Artificial Neural Networks

Examination, June/August 2003

Instructions
There are SIXTY questions (worth up to 60 marks). The exam mark (maximum 60) will be added to the
mark obtained in the laborations (maximum 5). The total pass mark for the course is 35 out of 65.
For each question, please select a maximum of ONE of the given answers (either A, B, C, D or E).
You should select the one answer that represents the BEST possible reply to the question (in some cases,
there may be no obvious “wrong” answers, but one answer should always be better than the others).
Every time you select the correct answer, you will be awarded +1 mark. However, every time you select
an incorrect answer, a penalty score will be subtracted from your total mark. This penalty depends on
the number of possible answers to the question, as follows:

Number of Score for Score for

possible correct incorrect
answers answer answer

2 +1 −1

3 +1 − 12

4 +1 − 13

5 +1 − 14

If you do not give any answer to a question, no marks will be added to your total score and there will be
no penalty. If you give more than one answer to a question, this will be counted as an incorrect answer.
So please be very careful, and make sure that ONLY one letter (A or B or C or D or E) is visible in each
of your written answers. Please write your answers very clearly so that they can be read by an average
examiner!

Advice: read all of the questions before you start to answer.

Tools required: calculator.

1
Artificial Neural Networks – Examination, June/August 2003 2

Questions
1 Pattern Recognition
What is a feature in a pattern classification problem?
A. An output variable.
B. An input variable.
C. A hidden variable or weight.

2 Pattern Recognition
What is a decision boundary in a pattern classification problem with two input variables?
A. A histogram defined for each image in the training set.
B. A line or curve which separates two different classes.
C. A plane or hypercurve defined in the space of possible inputs.

3 Pattern Recognition
What is a statistically optimal classifier?
A. A classifier which calculates the nearest neighbour for a given test example.
B. A classifier which gives the lowest probability of making classification errors.
C. A classifier which minimises the sum squared error on the training set.

4 Pattern Recognition
Is the following statement true or false? “In supervised learning, there is a target output vector which
tells the pattern recognition system the correct class for a given input vector.”
A. TRUE.
B. FALSE.

5 Classical Pattern Recognition

During training, which parameters must be calculated in a minimum distance classifier?
A. The mean vector of each class.
B. The mean vector and covariance matrix of each class.
C. The weights, connections and bias values of each class.

6 Classical Pattern Recognition

During training, which parameters must be calculated in a Bayes optimal classifier?
A. The mean vector of each class.
B. The mean vector and covariance matrix of each class.
C. The weights, connections and bias values of each class.
Artificial Neural Networks – Examination, June/August 2003 3

7 Classical Pattern Recognition

Design a minimum distance classifier with three classes using the following training data:
           
−2.0 −2.0 −2.0 −2.0 2.0 2.0
Class 1: , Class 2: , Class 3: ,
−1.0 −3.0 1.0 3.0 1.0 −1.0

What are the prototype vectors of the trained classifier?

     
−2.0 −2.0 2.0
A. m1 = ,m2 = ,m3 = .
−2.0 2.0 0.0
     
2.0 −2.0 −2.0
B. m1 = ,m2 = ,m3 = .
0.0 −2.0 2.0
     
−2.0 2.0 −2.0
C. m1 = ,m2 = ,m3 = .
2.0 0.0 −2.0

8 Classical Pattern Recognition

T
Classify the test vector [1.0, −2.0] with the trained classifier in question 7. Which class does this vector
belong to?
A. Class 1.
B. Class 2.
C. Class 3.

9 Classical Pattern Recognition

The decision function for a minimum distance classifier is dj (x) = xT mj − 12 mTj mj where mj is the
prototype vector for class j. What is the value of the decision function for each of the classes in question 7
T
for the test vector [−1.0, 2.0] ?
A. d1 (x) = −10.0, d2 (x) = −2.0, d3 (x) = 0.0.
B. d1 (x) = −6.0, d2 (x) = 2.0, d3 (x) = −4.0.
C. d1 (x) = −2.0, d2 (x) = −10.0, d3 (x) = 0.0.
D. d1 (x) = 2.0, d2 (x) = −6.0, d3 (x) = −4.0.

