AD-601 Deep Learning - Jun 2025 Question Paper

Under each question, How to answer lists the points an answer should cover, outlined by CampusPrep from the syllabus. Check the details against your textbook.

1. 7 Marks
   aExplain the significance of depth in deep neural networks. How does increasing the number of layers affect model performance?
   How to answer

   • Define depth as number of hidden layers
   • Hierarchical feature learning with depth
   • Effect on representational capacity and performance
   • Trade-offs: vanishing gradients, overfitting, computation
2. 7 Marks
   bCompare biological neural networks with artificial neural networks. What are the limitations of current deep learning models compared to the human brain?
   How to answer

   • Structure of biological vs artificial neurons
   • Learning mechanism comparison
   • Energy, generalization and robustness differences
   • Limitations vs human brain: reasoning, data efficiency
3. 7 Marks
   aWhat is a multilayer perceptron (MLP)? How does it overcome the limitations of single-layer perceptrons?
   How to answer

   • Definition and architecture of MLP
   • Hidden layers, activation functions
   • Limitation of single-layer perceptron: linear separability
   • How MLP solves XOR and non-linear problems
4. 7 Marks
   bDiscuss different optimization techniques used in training feedforward networks, such as gradient descent, stochastic gradient descent, and Adam optimizer.
   How to answer

   • Gradient descent update rule
   • Stochastic gradient descent characteristics
   • Adam optimizer: momentum and adaptive rates
   • Comparison of convergence and use cases
5. 7 Marks
   aExplain the process of initializing weights in a neural network. Why is proper initialization important?
   How to answer

   • Methods: random, Xavier, He initialization
   • Steps in weight initialization process
   • Problems of zero/large initialization
   • Importance for convergence and vanishing/exploding gradients
6. 7 Marks
   bExplain the significance of the convolution operation in CNNs. How does it capture spatial features from images?
   How to answer

   • Definition of convolution operation
   • Filters, feature maps and weight sharing
   • How spatial locality and translation invariance captured
   • Example of edge to high-level features
7. 7 Marks
   aWrite the difference between Convolutional Neural Network and Recurrent Neural Network.
   How to answer

   • CNN: spatial, feedforward, convolution and pooling
   • RNN: sequential, recurrent connections, memory
   • Differences in data type and architecture
   • Applications of each
8. 7 Marks
   bWhat is Long Short Term Memory in Neural Network? Explain.
   How to answer

   • Definition and need for LSTM
   • Cell state and gates: forget, input, output
   • How long-term dependencies handled
   • Diagram of LSTM cell
9. 7 Marks
   aDiscuss how batch gradient descent and stochastic gradient descent are different.
   How to answer

   • Batch GD: full dataset gradient
   • Stochastic GD: single sample update
   • Differences in speed, stability and convergence
   • Mini-batch as trade-off
10. 7 Marks
    bHow are deep generative models used in machine learning, and what are some of their key applications?
    How to answer

    • Definition of deep generative models
    • Types: GANs, VAEs, autoregressive models
    • How they learn data distribution
    • Key applications: image, text, drug discovery
11. 7 Marks
    aWhat is the role of pooling layers in CNNs? Explain max pooling and average pooling.
    How to answer

    • Purpose of pooling: downsampling, invariance
    • Max pooling operation with example
    • Average pooling operation with example
    • Effect on parameters and overfitting
12. 7 Marks
    bWhat is difference between artificial intelligence, machine learning and deep learning.
    How to answer

    • Definitions of AI, ML, DL hierarchy
    • Differences in scope and techniques
    • Data and compute requirements
    • Examples of each
13. 7 Marks
    aWhat is the role of filters or kernels in CNNs?
    How to answer

    • Definition of filters/kernels
    • Feature extraction via convolution
    • Learnable parameters and different features
    • Effect of size, stride and number
14. 7 Marks
    bDiscuss hybrid models that combine the benefits of VAEs and GANs. How do VAE-GAN hybrid models work?
    How to answer

    • Basics of VAE and GAN
    • Motivation for hybrid VAE-GAN
    • Architecture and loss components
    • How encoder-generator-discriminator work together
15. 14 Marks
    Explain any two of the following :

    1. i)Structured output in Convolutional Network
    2. ii)Alexnet
    3. iii)Deep Reinforcement
    4. iv)Autoencoder Architecture
    How to answer

    • Select any two topics from list
    • Definition and key architecture/components
    • Working principle with diagram if needed
    • Applications and advantages