Deep Learning (DL) SPPU Endsem Question Paper Nov 2025
All 8 questions from this SPPU endsem paper, as printed. Read them here or open the PDF.
- Subject code
- 410251
- Pattern
- 2019 pattern
- Exam
- Endsem
- Session
- Nov 2025
- Max. marks
- 70
- Time
- 2½ Hours
- Semester
- VIII
- Paper code
- [6584]-82
Printed for: B.E. (Computer Engineering)
Instructions to the candidates
- Attempt Q.1 or Q.2, Q.3 or Q.4, Q.5 or Q.6, Q.7 or Q.8.
- Figures to the right indicate full marks.
- Neat diagrams must be drawn wherever necessary.
- Assume suitable data if necessary.
Questions
Q1
- a)[6]
List the main steps involved in training a Convolutional Neural Network (CNN) for image classification? Why normalization is important before training a CNN.
- b)[6]
Given an input of size 64 × 64, a kernel size of 5 × 5, stride = 2, and ‘same’ padding : i) What will be the size of the output feature map? ii) How does padding help retain the spatial size?
- c)[6]
What are Convolutional Neural Networks (CNNs) primarily used for in deep learning? List at least four real-world applications.
OR- a)
Q2
- a)[6]
What is Interleaving Between Layers in CNN? Why is Interleaving Important? Explain the role of each interleaving layer?
- b)[6]
What is the ReLU activation function? Write its mathematical expression and describe how it transforms the input x=[–3,–2,–1,0,2,5,8]
- c)[6]
Explain how input data flows through a typical CNN architecture from the raw image to the final output layer?
- a)
Q3
- a)[6]
How is the computational graph of an RNN different from that of a feedforward neural network?
- b)[6]
List the types of Recurrent Neural Network (RNN) and explain Long Short-Term Memory (LSTM) three gates?
- c)[6]
What is Encoder-Decoder architecture, and how does it work in sequence- to- sequence learning? 2
OR- a)
Q4
- a)[6]
What are limitations of Bidirectional RNNs, and how do they differ from standard RNNs?
- b)[6]
Explain any Seven Challenges of Long-Term Dependencies?
Also asked:May 2025
- c)[6]
How Echo State Network Differ from Traditional RNNs?
- a)
Q5
- a)[6]
What is a Boltzmann Machine? Describe its structure and components.
- b)[6]
List at least five real-world applications of Generative Adversarial Network (GANs) and describe any one in detail.
- c)[5]
Describe the difference between generative and discriminative phases in Deep Belief Networks (DBNs).
OR- a)
Q6
- a)[6]
What is the role of the discriminator in a Generative Adversarial Network (GAN)? What are the inputs and outputs of a discriminator network?
- b)[6]
Explain the following Term : i) Deep Belief Network ii) Deep Generative Model
- c)[5]
Discuss the role of GANs in anomaly detection. How do they help identify outliers in data?
- a)
Q7
- a)[6]
What is dynamic programming in the context of reinforcement learning? How does it differ from traditional Dynamic Programming (DP) in computer science?
- b)[6]
Define the terms: state, action, reward, and policy in the context of Reinforcement Learning (RL).
- c)[5]
What is a Markov Decision Process (MDP)? Define its components. 3
OR- a)
Q8
- a)[6]
How does the recurrent layer in a Deep Recurrent Q-Networks (DQRN) help in decision-making over sequences?
- b)[6]
What is Q-learning? How does it differ from other reinforcement learning algorithms?
- c)[5]
How can the game of Tic-Tac-Toe be formulated as a reinforcement learning problem? ♦♦♦
- a)