Statistics & Data Analysis using Neural

Network

www.techsource.com.my

Statistical & Data Analysis

Using Neural Network

TechSource Systems Sdn. Bhd.

©2005 Systems Sdn. Bhd.

www.techsource.com.my

Course Outline:
1. Neural Network Concepts
a) Introduction
b) Simple neuron model
c) MATLAB representation of neural network

2. Types of Neural Network

a) Perceptrons
b) Linear networks
c) Backpropagation networks
d) Self-organizing maps

3. Case Study: Predicting Time Series

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 1
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

Section Outline:
1. Introduction
ƒ Definition of neural network
ƒ Biological perspective of neural network
ƒ Neural network applications

2. Simple neuron model

ƒ Components of simple neuron

3. MATLAB representation of neural network

ƒ Single neuron model
ƒ Neural network with single-layer of neurons
ƒ Neural network with multiple-layer of neurons

©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

Definition of Neural Network

“A neural network is an interconnected assembly of simple

processing elements, units or nodes, whose functionality is
loosely based on the animal neuron. The processing ability of
the network is stored in the inter-unit connection strengths, or
weights, obtained by a process of adaptation to, or learning
from, a set of training patterns.”

Inputs Outputs

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 2
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

The Biological Perspective of Neural Networks

©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

Neural Network Applications:

ƒ Aerospace ƒ Insurance
ƒ Automotive ƒ Manufacturing
ƒ Banking ƒ Medical
ƒ Credit Card Activity ƒ Oil & Gas
Checking ƒ Robotics
ƒ Defense ƒ Speech
ƒ Electronics ƒ Securities
ƒ Entertainment ƒ Telecommunications
ƒ Financial ƒ Transportation
ƒ Industrial

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 3
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

Components of Simple Neuron

Cell
Dendrites Body Axon

p1
w1
p2 w2
(Inputs) w3 a (Output)
p3
• wR
•
• b Bias
pR
Synapses
1

©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

Components of Simple Neuron

p1 Transfer
w1 Summation Function
p2
w2

Σ ƒ
n a
(Inputs) p3 w3
(Output)
•
• wR
•
b
pR
1

a = f(w1p1+ w2p2 + w3p3 + … + wRpR + b)

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 4
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

Example:
p1
w1

Σ
n a
Inputs w2 Output
p2
b

1
If p1 = 2.5; p2 = 3; w1 = 0.5; w2 = -0.7; b = 0.3. Let’s assume the transfer
function of the neuron is hardlimit, where,

0 for n < 0
a = hardlim(n) =
1 for n ≥ 0

∴ a = hardlim(n) = hardlim( w1p1 + w2p2 + b)

= hardlim( 0.5×2.5 + (-0.7)×3 + 0.3 )
= hardlim( -0.55)
=0
©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

MATLAB® Representation of
the Simple Neuron Model

Input A Single-Neuron Layer Output

Input p = p1
Vector
p a p2
W •
R×1 1×1
•
1×R
n
ƒ •
1×1
pR
1 b R×1
1×1 1
R
Weight
Vector
a = f(Wp + b) W = [w1 w2 • • • wR] 1 × R

R = number of input elements

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 5
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

Single Layer of Neurons

Input Layer 1 with S Neurons Output Where,

• S1: Number of neurons
p a1 in Layer 1
R×1 IW1,1 • IW1,1: Input Weight
S1 × 1
S1 × R n1
f1 matrix for connection
S1 × 1 from Input to Layer 1
1 b1
S1 × 1 S1 iw1,11,1 iw1,11,2 … iw1,11,R
R
iw1,12,1 iw1,12,2 … iw1,12,R
a1 = f1(IW1,1p + b1) IW1,1 =
.
.
.
.
.
.
. . .
iw1,1S,1 iw1,1S,2 … iw1,1S,R
R = number of input elements S×R

©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

Multiple Layer of Neurons

Hidden Layers Output Layer

Input Layer 1 Layer 2 Layer 3 Output

p a1 a2 a3 = y
R×1
IW1,1 LW2,1 LW3,2
n1 S1 × 1 n2 S2 × 1 n3 S3 × 1
S1 × R S2 × S1 S3 × S2
S1 × 1
f1 S2 × 1
f2 S3 × 1
f3
1 b1 1 b2 1 b3
R S1 S2 S3
S1 × 1 S2 × 1 S3 × 1

a1 = f1(IW1,1p + b1) a2 = f2(LW2,1a1 + b2) a3 = f3(LW3,2a2 + b3)

∴ a3 = f3(LW3,2 f2(LW2,1 f1(IW1,1p + b1) + b2) + b3) = y

• S1, S2, S3: Number of neurons in Layer 1, Layer 2, Layer 3 respectively

• IW1,1: Input Weight matrix for connection from Input to Layer 1
• LW2,1: Layer Weight matrix for connection from Layer 1 to Layer 2
• LW3,2: Layer Weight matrix for connection from Layer 2 to Layer 3

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 6
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

Example:
Neural network with 2 layers. 1st layer (hidden layer) consists of 2 neurons
with tangent-sigmoid (tansig) transfer functions; 2nd layer (output layer)
consists of 1 neuron with linear (purelin) transfer function.

Hidden Layer Output Layer

iw1,11,1
Σ
n11 a11
p1 lw2,11,1
iw1,1 2,1 a21
b1 n21
Σ Output
1
Inputs 1
iw1,11,2
b21
Σ
p2 iw1,12,2 n12 a12 lw2,11,2
1
b12
1

©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

In MATLAB® abbreviated notation, the neural network is represented by the

diagram below.
tansig purelin
p a1 a2 = y
2×1 IW1,1 2×1 LW2,1 1×1
2×2 n1 1×2
n2
2×1 1×1
1 b1 1 b2
2×1 1×1
2 2 1

∴ a2 = purelin(LW2,1 tansig(IW1,1p + b1) + b2) = y

p1 iw1,11,1 iw1,11,2
b11 b21
p= IW1,1 = b1 = LW2,1 = lw2,11,1 lw2,11,2 b2 =
p2 b12
iw1,1 2,1 iw1,1 2,2

n11 a11 n21 a21

a1 = n2 = a2 = =y
n1 = n12 a12

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 7
Statistics & Data Analysis using Neural
Network

Neural Network Concepts www.techsource.com.my

tansig(n) = 2 / ((1 + e-2n) -1)

purelin(n) = n
∴ a2 = purelin(LW2,1 tansig(IW1,1p + b1) + b2) =y

For,
0.3 -0.7
1 0.1
p= IW1,1 = b1 = 0.2
LW2,1 = 0.1 -0.2 b2 = 0.3
2 -0.2 0.5

∴y = a2 = purelin(LW2,1 tansig(IW1,1p + b1) + b2)

= purelin([0.1 -0.2] × tansig([0.3 -0.7 ; -0.2 0.5] × [1 ; 2] + [0.1 ; 0.2]) + 0.3)
= purelin([0.1 -0.2] × tansig([-1 ; 1]) + 0.3)
= purelin(0.0715)
= 0.0715

©2005 Systems Sdn. Bhd.

