Artificial Neural Network

Chapter 8:Self Organizing Map

 Introduction
The main property of a neural network is an
ability to learn from its environment, and to
improve its performance through learning. So
far we have considered supervised or active
learning - learning with an external “teacher”
or a supervisor who presents a training set to the
network. But another type of learning also
exists: unsupervised learning.

2
 In contrast to supervised learning, unsupervised or
self-organised learning does not require an
external teacher. During the training session, the
neural network receives a number of different
input patterns, discovers significant features in
these patterns and learns how to classify input data
into appropriate categories. Unsupervised
learning tends to follow the neuro-biological
organisation of the brain.
n Unsupervised learning algorithms aim to learn
rapidly and can be used in real-time.

3
 Hebbian learning
In 1949, Donald Hebb proposed one of the key
ideas in biological learning, commonly known as
Hebb’s Law. Hebb’s Law states that if neuron i is
near enough to excite neuron j and repeatedly
participates in its activation, the synaptic connection
between these two neurons is strengthened and
neuron j becomes more sensitive to stimuli from
neuron i.

4
Hebb’s Law can be represented in the form of two
rules:
1. If two neurons on either side of a connection
are activated synchronously, then the weight of
that connection is increased.
2. If two neurons on either side of a connection
are activated asynchronously, then the weight
of that connection is decreased.
Hebb’s Law provides the basis for learning
without a teacher. Learning here is a local
phenomenon occurring without feedback from
the environment.
5
Hebbian learning in a neural network

i j

6
Using Hebb’s Law we can express the adjustment
applied to the weight w ijat iteration p in the
following form:
i
As a special case, we can represent Hebb’s Law as
follows:

where α is the learning rate parameter.

This equation is referred to as the activity product
rule.

7
 • Hebbian learning implies that weights can only
increase. To resolve this problem, we might
impose a limit on the growth of synaptic weights.
It can be done by introducing a non-linear
forgetting factor into Hebb’s Law:

• Forgetting factor usually falls in the interval

between 0 and 1, typically between 0.01 and 0.1,
to allow only a little “forgetting” while limiting
the weight growth.

Ó Negnevitsky, Pearson Education, 2005 8

 Hebbian learning algorithm
Step 1: Initialisation.
Set initial synaptic weights and thresholds to small
random values, say in an interval [0, 1].

Step 2: Activation.
Compute the neuron output at iteration p

where n is the number of neuron inputs, and Ɵj is the

threshold value of neuron j.

Ó Negnevitsky, Pearson Education, 2005 9

 Step 3: Learning.
Update the weights in the network:
ij ij ij
where Dwij(p) is the weight correction at iteration p.
The weight correction is determined by the
generalised activity product rule:

ij j i ij
Step 4: Iteration.
Increase iteration p by one, go back to Step 2.

Ó Negnevitsky, Pearson Education, 2005 10

 Hebbian learning example
To illustrate Hebbian learning, consider a fully
connected feedforward network with a single layer
of five computation neurons. Each neuron is
represented by a McCulloch and Pitts model with
the sign activation function. The network is trained
on the following set of input vectors:
é0ù é0ù é0ù é0ù é0ù
ê0ú ê1ú ê0ú ê0ú ê1ú
êú êú êú êú êú
X1= ê0ú X = 2ê0ú X = ê0ú
3 X = ê1ú X
4 = ê0ú 5
êú êú êú êú êú
êú0 êú0 êú1 ê0ú ê0ú
êë0ú
û êë1ú
û êë0ú
û êë0ú
û êë1ú
û

Ó Negnevitsky, Pearson Education, 2005 11

 Initial and final states of the network
x1 1 1
1 y1 x1 1 1
0 y1
1 1

x2 0 2 2 0 y2 x2 0 2 2 1 y2

x3 0 3 3 0 y3 x3 0 3 3 0 y3

x4 0 4 4 0 y4 x4 0 4 4 0 y4

x5 1 5 5 1 y5 x5 1 5 5 1 y5
Input layer Output layer Input layer Output layer

Ó Negnevitsky, Pearson Education, 2005 12

 Initial and final weight matrices

Outputlayer Outputlayer
1 23 4 5 1 2 3 4 5
1 é1 0 0 0 0ù 1 é0 0 0 0 0ù
ê0 1 0 0 0 ú ê
2 0 2.0204 0
ú
2
ê ú ê 0 2.0204 ú
3 ê0 0 1 0 0ú 3 0ê 0 1.0200 0 0ú
ê ú ê ú
4 00010 4 ê0 0 0 0.9996 0 ú
ê ú
5 0ê
ë0 0 0 1 úû 5 0ê
ë2.0204 0 0 2.0204 úû