10 Classical Pattern Recognition

Give the equation of the decision boundary between classes 1 and 2 for the minimum distance classifier
in question 7.
A. x2 = 0

B. x2 = x1
C. x2 = −2x1 − 1
D. x2 = 2x1 + 1
Artificial Neural Networks – Examination, June/August 2003 4

11 Classical Pattern Recognition

Give the equation of the decision boundary between classes 2 and 3 for the minimum distance classifier
in question 7.
A. x2 = 0
B. x2 = x1
C. x2 = −2x1 − 1
D. x2 = 2x1 + 1

12 Training and Testing

What is generalization?
A. The ability of a pattern recognition system to approximate the desired output values for pattern
vectors which are not included in the test set.
B. The ability of a pattern recognition system to approximate the desired output values for pattern
vectors which are not included in the training set.
C. The ability of a pattern recognition system to extrapolate on pattern vectors which are not included
in the training set.
D. The ability of a pattern recognition system to interpolate on pattern vectors which are not included
in the test set.

13 Training and Testing

What is an outlier?
A. An input pattern which is not included in the test set.
B. An input pattern which produces a classification error.
C. An input pattern which is not included in the training set.
D. An input pattern which is very similar to the prototype vector of the patterns in the same class.
E. An input pattern which is very different from the prototype vector of the patterns in the same class.

14 Artificial Neural Networks

Which of the following statements is the best description of the Delta Rule?
A. “If a particular input stimulus is always active when a neuron fires then its weight should be
increased.”

B. “If a stimulus acts repeatedly at the same time as a response then a connection will form between
the neurons involved. Later, the stimulus alone is sufficient to activate the response.”
C. “The connection strenghts of the neurons involved are modified to reduce the error between the
desired and actual outputs of the system.”
Artificial Neural Networks – Examination, June/August 2003 5

15 Artificial Neural Networks

Which of the following artificial neural networks uses supervised learning?
A. Kohonen’s self-organizing feature map.
B. Elman’s simple recurrent network.

C. The Hopfield network.

D. Hard competitive learning.
E. A genetic algorithm.

16 Perceptrons
A perceptron with a unipolar step function has two inputs with weights w1 = 0.2 and w2 = −0.5, and a
threshold θ = −0.2 (θ can therefore be considered as a weight for an extra input which is always set to
T
-1). For a given training example x = [1, 1] , the desired output is 0 (zero). Does the perceptron give the
correct answer (that is, is the actual output the same as the desired output)?
A. Yes.
B. No.

17 Perceptrons
The perceptron in question 16 is trained using the learning rule 4w = η (d − y) x, where x is the input
vector, η is the learning rate, w is the weight vector, d is the desired output, and y is the actual output.
What are the new values of the weights and threshold after one step of training with the input vector
T
x = [0, 1] and desired output 1, using a learning rate η = 0.2?
A. w1 = 0.2, w2 = −0.3, θ = −0.4.
B. w1 = 0.2, w2 = −0.5, θ = −0.2.
C. w1 = 0.2, w2 = −0.3, θ = 0.0.
D. w1 = 0.4, w2 = −0.5, θ = −0.4.
E. w1 = 0.0, w2 = −0.5, θ = 0.0.

18 Perceptrons
A single-layer perceptron has 6 input units and 3 output units. How many weights does this network
have?
A. 6

B. 9
C. 18
D. 25
Artificial Neural Networks – Examination, June/August 2003 6

19 Perceptrons
The following network is a multi-layer perceptron, where all of the units have binary inputs (0 or 1),
unipolar step functions and binary outputs (0 or 1).

