M. Tech /Even/(13-14) / Regular.

2013-14
KNOWLEDGE BASED SYSTEMS
ME 9029
Time: 3Hours

Full Marks: 70
28/04/2014

Date of exam:

Group A
Answer any three questions.
Q1. Figure 1 shows the schematic diagram of an NN having two inputs and one output. The
neurons lying on the input, hidden and output layers have the transfer functions represented by y
1
1
, y=
. The initial values of weights are assumed to be as follows:
= x, y =
x
1+ e
1 + e x
0.4
0.2 0.3 0.2

0.6 0.3 0.2 and 0.3 .

0.4
Using an incremental mode of training for the data given in Table 1, calculate changes in the
weights. Consider back-propagation learning method with learning rate = 0.4. Show only one
iteration.

v11

w11

I1
1

v12

v21
w21

v13 v22

O1
1

I2
2

w31
v23

Input Layer

3
Hidden Layer

Output Layer

Figure 1: Schematic diagram of a neural network.

Table 1: Training Case.

Sl. No.
1

I1
0.6

I2
0.4

O1
0.9
[13]

Q2. Minimize f ( x1 , x 2 ) = x1 + x 2 2 x 12 x 22 + x1 x 2 in the range of 0 x1 , x 2 5 using steepest

0.0
descent method. Take the initial solution X 1 = .
0.0
[13]
32
Q3. Use a binary-coded GA to minimize f ( x) = 2 + x in the range of 0 x 10 . Use a random
x
population of size N = 4, a single point crossover with probability = 1.0 and mutation with
probability = 0.01. Show only one iteration by a hand calculation.
[13]
Q4. There are two inputs (I1 and I2) and one output (O) of a process. It is required to develop a
fuzzy logic-based expert system based on Mamdani Approach. Let us assume that the inputs and
output are expressed using three linguistic terms, namely Low (LW), Medium (M) and High (H).
The membership function distributions of the above inputs and output are shown in Fig. 2. The
rule base of the fuzzy logic controller is shown in Table 2. Determine the output of the controller
for a set of inputs: I1 = 1.6 and I2 = 22.0. Use centroid method of defuzzification.

1.0

0.0

1.0

2
I1

0.0

LW

1.0

10

20

30

0.0

I2

10

15

Figure 2: Data Base of the fuzzy Logic Controller.

Table 2: Rule Base of the Fuzzy Logic Controller.
I2
LW M
H
LW LW LW M
I1
M
LW M
H
H
M
H
H
[13]

Q5. An expert system based on Takagi and Sugenos approach of FLC is to be developed to
predict the output of a process. The membership function distributions of the input and output
variables are shown in Fig. 3. As there are two inputs: I1 and I2 and each input is represented
using three linguistic terms (such as LW, M and H), there is a maximum of 3 X 3 = 9 feasible
rules. The output of ith rule, that is, yi (i = 1, 2, . 9) is expressed as follows:

y i = f (I 1 , I 2 ) = a ij I 1 + bki I 2 ,
Where j,k = 1,2,3; a1i = 1.1, a 2i = 1.2, and a3i = 1.3 if I1 is found to be LW, M and H, respectively;
b1i = 1.5, b2i = 1.7, and b3i = 1.9 if I2 is considered to be LW, M and H, respectively. Calculate the

output of the FLC for the inputs: I1 = 8.0 and I2 = 2.5.

1.0

1.0

0.0

10

0.0

15

I1

I2

Figure 3: Data Base of the fuzzy Logic Controller.

[14]
GROUP B
Answer any five questions. All are of equal marks (5x6 = 30).
Q6. (a) In a Genetic Algorithm, suppose that two potential parents are given by
1

Assuming the numbering goes from left to right and that two crossover sites are at 4th and
8th locations, show the result of two-point crossover.
(b) To solve an optimization problem using a real coded GA, determine the mutated value
corresponding to an original solution Proriginal = 19.68 using random mutation. Assume random
number r = 0.7 and the maximum perturbation = 2.0 .

Q7. To solve an optimization problem using a real coded GA, let us assume that a mating pair
(consisting of two solutions) is found to be as follows:
Pr1 = 16.85, Pr2 = 19.50
Determine the children solutions using different crossover operators given below.
(a) Linear crossover
(b) Simulated Binary Crossover (SBX) assuming the probability distributions for the
contracting and expanding zones as follows:
C ( ) = 0.5(q + 1) q
1
Ex( ) = 0.5(q + 1) q + 2

Where is the spread factor and q = 4 . Assume the random number r = 0.7 .

Q8. Let us consider a Traveling Salesman Problem (TSP) involving eight cities A, B, C, D, E, F,
G, H. A scheduling GA with various types of crossover operator is to be used to solve the said
optimization problem. Determine children solutions of the following two parents:
Pr1:
Pr2:

A
C

B
A

C
D

D
B

E
F

F
H

G
E

H
G

(a) Use edge recombination. The edge table is shown in Table 4.

(b) Order Crossover, assuming 2nd and 5th sites as the crossover sites.
(c) Partially Mapped Crossover (PMX) assuming 2nd and 5th sites as the crossover sites.

