Exam 2020

1.
a) Using your own intuition and your own definitions of the Universe of
Discourse, plot fuzzy membership functions for the age of people:
(i) Very young
(ii) Young
(iii) Middle-aged
(iv) Old
(v) Very old
b) The results of a price survey for 30 vehicles is presented here (unit price
in £1,000):
Economy 5.5 5.8 7.5 7.9 8.2 8.5 9.2 10.4 11.2 13.5
Mid Size 11.9 12.5 13.2 14.9 15.6 17.8 18.2 19.5 20.5 24.0
Luxury 22.0 23.5 25.0 26.0 27.5 29.0 32.0 37.0 43.0 47.5
Consider the automobile prices as a variable and the classes as economy,

midsize, and luxury automobiles. Develop three membership function
envelopes for car prices using the method of inductive reasoning.
c) A constant force, F, acts on a body with mass, m, moving on a smooth

surface at velocity, v. The effective power (EP) of this force will be
EP = F × v × cos θ, where θ is the angle between the force direction and
the horizon.
i) Partition the input variable, θ into 5 labels within the range of ±180°,
and the output variable, EP, into 3 labels in the range of ±1. Plot the
membership functions of both.
ii) For the following rules:

IF θ is Zero THEN EP is Most Efficient
IF θ Negative Small OR θ is Positive Small THEN EP is Not Efficient
IF θ Positive Big OR θ is Negative Big THEN EP is Negative Most
Efficient (braking)
Conduct a graphical inference and estimate the results of EP for an
input of θ = 33 degree.
2.
a) The ANFIS system is used to model a two input XOR function. The ANFIS
structure with 2 rules is shown in Figure Q2.
R1
R2
Figure Q2
The input layer (layer 1) is described by a Gaussian membership function
represented by the following equation:
−(𝑥𝑥 − 𝑐𝑐)2
𝑦𝑦 = exp ( )
2𝜎𝜎 2
The weights of the input layer are listed in Table Q2 a)
A1 A2 B1 B2
σ c σ c σ c σ c
0.2345 -0.0597 0.2383 1.059 100 1 100 1
Table Q2 a)
The system is trained and the following rule base is generated:

R1: IF x is A1 AND y is B1 THEN f(x,y) = a1 x + b1 y + c1
R2: IF x is A2 AND y is B2 THEN f(x,y) = a2 x + b2 y + c2
The weight of the fourth layer are listed in Table Q2 b)

a b c
R1 1.894E-5 1 -2.541E-5
R2 0.5 -1 0.5
Table Q2 b)
i) Evaluate the ANFIS system output for the following inputs:
x = 0.1, y = 0.9
x = 0.2, y = 0.1
(Continued on next page)

ii) Calculate the error (actual – predicted) for the two input data cases
listed in part i) and comment on the results.
b) Let A = {0.5/-7, 0.9/-4, 0.6/0, 0.2/4, 0.1/7} be a fuzzy set defined in the
Universe of Discourse X. Let B = {0.25/-3, 0.45/1, 0.75/0, 0.8/1, 0.5/3}
be another fuzzy set defined in the Universe of Discourse Y. Find the
Fuzzy set C defined in the Universe of Discourse Z such that:
Z = f(X,Y) = X ×Y, where × is the multiplication sign.
3. Develop a simple genetic algorithm that minimizes the function
𝑓𝑓(𝑥𝑥1 , 𝑥𝑥2 ) = 100 × (𝑥𝑥12 − 𝑥𝑥2 )2 + (1 − 𝑥𝑥1 )2. Use a binary representation of
integers. Use one-point crossover, bit-mutation and tournament selection
(choose the most suitable the crossover and mutation rates). Minimize 𝑓𝑓(𝑥𝑥) in
the range of 0 ≤ 𝑥𝑥𝑖𝑖 ≤ 1023.
a) Identify the phenotype and genotype.
b) Describe one-point crossover for chosen chromosome representation.

Then describe the bit-mutation.
c) Write a pseudo-code of the simple genetic algorithm.
d) Explain how the genetic algorithm will be changed if elitism is used.