6

4

 6@
I
@
@
@
w43

3
w41  w42



J
]J

J

w31 w32J


J
1 2
 
6 6

The weights for this network are w31 = 1, w32 = 1, w41 = 1, w42 = 1 and w43 = −2. The threshold of the
hidden unit (3) is 1.5 and the threshold of the output unit (4) is 0.5. The threshold of both input units
(1 and 2) is 0.5, so the output of these units is exactly the same as the input. Which of the following
Boolean functions can be computed by this network?
A. AND.
B. OR.

C. XOR.
D. All of the above answers.
E. None of the above answers.

20 Perceptrons
Would it be possible to train the multi-layer perceptron in question 19 to solve the XOR problem using
the back-propagation algorithm?

A. Yes.
B. No.

21 Multi-Layer Feedforward Networks

Is the following statement true or false? “For any feedforward network, we can always create an equivalent
feedforward network with separate layers.”
A. TRUE.

B. FALSE.
Artificial Neural Networks – Examination, June/August 2003 7

22 Multi-Layer Feedforward Networks

Is the following statement true or false? “A multi-layer feedforward network with linear activation func-
tions is more powerful than a single-layer feedforward network with linear activation functions.”
A. TRUE.
B. FALSE.

23 Multi-Layer Feedforward Networks

A multi-layer feedforward network has 4 input units, one hidden layer with 3 units, and 2 output units.
How many weights does this network have?
A. 9
B. 12
C. 18
D. 24

24 Multi-Layer Feedforward Networks

What is the credit assignment problem in a multi-layer feedforward network?
A. The problem of adjusting the weights for the hidden units.
B. The problem of defining an error function for linearly separable problems.
C. The problem of avoiding local minima in the error function.
D. The problem of adjusting the weights for the output units.

25 Multi-Layer Feedforward Networks

A common technique for training MLFF networks is to calculate the generalization error on a separate
data set after each epoch of training. Training is stopped when the generalization error starts to increase.
This technique is called
A. momentum.
B. backpropagation.
C. early stopping.
D. extrapolation.
E. hold-one-out.
Artificial Neural Networks – Examination, June/August 2003 8

26 Multi-Layer Feedforward Networks

T T
A training pattern, consisting of an input vector x = [x1 , x2 , x3 ] and desired outputs t = [t1 , t2 , t3 ] , is
presented to the following neural network. What is the usual sequence of events for training the network
using the backpropagation algorithm?
 
x1 - I1 - z
1
 
HH
J v11 w11 
*

J HH
HH   
v21J y1 

j
H
H 
w 12
J *
 1 H
J 
J HH

 v12   J H
H w21 
J
J HH
 J

j
x2 -  J I2 - z
2
 *
H J

HH

v22 H
H J
J w22
HHJ^
J

j  J
v13
H2 H

Jw31
  y2 HH
*



v 23
w32 H J 
H
JJ
HH
j ^
x3 - 

I3 - z
3
 

A. (1) calculate yj = f (Hj ), (2) calculate zk = f (Ik ), (3) update wkj , (4) update vji .
B. (1) calculate yj = f (Hj ), (2) calculate zk = f (Ik ), (3) update vji , (4) update wkj .
C. (1) calculate yj = f (Hj ), (2) update vji , (3) calculate zk = f (Ik ), (4) update wkj .
D. (1) calculate zk = f (Ik ), (2) update wkj , (3) calculate yj = f (Hj ), (4) update vji .

27 Multi-Layer Feedforward Networks

After some training, the units in the neural network of question 26 have the following weight vectors:
   
−0.7 −1.2      
1.0 0.5 0.3
v1 = 1.8 , v2 = −0, 6 , w1 =
    , w2 = and w3 = .
−3.5 −1.2 0.6
2.3 2.1

Assume that all units have sigmoid activation functions given by

1
f (x) =
1 + exp(−x)
T
and that each unit has a bias θ = 0 (zero). If the network is tested with an input vector x = [2.0, 3.0, 1.0]
then the output of the first hidden neuron y1 will be
A. 0.1091
B. 0.5000
C. 0.8909
D. 0.0018
E. 0.9982
(Hint: on some calculators, exp(x) = ex where e = 2.7182818)
Artificial Neural Networks – Examination, June/August 2003 9

28 Multi-Layer Feedforward Networks

For the same neural network described in questions 26 and 27, the output of the third output neuron z3
will be
A. 0.6494
B. 0.0570
C. 0.5902
D. 0.5910
E. 0.2093
(Assume exactly the same weights, activation functions, bias values and input vector as in question 27.)