City
A
B
C
D
E
F
G
H

Table 4: Edge Table.

Connectivity information/Links
D,F,H,G,B
A,G,E,C,H
B,D,F
A,C,G,E
B,D,F,H
A,C,E
A,B,D,H
A,B,E,G

Q9. Explain briefly the principles of Elman and Jordon Networks.

Q10. Discuss briefly the principle of a genetic-fuzzy system with a suitable example.
Q11.

(a) Is GA-NN is preferred to the BPNN always? Explain.

(b) Do you prefer a radial basis function network to a conventional multi-layer feedforward network? Explain it.

Q12. Explain the principle of Mamdani-type neuro-fuzzy system.

Q13. An initial population of size N= 6 of a binary-coded GA is created at random as shown
below, while optimizing a function.
1
0
0
1
0
0
0
1
0
1
0
0
0
1
0
1
1
1
0
1
0
1
1
0
1
1
0
0
1
1
0
1
1
0
0
1
The fitness of a GA-string is assumed to be equal to its decoded value. Calculate the expected
number of strings to be represented by the schema H: *1**1*, at the end of first generation,
considering a single-point crossover of probability pc = 0.9 and a bit-wise mutation of probability
pm = 0.01. Make comments on the result.
--xxx

SAMPLE QUESTIONS
1. Explain briefly the role of different operators, namely reproduction, crossover and mutation
used in a GA.
2. The performance of a GA depends on the balance between selection pressure and population
diversity justify the statement.
3. Are uniform crossover with probability of 0.5 and bitwise mutation with probability of 0.5 the
same? Explain it?
4. Explain Elitist strategy used in GA.
5. Why do we prefer ranking selection to a Roulette-Wheel selection in GA?
6. Why do we use a high value of crossover probability and a low value of mutation probability
in GA-applications?
7. Do you prefer a Gray-coded GA to a Binary-coded GA? Explain it.
8. Explain the building block hypothesis of a binary coded GA.
9. Can you declare GA a global optimizer? Explain it.
10. State the advantages and disadvantages of a GA.
11. Use a binary coded GA to minimize the function f ( x1 , x 2 ) = x1 + x 2 2 x 12 x 22 + x1 x 2 in the
range of 0 x1 , x 2 5 . Use a random population of size N = 6, crossover with probability = 1.0
and mutation with probability = 0.01. Assume 3 bits for each variable and thus, the GA-string
will be 6-bits long. Show only one iteration by a hand calculation.
12. An initial population of size N=10 of a binary coded GA is created at random as shown
below, while optimizing a function.
100100 010111 010110 011001 110011
010100 111000 100110 100101 011001
The fitness of a GA string is assumed to be equal to its decoded value. Calculate the expected
number of strings to be represented by the schema H: *1**1*, at the end of first generation,
considering a single point crossover of probability pc = 0.9 and a bit-wise mutation of
probability pm = 0.01. Make comments on the results.
13. Why doe we need a special type of crossover operator for solving a scheduling problem?
Explain briefly different crossover operators used in the scheduling GA.
14. To solve an optimization problem using a real coded GA, let us assume that a mating pair
consisting of two parent solutions is found to be as follows:
Pr1 = 16.85, Pr2 = 19.50
Determine the children solution using different crossover operators given below.
(a) Linear crossover,
(b) Blend crossover with = 0.6, that is, BLX-0.6, assume the random number r = 0.7.
(c) Simulated Binary Crossover (SBX) assuming the probability distributions for the contracting
and expanding zones as follows:

C ( ) = 0.5(q + 1) q
Ex( ) = 0.5(q + 1)

1 ,

q+2

Where is the spread factor and q = 4. Assume the random number r = 0.7.
15. To solve an optimization problem utilizing a real-coded GA, determine the mutated value
corresponding to an original solution Proriginal = 19.68 under the following conditions:
(a) Use random mutation, assuming random number r = 0.6 and the maximum perturbation =
2.0
(b) Use polynomial mutation, assuming the random number r = 0.6.
16. Let us consider a Traveling Salesman Problem (TSP) involving eight cities A, B, C, D, E, F,
G, H. A scheduling GA with various types of crossover operator is to be used to solve the said
optimization problem. Determine children solutions of the following two parents:
Pr1:
Pr2:
(a)
(b)
(c)
(d)
(e)
(f)

A
C

B
A

C
D

D
B

E
F

F
H

G
E

H
G

Use edge recombination. The edge table is shown in Table 1.

Order Crossover #1, assuming 2nd and 5th sites as the crossover sites.
Order Crossover #2, considering 3rd, 4th and 7th as the key positions.
Cycle crossover assuming 4-th as starting position.
Position-based Crossover considering 2nd, 4th, and 6th as the crossover points.
Partially Mapped Crossover (PMX) assuming 2nd and 5th sites as the crossover sites.

City
A
B
C
D
E
F
G
H

Table 4: Edge Table.