4.
a) Name and describe the main features of components of fuzzy system.
b) Describe how ANFIS work by giving details of each layer in the system
with the aid of a block diagram.
5.
a) Describe how Euclidean Distance can be used as a strategy for classifying
patterns in the form of vectors. Illustrate your answer with a 2-
dimensional example showing TWO reference patterns, each one
representing a class of data and a third test pattern and its classification.
b) Suggest a simplification to the Euclidean distance classifier which results

in an approximate distance which is faster to calculate and show the
contour of equidistance (in 2 dimensions) of 1 unit from the reference
pattern X = (2, 2)
Two distinct data classes are represented by the 5-dimensional vectors
A = (7, 3 4, 6, 2)
and B = (3, 6, 1, 2, 7)
determine whether pattern
X = (4, 4, 3, 6 1)
belongs to class A or Class B, using
i) Euclidean distance
ii) The approximate distance you have used in part b) above.
c) Show that under certain circumstances your approximate distance

decision surface in 2 dimensions, which should be a straight line, becomes
an area, and what effect does this have on the performance of the
classifier?
6.
a) Justify the use of a logic function as an approximate model for a natural
neuron and state the principal advantage during training, of using a logic
function compared with the analogue vector model found in the
McCulloch and Pitts neuron model.
b) A Digital Neural Network contains THREE logic functions and each one
has THREE inputs. Each function is connected to a different row of a 3 by
3 pixel input space.
If the network is trained on the following patterns;
XXX .XX . . X
. . . X. . . X .
. . . . . . x . .
determine the equations of the three logic functions after training.
c) What is the size of the AND generalisation set (threshold =3) and the OR
generalisation set (threshold = 1)?
d) If the patterns;
. . X . X X
XX . . X .
X . . X . .
are considered to belong to the same class as the training set, what
output threshold would be appropriate for the network to recognise
them?
7.
a) Define a 3 by 3 window operator that can be used to LOW pass filter a
grey-level image and state what correction factor needs to be applied in
order to display the filtered image
b) Apply the LOW pass filter to the following image.
0 1 2 3 12
1 2 3 12 14
2 3 12 14 15
3 12 13 15 15
12 13 15 15 15
c) The MEDIAN filter is an alternative low pass filter. Describe its operation
and apply it to the image in part b) above.
d) If low pass filtering is to be used to remove “spot” noise for example a

single white pixel surrounded by black pixels, apply the LOW pass and
MEDIAN filters to the following image sample;
0 0 0
0 15 0
0 0 0
and determine the resulting output pixel for both filters. In this application
which gives the better result?
8.
a) Define the single pixel operators for Intensity Shift and Intensity
Multiply giving details of any correction factors needed to keep the
processed image within the intensity range O to W.
b) How may Intensity Shift and Intensity Multiply be combined to

NORMALISE a grey level image?
Apply Image NORMALISATION to the following 4-bit grey-level image
5 5 7 9
5 6 10 10
6 7 11 11
7 10 10 12
Present your solution as a 3 by 3 pixel grey level enhanced image.
c) How may the Normalisation strategy be modified to enhance the darker

50% of an image?
Apply the modified Normalisation process to the image in part b) in order

to enhance the darker 50% of the image.

Exam 2020

Uploaded by

Copyright:

Available Formats

Exam 2020

Uploaded by

Document Information

Original Description:

Copyright

Available Formats

Share this document

Share or Embed Document

Sharing Options

Did you find this document useful?

Is this content inappropriate?

Copyright:

Available Formats

Exam 2020

Uploaded by

Copyright:

Available Formats

1.

Consider the automobile prices as a variable and the classes as economy,

c) A constant force, F, acts on a body with mass, m, moving on a smooth

ii) For the following rules:

The weights of the input layer are listed in Table Q2 a)

The system is trained and the following rule base is generated:

The weight of the fourth layer are listed in Table Q2 b)

(Continued on next page)

a) Identify the phenotype and genotype.

b) Describe one-point crossover for chosen chromosome representation.

c) Write a pseudo-code of the simple genetic algorithm.

d) Explain how the genetic algorithm will be changed if elitism is used.

a) Name and describe the main features of components of fuzzy system.

b) Suggest a simplification to the Euclidean distance classifier which results

Two distinct data classes are represented by the 5-dimensional vectors

determine whether pattern

belongs to class A or Class B, using

c) Show that under certain circumstances your approximate distance

If the network is trained on the following patterns;

determine the equations of the three logic functions after training.

b) Apply the LOW pass filter to the following image.

d) If low pass filtering is to be used to remove “spot” noise for example a

b) How may Intensity Shift and Intensity Multiply be combined to

Apply Image NORMALISATION to the following 4-bit grey-level image

Present your solution as a 3 by 3 pixel grey level enhanced image.

c) How may the Normalisation strategy be modified to enhance the darker

Apply the modified Normalisation process to the image in part b) in order

You might also like