Neural Network Concepts www.techsource.com.my

Section Summary:
1. Introduction
ƒ Definition of neural network
ƒ Biological perspective of neural network
ƒ Neural network applications

2. Simple neuron model

ƒ Components of simple neuron

3. MATLAB representation of neural network

ƒ Single neuron model
ƒ Neural network with single-layer of neurons
ƒ Neural network with multiple-layer of neurons

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 8
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

Section Outline:
1. Perceptrons 4. Self-Organizing Maps
ƒ Introduction ƒ Introduction
ƒ The perceptron architecture ƒ Competitive learning
ƒ Training of perceptrons ƒ Self-organizing maps
ƒ Application examples ƒ Application examples

2. Linear Networks
ƒ Introduction
ƒ Architecture of linear networks
ƒ The Widrow-Hoff learning algorithm
ƒ Application examples

3. Backpropagation Networks
ƒ Introduction
ƒ Architecture of backprogation network
ƒ The backpropagation algorithm
ƒ Training algorithms
ƒ Pre- and post-processing
ƒ Application examples
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

Perceptrons

ƒ Invented in 1957 by Frank Rosenblatt at Cornell Aeronautical

Laboratory.

ƒ The perceptron consists of a single-layer of neurons whose

weights and biases could be trained to produce a correct
target vector when presented with corresponding input vector.

ƒ The output from a single perceptron neuron can only be in one

of the two states. If the weighted sum of its inputs exceeds a
certain threshold, the neuron will fire by outputting 1;
otherwise the neuron will output either 0 or -1, depending on
the transfer function used.

ƒ The perceptron can only solve linearly separable problems.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 9
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

Linearly Separable Problems

ƒ If a straight line can be drawn to separate the input vectors

into two categories, the input vectors are linearly separable, as
illustrated in the diagram below. If need to identify four
categories, we need to use two perceptron neurons.

p2 p2

p1

p1

(a) Two categories of (b) Four categories of

input vectors input vectors

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

The Perceptron Neuron

p1
p2 w1 Summation Hardlimit
w2

Σ
p3 w3 n a Output
Inputs
•
• wR
•
b
pR
1

a = hardlim(w1p1+ w2p2 + w3p3 + … + wRpR + b)

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 10
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

MATLAB® Representation of
the Perceptron Neuron

Input A Single-Neuron Layer Output

Transfer Function
p a a
R×1 W 1×1 +1
1×R
n
1×1
n
1 b 0
1×1 1
R -1
0 for n < 0
a = hardlim(n) =
a = hardlim(Wp + b) 1 for n ≥ 0

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

The Perceptron Architecture

Input The Perceptron Layer Output

Σ
n11 a11
iw1,11,1
Input Single Layer of S Neurons Output
b11
p1 1 p
IW1,1
R×1 n1 a1
S1 × R

Σ
n12 a12
S1 × 1
S1 × 1
p2 1 b1
b12 • • S1
• S1 × 1
• 1 • • R
• • •

Σ
n1S1 a1S1
pR
iw1,1S1,R
b1S1
1

a1 = hardlim(IW1,1p + b1)
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 11
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

Creating a Perceptron:
Command-Line Approach

[-2:2] p1 w1 p
IW1,1
2×1 n1 a1
n a
Σ 1×2
1×1 1×1
[-2:2] p2 w2 b1
1
b 1
1×1
2
1

The example below illustrates how to create a two-input, single-ouput

perceptron. The input values range from -2 to 2.

% Creating a perceptron
>> net = newp([-2 2; -2 2], 1);

% Checking properties and values of Input Weights

>> net.inputWeights{1,1} % properties
>> net.IW{1,1} % values
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

% Checking properties and values of bias

>> net.biases{1} % properties
>> net.b{1} % values

% Note that initial weights and biases are initialized to zeros using “initzero”
>> net.inputWeights{1,1}.initFcn
>> net.biases{1}.initFcn

% To compute the output of perceptron from input vectors [p1; p2], use the
“sim” command
>> p = [ [2; 2] [1; -2] [-2; 2] [-1; 1] ]
>> a = sim(net, p)
>> a =
1 1 1 1

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 12
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

The Perceptron Learning Rule

ƒ Perceptrons are trained on examples of desired behavior, which
can be summarized by a set of input-output pairs

{p1, t1}, {p2, t2},…, {pQ, tQ}

ƒ The objective of training is to reduce the error e, which is the

difference t – a between the perceptron output a, and the target
vector t.

ƒ This is done by adjusting the weights (W) and biases (b) of the
perceptron network according to following equations

Wnew = Wold + ∆W = Wold + epT

bnew = bold + ∆b = bold + e

Where e = t – a
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

Training of Perceptron
ƒ If the Perceptron Learning Rule is used repeatedly to adjust the
weights and biases according to the error e, the perceptron wil
eventually find weight and bias values that solve the problem,
given that the perceptron can solve it.

ƒ Each traverse through all the training vectors is called an epoch.

ƒ The process that carries out such a loop of calculation is called

training.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 13
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

10

Example:
We can train a Perceptron network to classify two groups of data, as
illustrated below

Data x1 x2 Group
x2
p1 -3 -0.5 0
p4 Gr
ou p2 -2 -1.2 0
p
p6 1
p3 p3 -1.5 0.7 1
p8
x1 p4 -1 3 1
p1
p5 -1 -3.5 0
Gr p2
ou p7 p6 0 2 1
p
0 p5
p7 0 -2.5 0
p8 1 0.7 1

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

11

Procedures:
% Load the data points into Workspace
>> load data

% Assign training inputs and targets

>> p = points; % inputs
>> t = group; % targets

% Construct a two-input, single-output perceptron

>> net = newp(minmax(p), 1);

% Train the perceptron network with training inputs (p) and targets (t)
>> net = train(net, p, t)

% Simulate the perceptron network with same inputs again

>> a = sim(net, p)
>> a =
0 0 1 1 0 1 0 1 % correct classification
>> t =
0 0 1 1 0 1 0 1

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 14
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

12

% Let’s be more adventurous by querying the perceptron with inputs it

never seen before
x2
t2
>> t1 = [-2; -3];
p4 Gr
>> t2 = [0.5; 4]; ou
p
p6 1
>> a_t1 = sim(net, t1)
p3
>> a_t1 = p8
x1
0 p1
p2
>> a_t2 = sim(net, t2) Gr
ou p7
p
>> a_t2 = 0 p5
1 t1
∴ The perceptron classifies t1 and t2 correctly.

