School of Information Technology

Indian Institute of Technology Kharagpur

IT60108: Soft Computing and Application
Class Test - I
F.M. 20 Session 2014 − 2015 Time: 20 mins

Q. 1

(a) µA (x) and µB (x) are the membership functions of the fuzzy sets A and B, respectively.
1
µA (x) = e 1+x
1
µB (x) = 1+( x−50 )4
10

Decide whether A and B are closed or open.

Answer : A membership function µ(x) is said to be closed iff limx→−∞ µ( x) =0= limx→α µ(x)
1
Case 1 : µA (x) = e 1+n : Here limx→−∞ µA (x) = limx→+∞ µA (x) = 1
Hence, it is neither closed nor open.
1
Case 2 : µB (x) = 1+( x−50 )4
: Here limx→−∞ µB (x) = limx→+∞ µA (x) = 0
10
Hence, it is closed.
[2+2]

(b) Given two fuzzy sets A and B defined over universe of discourses X and Y , respectively.
A = {(20, 0.2), (25, 0.4), (30, 0.6), (35, 0.6), (40, 0.7), (45, 0.8),
(50, 0.8)}
B = {(1, 0.8), (2, 0.8), (3, 0.6), (4, 0.4)}
X = {10, 15, 20, 25, 30, 35, 40, 45, 50, 55}
Y = {0, 1, 2, 3, 4, 5}
Draw the graphs for the following.

i. A × B
ii. A =⇒ B

(i) A × B

1 2 3 4
 
20 0.2 0.2 0.2 0.2
25 
 0.4 0.4 0.4 0.4 

30 
 0.6 0.6 0.6 0.4 

µA×B (x, y) = min{µA (x), µB (y)} = 35 
 0.6 0.6 0.6 0.4 

40 
 0.7 0.7 0.6 0.4 

45  0.8 0.8 0.6 0.4 
50 0.8 0.8 0.6 0.4

(ii) A ⇒ B
For this many interpretation are possible.
A ⇒ B ≡ Ā ∪ B or A × B or (A × B) ∪ (Ā × Y )
Accordingly answer will be different.
[4+4]
Q. 2

(a) Suppose, a fuzzy relation is ‘If x is A then y is B’. How to find the following:

i. x is C, given that y is D
ii. y is D, given that x is C

Answer :
GPM
if x is A then y is B
x is A0
− − − − − − − − − − − − − − −−
y is B 0
GMT is
if x is A then y is B
y is B 0
− − − − − − − − − − − − − − −−
x is A0
Since C and D are not mentioned as A0 or B 0 and (vice-versa) none of the GMP and GMT are
applicable in this case and hence we can not deduce anything.
[3+3]

(b) Two fuzzy sets P and Q are defined on x∈ X as follows.

x1 x2 x3 x4 x5
P 0.1 0.2 0.7 0.5 0.4
Q 0.9 0.6 0.3 0.2 0.8

Find (i.) (P ∩ Q)0.4 (ii.) (P × Q)0.4

Answer :
(i)
P ∩ Q = (x1 , 0.1), (x2 , 0.2), (x3 , 0.7), (x4 , 0.5), (x5 , 0.2)

2
∴ (P ∩ Q)0.4 = {x|µ(x) ≥ 0.4} = {x3 , x4 }
(ii)
x1 x2 x3 x4 x5 x1 x2 x3 x4 x5
   
x1 0.1 0.1 0.1 0.1 0.1 x1 0 0 0 0 0
x2 
 0.2 0.2 0.2 0.2 0.2 
 x2 
 0 0 0 0 0 

P ×Q= x3 
 0.7 0.6 0.3 0.2 0.7 , (P × Q)0.4 =
 x3 
 1 1 0 1 1 

x4  0.5 0.5 0.3 0.2 0.4  x4  1 1 0 0 1 
x5 0.4 0.4 0.3 0.2 0.4 x5 1 1 0 0 1
[3+3]

3
Q. 3

(a) The membership functions of two fuzzy sets A and B are shown in the following graph.
A: climate is Hot.
B: climate is Cold.

1.0
B A

-15 -10 -5 0 5 10 15 20 25 30 35 40

i. Draw the graph of the membership function, which represents the fuzzy set C: climate is
Extreme.
Answer : Climate is Extreme ≈ Climate is Hot OR Climate is Cold.

