ML - Compre - Question - Paper - 2022 - 23 - Marking Scheme

You might also like

Download as pdf or txt
Download as pdf or txt
You are on page 1of 6

BIRLA INSTITUTE OF TECHNOLOGY & SCIENCE, PILANI, DUBAI CAMPUS

DUBAI INTERNATIONAL ACADEMIC CITY, DUBAI


FIRST SEMESTER 2022 – 2023 Questions: 8
DEPARTMENT OF COMPUTER SCIENCE Pages:
Evalaution Scheme
COURSE : Machine Learning (BITS F464)

COMPONENT : Compre Exam CLOSED BOOK WEIGHTAGE : 40% (40 Marks)

DATE : 5-January-2023 DURATION : 3 hrs (8:30 am- 11:30 am)


Note:
1. Please write down any assumptions that you make.
2. Show the intermittant steps clearly for all answers.

Q.1 The State hospital is working on developing classification model using Gaussian classifier 6 M
for classifying the two kinds of diseases using Parameter 1 and Parameter 2. The data
consider for the analysis is as follows,
Table Q.1: Parameter 1 and Parameter 2 data for two diseases
Disease Parameter 1 Parameter 2
Normal-Flu 8 9
Normal-Flu 10 10
Normal-Flu 7 4
COVID-19 10 12
COVID-19 10 15
COVID-19 13 15
COVID-19 11 13
What will be the class label for following test data? (Show all steps of calculation.)
• Parameter 1 = 10, Parameter 2 = 11

Ans Assume, COVID-19 = 1 and Normal-Flu = 0


Prior probabilities, P(1) = 0.57, P(0) = 0.42 => 1 M
11 8.33
𝑀𝑒𝑎𝑛𝑠, 𝜇1 = ( ) , 𝜇0 = ( ) => 1 M
13.75 7.66
𝑉𝑎𝑟𝑖𝑎𝑛𝑐𝑒𝑠, 𝜎12 = 0.1875, , 𝜎02 = 2.7812 => 1 M
Log posterior probability ratio is, => 2 M

For Parameter 1 = 10, Parameter 2 = 11 => 1 M

Class 0 (Normal-Flu)

Q.2 a. Define reinforcement learning. Write the generalized equation for Q-learning for 2+2
calculating Q-value, Q(s,a) and describe each of the parameter in equation. M
b. What is case-based reasoning? Draw the architecture of case-based reasoning.
Ans a. Define reinforcement learning: 1 M

1
Generalized equation of Q-learning: 1 M

b. Definition Case-based reasoning: 1 M


Architecture of case-based reasoning: 1 M

Q.3 a. Write step by step execution of forward and backward propagation of following 3 + 3
sigmoid neuron with n inputs. M

b. How values of mean, mixing coefficient, and covariance are adjusted during
maximization step of EM for Gaussian Mixture clustering?
Ans a.

2
b.

3
Q.4 A patient goes to the doctor for a medical condition, the doctor suspects three diseases as 4 M
the cause of the condition. The three diseases are D1, D2, D3, which are marginally
independent from each other. There are four symptoms S1, S2, S3, S4 which the doctor
wants to check for presence in order to find the most probable cause of the condition. The
symptoms are conditionally dependent to the three diseases as follows: S1 depends only on
D1, S2 depends on D1 and D2. S3 is depends on D1 and D3, whereas S4 depends only on
D3. Assume all random variables are Boolean, they are either ‘true’ or ‘false’.
i. Draw the Bayesian Belief Network (BNN) for above mention scenario.
ii. Express the probability of P (D1, D2, D3, S1, S2, S3, S4) based on constructed BNN
in (i).
Ans i.

ii.
P (D1;D2;D3;S1;S2;S3;S4) = P (D1) P (D2) P (D3) P (S1 jD1) P (S2 jD1;D2) P (S3 jD1;D3) P (S4 jD3)

Q.5 Consider the following data related two samples for four different genes, 6M
Table Q5: Gene Sample Dataset
Sample 1 Sample 2
Gene A 4 5
Gene B 1 4
Gene C 8 8
Gene D 5 7

Use Principal Component Analysis (PCA) to reduce the dimension of above data from 2 to
1. Show all steps of calculation of covariance matrix, eigen values, normalized eigen
vectors, and draw sample points in one dimension with respect to Principal Component 1.
4
Ans Assume, Sample 1 = x and Sample 2 = y
Means, 𝑥̅ = 4.5, 𝑦̅ = 6 => 1M
8.33 5
Covariance matrix, [ ] => 1M
5 3.231
Eigen values, 𝜆1 = 11.395, 𝜆2 = 0.169 => 1M
Normalized eigen vectors, 𝑉1 = 0.853, 𝑉2 = 0.53 => 1M
Sample points with respect to PC1 are, -0.955, -4.0295, 4.0295, 0.9485 => 1M
Plot of one dimension with respect to PC1 => 1 M

Q.6 Consider the following two Hidden Markov Models, M1 and M2. 4M

(a) (b)
Figure Q6: (a) M1 (b) M2
In both M1 and M2, transition probabilities and starting probabilities are mention near to
the respective edges. For example, in M1 start probability is 0.5 for state A and start
probability for state B is 0.5. Similar assumption is considered for M2.
i. What is P(S100 = B; S101 = B; S102 = B) for M2 ? Here, S100 means 100th symbol
emitted by M2 is B, and similarly for others. (Show all steps of calculation)
ii. Consider, P1 be: P1 = P(S100 = A; S101 = B; S102 = A; S103 = B) for M1 and let P2 be:
P2 = P(S100 = A; S101 = B; S102 = A; O103 = B) for M2. Is P1>P2, P1<P2, or P1=P2?
(Show all steps of calculation)

Ans i. 0.32
ii. P1 = 0.0625, P2 = 0.004 P1 > P2

Q.7 Consider the following data related (shown in Figure Q7) to results of PhD qualifying exam 6 M
of PhD research scholar based on two subjects’ mark. 1 indicate pass and 0 indicate fail.
Find the discriminating hyperplane to classify the sample into pass and fail class using
Support Vector classifier.

Figure Q7: PhD Qualifying exam result visulaization

5
Ans Identifying correct support vector, (4,6), (1,5), (7,2) => 1 M
126 −128 −53
Calculate values of α , 𝛼1 = , 𝛼2 = , 𝛼3 = => 3 M
25 25 25
Values of w and b, 𝑤1 = 1, 𝑤2 = 2, 𝑏 = −11

Q.8 i. Differentiate between linear regression and logistic regression (Minimum 4 correct 2 + 2
differences). M
ii. Consider the problem of binary classification using the Naive Bayes classifier. You
are given two dimensional features (X1, X2) and the categorical class conditional
distributions in the tables below. The entries in the tables correspond to P(X1 = x1|Ci)
and P(X2 = x2|Ci) respectively. The two classes are equally likely.

Given the datapoint (-1,1), calculate the following posterior probabilities.


a. P(C1 | X1 = -1 , X2 = 1)
b. P(C2 | X1 = -1 , X2 = 1)
Ans i. Each correct difference carries 0.5 M.
ii. Probabilities are as follows,

********** ALL THE BEST ********

You might also like