[Link]
com
Name : ……………………………………………………………
Roll No. : ………………………………………………………..
Invigilator’s Signature : ………………………………………..
CS/[Link] (CSE)/SEM-8/CS-801F/2013
2013
PATTERN RECOGNITION
Time Allotted : 3 Hours Full Marks : 70
The figures in the margin indicate full marks.
Candidates are required to give their answers in their own words
as far as practicable.
GROUP – A
( Multiple Choice Type Questions )
1. Choose the correct alternatives for the following :
10 1 = 10
i) Clustering algorithm usually employ
a) supervised learning
b) unsupervised learning
c) reinforcement learning
d) competitive learning.
8207 [ Turn over
CS/[Link] (CSE)/SEM-8/CS-801F/2013
ii) The likelihood of class w 1 and w 2 followed normal
distribution N ( – 0·5, 2 ) and N ( 0·5, 2 ), respectively.
For equal prior, a pattern X = 1·0 belongs to
a) class w 1
b) class w 2
c) either class w 1 or class w 2
d) both the classes.
iii) If the covariance matrices for all of the classes are
identical, then the discriminant functions will be
a) Linear b) Quadratic
c) Polynomial d) None of these.
iv) For uniform prior we can estimate the parameter of a
density function by using
a) maximum likelihood ( ML )
b) maximum a posteriority ( MAP )
c) either ML or MAP
d) none of these.
v) K-Nearest Neighbor based classifier is
a) linear and optimal
b) linear and suboptimal
c) nonlinear and optimal
d) nonlinear and suboptimal.
8207 2
CS/[Link] (CSE)/SEM-8/CS-801F/2013
vi) If P NN is the classification error probability for the
Nearest Neighbor rule and P B is the Bayes error then
a) P B P NN 2P B b) P NN P 2B
c) P NN P 2B d) P NN P B..
vii) Gradient descent search is not applicable to find optima
on a
a) rough surface
b) smooth surface
c) surface with single optima
d) surface with multiple optima.
viii) Perceptron is not able to implement
a) OR gate b) AND gate
c) XOR gate d) NOT gate.
ix) Given two fuzzy clusters A 1 and A 2 . A data point X in
two-class ( fuzzy C-means clustering ) then satisfies
a) µA ( x)+µA ( x)=1
1 2
b) µA ( x)+µA ( x)<1
1 2
c) µA ( x)+µA ( x)>1
1 2
d) µA ( x)+µA ( x) 1.
1 2
x) Principal component analysis is one important step in
a) Data dimension reduction
b) Data encryption
c) Noise filtering
d) Data communication.
8207 3 [ Turn over
CS/[Link] (CSE)/SEM-8/CS-801F/2013
GROUP – B
( Short Answer Type Questions )
Answer any three of the following. 3 5 = 15
2. Compare and contrast supervised and unsupervised
learning.
3. Design a Bayes classifier in terms of a set of discriminant
functions.
4. A sample from class-A is located at ( X, Y, Z ) = ( 1, 2, 3 ), a
sample from class-B is at ( 7, 4, 5 ) and a sample from
class-C is at ( 6, 2, 1 ). How would a sample at ( 3, 4, 5 ) be
classified using the Nearest Neighbor technique and
Euclidean distance ?
5. Write a short note on generalized linear discriminant
function.
6. Consider the following proximity matrix :
x1 x2 x3 x4 x5
P= x1 0 6 8 2 7
x2 0 1 5 3
x3 0 10 9
x4 0 4
x5 0
Draw the resulting dendrogram by applying single link
clustering algorithm.
8207 4
CS/[Link] (CSE)/SEM-8/CS-801F/2013
GROUP – C
( Long Answer Type Questions )
Answer any three of the following. 3 15 = 45
7. a) Describe the basic steps involved in the design of
pattern recognition system.
b) What is maximum likelihood ( ML ) estimation ? Show
that if the likelihood function is univariate Gaussian
2,
with unknowns the mean µ as well as variance then
ML estimate are given by
N
1 2 1 2
µ =
N
X k , and =
N (Xk–µ) ,
k=1
where X k is the k th pattern and N is the total number
of training patterns.
c) Compare parametric and non-parametric technique.
6+5+4
8207 5 [ Turn over
CS/[Link] (CSE)/SEM-8/CS-801F/2013
8. a) What is Bayesian classifier ? Prove that it is an optimal
classifier.
b) In a two class problem with single feature X the pdf’s
1
are Gaussians with variance 2 = for both classes
2
and mean value 0 and 1 respectively. If
1
P (w1 )= P (w2 ) = , compute the threshold value
2
X 0 for minimum error probability. 4+5+6
9. a) What is density estimation ? What are the necessary
conditions for its convergence ?
b) Compare Parzen Windows and k-Nearest Neighbor
density estimation technique.
c) What is perceptron ? Discuss briefly the perceptron
based learning algorithm. 4+4+7
10. a) What is clustering ? Categorize the different clustering
algorithms of the pattern recognition domain.
b) Explain Fuzzy-C-means clustering algorithm. Write a
short note about its criterion function. 6+9
8207 6
CS/[Link] (CSE)/SEM-8/CS-801F/2013
11. a) What is feature selection ? What is optimal and
suboptimal feature subset selection ?
b) Explain one suboptimal feature subset selection
technique.
c) What is feature generation ? Write a short note on
principal component analysis. 4+5+6
8207 7 [ Turn over