1.0  A B

-15 -10 -5 0 5 10 15 20 25 30 35 40

ii. What would be the graph of the membership function µD of the fuzzy set D = (A ∩ C)?
State D in terms of fuzzy linguistic.
Answer :
D
1.0

B A

-15 -10 -5 0 5 10 15 20 25 30 35 40
¯ B = Climate is not Pleasant
Lingustic interpretation : A ∩ B = Climate is Pleasant, D = A ∩
[3+3]
(b) Two fuzzy relations ‘likes’ and ‘earns’ are defined below.
F ootball Hockey Cricket
 
Dhoni 0.1 0.3 0.8
V irat  0.2 0.7 0.5 
likes =  
Rohit  0.5 0.4 0.2 
Sekhar 0.4 0.5 0.6
For example, x likes Game.
10L 50L 100L
 
Dhoni 0.6 0.3 0.2
V irat  0.4 0.7 0.8 
earns = 
 0.1

Rohit 0.3 0.2 
Sekhar 0.5 0.2 0.6

4
For example, x earns Money.
Obtain the relation between a game to a money? [6]
Answer :
This relation can be obtained as likesT ◦ earns. That is,

10L 50L 100L

Dhoni V irat Rohit Sekhan  
  Dhoni 0.6 0.3 0.2
F ootboll 0.1 0.2 0.5 0.4  0.4
 0.3 ◦ V irat 0.7 0.8 
Hockey 0.7 0.4 0.5 
 0.1
=
Rohit 0.3 0.2 
Cricket 0.8 0.5 0.2 0.6
Sekhan 0.5 0.2 0.6
10L 50L 100L
 
F ootball 0.4 0.3 0.4
Hockey  0.5 0.7 0.7 
Circket 0.6 0.5 0.6

5
Q. 4

(a) What are the components you should consider in order to mathematically model an artificial
neuron ?
[4]
Answer : Two components namely summation until and threshold unit are required to mathe-
matically model as artificial neuron. Two components can be defined as follows.

x1 Ø(I)
w1

x2 w2

w3
I y
x3
…..

wn

xn

Summation Threshold unit output

input weight
unit

Pn
Summation unit I = i=1 xi · wi . Threshold unit = φ(I), where φ is some transfer function.
1
(b) If φ(I) = 1+e−αI
is a transfer function in a perceptron, then show that
Answer :
1−φ(I)
φ(I) = 1
1+e−αI
; Let z = 1 + e−αI . ∴ ∂z
∂I = −α · e−αI = −α · φ(I)
∂φ(I) ∂φ(I) ∂z
∂I = ∂z · ∂I

1−φ(I)
= − z12 · −α · φ(I)
1−φ(I) 1
= φ(I)2 · α · φ(I) ,∵ z = φ(I)
= α(1 − φ(I)) · φ(I) [3]

(c) Draw a schematic diagram of a multi-layer feed-forward artificial neural network architecture and
clearly label the different elements in it.
Give one application, where you should apply such an ANN architecture.
Answer :

Application : Such an ANN would be applied to problems whose output are non-separable with
resepct to input. [4+1]

6
 Io O
I OI [V] IH OH [W]
21 31

v11

......
wj1

......
1 11
I

.
1
1

vi
v1i

......
vl1
wjk
vij
I i1 1i
v1m
2j 3k
......

vlj

......
vlm

......
wjn

.
1 1l
I l

vlm

2m 3n

OI IH OH Io

I N1 N2 N3 O

[V] [W]
Input layer Hidden Layer Output Layer

|N2|=m |N3|=n
|N1|=l

Linear transfer Log-Sigmoid transfer Tan-Sigmoid transfer

function, ɵ function function

f jm  ( I Hj ,  j ) 
1 e   o I o  e  o I o
fi l  ( I i , i )  j I Hj f kn  ( I ko ,  k ) 
1 e e   o I o  e  o I o

Q. 5

(a) Show how the computations in input, hidden and output layers of an ANN can be accomplished
in terms of matrix algebra.
[2+2+2]
Answer :
Whole learning method consists of the following three computations:

(a) Input layer computation

(b) Hidden layer computation
(c) Output layer computation

In our computation, we assume that < T0 , TI > be the training set of size |T |.