©2005 Systems Sdn. Bhd.

Types of Neural Network

www.techsource.com.my
13

Using “nntool” GUI

ƒ The “nntool” GUI can be used to create and train different types of neural
network available under MATLAB® Neural Network Toolbox

ƒ The GUI can be invoked by typing at the command window,

>> nntool

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 15
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

14

ƒ First, define the training inputs by clicking “Import…”, select group from
the list of variables. Assign a name to the inputs and indicate that this
variable should be imported as inputs.
ƒ Define the targets similarly.

ƒ Create a new perceptron network by clicking “New Network…”, a new

window appears where network architecture can be defined. Click “Create”
to create the network.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

15

ƒ Next, select the “Train…” tab and set Inputs to p and Targets to t. Click on
“Train Network” to start the training.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 16
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

16

ƒ The network completed training in 4 epochs, which is 4 complete passes

through all training inputs.
ƒ Now, we can test the performance of the trained network by clicking
“Simulate...”. Set the Inputs to p.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

17

Exercise 1: Modeling Logical AND Function

The Boolean AND function has the following truth table:

X Y X AND Y
0 0 0
0 1 0
1 0 0
1 1 1

The problem is linearly-separable, try to build a one-

neuron perceptron network with following inputs and
output:
p1
p1 p2 a
0 0 0
0 1
0 1 0
1 0 0
0 0
1 1 1 p2
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 17
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

18

Solution:
Command-line approach is demonstrated herein. The “nntool” GUI can be used alternatively.

% Define at the MATLAB® command window, the training inputs and targets
>> p = [0 0 1 1; 0 1 0 1]; % training inputs, p = [p1; p2]
>> t = [0 0 0 1]; % targets

% Create the perceptron

>> net = newp([0 1; 0 1], 1);

% Train the perceptron with p and t

>> net = train(net, p, t);

% To test the performance, simulate the perceptron with p

>> a = sim(net, p)
>> a =
0 0 0 1
>> t =
0 0 0 1

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

19

Exercise 2: Pattern Classification

Build a perceptron that can differentiate between two group of images:

Group 0 Group 1 Hint:

Use Boolean values 1’s and 0’s to
0 1 0 represent the image.
1 0 1 Example for image_1 is shown.
1 0 1
0 1 0
∴ image_1 = [ 0 1 0
img1 img2 101
101
0 1 0]’;

Load the training vectors into

workspace:
>> load train_images
img3 img4

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 18
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

20

Try testing the trained perceptron on following images:

timg1 timg2 timg3

>> load test_images

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

21

Solution:
Command-line approach is demonstrated herein. Tne “nntool” GUI can be used alternatively.

% Define at the MATLAB® command window, the training inputs and targets
>> load train_images
>> p = [img1 img2 img3 img4];
>> t = targets;

% Create the perceptron

>> net = newp(minmax(p), 1);

% Training the perceptron

>> net = train(net, p, t);

% Testing the performance of the trained perceptron

>> a = sim(net, p)

% Load the test images and ask the perceptron to classify it

>> load test_images
>> test1 = sim(net, timg1) % to do similarly for other test images

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 19
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

Linear Networks

ƒ Linear networks are similar to perceptron, but their transfer

function is linear rather than hard-limiting.

ƒ Therefore, the output of a linear neuron is not limited to 0 or 1.

ƒ Similar to perceptron, linear network can only solve linearly

separable problems.

ƒ Common applications of linear networks are linear

classification and adaptive filtering.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

The Linear Neuron

p1
p2 w1 Summation Linear
w2

Σ
p3 w3 n a Output
Inputs
•
• wR
•
b
pR
1

a = purelin(w1p1+ w2p2 + w3p3 + … + wRpR + b)

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 20
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

MATLAB® Representation of
the Linear Neuron

Input A Single-Neuron Layer Output

Transfer Function
p a a
R×1 W 1×1 +1
1×R
n
1×1
n
1 b 0
1×1 1
R -1

a = purelin(n) = n
a = purelin(Wp + b)

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

Architecture of Linear Networks

Input Layer of Linear Neurons Output

Σ
n11 a11
iw1,11,1
Input Layer of S1 Linear Neurons Output
b11
p1 1 p
IW1,1
R×1 n1 a1
S1 × R

Σ
n12 a12
S1 ×1
S1 × 1
p2 1 b1
b1 • • S1
• 2
S1 × 1
• 1 • • R
• • •

Σ
n1S1 a1S1
pR
iw1,1S1,R
b1S1
1

a1 = purelin(IW1,1p + b1)
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 21
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

5

Creating a Linear Network:

Command-Line Approach

[-2:2] p1 w1 p
IW1,1
2×1 n1 a1
n a
Σ 1×2
1×1 1×1
[-2:2] p2 w2 b1
1
b 1
1×1
2
1

The example below illustrates how to create a two-input, single-ouput linear

network via command-line approach. The input values range from -2 to 2.

% Creating a linear network

>> net = newlin([-2 2; -2 2], 1);

% Checking properties and values of Input Weights

>> net.inputWeights{1,1} % properties
>> net.IW{1,1} % values
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

6

% Checking properties and values of bias

>> net.biases{1} % properties
>> net.b{1} % values

% Note that initial weights and biases are initialized to zeros using “initzero”
>> net.inputWeights{1,1}.initFcn
>> net.biases{1}.initFcn

% To compute the output of linear network from input vectors [p1; p2], use
the “sim” command
>> p = [ [2; 2] [1; -2] [-2; 2] [-1; 1] ]
>> a = sim(net, p)
>> a =
0 0 0 0

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 22
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

The Widrow-Hoff Learning Algorithm

ƒ Similar to perceptron, the Least Mean Square (LMS) algorithm,
alternatively known as the Widrow-Hoff algorithm, is an example
of supervised training based on a set of training examples.

{p1, t1}, {p2, t2}, …, {pQ, tQ}

ƒ The LMS algorithm adjusts the weights and biases of the linear
networks to minimize the mean square error (MSE)

1 Q 1 Q
∑ e ( k ) = ∑ ( t ( k ) − a ( k ))
2 2
MSE =
Q k =1 Q k =1

ƒ The LMS algorithm adjusts the weights and biases according to

following equations

W(k + 1) = W(k) + 2αe(k)pT(k)

b(k + 1) = b(k) + 2αe(k)
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

Linear Classification (train)

ƒ Linear networks can be trained to perform linear classification
with the function train.

