Title: Linear%20and%20Nonlinear%20Data%20Dimensionality%20Reduction
1Linear and NonlinearData Dimensionality Reduction
David Gering
- Eigenfaces
- Turk, Pentland, 1991
- Locally Linear Embedding
- Saul, Roweis, 2000
- Isomap
- Tennenbaum, Langford, 2000
2Agenda
- Introduction to the Problem
- Background on PCA, MDS
- Eigenfaces
- LLE
- Isomap
- Summary
3Data Dimensionality Reduction for Images
D gtgt d
4Agenda
- Introduction to the Problem
- Background on PCA, MDS
- Eigenfaces
- LLE
- Isomap
- Summary
5Principle Component Analysis
- 3 Appoaches
- Least Squares Dist.
- Change of Variables
- Matrix Factorization
6PCA Approach 1 Least Squares Dist.
Hotelling, 1901
0-Dimensional single point
D2 xkxk1 xk2T
7PCA Approach 1 Least Squares Dist.
1-Dimensional single line
I2
u
m
I1
D2 xkxk1 xk2T
d1 ykyk1T
8PCA Approach 1 Least Squares Dist.
d-Dimensional d lines
I2
u1
u2
m
I1
Ii standard basis (axes) xij standard
components ui principle axes yij principle
components
D2 xkxk1 xk2T d2 ykyk1 yk2T
9PCA Approach 1 Least Squares Dist.
d-Dimensional d lines
I2
I2
u1
u2
u2
m
m
u1
I1
I1
10PCA Approach 2 Change of Variables
Pearson 1933
kTH principle component linear combination
ukTx that maximizes Var(ukTx) subject to
uTu1 Cov(ukx, ujx) 0 for k lt j
11PCA Approach 3 Matrix Factorization
12PCA Approach 3 Matrix Factorization
13MDS Classical Multidimensional Scaling
14PCA Summary
Compression
15Agenda
- Introduction to the Problem
- Background on PCA, MDS
- Eigenfaces
- LLE
- Isomap
- Summary
16Eigenfaces
- PCA
- Project faces onto a face space that spans the
significant variations of known faces - Projected Face weighted sum of Eigenfaces
- Eigenfaces are the eigenvectors (principle axes)
of the scatter matrix that span face space
17Eigenfaces Algorithm
18Eigenfaces Experimental Results
19Eigenfaces Applications
- (Training)
- Calculate the basis from the training set images.
- Project the training images into FaceSpace
- Compression
- Project the test image into FaceSpace
- Detection
- Determine if the image is a face by measuring its
distance from FaceSpace - Recognition
- If it is a face, compare it to the training
images (using FaceSpace coordinates) - Knobification
20Eigenfaces Advantages
- Discovers structure of data lying near a linear a
subspace of the input space - Unsupervised learning
- Linear nature easily visualized
- Simple implementation
- No assumptions regarding statistical distribution
of data - Non-iterative, globally optimal solution
- Polynomial time complexity
- Training O(N3)
- Test O(DND)
21Eigenfaces Disadvantages
- Not capable of discovering nonlinear degrees of
freedom - Optimal only when xi form a hyperellipsoid cloud
- Multi-dimensional Gaussian distribution
- Consequence of least-squares derivation
- Unable to answer how well new data are fit
probabilistically - Registration and scaling issues
- Compression gtgt Detection gtgt Recognition
- Eigenfaces are not logical face components
22Agenda
- Introduction to the Problem
- Background on PCA, MDS
- Eigenfaces
- LLE
- Isomap
- Summary
23Locally Linear Embedding (LLE)
Main Idea Overlapping local structure
collectively analyzed can provide information
about global geometry
24LLE Algorithm
25LLE Algorithm
Step Name Description
1 O(DN2) K neighbors Compute the neighbors of each data point xi
2 O(DNK3) Wij Compute the weights Wij that best reconstruct each data point xi from its neighbors
3 O(dN2) yi Compute the vectors yi that are best reconstructed by the weights Wij
26LLE Advantages
- Ability to discover nonlinear manifolds of
arbitrary dimension - Non-iterative
- Global optimality
- Few parameters K, d
- Captures context