29 Multi-Layer Feedforward Networks

The following figure shows part of the neural network described in questions 26 and 27. In this question,
a new input pattern is presented to the network and training continues as follows. The actual outputs
T
of the network are given by z = [0.35, 0.88, 0.57] and the corresponding target outputs are given by
T
t = [1.00, 0.00, 1.00] . The weights w12 , w22 and w32 are also shown below.

.. ..
..
..
. . I1 - z = 0.35
1
J ...... . .. 
.. . .

. . 
J . .. . .

.. ..
.. ..

J ..
J . .. .
. . ...
. .
. .

−3.5
J. . . . . ... 
. . . . ..
. ..

. ..
.
..

. . ..
. . . .
..
.
J . .
. . ..
.. . .

. ..
.. J . . ..
HH .
. J
.

... *
 I2 - z = 0.88
2
. . 
.
HH .
. −1.2

. J .
.
.
.
. H J
 .
.
.
. HH j^ .
.
.
. H 2

 .
.
.
.
 * 
 H
y 2
H .
.
.
. H .
. 
0.6 . 
.  H .
. H .
.  .
.  H .
. j
H .
.
I3 - z = 0.57
3

For the output units, the Generalized Delta Rule can be written as
4wkj = ηδk yj
where
δk = f 0 (Ik )(tk − zk )
where 4wkj is the change to the weight from unit j to unit k, η is the learning rate, δk is the error for
unit k, and f 0 (net) is the derivative of the activation function f (net).
For the sigmoid activation function given in question 27, the derivative can be rewritten as
f 0 (Ik ) = f (Ik )[1 − f (Ik )].
What is the error for each of the output units?
A. δoutput 1 = 0.4225, δoutput 2 = −0.1056, and δoutput 3 = 0.1849.
B. δoutput 1 = 0.1479, δoutput 2 = −0.0929, and δoutput 3 = 0.1054.
C. δoutput 1 = −0.1479, δoutput 2 = 0.0929, and δoutput 3 = −0.1054.
D. δoutput 1 = −0.4225, δoutput 2 = 0.1056, and δoutput 3 = −0.1849.
(Assume exactly the same weights, activation functions and bias values as described in question 27.)
Artificial Neural Networks – Examination, June/August 2003 10

30 Multi-Layer Feedforward Networks

For the hidden units of the same network, the Generalized Delta Rule can be written as

4vji = ηδj xi

where
X
δj = f 0 (Hj ) δk wkj
k

where 4vji is the change to the weight from unit i to unit j, η is the learning rate, δj is the error for unit
j, and f 0 (net) is the derivative of the activation function f (net).
For the sigmoid activation function given in question 27, the derivative can be rewritten as

f 0 (Hj ) = f (Hj )[1 − f (Hj )].

What is the error for hidden unit 2 given that its activation for the pattern being processed is currently
y2 = 0.74?
A. δhidden 2 = −0.0660
B. δhidden 2 = 0.0000
C. δhidden 2 = −0.2388
D. δhidden 2 = 0.2388
E. δhidden 2 = 0.0660
(Assume exactly the same weights, activation functions and bias values as described in question 27, and
exactly the same output vectors t and z as described in the previous question.)

31 Multi-Layer Feedforward Networks

Which of the following techniques is NOT a strategy for dealing with local minima in the backpropagation
algorithm?
A. Add random noise to the weights or input vectors during training.
B. Train using the Generalized Delta Rule with momentum.
C. Train and test using the hold-one-out strategy.
D. Test with a committee of networks.