ƒ The train function applies each vector of a set of input vectors

and calculates the network weight and bias increments due to
each of the inputs according to the LMS (Widrow-Hoff)
algorithm.

ƒ The network is then adjusted with the sum of all these

corrections.

ƒ A pass through all input vectors is called an epoch.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 23
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

Example:
Let’s re-visit Exercise 2: Pattern Classification of the Perceptrons.
We can build a Linear Network to perform not only pattern classification but
also association tasks.

Training Images Testing Images

0 1 0
Group 1 0 1
0 1 0 1
0 1 0
img1 img3 timg1 timg2

Group
1

img2 img4 timg3

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

Solution:
Command-line approach is demonstrated herein. Tne “nntool” GUI can be used alternatively.

% Define at the MATLAB® command window, the training inputs and targets
>> load train_images
>> p = [img1 img2 img3 img4];
>> t = targets;

% Create the linear network

>> net = newlin(minmax(p), 1);

% Train the linear network

>> net.trainParam.goal = 10e-5; % training stops if goal achieved
>> net.trainParam.epochs = 500; % training stops if epochs reached
>> net = train(net, p, t);

% Testing the performance of the trained linear network

>> a = sim(net, p)
>> a =
-0.0136 0.9959 0.0137 1.0030

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 24
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

% Comparing actual network output, a, with training targets, t:

>> a =
-0.0136 0.9959 0.0137 1.0030
>> t =
0 1 0 1

∴ The actual network output, a, closely resembles that of target, t. It is

because the output from Linear Network is not straightly 0 or 1, the
output can be a range of values.

% Now, test the Linear Network with 3 images not seen previously
>> load test_images
>> test1 = sim(net, timg1)
>> test1 =
0.2271
>> test2 = sim(net, timg2)
>> test2 =
0.9686
>> test3 = sim(net, timg3)
test3 =
0.8331
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

How should we interpret the network outputs test1, test2 and test3? For that we
need to define a Similarity Measure, S

S = t − test
Where t is the target-group (i.e. 0 or 1) and test is the network output when
presented with test images.

The smaller the S is, the more similar is a test image to a particular group.

∴ timg1 belongs to Group 0 while

Similarity Measure, S timg2 and timg3 belong to Group 1.
test image wrt. Group 0 wrt. Group 1 These results are similar to what we
timg1 0.2271 0.7729 obtained previously using
Perceptron. By using Linear
timg2 0.9686 0.0314 Network we have the added
advantage of knowing how similar
timg3 0.8331 0.1669 is it a test image is to the target
group it belonged.
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 25
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

Exercise 1: Simple Character Recognition

Create a Linear Network that will differentiate between a Letter ‘U’ and Letter
‘T’. The Letter ‘U’ and ‘T’ are represented by a 3×3 matrices:

Group 0 Group 1
T = [1 1 1 U = [1 0 1
010 101
0 1 0]’ 1 1 1]’
>> load train_letters

Test the trained Linear Network with following test images:

U_odd = [1 1 1 T_odd = [1 1 1
101 010
1 1 1]’ 0 1 1]’

>> load test_letters

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

14

Solution:
Command-line approach is demonstrated herein. Tne “nntool” GUI can be used alternatively.

% Define at the MATLAB® command window, the training inputs and targets
>> load train_letters
>> p = [T U];
>> t = targets;

% Create the linear network

>> net = newlin(minmax(p), 1);

% Train the linear network

>> net.trainParam.goal = 10e-5; % training stops if goal achieved
>> net.trainParam.epochs = 500; % training stops if epochs reached
>> net = train(net, p, t);

% Testing the performance of the trained linear network

>> a = sim(net, p)
>> a =
0.0262 0.9796

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 26
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

15

% Comparing actual network output, a, with training targets, t:

>> a =
0.0262 0.9796
>> t =
0 1

% Now, test the Linear Network with odd-shapes of T and U

>> load test_letters
>> test1 = sim(net, T_odd)
>> test1 =
0.2066 % more similar to T
>> test2 = sim(net, U_odd)
>> test2 =
0.8637 % more similar to U

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

1

Backpropagation (BP) Networks

ƒ Backpropagation network was created by generalizing the
Widrow-Hoff learning rule to multiple-layer networks and non-
linear differentiable transfer functions (TFs).

ƒ Backpropagation network with biases, a sigmoid TF layer, and a

linear TF output layer is capable of approximating any function.

ƒ Weights and biases are updated using a variety of gradient

descent algorithms. The gradient is determined by propagating
the computation backwards from output layer to first hidden
layer.

ƒ If properly trained, the backpropagation network is able to

generalize to produce reasonable outputs on inputs it has never
“seen”, as long as the new inputs are similar to the training
inputs.
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 27
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

2

Architecture of Feedforward BP Network

Hidden Layer Output Layer

iw1,11,1 n11
p1
iw1,12,1
Σ lw2,11,1
b1 n21 a21
lw2,12,1
Σ
1
1

p2 n12
Inputs Σ 1
b21 Outputs
b12
1
Σ
n22 a22
• •
• •
• •
b22
n1S1
Σ
lw2,12,S1 1
pR
iw1,1S1,R
b1S1
1

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

3

MATLAB® Representation of
the Feedforward BP Network

tansig purelin
p a1 a2 = y
R×1 IW1,1 S1 ×1 LW2,1 2×1
S1 × R n1 2 × S1
n2
S1 × 1 2×1
1 b1 1 b2
S1 × 1 2×1
R S1 2

∴ a2 = purelin(LW2,1 tansig(IW1,1p + b1) + b2) = y

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 28
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

4

Log-Sigmoid Transfer Functions for BP Networks

a

+1 Linear
logsig(n) = 1 / (1 + exp(-n)) a
n
0 +1

-1
n
0
Tangent-Sigmoid
a -1

+1 purelin(n) = n
tansig(n) = 2/(1+exp(-2*n))-1
n
0

-1
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

5

The Backpropagation Algorithm

ƒ The backpropagation algorithm is used to update the weights
and biases of the neural networks. Full details of the algorithm
is given in Appendix 1.

