B.TECH/CSE/8TH SEM/CSEN 4264/2018 B.

TECH/CSE/8TH SEM/CSEN 4264/2018

B.TECH/CSE/8TH SEM/CSEN 4264/2018 (vi) After SVM learning, each Lagrange multiplier αi takes either zero or
MACHINE LEARNING non-zero value. What does it indicate in each situation?
(CSEN 4264) (a) A non-zero αi indicates the data point i is a support vector,
meaning it touches the margin boundary.
Time Allotted : 3 hrs Full Marks : 70 (b) A non-zero αi indicates that the learning has not yet converged to a
global minimum.
Figures out of the right margin indicate full marks. (c) A zero αi indicates that the data point i has become a support
Candidates are required to answer Group A and vector data point, on the margin.
any 5 (five) from Group B to E, taking at least one from each group. (d) A zero αi indicates that the learning process has identified support
for vector i.
Candidates are required to give answer in their own words as far as
practicable. (vii) For a neural network, which one of these structural assumptions is the
one that most affects the trade- off between under-fitting and over-
Group – A fitting?
(Multiple Choice Type Questions) (a) Number of hidden nodes (b) learning rate
1. Choose the correct alternative for the following: 10 × 1 = 10 (c) Initial choice of weights (d) Use of constant term unit input.

(i) An artificial neuron receives n inputs x1, x2, ........, xn with weights w1, (viii) Statement- 1: The SVM learning algorithm is guaranteed to find the
w2, .........., wn attached to the input links. The weighted sum globally optimum hypothesis with respect to its object function.
_____________ is computed to be passed on to a non-linear filter Φ Statement- 2: The VC dimension of a perception is smaller than VC
called activation function to release the output. dimension of a linear SVM.
(a) Σ wi (b)Σ xi (c) Σ wi + Σ xi (d)Σ wi* xi. (a) Only statement- 1 is true (b) Only statement -2 is true
(ii) The growth function h(N) for positive rays (h(X) = 1 when X > a and (c) Both are true. (d) Both are false.
h(X) = -1 otherwise) is (ix) Statement-1: Kernel density estimator can be used to perform
(a) N+1 (b) N (c) 2N (d) ∞ (In inity). classification.
(iii) Perceptron can learn Statement -2: Cross validation can be used to select the number of
(a) AND (b) XOR (c) both (a) and (b) (d)none of these. iterations in boosting; this procedure may help reduce over fitting.
(iv) Back propagation is a learning technique that adjusts weights in the (a) Only Statement -1 is true (b) Only statement- 2 is true
neural network by propagating weight changes (c) Both are true (d) Both are false.
(a) forward from source to sink (x) Consider the following data set, where X1 and X2 are input and Y is the
(b) backward from sink to source output class.
(c) forward from source to hidden nodes X1 0 0 1 1
(d) backward from sink to hidden nodes. X2 0 1 0 1
(v) Which technique uses mean and standard deviation scores to Y +1 -1 -1 +1
transform real-valued attributes?
(a) Decimal scaling (b) Min-max normalization Minimum training error can be achieved by
(c) Z-score normalization (d) Logarithmic normalization (a) 3NN Classifier (b) SVM (Quadratic Kernel)
(c) Both (a) and (b) (d) Neither (a) nor (b).

CSEN 4264 1 CSEN 4264 2

B.TECH/CSE/8TH SEM/CSEN 4264/2018 B.TECH/CSE/8TH SEM/CSEN 4264/2018
Group – B We have decided to use a neural network to solve this problem. We have
2. (a) Write short notes on any two of the followings: two choices: (i) either to train a separate neural network for each of the
(i) Hoeffding’s inequality in learning theory diseases or (ii) to train a single neural network with one output neuron for
(ii) Error vs. Noise each disease, but with a shared hidden layer. Which method do you prefer?
(iii) Bias-Variance tradeoff Justify your answer.
(b) Explain how MLP can be used as an estimator.
(b) Write the Perceptron Learning Algorithm (PLA) and briefly explain the 8 + 4 = 12
working principle of the algorithm. Group – E
6 + 6 = 12
3. (a) Derive the linear regression formula for multiple dependent variables. 8. A linearly separable dataset is given in the table below. Predict the class of
(0.6, 0.8) using a support vector machine classifier. Show all the relevant
(b) Explain briefly the difference between input space and feature space. Also
computations.
explain how the derived linear regression formula can be used for
X1 X2 Y Lagrange
nonlinear cases.
Multiplier
6 + 6 = 12
0.3858 0.4687 +1 65.5261
Group – C
0.4871 0.611 -1 65.5261
4. (a) Explain the importance of VC dimension in machine learning. 0.9218 0.4103 -1 0
(b) “Sum of square error is not always suitable.” Justify the statement with 0.7382 0.8936 -1 0
appropriate example. Explain the error and noise in the context of 0.1763 0.0579 +1 0
learning. 0.4057 0.3529 +1 0
(c) Explain the Bias-Variance trade- off in the context of learning. 0.9355 0.8132 -1 0
4 +4 + 4 =12 0.2146 0.0099 +1 0
5. (a) Define growth function and break point. 12
(b) Find the growth function and break point for the positive intervals
F(x) = +1 (when a ≤ x ≤ b); otherwise, -1. 9. Sometimes data is just nonlinearly separable or data has errors and one
(c) Consider a hypothesis class H as a set of all perceptrons in 2D plane, i.e., wants to ignore them to obtain a better solution. In fact, this is achieved by
H={hk | hk = f(w0+w1x1+w2x2), where f(x)=1, iff x>=0, otherwise f(x)=0} relaxing the margin, in other words, using a soft margin.
Find the VC dimension of this hypothesis H. Derive the Lagrangian for the optimization problem as defined by linear
3 + 3 + 6 = 12 SVM – soft margin classification.
Group – D 12
6.(a) Derive the weight update equation of a feed forward multi-layered
Perceptron network using back propagation learning.
(b) Briefly explain the momentum and how is it being incorporated in the
back propagation learning technique.
10 + 2 = 12
7.(a) Assume we have a set of data from patients who have visited Heritage
hospital during the year 2017. A set of features (e.g., temperature, height)
have also been extracted for each patient. Our goal is to decide whether a
new visiting patient has diabetes, heart disease, or Alzheimer (a patient
can have one or more of these diseases).

CSEN 4264 3 CSEN 4264 4