32 Multi-Layer Feedforward Networks

Training with the “1-of-M” coding is best explained as follows:
A. Set the actual output to 1 for the correct class, and set all of the other actual outputs to 0.
B. Set the actual outputs to the posterior probabilities for the different classes.
C. Set the target output to 1 for the correct class, and set all of the other target outputs to 0.
D. Set the target outputs to the posterior probabilities for the different classes.
Artificial Neural Networks – Examination, June/August 2003 11

33 Multi-Layer Feedforward Networks

Which of the following statements is the best description of underfitting?
A. The network becomes “specialized” and learns the training set too well.
B. The network can predict the correct outputs for test examples which lie outside the range of the
training examples.
C. The network does not contain enough adjustable parameters to find a good approximation to the
unknown function which generated the training data.
D. None of the above answers.

34 Multi-Layer Feedforward Networks

Which of the following statements is the best description of overfitting?
A. The network becomes “specialized” and learns the training set too well.
B. The network can predict the correct outputs for test examples which lie outside the range of the
training examples.
C. The network does not contain enough adjustable parameters to find a good approximation to the
unknown function which generated the training data.
D. None of the above answers.

35 Recurrent Artificial Neural Networks

A simple recurrent network has 4 input units, 3 hidden units and 2 output units. How many weights does
this network have?
A. 15
B. 18
C. 27
D. 32

36 Genetic Algorithms
Which operator in a genetic algorithm can be defined as follows: “Randomly change a small part of some
strings.”?
A. reproduction
B. cross-over
C. mutation

D. ranking
Artificial Neural Networks – Examination, June/August 2003 12

37 Genetic Algorithms
Which operator in a genetic algorithm can be defined as follows: “Randomly combine the genetic infor-
mation from two strings.”?
A. reproduction
B. cross-over
C. mutation
D. ranking

38 Genetic Algorithms
Which operator in a genetic algorithm can be defined as follows: “Randomly pick strings to make the
next generation.”?
A. reproduction
B. cross-over
C. mutation
D. ranking

39 Genetic Algorithms
The weighted roulette wheel is a technique used for
A. selecting the best string.
B. randomly selecting the strings.
C. crossing-over the selected strings.
D. mutating the fittest strings.
E. measuring the fitness of the strings.

40 Genetic Algorithms
Which of the following applications can be implemented by a genetic algorithm?
A. Feature selection for pattern recognition problems.

B. Learning the weights in a neural network.

C. Robot path planning.
D. All of the above answers.

E. None of the above answers.

41 Unsupervised Learning
Is the following statement true or false? “Patterns within a cluster should be similar in some way.”

A. TRUE.
B. FALSE.
Artificial Neural Networks – Examination, June/August 2003 13

42 Unsupervised Learning
Is the following statement true or false? “Clusters that are similar in some way should be far apart.”
A. TRUE.
B. FALSE.

43 Unsupervised Learning
A self-organizing feature map (SOFM) has 8 input units, and 25 output units arranged in a two-
dimensional grid. How many weights does this network have?
A. 20
B. 25
C. 200
D. 250
E. 800

44 Unsupervised Learning
Which of the following statements is NOT true for hard competitive learning (HCL)?
A. There is no target output in HCL.
B. There are no hidden units in a HCL network.
C. The input vectors are often normalized to have unit length — that is, k x k= 1.
D. The weights of the winning unit k are adapted by 4wk = η (x − wk ), where x is the input vector.
E. The weights of the neighbours j of the winning unit are adapted by 4wj = ηj (x − wj ), where
ηj < η and j 6= k.