ƒ The weights are updated according to following formulae:

∂E
wj , i ← wj , i + ∆wj , i where, ∆wj , i = −α = −αδ jxj , i
∂w j , i
For output neuron k,

δ k = − ak ( 1 − ak )( tk − ak )
Please see
For hidden neuron h, Appendix 1 for
full derivation of
δ h = a h ( 1 − ah ) ∑ δ kwk , h the algorithm
k∈ Downstream ( h )

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 29
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

6

Training of Backpropagation Networks

ƒ There are generally four steps in the training process:

1. Assemble the training data;

2. Create the network object;
3. Train the network;
4. Simulate the network response to new inputs.

ƒ The MATLAB® Neural Network Toolbox implements some of

the most popular training algorithms, which encompass both
original gradient-descent and faster training methods.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

7

Batch Gradient Descent Training

ƒ Batch Training: the weights and biases of the network are
updated only after the entire training data has been applied to
the network.

ƒ Batch Gradient Descent (traingd):

• Original but the slowest;

• Weights and biases updated in the direction of the

negative gradient (note: backprop. algorithm);

• Selected by setting trainFcn to traingd:

net = newff(minmax(p), [3 1], {‘tansig’, ‘purelin’}, ‘traingd’);

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 30
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

8

Batch Gradient Descent with Momentum

ƒ Batch Gradient Descent with Momentum (traingdm):

• Faster convergence than traingd;

• Momentum allows the network to respond not only the

local gradient, but also to recent trends in the error
surface;

• Momentum allows the network to ignore small features in

the error surface; without momentum a network may get
stuck in a shallow local minimum.

• Selected by setting trainFcn to traingdm:

net = newff(minmax(p), [3 1], {‘tansig’, ‘purelin’}, ‘traingdm’);

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

9

Faster Training
ƒ The MATLAB® Neural Network Toolbox also implements some
of the faster training methods, in which the training can
converge from ten to one hundred times faster than traingd
and traingdm.

ƒ These faster algorithms fall into two categories:

1. Heuristic techniques: developed from the analysis of the

performance of the standard gradient descent algorithm,
e.g. traingda, traingdx and trainrp.

2. Numerical optimization techniques: make use of the

standard optimization techniques, e.g. conjugate
gradient (traincgf, traincgb, traincgp, trainscg), quasi-
Newton (trainbfg, trainoss), and Levenberg-Marquardt
(trainlm).

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 31
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

10

Comparison of Training Algorithms

Training Algorithms Comments
traingd Gradient Descent (GD) Original but slowest
traingdm GD with momentum Faster than traingd
traingda GD with adaptive α
Faster than traingd, but can use
GD with adaptive α and with for batch mode only.
traingdx
momentum
trainrp Resilient Backpropagation Fast convergence
traincgf Fletcher-Reeves Update
traincgp Polak-Ribiére Update Conjugate Gradient Algorithms
traincgb Powell-Beale Restarts with fast convergence

trainscg Scaled Conjugate Gradient

trainbfg BFGS algorithm Quasi-Newton Algorithms with
trainoss One Step Secant algorithm fast convergence

Fastest training. Memory

trainlm Levenberg-Marquardt
reduction features
trainbr Bayesian regularization Improve generalization capability
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

11

Pre- and Post-Processing Features

Function Description
premnmx Normalize data to fall into range [-1 1].
postmnmx Inverse of premnmx. Convert data back into original range of values.
Preprocess new inputs to networks that were trained with data
tramnmx
normalized with premnmx.
prestd Normalize data to have zero mean and unity standard deviation
poststd Inverse of prestd. Convert data back into original range of values.
Preprocess new inputs to networks that were trained with data
trastd
normalized with prestd.
prepca Principal component analysis. Reduces dimension of input vector
Preprocess new inputs to networks that were trained with data
trapca
transformed with prepca.
Linear regression between network outputs and targets. Use to
postreg
determine adequacy of network fit.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 32
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

12

Example: Modeling Logical XOR Function

The XOR Problem is highly non-linear, thereby cannot be solved using
Perceptrons or Linear Networks. In this example, we will construct a simple
backpropagation network to solve this problem.

X Y X XOR Y
0 0 0 Y
0 1 1
1 0 1 1 0

1 1 0

p1 p2 a
0 0 0 0 1

0 1 0
X
1 0 0
1 1 1
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

13

Solution:
Command-line approach is demonstrated herein. Tne “nntool” GUI can be used alternatively.

% Define at the MATLAB® command window, the training inputs and targets
>> p = [0 0 1 1; 0 1 0 1];
>> t = [0 0 0 1];

% Create the backpropagation network

>> net = newff(minmax(p), [4 1], {‘logsig’, ‘logsig’}, ‘traingdx’);

% Train the backpropagation network

>> net.trainParam.epochs = 500; % training stops if epochs reached
>> net.trainParam.show = 1; % plot the performance function at every epoch
>> net = train(net, p, t);

% Testing the performance of the trained backpropagation network

>> a = sim(net, p)
>> a =
0.0002 0.0011 0.0001 0.9985
>> t =
0 0 0 1
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 33
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

14

Improving Generalization with Early Stopping

ƒ The generalization capability can be improved with the early

stopping feature available with the Neural Network toolbox.

ƒ In this technique the available data is divided into three

subsets:

1. Training set
2. Validation set
3. Testing set

ƒ The early stopping feature can be invoked when using the

train command:

[net, tr] = train(net, p, t, [ ], [ ], VV, TV)

VV: Validation set structure; TV: Test set structure.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

15

Example: Function Approximation with Early Stopping

% Define at the MATLAB® command window, the training inputs and targets
>> p = [-1: 0.05: 1];
>> t = sin(2*pi*p) + 0.1*randn(size(p));

% Construct Validation set

>> val.P = [-0.975: 0.05: 0.975]; % validation set must be in structure form
>> val.T = sin(2*pi*val.P) + 0.1*randn(size(val.P));

% Construct Test set (optional)

>> test.P = [-1.025: 0.05: 1.025]; % validation set must be in structure form
>> test.T = sin(2*pi*test.P) + 0.1*randn(size(test.P));

% Plot and compare three data sets

>> plot(p, t), hold on, plot(val.P, val.T,‘r:*’), hold on, plot(test.P, test.T, ‘k:^’);
>> legend(‘train’, ‘validate’, ‘test’);

% Create a 1-20-1 backpropagation network with ‘trainlm’ algorithm

>> net = newff(minmax(p), [20 1], {‘tansig’, ‘purelin’}, ‘trainlm’);

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 34
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

16

>> net.trainParam.show = 1;
>> net.trainParam.epochs = 300;

% First, train the network without early stopping

>> net = init(net); % initialize the network
>> [net, tr] = train(net, p, t);
>> net1 = net; % network without early stopping