Connectivity information/Links
D,F,H,G,B
A,G,E,C,H
B,D,F
A,C,G,E
B,D,F,H
A,C,E
A,B,D,H
A,B,E,G

17. Explain briefly the principle of Mamdani Approach of fuzzy logic controller. State its
advantages and disadvantages.
18. Discuss briefly various methods of defuzzification,
19. Explain the principle of Takagi and Sugenos Approach of FLC.
20. When and why do we go for hierarchical FLC?
21. There are two inputs (I1 and I2) and one output (O) of a process. It is required to develop a
fuzzy logic-based expert system based on Mamdani Approach. Let us assume that the inputs and
output are expressed using three linguistic terms, namely Low (LW), Medium (M) and High (H).
The membership function distributions of the above inputs and output are shown in Fig. 2. The

rule base of the fuzzy logic controller is shown in Table 2. Determine the output of the controller
for a set of inputs: I1 = 1.6 and I2 = 22.0. Use centroid method of defuzzification.

1.0

0.0

1.0

2
I1

0.0

LW

1.0

10

20

30

0.0

I2

10

15

Figure 2: Data Base of the fuzzy Logic Controller.

Table 2: Rule Base of the Fuzzy Logic Controller.
I2
LW M
H
LW LW LW M
I1
M
LW M
H
H
M
H
H
22. An expert system based on Takagi and Sugenos approach of FLC is to be developed to
predict the output of a process. The membership function distributions of the input and output
variables are shown in Fig. 3. As there are two inputs: I1 and I2 and each input is represented
using three linguistic terms (such as LW, M and H), there is a maximum of 3 X 3 = 9 feasible
rules. The output of ith rule, that is, yi (i = 1, 2, . 9) is expressed as follows:

y i = f (I 1 , I 2 ) = a ij I 1 + bki I 2 ,
Where j,k = 1,2,3; a1i = 1.1, a 2i = 1.2, and a3i = 1.3 if I1 is found to be LW, M and H, respectively;
b1i = 1.5, b2i = 1.7, and b3i = 1.9 if I2 is considered to be LW, M and H, respectively. Calculate the

output of the FLC for the inputs: I1 = 8.0 and I2 = 2.5.

1.0

0.0

1.0

10

15

0.0

I1

I2

Figure 3: Data Base of the fuzzy Logic Controller.

23. What do you mean by a feed forward neural network? How does it differ from a recurrent
neural network?
24. Explain the principle of a back propagation algorithm. State its advantages and
disadvantages.
25. Do you prefer a radial basis function network to a conventional multi-layer feed-forward
network? Explain it.
26. Explain briefly the principles of Elman and Jordon Networks.
27. Figure 1 shows the schematic diagram of an NN having two inputs and one output. The
neurons lying on the input, hidden and output layers have the transfer functions represented by y
1
1
= x, y =
, y=
. The initial values of weights are assumed to be as follows:
x
1+ e
1 + e x
0 .4
0.2 0.3 0.2

0.6 0.3 0.2 and 0.3 .

0.4
Using an incremental mode of training for the data given in Table 1, calculate changes in the
weights. Consider back-propagation learning method with learning rate = 0.4. Show only one
iteration.
w11
v11

I1
1

v12

v21
w21
v13

v22

O1
1

I2
2
w31
v23

Input Layer

3
Hidden Layer

Output Layer

Figure 1: Schematic diagram of a neural network.

Table 1: Training Case.

Sl. No.
1

I1
0.6

I2
0.4

O1
0.9

28. Solve the previous question with four neurons in the hidden layer. The initial values of
weights are assumed to be as follows:

0.4
0.2 0.3 0.2 0.4

0.6 0.3 0.5 0.2 and 0.3 .
0.4

0.2
29. Discuss briefly the principle of a genetic-fuzzy system with a suitable example.
30. There are two inputs I1 and I2 and one output O of a Mamdani Type Fuzzy Logic Controller.
Both the DB and RB of the FLC is optimized using a binary coded GA simultaneously. Training
scenario is mentioned below.
Table 1: Training Case.
Sl. No. I1

I2

18.0

9.0

1.6

Ranges of the variables are also given below.

1.0 I 1 4.5
10.0 I 2 40
5.0 O 21.0

Generate randomly a binary string and determine the deviation in prediction.

31. How do you design a genetic-neural system? Explain with an example.
32. Is GA-NN preferred to the BPNN always? Explain.
33. A binary-coded GA is used to optimize the connecting weights of a partially-connected NN shown
below.
w11

v11

I1
1

w21

v12

O1
1

v22
w31
I2

v23
Output Layer

Hidden Layer

Input Layer

Figure 2: Schematic diagram of a neural network.

The neurons of input, hidden and output layers are assumed to have transfer function of the forms: y = x2,
e x ex
1
,
. The weights [V] and [W] vary in the range of 0.0 and 1.0. One of the training
y
=
e x + e x
1 + e x
cases is shown below.
Table 1: Training Case.
y=

Sl. No. I1

I2

0.7

0.8

0.6

Determine the error in prediction for the said training case, corresponding to a GA-string shown below, in
which four bits have been assigned to represent each real variable.

1010
{ 0111
{ 1100
{ 0101
{ 1111
{ 1001
{ 0111
{
v11

v12

v 22

v 23

w11

w 21

w 31

34. Explain the principle of a neuro-fuzzy system.

11