenmap)

yi with yit = v ti
Outline
 Introduction
 Supervised Global Distance Metric Learning
 Supervised Local Distance Metric Learning
 Unsupervised Distance Metric Learning
 Distance Metric Learning based on SVM
 Kernel Methods for Distance Metrics Learning
 Conclusions
Distance Metric Learning based on SVM
 Large Margin Nearest Neighbor Based Distance Metric
Learning
 Objective Function

 Reformulation as SDP

 Cast Kernel Margin Maximization into a SDP Problem

 Maximum Margin

 Cast into SDP problem

 Apply to Hard Margin and Soft Margin

 Large Margin Nearest Neighbor
Based Distance Metric Learning
[K. Weinberger et al., 2006]
 Learns a Mahanalobis distance metric in the kNN classification
setting by SDP, that
 Enforces the k-nearest neighbors belong to the same class
 examples from different classes are separated by a large margin

 After training
 k=3 target neighbors lie within a smaller radius
 differently labeled inputs lie outside this smaller radius with a
margin of at least one unit distance.
 Large Margin Nearest Neighbor
Based Distance Metric Learning
 Cost function:
 (L) = ij L(x i -x j ) 2  Cij (1  yil )[1  L(x i -x j ) 2  L(x i -x l ) 2 ]
2 2 2

ij ijl

[ z ]  max(z,0) denotes the standard hinge loss and the constant C > 0.
yij  {0,1} indicate whether or not the label y i and y j match
ij  {0,1} indicate whether x j is a target neighbor of x i

 Penalize large distances between each input and its target neighbors

 The hinge loss is incurred by differently labeled inputs whose

distances do not exceed the distance from input x i to any of its target
neighbors by one absolute unit of distance
-> do not threaten to invade each other’s neighborhoods
 Reformulation as SDP
2
Let L(x i -x j )  (x i -x j )T M(x i -x j ), and introducing slack variable ijl
2

The resulting SDP is :

min ij (x i  x j )T M(x i  x j )  Cij (1  yil )ijl
M
ij ijl

s.t. (x i  x j )T M(x i  x j )  (x i  x l ) T M(x i  x l )  1  ijl

ijl  0, M  =0
 Cast Kernel Margin Maximization
into a SDP Problem
[G. R. G. Lanckriet et al, 2004]
 Maximum margin : the decision boundary has the maximum
minimum distance from the closest training point.
 Hard Margin: linearly separable

 Soft Margin: nonlinearly separable

 The performance measure, generalized from dual solution of

different maximizing margin problem
wC , (K)  max 2 T e   T (G ( K )   I ) : C    0,  T y  0


with   0 on the training data w.r.t K. G is Gram matrix.

 Cast into SDP Problem
min t
K,t, , ,

s.t. trace(K)=c, K  =0,   0,   0,

min wC , (K) s.t. trace(K) = c 
K =0
 G(K tr   I ntr ) e       y 
   0
 (e      y) t-2C e 
T T

 Hard Margin
min w(K tr ) s.t. trace(K)=c. Here w(K tr ) =w ,0 (K tr )
K =0

 1-norm soft margin

min wS 1 (K tr ) s.t. trace(K)=c. Here wS 1 (K tr ) =w C,0 (K tr )
K =0

 2-norm soft margin

min wS 2 (K tr ) s.t. trace(K)=c. Here wS 2 (K tr ) =w , (K tr )
K =0
Outline
 Introduction
 Supervised Global Distance Metric Learning
 Supervised Local Distance Metric Learning
 Unsupervised Distance Metric Learning
 Distance Metric Learning based on SVM
 Kernel Methods for Distance Metrics Learning
 Conclusions
Kernel Methods for
Distance Metrics Learning
 Learning a good kernel is equivalent to distance metric learning

 Kernel Alignment

 Kernel Alignment with SDP

 Learning with Idealized Kernel

 Ideal Kernel
 The Idealized Kernel
 Kernel Alignment
[N. Cristianini,2001]
 A measure of similarity between two kernel functions or between
a kernel and a target function

 The inner product between n

two kernel matrices based on kernel k1
and k2. K1 , K 2 F   K1 (x i , x j )K 2 (x i , x j )
i , j 1
^ K1 , K 2
 The alignment of K1 and K2 w.r.t S: A(S, k1 , k 2 )  F

K 1 , K1 F
K2 , K2 F

 Measure the degree of agreement between a kernel and a given

learning task.
^ K1 , yyT
A(S, k1 , k 2 )  F
, y {  1}m
m K1 , K1 F
Kernel Alignment with SDP
[G. R. G. Lanckriet et al, 2004]
 Optimizing the alignment between a set of labels and a
kernel matrix using SDP in a transductive setting.
 K tr K tr,t 
K=  T  , where K ij  (x i ), (x j ) ,i, j =1, , n tr  n t .
K K 
 tr,t

 Optimizing an objective function over the training data

block -> automatic tuning of testing data block
^
max A( S , K1 , yy T ) s.t. K  =0, trace(K) =1
K

 Introduce A with K T K  =A and trace(A)  1 , this reduces to

max K tr , yy T
A,K F

 A KT 
s.t. trace(A)  1, K  =0,    =0.
 K In 
Learning with Idealized Kernel
[J. T. Kwok and I.W. Tsang,2003]
 Idealize a given kernel by making it more similar to the
ideal kernel matrix. 1, y(x i )  y(x j )
 Ideal kernel: k (x i , x j )  
*

0, y(x i )  y(x j )

 *
~
 Idealized kernel: k = k + k
2
~ K,K *
 The alignment of k will be greater than k, if    n 2  n 2
n , n are the number of positive and negative samples. 

 Under the original distance metric M:

k(x i , x j ) = x iT Mx j , M  =0; d ij2  (x i - x j )T M(x i - x j )
~ ~ ~ d ij2 yi =y j
K ii  K jj  2 K ij   2
d ij   yi  y j
 Idealized kernel
~2
 We modify d ij  (x i - x j ) A A(x i - x j )
T T

 Search for a matrix A under which

 different classes : pulled apart by an amount of at least 
 same class :getting close together.
~ d ij2 yi = y j
d  2
2
ij
 d ij   yi  y j
 Introduce slack variables for error tolerance
1 2 C  1 
min B 2  S
B,  , ij 2 NS
 ij D    C   
ND
 ij  , where B= AA T

(x i ,x j )S  ( x i ,x j )D 
 2 ~2 ~

dij  dij    ij , (x i , x j )  D

s.t.  ~
, ij  0,   0,
d 2  d 2   , (x , x )  S
 ij ij ij i j
Conclusions
 A comprehensive review, covers:
 Supervised distance metric learning

 Unsupervised distance metric learning

 Maximum margin based distance metric learning approaches

 Kernel methods towards distance metrics

 Challenge:
 Unsupervised distance metric learning.

 Going local in a principle manner.

 Learn an explicit nonlinear distance metric in the local sense.

 Efficiency issue.