uiz Space

January 2026 term · Deep Learning · BSCS3004

Deep Learning End Term: 10 May 2026 (January 2026 term)

The IIT Madras BS Deep Learning (Deep Learning) End Term paper sat on 10 May 2026, in the January 2026 term: 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
MCQ
6
Numerical
10
MSQ
4
Take as mock testPractise with answers

Updated

Official paper: Deep Learning 06 May 26 · No negative marking.

Question 1

+2 marksOne correct option
  1. A

    (A figure from the original paper is missing from the source site.)

  2. B

    (A figure from the original paper is missing from the source site.)

  3. C

    (A figure from the original paper is missing from the source site.)

  4. D
Show answer

Correct answer

  • B

    (A figure from the original paper is missing from the source site.)

Question 1 on its own page

Question 2

+2 marksOne correct option

Consider a vanilla Recurrent Neural Network (RNN) and an LSTM network with the same input dimension, hidden dimension, and output dimension.
Statement 1: The number of trainable parameters in a vanilla RNN is higher than that in an LSTM network. Statement 2: An LSTM has a higher number of parameters because it contains multiple gates (input gate, forget gate, output gate, and candidate state), each having separate weight matrices and biases.
Choose the correct option from the following.

  1. A

    Statement 1 is true, and Statement 2 is the correct reason.

  2. B

    Statement 1 is true, but Statement 2 is false.

  3. C

    Statement 1 is false, but Statement 2 is true.

  4. D

    Statement 1 is false, and Statement 2 is false.

Show answer

Correct answer

  • C

    Statement 1 is false, but Statement 2 is true.

Question 2 on its own page

Question 3

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

Correct answer

  • D
Question 3 on its own page

Question 4

+3 marksOne correct option
  1. A

    Yes

  2. B

    No

Show answer

Correct answer

  • A

    Yes

Question 4 on its own page

Question 5

+3 marksOne correct option

Match the attention mechanism in Column I with the correct description in Column II.

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

Correct answer

  • C
Question 5 on its own page

Question 6

+2 marksNumerical answer
Show answer

Correct answer: 0.5

Question 6 on its own page

Question 7

+2 marksNumerical answer
Show answer

Correct answer: 0

Question 7 on its own page

Question 8

+3 marksOne or more correct options

Select all that apply.

  1. A

    It remains the same after the first update using stochastic gradient descent.

  2. B

    Different neurons in the first hidden layer will receive different gradient updates.

  3. C
  4. D
Show answer

Correct answer

  • D
Question 8 on its own page

Question 9

+2 marksNumerical answer

Based on the above data, answer the given subquestions.

Find the total number of parameters in the network.

Show answer

Correct answer: 9

Question 9 on its own page

Question 10

+3 marksNumerical answer

Based on the above data, answer the given subquestions.

The loss is defined as

. If

then compute

.

Show answer

Correct answer: 0

Question 10 on its own page

Question 11

+2 marksOne or more correct options

After training a neural network, the training error is observed to be 10%. It gives the test error to be 56%. Which of the following methods can be used to reduce the test error?

Select all that apply.

  1. A

    Adam

  2. B

    ReLu activation

  3. C

    Injecting noise at input

  4. D

    Maxout

  5. E

    L2 regularization

Show answer

Correct answers

  • C

    Injecting noise at input

  • E

    L2 regularization

Question 11 on its own page

Question 12

+3 marksOne or more correct options

Consider a transformer model using scaled dot-product attention:

Select all that apply.

  1. A
  2. B
  3. C

    If all attention heads learn identical projection matrices, multi-head attention behaves equivalently to single-head attention.

  4. D

    For a fixed embedding dimension, increasing the number of heads always increases the total number of parameters in the query, key, and value projections.

Show answer

Correct answers

  • A
  • C

    If all attention heads learn identical projection matrices, multi-head attention behaves equivalently to single-head attention.

Question 12 on its own page

Question 13

+3 marksNumerical answer
Show answer

Correct answer: 2

Question 13 on its own page

Question 14

+3 marksOne correct option
  1. A

    Positive

  2. B

    Zero

  3. C

    Negative

Show answer

Correct answer

  • A

    Positive

Question 14 on its own page

Question 15

+3 marksOne or more correct options

Which of the following problems of the co-occurrence matrix are addressed by applying Singular Value Decomposition (SVD)? Choose all the correct options.

Select all that apply.

  1. A

    High dimensional representation

  2. B

    Sparse matrix with many zeros

  3. C

    Captures latent semantic similarity between words

  4. D

    Increases vocabulary size

  5. E

    Produces dense low-dimensional embeddings

Show answer

Correct answers

  • B

    Sparse matrix with many zeros

  • C

    Captures latent semantic similarity between words

  • E

    Produces dense low-dimensional embeddings

Question 15 on its own page

Question 16

+2 marksNumerical answer

•

Compute the total number of elements in the resulting attention output.

Show answer

Correct answer: 512

Question 16 on its own page

Question 17

+2 marksNumerical answer

•

Show answer

Correct answer: 12288

Question 17 on its own page

Question 18

+2 marksNumerical answer

•

Assume a Feed-Forward Network (FFN) follows the Multi-Head Attention layer in the encoder. The FFN consists of two linear transformations:

Show answer

Correct answer: 33088

Question 18 on its own page

Question 19

+2 marksNumerical answer

Assume that your CBOW model outputs a probability distribution over a vocabulary of 20,000 words for a given context. If the correct target word is word number 150, and the model's predicted probability for this word is 0.2, calculate the cross-entropy loss for this prediction. Enter the answer correct to two decimal places. (use natural log)

Show answer

Correct answer: 1.6 (accepted within ±0.03)

Question 19 on its own page

Question 20

+3 marksNumerical answer

In the Skip-gram model with a window size of 2 (on each side), how many unique pairs of target and context words will be generated for the following sentence. Note: Ignore punctuation in your calculation
'Learning from data helps build intelligent systems for society'

Show answer

Correct answer: 30

Question 20 on its own page