44
1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

Embed Size (px)

Citation preview

Page 1: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

1

Pattern Recognition:Statistical and Neural

Lonnie C. Ludeman

Lecture 21

Oct 28, 2005

Nanjing University of Science & Technology

Page 2: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

2

Lecture 21 Topics

1.Example – Analysis of simple Neural Network

2.Example - Synthesis of special forms of Artificial Neural Networks

3. General concepts of Training an Artificial Neural Network- Supervised and unsupervised,training sets

4. Neural Networks Nomenclature and Notation

5. Derivation and Description of the Backpropagation Algorithm for Feedforward Neural Networks

Page 3: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

3

Example: Analyze the following Neural Network

-1

1

-1

110

00

1

Page 4: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

4

Solution: Outputs of layer 1 ANEs

Page 5: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

5

Output of layer 2 ANE is

Thus from layer 1 we have

- 2 ≥ 0 < 0

Page 6: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

6

Page 7: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

7

Final Solution: Output Function for Given Neural Network

Page 8: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

8

Example: Synthesize a Neural Network

Given the following decision regions build a neural network to perform the classification process

Solution: Use Hyperplane-AND-OR structure

Page 9: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

9

Each gk(x) specifies a

hyperplane boundary

Page 10: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

10

Hyperplane Layer AND Layer OR Layer

all f(·) = μ(·)

Solution:

Page 11: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

11

Training a Neural Network

“With a teacher” “Without a teacher”

Page 12: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

12

Page 13: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

13

Training Set

xj are the training samples

dj is the class assigned to training sample xj

Page 14: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

14

Example of a training set:

( x1 = [ 0, 1 ,2 ]T , d1 = C1 ) ,

( x2 = [ 0, 1 ,0 ]T , d2 = C1 ) ,

( x3 = [ 0, 1 ,1 ]T , d3 = C1 ) ,

( x4 = [ 1, 0 ,2 ]T , d4 = C2 ) ,

( x5 = [ 1, 0 ,3 ]T , d5 = C2 ) ,

( x6 = [ 0, 0 ,1 ]T , d6 = C3 ) ,

( x7 = [ 0, 0 ,2 ]T , d7 = C3 )

( x8 = [ 0, 0 ,3 ]T d8 = C3 )

( x9 = [ 0, 0 ,3 ]T d9 = C3 )

( x10 = [ 1, 1 ,0 ]T d10 = C4 )

( x11 = [ 2, 2 ,0 ]T d11 = C4 )

( x12 = [ 2, 2 ,2 ]T d12 = C5 )

( x13 = [ 3, 2, 2 ]T d13 = C6 )

{

}

Page 15: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

15

General Weight Update Algorithm

x(k) is the training sample for the k th iteration

d(k) is the class assigned to training sample x(k) y(k) is the output vector for the k th training sample

Page 16: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

16

Training with a Teacher( Supervised)

1. Given a set of N ordered samples with their known class assignments.

2. Randomly select all weights in the neural network.

3. For each successive sample in the total set of samples, evaluate the output.

4. Use these outputs and the input sample to update the weights

5. Stop at some predetermined number of iterations or if given performance measure is satisfied. If not stopped go to step 3

Page 17: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

17

Training without a Teacher( Unsupervised)

1. Given a set of N ordered samples with unknown class assignments.

2. Randomly select all weights in the neural network.

3. For each successive sample in the total set of samples, evaluate the outputs.

4. Using these outputs and the inputs update the weights

5. If weights do not change significantly stop with that result. If weights change return to step 3

Page 18: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

18

Supervised Training of a Feedforward Neural Network

Nomenclature

Page 19: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

19

Output vector of layer m

Output vector of layer L

Node Number Layer m

Node Number Layer L

1

Page 20: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

20

Weight Matrix for layer m

Node 1 Node 2 Node Nm

N

Nm

Page 21: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

21

fix

Layers, Nets, Outputs, Nonlinearities

Page 22: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

22

Define the performance Ep for sample x(p) as

We wish to select weights so that Ep is

Minimized – Use Gradient Algorithm

Page 23: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

23

Gradient Algorithm for Updating the weights

p w(p)

px(p)

Page 24: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

24

Derivation of weight update equation for Last Layer (Rule #1) Backpropagation Algorihm

The partial of ym(L)

with respect to wkj(L) is

Page 25: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

25

General Rule #1 for Weight Update

Therefore

Page 26: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

26

Derivation of weight update equation for Next to Last Layer (L-1) Backpropagation Algorithm

Page 27: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

27

Page 28: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

28

General Rule #2 for Weight Update- Layer L-1 Backpropagation Algorithm

Therefore

and the weight correction is as follows

Page 29: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

29

where weight correction (general Rule #2) is

w

(L-1)

Page 30: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

30

Backpropagation Training Algorithm for Feedforward Neural networks

Page 31: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

31

Input pattern sample xk

Page 32: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

32

Calculate Outputs First Layer

Page 33: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

33

Calculate Outputs Second Layer

Page 34: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

34

Calculate Outputs Last Layer

Page 35: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

35

Check Performance

ETOTAL(p) ½ (d[x(p-i)] – f( wT(p-i)x(p-i) )2

i = 0

Ns - 1

ETOTAL(p+1) = ETOTAL(p) + Ep+1 (p+1) – Ep-Ns (p-Ns )

Single Sample Error

Over all Samples Error

Can be computed recursively

Page 36: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

36

Change Weights Last Layer using Rule #1

Page 37: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

37

Change Weights previous Layer using Rule #2

Page 38: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

38

Change Weights previous Layer using Modified Rule #2

Page 39: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

39

Input pattern sample xk+1

Continue Iterations Until

Page 40: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

40

Repeat process until performance is satisfied or maximum number of iterations are reached.

If performance not satisfied at maximum number of iterations the algorithm

stops and NO design is obtained.

If performance is satisfied then the current weights and structure provide the

required design.

Page 41: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

41

Freeze Weights to get Acceptable Neural Net Design

Page 42: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

42

Backpropagation Algorithm for Training Feedforward Artificial Neural Networks

Page 43: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

43

Summary Lecture 21

1.Example – Analysis of simple Neural Network

2.Example - Synthesis of special forms of Artificial Neural Networks

3. General concepts of Training an Artificial Neural Network- Supervised and unsupervised,and description of training sets

4. Neural Networks Nomenclature and Notation

5. Derivation and Description of the Backpropagation Algorithm for Feedforward Neural Networks

Page 44: 1 Pattern Recognition: Statistical and Neural Lonnie C. Ludeman Lecture 21 Oct 28, 2005 Nanjing University of Science & Technology

44

End of Lecture 21