% Then, train the network with early stopping with both Validation & Test sets
>> net = init(net);
>> [net, tr] = train(net, p, t, [], [], val, test);
>> net2 = net; % network with early stopping

% Test the modeling performance of net1 & net2 on Test sets

>> a1 = sim(net1, test.P); % simulate the response of net1
>> a2 = sim(net2, test.P); % simulate the response of net2
>> figure, plot(test.P, test.T), xlim([-1.03 1.03]), hold on
>> plot(test.P, a1, ‘r’), hold on, plot(test.P, a2, ‘k’);
>> legend(‘Target’, ‘Without Early Stopping’, ‘With Early Stopping’);

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

17

∴Network with early stopping can better fit the Test data set with
less discrepancies, therefore the early stopping feature can be
used to prevent overfitting of network towards the training data.
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 35
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

18

Exercise 1: Time-Series Prediction

Create a Neural Network that can predict the next-day 24-hour time-series
based on current-day 24-hour time-series.

d-1 d d+1 d+2

current day next day
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

19

The Neural Network structure is as follow:

data(d, t=1) data(d+1, t=1)

data(d, t=2) data(d+1, t=2)
data(d, t=3) data(d+1, t=3)
data(d, t=4) data(d+1, t=4)
Inputs Backprop. Output
(Current (Next
Day) data(d, t=21) Network data(d+1, t=21)
Day)
data(d, t=22) data(d+1, t=22)
data(d, t=23) data(d+1, t=23)
data(d, t=24) data(d+1, t=24)

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 36
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

20

Network details:
ƒ Architecture: 24-48-24 network, with tansig TF and purelin TF in hidden and
output layer respectively.
ƒ Training: trainlm algorithm with 7 epochs and plot the performance function
every 1 epoch.

Data details:
ƒ Load timeseries.mat into MATLAB® workspace.
ƒ Training Data (1st to 37th days): TrainIp (inputs), TrainTgt (targets)
ƒ Testing Data (38th to 40th days): TestIp (query inputs), TestTgt (actual values)

Hint: use pre- and post-processing functions premnmx, tramnmx, postmnmx to

have more efficient training.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

21

Solution:
% Load the Time Series data into MATLAB® Workspace
>> load timeseries

% Prepare the data for the network training

>> [PN, minp, maxp, TN, mint, maxt] = premnmx(TrainIp, TrainTgt);

% Create the backpropagation network

>> net = newff(minmax(PN), [48 24], {‘tansig’, ‘purelin’}, ‘trainlm’);
>> net.trainParam.epochs = 7;
>> net.trainParam.show = 1;

% Training the neural network

>> [net, tr] = train(net, PN, TN);

% Prepare the data for testing the network (predicting 38th to 40th days)
>> PN_Test = tramnmx(TestIp,minp,maxp);

% Testing the neural network

>> TN_Test = sim(net, PN_Test);
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 37
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

22

% Convert the testing output into prediction values for comparison purpose
>> [queryInputs predictOutputs] = postmnmx(PN_Test, minp, maxp, …
TN_Test, mint, maxt);

% Plot and compare the predicted and actual time series

>> predictedData = reshape(predictOutputs, 1, 72);
>> actualData = reshape(TestTgt, 1, 72);
>> plot(actualData, ‘-*’), hold on
>> plot(predictOutputs, ‘r:’);

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

23

Homework: Try to subdivide the training data [TrainIp TrainTgt] into Training
& Validation Sets to accertain whether the use of early stopping would
improve the prediction accuracy.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 38
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

24

Exercise 2: Character Recognition

Create a Neural Network that can recognize 26 letters
of the alphabet. An imaging system that digitizes
each letter centered in the system field of vision is
available. The result is each letter represented as a 7
by 5 grid of boolean values. For example, here are the
Letters A, G and W:

A
A G
W P
Q Z

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

25

¾ For each alphabetical letter, create a 35-by-1 input vector containing

boolean values of 1’s and 0’s. Example for Letter A:

0 0 1 0 0 Letter A = [0 0 1 0 0 … Target A = [1 0 0 0 0 …
0 1 0 1 0 01010… 00000…
01010… 00000…
0 1 0 1 0 00000…
10001…
1 0 0 0 1 11111… 00000…
10001… 0]’;
1 1 1 1 1
1 0 0 0 1]’;
1 0 0 0 1
1 0 0 0 1

¾ Corresponding Output is a 26-by-1 vector, indicating a 1 (i.e. TRUE) at

the correct alphabetical sequence. Example for Target A:

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 39
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

26

The Neural Network structure is as follow:

0 1 0 A B C
1 1 0
1 1 1 1 0 0
Inputs 1 1 0 0 1 0 Outputs
Neural
0 0 1
(Alphabets) (Targets)
Network
1 1 0
0 0 0 0
1 1
0 0 0 0
1 1
0 0 1

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

27

Network details:
ƒ Architecture: 35-10-26 network, with logsig TFs in hidden and output layers.
ƒ Training: traingdx algorithm with 500 epochs and plot the performance
function every 1 epoch. Performance goal is 0.001.

Data details:
ƒ Load training inputs and targets into workspace by typing
[alphabets, targets] = prprob;

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 40
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

28
Solution:
% Load the training data into MATLAB® Workspace
>> [alphabets, targets] = prprob;

% Create the backpropagation network

>> net = newff(minmax(alphabets), [10 26], {‘logsig’, ‘logsig’}, ‘traingdx’);
>> net.trainParam.epochs = 500;
>> net.trainParam.show = 1;
>> net.trainParam.goal = 0.001;

% Training the neural network

>> [net, tr] = train(net, alphabets, targets);

% First, we create a normal ‘J’ to test the network performance

>> J = alphabets(:,10);
>> figure, plotchar(J);
>> output = sim(net, J);
>> output = compet(output) % change the largest values to 1, the rest 0s
>> answer = find(compet(output) == 1); % find the index (out of 26) of network
output
>> figure, plotchar(alphabets(:, answer));
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

29

% Next, we create a noisy ‘J’ to test the network can still identify it correctly…
>> noisyJ = alphabets(:,10)+randn(35,1)*0.2;
>> figure; plotchar(noisyJ);
>> output2 = sim(network1, noisyJ);
>> output2 = compet(output2);
>> answer2 = find(compet(output2) == 1);
>> figure; plotchar(alphabets(:,answer2));

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 41
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

1

Self-Organizing Maps
ƒ Self-organizing in networks is one of the most fascinating
topics in the neural network field. Such networks can learn to
detect regularities and correlations in their input and adapt their
future responses to that input accordingly.

