Deep Learning (DL) SPPU Endsem Question Paper Nov 2024

Questions

1. Q1

   1. a)

      How would you describe the architecture of a typical Convolutional Neural Network (CNN)? Explain the main components and their roles.

      [6]
   2. b)

      What is the purpose of pooling layers in Convolutional Neural Networks? Explain how max pooling and average pooling contribute to down- sampling the input data.

      [6]
   3. c)

      How do data augmentation and dropout regularization techniques contribute to training CNNs?

      [6]
   OR

2. Q2

   1. a)

      What role does padding play in Convolutional Neural Networks? How does it impact the size of the output feature maps?

      [6]
   2. b)

      Explain typical process involved in training a Convolutional Neural Network, including the key steps and optimization techniques used?

      [6]
   3. c)

      Explain the purpose of LRN in enhancing the performance of CNNs.

      [6]

3. Q3

   1. a)

      Explain the concept of unfolding computational graphs in the context of recurrent networks.

      [6]
   2. b)

      Explain the challenge of long-term dependencies in the context of recurrent neural networks, including the issues of vanishing and exploding gradients.

      [6]
   3. c)

      What are some common performance metrics used to evaluate the effectiveness of recurrent neural networks? 2

      [5]
   OR

4. Q4

   1. a)

      Describe the architecture and usage of Bidirectional RNNs. What advantages do they offer in sequence modeling tasks?

      [6]
   2. b)

      What are Leaky Units, and how do they contribute to handling multiple time scales in the context of recurrent neural networks?

      [6]
   3. c)

      What are Echo State Networks (ESNs), and how do they address the challenge of long-term dependencies?

      [5]

5. Q5

   1. a)

      How do deep generative models differ from discriminative models in terms of their learning and inference mechanisms?

      [6]
   2. b)

      How are Deep Belief Networks structured, and how do they leverage the concept of restricted Boltzmann machines in theft architecture?

      [6]
   3. c)

      What is a Generative Adversarial Network (GAN), and how does it work?

      Also asked:May 2023

      [6]
   OR

6. Q6

   1. a)

      Explain the concept of latent variables in deep generative models and their role in capturing complex data distributions.

      [6]
   2. b)

      What are the fundamental characteristics of Boltzmann Machines, and how do they model probabilistic dependencies among variables?

      [6]
   3. c)

      How can GANs be used in generating realistic images, such as human faces or artistic images?

      [6]
   OR

7. Q7

   1. a)

      Explain objectives and challenges of deep reinforcement learning in comparison to traditional reinforcement learning methods.

      [6]
   2. b)

      Describe the concept of Deep Q-Networks (DQN) and how they combine Q-learning with deep neural networks.

      [6]
   3. c)

      Explain how dynamic programming algorithms, such as policy iteration and value iteration, are used in reinforcement learning.

      Also asked:May 2024 ·Nov 2023 ·May 2023

      [5]
   OR

8. Q8

   1. a)

      What are the key components of an Markov Decision Process, and how do they relate to the decision-making process of an agent?

      [6]
   2. b)

      How do Deep Q Recurrent Networks extend the capabilities of Deep Q-Networks in handling sequential decision-making problems?

      [6]
   3. c)

      This part has equations that don’t show as text. Open the PDF to read it.

      [5]