Ó Negnevitsky, Pearson Education, 2005 13

 n A test input vector, or probe, is defined as
é1ù
ê0ú
êú
X = ê0ú
êú
ê0ú
êë1úû

n When this probe is presented to the network, we

obtain:
ì 0é 0 0 0 ùé ù é 01
ïê úê ú ê úï 0.4940
êú ü
ïê 0 2.0204 0 0 2.0204 0 0.2661 1
ï úê ú ê úïï ê0ú
Y = sign ê0
í 0 1.0200 0 0 úê0ú- ê0.0907ú = ýê0ú ùéù
ï 0 ê úê ú ê ú ï ú0
ê
0 0 0.9996 0 0 0.9478
ïê úê ú ê úï ê ú
ïî êë0 2.0204 0 úê
ûë úû êë ûï 1ë û úþê ú
0 2.0204 1 0.0737

Ó Negnevitsky, Pearson Education, 2005 14

 Competitive learning
• In competitive learning, neurons compete among
themselves to be activated.
• While in Hebbian learning, several output neurons
can be activated simultaneously, in competitive
learning, only a single output neuron is active at
any time.
• The output neuron that wins the “competition” is
called the winner-takes-all neuron.

Ó Negnevitsky, Pearson Education, 2005 15

 n The basic idea of competitive learning was
introduced in the early 1970s.

In the late 1980s, Teuvo Kohonen introduced a

special class of artificial neural networks called
self-organising feature maps. These maps are
based on competitive learning.

Ó Negnevitsky, Pearson Education, 2005 16

 What is a self-organising feature map?
Our brain is dominated by the cerebral cortex, a
very complex structure of billions of neurons and
hundreds of billions of synapses. The cortex
includes areas that are responsible for different
human activities (motor, visual, auditory,
somatosensory, etc.), and associated with different
sensory inputs. We can say that each sensory
input is mapped into a corresponding area of the
cerebral cortex. The cortex is a self-organising
computational map in the human brain.

Ó Negnevitsky, Pearson Education, 2005 17

 Feature-mapping Kohonen model
Kohonen layer Kohonen layer

Input layer Input layer

10 01
(a) (b)
Ó Negnevitsky, Pearson Education, 2005 18
 The Kohonen network
n The Kohonen model provides a topological
mapping. It places a fixed number of input
patterns from the input layer into a higher-
dimensional output or Kohonen layer.
n Training in the Kohonen network begins with the
winner’s neighbourhood of a fairly large size.
Then, as training proceeds, the neighbourhood size
gradually decreases.

Ó Negnevitsky, Pearson Education, 2005 19

 Architecture of the Kohonen Network

y1
x1
y2
x2
y3

Input Output
layer layer

Ó Negnevitsky, Pearson Education, 2005 20

 n The lateral connections are used to create a
competition between neurons. The neuron with
the largest activation level among all neurons in
the output layer becomes the winner. This neuron
is the only neuron that produces an output signal.
The activity of all other neurons is suppressed in
the competition.
n The lateral feedback connections produce
excitatory or inhibitory effects, depending on the
distance from the winning neuron. This is
achieved by the use of a Mexican hat function
which describes synaptic weights between neurons
in the Kohonen layer.
Ó Negnevitsky, Pearson Education, 2005 21
 The Mexican hat function of lateral connection

Connection
1 strength
Excitatory
effect

0 D istance
Inhibitory Inhibitory
effect effect

Ó Negnevitsky, Pearson Education, 2005 22

 n In the Kohonen network, a neuron learns by
shifting its weights from inactive connections to
active ones. Only the winning neuron and its
neighbourhood are allowed to learn. If a neuron
does not respond to a given input pattern, then
learning cannot occur in that particular neuron.
n The competitive learning rule defines the change
Dwij applied to synaptic weight w ijas
ì α(x - wi ), ifijneuron j wins the competitio n
Dwij= í
î 0, if neuron j loses the competitio n
where xi is the input signal and α is the learning
rate parameter.
Ó Negnevitsky, Pearson Education, 2005 23
 n The overall effect of the competitive learning rule
resides in moving the synaptic weight vector W jof
the winning neuron j towards the input pattern X.
The matching criterion is equivalent to the
minimum Euclidean distance between vectors.
n The Euclidean distance between a pair of n-by-1
vectors X and Wj is defined by
1/ 2
én 2ù
d = X - W =j (x -êåw) i ij ú
êëi=1 ú
û
where xi and w ijare the ith elements of the vectors
X and Wj, respectively.
Ó Negnevitsky, Pearson Education, 2005 24
 n To identify the winning neuron, j ,Xthat best
matches the input vector X, we may apply the
following condition:
jX= min X - W , j j = 1, 2, . . ., m
j
where m is the number of neurons in the Kohonen
layer.