45 Unsupervised Learning
Which of the following statements is NOT true for a self-organizing feature map (SOFM)?
A. The size of the neighbourhood is decreased during training.
B. The SOFM training algorithm is based on soft competitive learning.
C. The cluster units are arranged in a regular geometric pattern such as a square or ring.
D. The learning rate is a function of the distance of the adapted units from the winning unit.
E. The network can grow during training by adding new cluster units when required.
Artificial Neural Networks – Examination, June/August 2003 14

46 Unsupervised Learning
A self-organizing feature map has four cluster units arranged in a one-dimensional ring, as shown in the
following diagram:

The weight vectors of the four units are given as follows:

T
w1 = [−1.00, −1.50, 0.50]
T
w2 = [2.00, −2.00, 5.20]
T
w3 = [1.50, 6.00, 4.30]
T
w4 = [−4.00, 7.00, 0.60]

T
An input vector x = [−1.40, 2.30, 0.20] is presented to the network. Which unit is nearest to x in terms
of Euclidean distance?
A. Unit 1.

B. Unit 2.
C. Unit 3.
D. Unit 4.

47 Unsupervised Learning
Adapt the weight vector of the winning unit in question 46 according to the SOFM learning algorithm
with a learning rate of 0.5, using the same input vector as before. What is the new weight vector?
T
A. wwinner = [−2.70, 4.65, 0.40]
T
B. wwinner = [−1.20, 0.40, 0.35]
T
C. wwinner = [0.05, 4.15, 2.25]
T
D. wwinner = [0.30, 0.15, 2.70]
Artificial Neural Networks – Examination, June/August 2003 15

48 Unsupervised Learning
Adapt the weight vectors of the neighbours of the winning unit in question 46 according to the SOFM
learning algorithm with a learning rate of 0.2, using the same input vector as before. What are the new
weight vectors for the two units?
T T
A. wneighbour1 = [−3.48, 6.06, 0.52] and wneighbour2 = [1.32, −1.14, 4.20]
T T
B. wneighbour1 = [−3.48, 6.06, 0.52] and wneighbour2 = [0.92, 5.26, 3.48]
T T
C. wneighbour1 = [−1.08, −0.74, 0.44] and wneighbour2 = [0.92, 5.26, 3.48]
T T
D. wneighbour1 = [−1.08, −0.74, 0.44] and wneighbour2 = [1.32, −1.14, 4.20]

49 Unsupervised Learning
What is the topological mapping in a self-organizing feature map (SOFM)?
A. A map which organizes the robots and tells them where to go.
B. A mapping where similar inputs produce similar outputs, which preserves the probability distribution
of the training data.
C. An approximation of a continuous function, which maps the input vectors onto their posterior
probabilities.
D. A mapping where similar inputs produce different outputs, which preserves the possibility distribu-
tion of the training data.

50 Associative Memory
Which of the following statements is NOT true for a Hopfield network?
A. A unit cannot be connected to itself — that is, wii = 0 for all units i.
B. The weight matrix is symmetric — that is, wij = wji for all units i and j.
C. The Hopfield network minimizes an energy function during recall.
D. The error function can be defined as E = − 21 i j wij Si Sj .
P P

E. Classification in a Hopfield network is decided using the “winner-takes-all” strategy.

51 Associative Memory
A Hopfield network has 10 units. How many adjustable parameters does this network have?
A. 90
B. 50

C. 100
D. 45
Artificial Neural Networks – Examination, June/August 2003 16

52 Associative Memory
 
Calculate the weight matrix for a Hopfield network to store the pattern −1 1 −1 .
 
1 −1 1
A. W = −1 1 −1
1 −1 1
 
−1 1 −1
B. W =  1 −1 1 
−1 1 −1
 
0 −1 1
C. W = −1 0 −1
1 −1 0
 
0 1 −1
D. W =  1 0 1 
−1 1 0

53 Associative Memory
   
Calculate the weight matrix for a Hopfield network to store two patterns −1 1 −1 and 1 −1 1 .
 
