Singular Value Decomposition Jonathan P. Bernick Department of Computer Science Coastal Carolina University.

Singular Value Decomposition

Jonathan P. Bernick

Department of Computer Science

Coastal Carolina University

2

Outline

Derivation Properties of the SVD Applications Research Directions

3

Matrix Decompositions

Definition: The factorization of a matrix M into two or more matrices M1, M2,…, Mn, such that M = M1M2…Mn.

Many decompositions exist… QR Decomposition LU Decomposition LDU Decomposition Etc.

One is special…

4

[Will] For an m by n matrix A:nm and any orthonormal basis {a1,...,an} of n, define

(1) si = ||Aai||

(2)

Theorem One

01

0

iii

i

i ss

s

Aah

0

Then…

5

Theorem One (continued)

nn

n

s

s

s

a

a

a

hhhA

2

1

2

1

21

000

000

000

000

|||

iii

iii sss AaAah

1Proof:

6

Theorem Two

[Will] For an m by n matrix A, there is an orthonormal basis {a1,...,an} of n such that for all i j, Aai Aaj = 0

Proof: Since ATA is symmetric, the existence of {a1,...,an} is guaranteed by the Spectral Theorem.

Put Theorems One and Two together, and we obtain…

7

Singular Value Decomposition

[Strang]: Any m by n matrix A may be factored such that

A = UVT

U: m by m, orthogonal, columns are the eigenvectors of AAT

V: n by n, orthogonal, columns are the eigenvectors of ATA

: m by n, diagonal, r singular values are the square roots of the eigenvalues of both AAT and ATA

8

SVD Example

From [Strang]:

10

01

0

3

0

0

0

2

100

010

001

0

3

0

0

0

2

9

SVD Properties

U, V give us orthonormal bases for the subspaces of A: 1st r columns of U: Column space of A Last m - r columns of U: Left nullspace of A 1st r columns of V: Row space of A 1st n - r columns of V: Nullspace of A

IMPLICATION: Rank(A) = r

10

Application: Pseudoinverse

Given y = Ax, x = A+y For square A, A+ = A-1

For any A…

A+ = V-1UT

A+ is called the pseudoinverse of A. x = A+y is the least-squares solution of y = Ax.

11

Rank One Decomposition

Given an m by n matrix A:nm with singular values {s1,...,sr} and SVD A = UVT, define

U = {u1| u2| ... |um} V = {v1| v2| ... |vn}T

Then…

i

r

iiis vuA

1 |

|A may be expressed as the sum of r rank one matrices

12

Matrix Approximation

Let A be an m by n matrix such that Rank(A) = r

If s1 s2 ... sr are the singular values of A, then

B, rank q approximation of A that minimizes ||A - B||F,

is

i

q

iiis vuB

1 |

|Proof: S. J. Leon, Linear Algebra with Applications, 5th Edition, p. 414 [Will]

13

Application: Image Compression

Uncompressed m by n pixel image: m×n numbers Rank q approximation of image:

q singular values The first q columns of U (m-vectors) The first q columns of V (n-vectors) Total: q × (m + n + 1) numbers

14

Example: Yogi (Uncompressed)

Source: [Will] Yogi: Rock photographed

by Sojourner Mars mission.

256 × 264 grayscale bitmap 256 × 264 matrix M

Pixel values [0,1] ~ 67584 numbers

15

Example: Yogi (Compressed)

M has 256 singular values Rank 81 approximation of

M: 81 × (256 + 264 + 1) = ~ 

42201 numbers

16

Example: Yogi (Both)

17

Application: Noise Filtering

Data compression: Image degraded to reduce size Noise Filtering: Lower-rank approximation used to

improve data. Noise effects primarily manifest in terms corresponding

to smaller singular values. Setting these singular values to zero removes noise

effects.

18

Example: Microarrays

Source: [Holter] Expression profiles for

yeast cell cycle data from characteristic nodes (singular values).

14 characteristic nodes Left to right: Microarrays

for 1, 2, 3, 4, 5, all characteristic nodes, respectively.

19

Research Directions

Latent Semantic Indexing [Berry] SVD used to approximate document retrieval matrices.

Pseudoinverse Applications to bioinformatics via Support Vector

Machines and microarrays.

20

References

[Berry]: Michael W. Berry, et. al., “Using Linear Algebra for Intelligent Information Retrieval,” CS 94-270, Department of Computer Science, University of Tennessee, 1994. Submitted to SIAM Review.

[Holter]: Neal S. Holter, et. al., “Fundamental patterns underlying gene expression profiles: Simplicity from complexity,” Proc. Natl. Acad. Sci. USA, 10.1073/pnas. 150242097, 2000 (preprint). Available online at www.pnas.org/doi/10.1073/pnas.150242097

21

References (continued)

[Strang]: Gilbert Strang, Linear Algebra and Its Applications, 3rd edition, Academic Press, Inc., New York, 1988.

[Will]: Todd Will, “Introduction to the Singular Value Decomposition,” Davidson College, http://www.davidson.edu/math/will/svd/index.html

22

Full Presentation Text

http://www.coastal.edu/~jbernick/

Comments Welcome!