Ó Negnevitsky, Pearson Education, 2005 25

 n Suppose, for instance, that the 2-dimensional input
vector X is presented to the three-neuron Kohonen
network,
é0.52ù
X=êú !
ë0.12
û
n The initial weight vectors, W ,j are given by

é0.27
ù é0.42
ù é0.43
ù
W1= ê ú W2= ê ú W3= ê ú
ë0.81
û ë0.70
û ë0.21
û

Ó Negnevitsky, Pearson Education, 2005 26

 n We find the winning (best-matching) neuron jX
using the minimum-distance Euclidean criterion:
2 2 2 2
d1= (x - w1 ) + (x
11 - w ) =2(0.5221- 0.27) + (0.12 - 0.81) = 0.73

2 2 2 2
d 2= (x - w1 ) + (x
12- w ) = (0.52
2 - 0.42) + (0.12 - 0.70) = 0.59
22

2 2 2 2
d 3= (x - w1 ) + (x
13 - w ) = (0.52
2 -
23 0.43) + (0.12 - 0.21) = 0.13

n Neuron 3 is the winner and its weight vector W is

3
updated according to the competitive learning rule.
Dw13= α (x - w
1 ) = 13
0.1 (0.52 - 0.43) = 0.01

Dw23
=α (x - w2) = 0.1
23 (0.12 - 0 .21) = - 0.01

Ó Negnevitsky, Pearson Education, 2005 27

 n The updated weight vector W at
3 iteration (p + 1)
is determined as:

é0.43ù é 0.01ù é0.44ù

W3( p +1) = W3( p) + DW3 (p) = ê ú ê + ú =ê ú
0.21
ë û ë -û0.01
ë û 0.20

n The weight vector W of

3 the wining neuron 3
becomes closer to the input vector X with each
iteration.

Ó Negnevitsky, Pearson Education, 2005 28

 Competitive Learning Algorithm
Step 1: Initialisation.
Set initial synaptic weights to small random
values, say in an interval [0, 1], and assign a small
positive value to the learning rate parameter a .

Ó Negnevitsky, Pearson Education, 2005 29

 Step 2: Activation and Similarity Matching.
Activate the Kohonen network by applying the
input vector X, and find the winner-takes-all (best
matching) neuron j Xat iteration p, using the
minimum-distance Euclidean criterion
1/2
ìï n üï
jXp(min
) =X W p - j ( ) = í[ å- ] 2
xi w (p)
ij ý ,
j ïî i=1 þï
j = 1, 2, . . ., m
where n is the number of neurons in the input
layer, and m is the number of neurons in the
Kohonen layer.
Ó Negnevitsky, Pearson Education, 2005 30
 Step 3: Learning.
Update the synaptic weights
wij( p +1) = w ij( p) + Dw ij( p)
where ΔΔi(p)
ij is the weight correction at iteration p.
The weight correction is determined by the
competitive learning rule:
ìï α [x iw- (p )]ij, j Î L (j ) p
Dwij( p) = í
ïî 0, j Ï L (j p)

where a is the learning rate parameter, and L (p)

j is
the neighbourhood function centred around the
winner-takes-all neuron j Xat iteration p.
Ó Negnevitsky, Pearson Education, 2005 31
 Step 4: Iteration.
Increase iteration p by one, go back to Step 2 and
continue until the minimum-distance Euclidean
criterion is satisfied, or no noticeable changes
occur in the feature map.

Ó Negnevitsky, Pearson Education, 2005 32

 Competitive learning in the Kohonen network
n To illustrate competitive learning, consider the
Kohonen network with 100 neurons arranged in the
form of a two-dimensional lattice with 10 rows and
10 columns. The network is required to classify
two-dimensional input vectors - each neuron in the
network should respond only to the input vectors
occurring in its region.
n The network is trained with 1000 two-dimensional
input vectors generated randomly in a square
region in the interval between –1 and +1. The
learning rate parameter a is equal to 0.1.
Ó Negnevitsky, Pearson Education, 2005 33
 Initial random weights
1

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
W(1,j)

Ó Negnevitsky, Pearson Education, 2005 34

 Network after 100 iterations
1

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
W(1,j)

Ó Negnevitsky, Pearson Education, 2005 35

 Network after 1000 iterations
1

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
W(1,j)

Ó Negnevitsky, Pearson Education, 2005 36

 Network after 10,000 iterations
1

0.8

0.6

0.4

0.2

-0.2

-0.4

-0.6

-0.8

-1
-1 -0.8 -0.6 -0.4 -0.2 0 0.2 0.4 0.6 0.8 1
W(1,j)
Ó Negnevitsky, Pearson Education, 2005 37