ƒ The neurons of competitive networks learn to recognize groups

of similar input vectors. Self-organizing maps learn to recognize
groups of similar input vectors in such a way that neurons
physically near each other in the neuron layer respond to
similar input vectors.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

2

Competitive Learning

Input Competitive Layer Output

S1 × R IW1,1

a1
p S1 × 1
|| ndist || S1 × 1
R×1 n1
S1 × 1
C
1 b1
R S1 × 1
S1

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 42
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

3

Learning Algorithms for Competitive Network

ƒ The weights of the winning neuron (represented by a row of the
input weight matrix) are adjusted with the Kohonen learning rule
(learnk).

ƒ Supposed that the ith neuron wins, the elements of the ith row of
the input weight matrix are adjusted according to following
formula:

iIW
1,1(q) = iIW1,1(q-1) + α(p(q) – iIW1,1(q-1))

ƒ One of the limitations of the competitive networks is that some

neurons may never wins because their weights are far from any
input vectors.

ƒ The bias learning rule (learncon) is used to allow every neuron in

the network learning from the input vectors.
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

4

Example: Classification using Competitive Network

We can use a competitive network to
classify input vectors without any
learning targets.
Let’s say if we create four two-element
input vectors, with two very close to (0 p(2)
0) and others close to (1 1).

p = [0.1 0.8 0.1 0.9

0.2 0.9 0.1 0.8];

p(1)
Let’s see whether the competitive
network is able to identify the
classification structure…

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 43
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

5

% Define at the MATLAB® command window, the four two-element vectors

>> p = [0.1 0.8 0.1 0.9; 0.2 0.9 0.1 0.8];

% Create a two-neuron competitive network

>> net = newc([0 1; 0 1], 2);

% The weights are initialized to the center of input ranges with ‘midpoint’ fcn
>>net.IW{1,1}
ans =
0.5 0.5
0.5 0.5

% The biases are computed by ‘initcon’ fcn, which gives

>> net.b{1}
ans =
5.4366
5.4366

% Let’s train the competitive network for 500 epochs

>> net.trainParam.epochs = 500;
>> net = train(net, p);
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

6

% Simulate the network with input vectors again

>> a = sim(net, p)
>> ac = vec2ind(a)
>> ac =
2 1 2 1

∴The network is able to classfy the input vectors into two classess, those
who close to (1,1), class 1 and those close to origin (0,0), class 2. If we look
at the adjusted weights,
>> net.IW{1,1}
ans =
0.8500 0.8500
0.1000 0.1501

∴Note that the first-row weight vector (associated with 1st neuron) is near to
input vectors close to (1,1), which the second-row weight vector (associated
with 2nd neuron) is near to input vectors close to (0,0).

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 44
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

7

Exercise 1: Classification of Input Vectors:

Graphical Example
First, generate the input vectors by using the built-in nngenc function:
>> X = [0 1; 0 1]; % Cluster centers to be in these bounds
>> clusters = 8; % Number of clusters
>> points = 10; % Number of points in each cluster
>> std_dev = 0.05; % Standard deviation of each cluster
>> P = nngenc(X,clusters,points,std_dev); % Number of clusters

Plot and show the generated clusters

>> plot(P(1,:),P(2,:),'+r');
>> title('Input Vectors');
>> xlabel('p(1)');
>> ylabel('p(2)');

Try to build a competitive network with 8 neurons and train for 1000
epochs. Superimpose the trained network weights onto the same figure.
Try to experiement with the number of neurons and conclude on the
accuracy of the classification.
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

8

Solution:
% Create and train the competitive network
>> net = newc([0 1; 0 1], 8, 0.1); % Learning rate is set to 0.1
>> net.trainParam.epochs = 1000;
>> net = train(net, P);

% Plot and compare the input vectors and cluster centres determined by the
competitive network
>> w = net.IW{1,1};
>> figure, plot(P(1,:),P(2,:),‘+r’);
>> hold on, plot(w(:,1), w(:,2), ‘ob’);

% Simulate the trained network to new inputs

>> t1 = [0.1; 0.1], t2 = [0.35; 0.4], t3 = [0.8; 0.2];
>> a1 = sim(net, [t1 t2 t3]);
>> ac1 = vec2ind(a1);
ac1 =
1 5 6
Homework: Try altering the number of neurons in the competitive layer and
observe how it affects the cluster centres.
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 45
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

9

Self-Organizing Maps
ƒ Similar to competitive neural networks, self-organizing maps
(SOMs) can learn the distribution of the input vectors. The
distinction between these two networks is that the SOM can also
learn the topology of the input vectors.

ƒ However, instead of updating the weight of the winning neuron i*,

all neurons within a certain neighborhood Ni*(d) of the winning
neuron are also updated using the Kohonen learning learnsom, as
follows:
iw(q) = iw(q – 1) + α(p(q) – iw(q – 1))

ƒ The neighborhood Ni*(d) contains the indices for all the neurons
that lie within a radius d of the winning neuron i*.
Ni(d) = {j, dij ≤ d}

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

10

Topologies & Distance Functions

ƒ Three different types of topology can be specified for the initial
location of the neurons:
1. Rectangular grid: gridtop
2. Hexagonal grid: hextop
3. Random grid: randtop
ƒ For example, to create a 5-by-7 hexagonal grid,
>> pos = hextop(5,7);
>> plotsom(pos);

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 46
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

11

ƒ Similarly, there are four different ways of calculating the distance

from a particular neuron to its neighbors:
1. Euclidean distance: dist
2. Box distance: boxdist
3. Link distance: linkdist
4. Manhattan distance: mandist

ƒ For example, with the 2-by-3 rectangular grid shown below,

5 6
>> d = boxdist(pos)
d=
3 0 1 1 1 2 2
4
1 0 1 1 2 2
1 1 0 1 1 1
1 1 1 0 1 1
1 2 2 2 1 1 0 1
2 2 1 1 1 0

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

12

The SOM Architecture

Input Self-Organizing Map Layer Output

S1 × R IW1,1

p n1 a1
R×1
|| ndist ||
S1 × 1
C S1 × 1

S1
R

ni1 = -|| iIW1,1 – p ||

a1 = compet(n1)

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 47
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

13

Creating and Training the SOM

ƒ Let’s load an input vector into MATLAB® workspace
>> load somdata
>> plot(P(1,:), P(2,:), ‘g.’, ‘markersize’, 20), hold on

ƒ Create a 2-by-3 SOM with following command, and superimpose

the initial weights onto the input space
>> net = newsom([0 2; 0 1], [2 3]);
>> plotsom(net.iw{1,1}, net.layers{1}.distances), hold off

ƒ The weights of the SOM are updated using the learnsom function,
where the winning neuron’s weights are updated proportional to α
and the weights of neurons in its neighbourhood are altered
proportional to ½ of α.