0 −1 1
A. W = −1 0 −1
1 −1 0
 
0 1 −1
B. W =  1 0 1 
−1 1 0
 
0 −2 2
C. W = −2 0 −2
2 −2 0
 
0 2 −2
D. W =  2 0 2 
−2 2 0

54 Associative Memory
   
Calculate
 the weight matrix for a Hopfield network to store three patterns −1 1 −1 , 1 −1 1 and
−1 −1 1 .
 
0 1 −1
A. W =  1 0 −1
−1 −1 0
 
0 −2 2
B. W = −2 0 −2
2 −2 0
 
0 −1 1
C. W = −1 0 −3
1 −3 0
 
0 3 −3
D. W =  3 0 1 
−3 1 0
Artificial Neural Networks – Examination, June/August 2003 17

55 Applications
Which type of artificial neural network was used for learning simple behaviours such as obstacle avoidance
and wall following on a mobile robot?

A. Linear feedforward network.

B. Multi-layer feedforward network.
C. Simple recurrent network.
D. The Hopfield network.
E. Self-organizing feature map.

56 Applications
Which type of artificial neural network was used as an adaptive filter for echo cancellation in telephone
circuits?

A. Linear feedforward network.

B. Multi-layer feedforward network.
C. Simple recurrent network.
D. The Hopfield network.
E. Self-organizing feature map.

57 Applications
Which type of artificial neural network can be used to recover a clean version of a stored image given a
noisy version of that image?

A. Linear feedforward network.

B. Multi-layer feedforward network.
C. Simple recurrent network.
D. The Hopfield network.
E. Self-organizing feature map.

58 Applications
Which type of artificial neural network was used to learn the angle to an odour source using gas sensors
on a mobile robot?

A. Linear feedforward network.

B. Multi-layer feedforward network.
C. Simple recurrent network.
D. The Hopfield network.

E. Self-organizing feature map.

Artificial Neural Networks – Examination, June/August 2003 18

59 Applications
Which type of artificial neural network was used to control the ALVINN autonomous land vehicle?

A. Linear feedforward network.

B. Multi-layer feedforward network.
C. Simple recurrent network.
D. The Hopfield network.
E. Self-organizing feature map.

60 Applications
Which type of artificial neural network was used to detect abnormalities in microscopic images of breast
tissue?

A. Linear feedforward network.

B. Multi-layer feedforward network.
C. Simple recurrent network.
D. The Hopfield network.
E. Self-organizing feature map.
Artificial Neural Networks – Examination, June/August 2003 19

Answers
1. B +1 − 12
2. B +1 − 12
3. B +1 − 12
4. A +1 − 12
5. A +1 − 12
6. B +1 − 12
7. A +1 − 12
8. C +1 − 12
9. B +1 − 13
10. A +1 − 13

11. D +1 − 13
12. B +1 − 13
13. E +1 − 14
14. C +1 − 12
15. B +1 − 14
16. A +1 −1
17. A +1 − 14
18. C +1 − 13
19. C +1 − 14
20. B +1 −1

21. A +1 −1
22. B +1 −1
23. C +1 − 13
24. A +1 − 13
25. C +1 − 14
26. A +1 − 13
27. E +1 − 14
28. C +1 − 14
29. B +1 − 13
30. A +1 − 14

31. C +1 − 13
32. C +1 − 13
33. C +1 − 13
34. A +1 − 13
35. C +1 − 13
36. C +1 − 13
37. B +1 − 13
38. A +1 − 13
39. B +1 − 14
40. D +1 − 14

41. A +1 −1
42. B +1 −1
43. C +1 − 14
44. E +1 − 14
45. E +1 − 14
46. A +1 − 13
47. B +1 − 13
48. A +1 − 13
49. B +1 − 13
50. E +1 − 14
Artificial Neural Networks – Examination, June/August 2003 20

51. D +1 − 13
52. C +1 − 13
53. C +1 − 13
54. C +1 − 13
55. A +1 − 14
56. A +1 − 14
57. D +1 − 14
58. C +1 − 14
59. B +1 − 14
60. E +1 − 14