• Let us consider an input training data at any instant be I I = [I11 , I21 , · · · , Ii1 , Il1 ] where I I ∈ TI
• Consider the outputs of the neurons lying on input layer are the same with the corresponding
inputs to neurons in hidden layer. That is,
OI = I I
[l × 1] = [l × 1] [Output of the input layer]
• The input of the j-th neuron in the hidden layer can be calculated as follows.

IjH = v1j oI1 + v2j oI2 +, · · · , +vij oIj + · · · + vij oIl

7
 where j = 1, 2, · · · m.
[Calculation of input of each node in the hidden layer]
• In the matrix representation form, we can write
I H = V T · OI
[m × 1] = [m × l] [l × 1]

• Let us consider any j-th neuron in the hidden layer.

• Since the output of the input layer’s neurons are the input to the j-th neuron and the j-th
neurons follows the log-sigmoid transfer function, we have
1
OjH = −α ·I H
1+e H j

where j = 1, 2, · · · , m and αH is the constant co-efficient of the transfer function.

Note that all output of the nodes in the hidden layer can be expressed as a one-dimensional column
matrix.
 
···

 ··· 

 .. 

 . 

1
OH =  −αH ·IjH 
 
1+e
..
 
 

 . 

 ··· 
··· m×1

Let us calculate the input to any k-th node in the output layer. Since, output of all nodes in the
hidden layer go to the k-th layer with weights w1k , w2k , · · · , wmk , we have

IkO = w1k · oH H H
1 + w2k · o2 + · · · + wmk · om

where k = 1, 2, · · · , n
In the matrix representation, we have

I O = W T · OH
[n × 1] = [n × m] [m × 1]

Now, we estimate the output of the k-th neuron in the output layer. We consider the tan-sigmoid
transfer function.
o o
eαo ·Ik −e−αo ·Ik
Ok = α ·I o −α ·I o
e o k +e o k

for k = 1, 2, · · · , n
Hence, the output of output layer’s neurons can be represented as
 
···

 ··· 

 .. 
 α ·I o . −α ·I o 
 
 e o k −e o k 
O =  αo ·I o −αo ·I o 
 e k +e k
 .. 

 . 

 ··· 
··· n×1

8
(b) Explain the basic principle of calculating error in supervised learning.
[2]

Answer :

• Let us consider any k-th neuron at the output layer. For an input pattern Ii ∈ TI (input in
training) the target output TOk of the k-th neuron be TOk .
• Then, the error ek of the k-th neuron is defined corresponding to the input Ii as

ek = 1
2 (TOk − OOk )2

where OOk denotes the observed output of the k-th neuron.

• For a training session with Ii ∈ TI , the error in prediction considering all output neurons can
be given as
Pn 1 Pn
e= k=1 ek = 2 k=1 (TOk − OOk )

where n denotes the number of neurons at the output layer.

• The total error in prediction for all output neurons can be determined considering all training
session < TI , TO > as
Pn
1
− OOk )2
P P
E= ∀Ii ∈TI e= 2 ∀t∈<TI ,TO > k=1 (TOk

(c) Derive the ‘delta rule’ according to the method of Steepest descent.

• For simplicity, let us consider the connecting weights are the only design parameter.
• Suppose, V and W are the wights parameters to hidden and output layers, respectively.
• Thus, given a training set of size N , the error surface, E can be represented as

E= N i
P
i=1 e (V, W, Ii )

where Ii is the i-th input pattern in the training set and ei (...) denotes the error computation
of the i-th input.

• Now, we will discuss the steepest descent method of computing error, given a changes in V
and W matrices.
~
• Suppose, A and B are two points on the error surface (see figure in Slide 30). The vector AB
can be written as

~ = (Vi+1 − Vi ) · x̄ + (Wi+1 − Wi ) · ȳ = ∆V · x̄ + ∆W · ȳ
AB
~ can be obtained as
The gradient of AB

∂E ∂E
~ =
eAB ∂V · x̄ + ∂W · ȳ

Hence, the unit vector in the direction of gradient is

1
 ∂E ∂E

ēAB
~ = |eAB · x̄ + · ȳ
~ | ∂V ∂W

• With this, we can alternatively represent the distance vector AB as

~ =η
 ∂E ∂E

AB ∂V · x̄ + ∂W · ȳ

9
 k
where η = |eAB and k is a constant
~ |

• So, comparing both, we have

∂E
∆V = η ∂V
∂E
∆W = η ∂W

This is also called as delta rule and η is called learning rate.

[2+2]

10