- O(DN2) and space efficient due to sparse matrix
27LLE Disadvantages
- Requires smooth, non-closed, densely sampled
manifold - Must choose parameters K, d
- Quality of manifold characterization dependent on
neighborhood choice - Fixed radius would allow K to vary locally
- Clustering would help high, irregular curvature
- Sensitive to outliers
- Weight less those Wi for points with poor
reconstructions (in least squares sense)
28Comparisons LLE vs Eigenfaces
- PCA find embedding coordinate vectors that
minimize distance to all data points - LLE find embedding coordinate vectors that best
fit local neighborhood relationships - Application to Recognition and Detection (Map
test image from input space to manifold space) - Determine novel points K neighbors
- Compute test points reconstruction as linear
combination of training points - Approximate points manifold coordinates by
applying its weights to Y from training stage
29Agenda
- Introduction to the Problem
- Background on PCA, MDS
- Eigenfaces
- LLE
- Isomap
- Summary
30Isomap
Main Idea Use approximate geodesic distance
instead of Euclidean distance
31Isomap Algorithm
Step Name Description
1 O(DN2) Construct neighborhood graph, G Compute matrix DGdX(i,j) dx(i,j) Euclidean distance between neighbors
2 O(DN2) Compute shortest paths between all pairs Compute matrix DGdG(i,j) dG(i,j) sequence of hops approx geodesic dist.
3 O(dN2) Construct d-dimensional coordinate vectors, yi Apply MDS to DG instead of DX
32Isomap Algorithm
33Isomap Comparison Eigenfaces
- Eigenfaces w/ MDS
- DX derived from X in Euclidean space
- Only difference with Isomap
- DG is computed in approximate Geodesic space
- Map test image to manifold
- Compute dx to neighbors
- Interpolate y of neighbors
34Isomap Advantages
- Nonlinear
- Non-iterative
- Globally optimal
- Parameters K, d
35Isomap Comparison to LLE
- Similarities
- Begin with preprocessing step to identify
neighbors - Preserve intrinsic geometry of data by computing
local measures, after which data can be discarded - Overcome limitations of attempts to extend PCA
- Difference in local measure
- Geodesic distance vs. neighborhood relationship
- Differences in application
- Depends how well local metrics characterize
manifold - Isomap more robust to outliers
- Isomap preserves distance, LLE angles
- LLE avoids complexity of pairwise distance
computation
36Choosing the Algorithm for the Application
- Smooth manifolds manifest themselves as slightly
noticeable changes between an ordered set of
examples - Video sequence
- Data that could conceptually be organized as an
aesthetically pleasing video sequence - Presence of nonlinear structure not known a
priori - Run both PCA and manifold learning to inspect
results for discrepency - Preservation of gedesic distances
37Agenda
- Introduction to the Problem
- Background on PCA, MDS
- Eigenfaces
- LLE
- Isomap
- Summary
38Summary
Eigenfaces LLE Isomap
Optimization Constraint Linear combination of original coordinates that accounts for most variance Local intrinsic geometry represented as linear combination of neighbors Geodesic distance approximated as neighbor-to-neighbor hopping distance
Parameters None K or ? K or ?
Global Optimality Yes Yes Yes
Nonlinear Manifolds No Yes Yes
Training Complexity O(N3) O(DN2) O(DN2)
39Isomap Disadvantages
- Guaranteed asymptotically to recover geometric
structure of nonlinear manifolds - As N increases, pairwise distances provide better
approximations to geodesics by hugging surface
more closely - Graph discreteness overestimates dM(i,j)
- K must be high to avoid linear shortcuts near
regions of high surface curvature - Mapping novel test images to manifold space