uiz Space

May 2026 term · Deep Learning · BSCS3004

Deep Learning End Term: 13 September 2026, Set S2 (May 2026 term)

The IIT Madras BS Deep Learning (Deep Learning) End Term paper sat on 13 Sept 2026, in the May 2026 term, set S2: 20 questions for 50 marks in 180 minutes. Every question is below with its answer. Take it as a timed mock test to be marked, or read it through first.

Questions
20
Marks
50
Duration
180 min
Numerical
13
MCQ
5
MSQ
2
Take as mock testPractise with answers

Updated

Official paper: Deep Learning 13 Sep 26 (Session 2) · No negative marking.

Question 1

+2 marksNumerical answer
Show answer

Correct answer: -0.1

Question 1 on its own page

Question 2

+2 marksNumerical answer
Show answer

Correct answer: 3

Question 2 on its own page

Question 3

+3 marksNumerical answer

Suppose we want to minimize the following function:

Show answer

Correct answer: -0.6 (accepted within ±0.02)

Question 3 on its own page

Question 4

+2 marksOne correct option

Suppose we want to minimize the following function:

  1. A
  2. B
  3. C

    Both coordinates diverge equally.

  4. D

    Neither; both converge at the same rate.

Show answer

Correct answer

  • B
Question 4 on its own page

Question 5

+3 marksNumerical answer

Suppose we want to minimize the following function:

Show answer

Correct answer: 0.104 (accepted within ±0.003)

Question 5 on its own page

Question 6

+2 marksOne correct option

Suppose we want to minimize the following function:

How does AdaGrad address the behavior observed with vanilla gradient descent?

  1. A
  2. B
  3. C
  4. D
Show answer

Correct answer

  • A
Question 6 on its own page

Question 7

+2 marksOne or more correct options

Select all that apply.

  1. A
  2. B

    Adding more hidden layers with identity activations increases the representational capacity beyond that of a single-layer linear model.

  3. C

    Replacing the identity activation with a nonlinear function such as sigmoid enables the network to learn nonlinear decision boundaries.

  4. D

    With identity activations, the network can still learn arbitrary decision boundaries provided enough hidden neurons are used.

Show answer

Correct answers

  • A
  • C

    Replacing the identity activation with a nonlinear function such as sigmoid enables the network to learn nonlinear decision boundaries.

Question 7 on its own page

Question 8

+3 marksNumerical answer
Show answer

Correct answer: 1048576

Question 8 on its own page

Question 9

+3 marksOne correct option

How many learnable parameters (including both weights and biases) are used to compute the queries, keys, and values?

  1. A
  2. B
  3. C
  4. D
Show answer

Correct answer

  • B
Question 9 on its own page

Question 10

+2 marksNumerical answer
Show answer

Correct answer: 10000

Question 10 on its own page

Question 11

+2 marksOne correct option
  1. A
  2. B
  3. C
  4. D
Show answer

Correct answer

  • B
Question 11 on its own page

Question 12

+2 marksNumerical answer

Based on the above data, answer the given subquestions.

Show answer

Correct answer: 2

Question 12 on its own page

Question 13

+3 marksNumerical answer

Based on the above data, answer the given subquestions.

Show answer

Correct answer: 0.486 (accepted within ±0.003)

Question 13 on its own page

Question 14

+3 marksNumerical answer

Based on the above data, answer the given subquestions.

Show answer

Correct answer: 3.975 (accepted within ±0.025)

Question 14 on its own page

Question 15

+3 marksOne or more correct options

Select all that apply.

  1. A

    The decoder's softmax layer has a fixed vocabulary size and therefore cannot generate longer sentences.

  2. B
  3. C

    Teacher forcing is valid only for source sentences shorter than the maximum training length.

  4. D

    Longer source sentences necessarily require a higher-dimensional one-hot representation, which increases the input noise.

Show answer

Correct answer

  • B
Question 15 on its own page

Question 16

+3 marksNumerical answer
Show answer

Correct answer: 1.6265 (accepted within ±0.0035)

Question 16 on its own page

Question 17

+3 marksNumerical answer
Show answer

Correct answer: 0.785 (accepted within ±0.005)

Question 17 on its own page

Question 18

+3 marksNumerical answer
Show answer

Correct answer: 3.91 (accepted within ±0.02)

Question 18 on its own page

Question 19

+1 markNumerical answer
Show answer

Correct answer: 0

Question 19 on its own page

Question 20

+3 marksOne correct option
  1. A

    Each update uses a gradient estimate with higher variance, while the number of updates per epoch decreases.

  2. B

    Each update uses a gradient estimate with lower variance, while the number of updates per epoch decreases.

  3. C

    The number of parameter updates per epoch increases by a factor of four, leading to faster convergence.

  4. D

    The larger batch size guarantees convergence to a better local minimum.

Show answer

Correct answer

  • B

    Each update uses a gradient estimate with lower variance, while the number of updates per epoch decreases.

Question 20 on its own page