©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

14

ƒ The training is divided into two phases:

1. Ordering phase: The neighborhood distance starts as the
maximum distance between two neurons, and decreases to
the tuning neighborhood distance. The learning rate starts at
the ordering-phase learning rate and decreases until it
reaches the tuning-phase learning rate. This phase typically
allows the SOM to learn the topology of the input space.

2. Tuning Phase: The neighborhood distance stays at the

tuning neighborhood distance (i.e., typically 1.0). The
learning rate continues to decrease from the tuning phase
learning rate, but very slowly. The small neighborhood and
slowly decreasing learning rate allows the SOM to learn the
distribution of the input space. The number of epochs for
this phase should be much larger than the number of steps
in the ordering phase.

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 48
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

15

ƒ The learning parameters for both phases of training are,

>> net.inputWeights{1,1}.learnParam
ans =
order_lr: 0.9000
order_steps: 1000
tune_lr: 0.0200
tune_nd: 1

ƒ Train the SOM for 1000 epochs with

>> net.trainParam.epochs = 1000;
>> net = train(net, P);

ƒ Superimpose the trained network structure onto the input space

>> plot(P(1,:), P(2,:), ‘g.’, ‘markersize’, 20), hold on
>> plotsom(net.iw{1,1}, net.layers{1}.distances), hold off

Try alter the size of the SOM and learning parameters and draw
conclusion on how it affects the result.
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

16

Exercise 2: Mapping of Input Space

In this exercise we will test whether the SOM can map out the topology and
distribution of an input space containing three clusters illustrated in the
figure below,

©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 49
Statistics & Data Analysis using Neural
Network

Types of Neural Network www.techsource.com.my

17
Solution:
% Let’s start by creating the data for the input space illustrated previously
>> d1 = randn(3,100); % cluster center at (0, 0, 0)
>> d2 = randn(3, 100) + 3; % cluster center at (3, 3, 3)
>> d3 = randn(3,100), d3(1,:) = d3(1,:) + 9; % cluster center at (9, 0, 0)
>> d = [d1 d2 d3];

% Plot and show the generated clusters

>> plot3(d(1,:), d(2,:), d(3,:), ‘ko’), hold on, box on

% Try to build a 10-by-10 SOM and train it for 1000 epochs,

>> net = newsom(minmax(d), [10 10]);
>> net.trainParam.epochs = 1000;
>> net = train(net, d);

% Superimpose the trained SOM’s weights onto the input space,

>> gcf, plotsom(net.IW{1,1}, net.layers{1}.distances), hold off

% Simulate SOM to get indices of neurons closest to input vectors

>> A = sim(net, d), A = vec2ind(A);
©2005 Systems Sdn. Bhd.

Types of Neural Network www.techsource.com.my

Section Summary:
1. Perceptrons 4. Self-Organizing Maps
ƒ Introduction ƒ Introduction
ƒ The perceptron architecture ƒ Competitive learning
ƒ Training of perceptrons ƒ Self-organizing maps
ƒ Application examples ƒ Application examples

2. Linear Networks
ƒ Introduction
ƒ Architecture of linear networks
ƒ The Widrow-Hoff learning algorithm
ƒ Application examples

3. Backpropagation Networks
ƒ Introduction
ƒ Architecture of backprogation network
ƒ The backpropagation algorithm
ƒ Training algorithms
ƒ Pre- and post-processing
ƒ Application examples
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 50
Statistics & Data Analysis using Neural
Network

Case Study www.techsource.com.my

Predicting the Future Time Series Data

The demand for electricity (in MW) varies according to seasonal changes and
weekday-weekend work cycle. How do we develop a neural-network based
Decision-Support System to forecast the next-day hourly demand?

Electricity Demand in Different Seasons Electricity Demand: Weekdays Vs Weekends

9000 9500
Summer
8500 9000
Winter Weekday

8000 8500
Demand (MW)

7500 8000
Weekends Sat

Demand (MW)
7000 Spring 7500

6500 7000 Sun

Autumn
6000 6500

5500 6000

5000 5500

4500 5000
0 5 10 15 20 25 30 35 40 45 50 0 5 10 15 20 25 30 35 40 45 50
Time in half-hourly records Time in half-hourly records

Note: NSW electricity demand data (1996 – 1998) courtesy of NEMMCO, Australia
©2005 Systems Sdn. Bhd.

Case Study www.techsource.com.my

Solution:
Step 1: Formulating Inputs and Outputs of Neural Network
By analysing the time-series data, a 3-input and 1-output neural network is
proposed to predict next-day hourly electricity demand,

Inputs: Output:
p1 = L(d,t) a1 = L(d+1, t)
p2 = L(d,t) - L(d-1,t)
p3 = Lm(d+1,t) - Lm(d,t)

Where,
L(d, t): Electricity demand for day, d, and hour, t
L(d+1, t): Electricity demand for next day, (d+1), and hour, t
L(d-1, t): Electricity demand for previous day, (d-1), and hour t
Lm(a, b) = ½ [ L(a-k, b) + L(a-2k, b)]
k = 5 for Weekdays Model & k = 2 for Weekends Model
©2005 Systems Sdn. Bhd.

Copyrighted 2005 TechSource Systems

Sdn Bhd 51
Statistics & Data Analysis using Neural
Network

Case Study www.techsource.com.my

Step 2: Pre-process Time Series Data to Appropriate Format

The time-series data is in MS Excel format and was date- and time-tagged. We
need to preprocess the data according to following steps:
1. Read from the MS Excel file [histDataIp.m]
2. Divide the data into weekdays and weekends [divideDay.m]
3. Remove any outliers from the data [outlierRemove.m]

Step 3: Constructing Inputs and Output Data for Neural Network

Arrange the processed data in accordance to neural-network format:
1. Construct Input-Output pair [nextDayIp.m]
2. Normalizing the training data for faster learning [processIp.m]

Step 4: Training & Testing the Neural Network

Train the neural network using command-line or NNTOOL. When training is
completed, proceed to test the robustness of the network against “unseen”
data.

©2005 Systems Sdn. Bhd.

www.techsource.com.my

The End
Kindly return your Evaluation Form

Tel: 603 – 8076 1953 Fax: 603 – 8076 1954

Email: [email protected] Web: www.techsource.com.my
©2005 Systems Sdn. Bhd.
Tech-
Tech-Support: [email protected]

Copyrighted 2005 TechSource Systems

Sdn Bhd 52