AL-503-B 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 Multilayer perception with neat diagram.
   How to answer

   • Definition and structure of MLP: input, hidden, output layers
   • Forward computation with weights, biases and activation functions
   • Training concept via backpropagation
   • Neat labelled diagram of MLP
2. 7 Marks
   bDiscuss about feed forward neural networks with suitable diagram.
   How to answer

   • Definition of feedforward neural networks
   • Layer-wise information flow without cycles
   • Role of weights and activation functions
   • Suitable labelled diagram
3. 7 Marks
   aExplain Gradient Descent method in deep learning.
   How to answer

   • Concept of loss function and gradient
   • Iterative weight update rule
   • Learning rate and variants
   • Role in deep learning optimization
4. 7 Marks
   bGive a brief discussion on regularization in auto encoders.
   How to answer

   • Purpose of regularization in autoencoders
   • Types: sparse, denoising, contractive, L1/L2
   • How overfitting is prevented
   • Effect on learned representation
5. 7 Marks
   aExplain the architecture of convolution neural networks with neat diagram.
   How to answer

   • Layers: convolution, activation, pooling, fully connected
   • Function of filters and feature maps
   • Neat labelled architecture diagram
   • Data flow from input image to output
6. 7 Marks
   bDefine Stochastic pooling. Explain it with example.
   How to answer

   • Definition of stochastic pooling
   • Probabilistic selection vs max/average pooling
   • Illustrative example with pooling region values
   • Advantages for generalization
7. 7 Marks
   aDraw the architecture of Deep Recurrent Neural Networks explain in detail.
   How to answer

   • Structure of deep RNN with multiple recurrent layers
   • Recurrent connections across time and depth
   • Neat architecture diagram
   • Detailed working explanation
8. 7 Marks
   bExplain about Directed graphical methods.
   How to answer

   • Concept of directed graphical models
   • Nodes, edges and conditional dependencies
   • Examples like Bayesian networks
   • Use in deep generative models
9. 7 Marks
   aExplain Restricted Boltzmann Machines with example.
   How to answer

   • Structure: visible and hidden layers with restricted connections
   • Energy-based model and learning principle
   • Example application e.g., feature learning
   • Training via contrastive divergence
10. 7 Marks
    bDefine Deep learning. Give different applications of deep learning.
    How to answer

    • Definition of deep learning with multi-layer networks
    • Difference from shallow machine learning
    • Various applications: vision, speech, NLP etc.
    • Significance of applications
11. 7 Marks
    aExplain how decomposition can be done in PCA and SVD.
    How to answer

    • Concept of dimensionality reduction
    • PCA via eigendecomposition / SVD
    • SVD formula $A = U\Sigma V^T$
    • Relation between PCA and SVD
12. 7 Marks
    bExplain about Nesterov Accelerated GD.
    How to answer

    • Motivation: acceleration of gradient descent
    • Look-ahead gradient concept
    • Update equations of Nesterov momentum
    • Advantage over classical momentum
13. 7 Marks
    aExplain Back propagation through time with examples.
    How to answer

    • Concept of unfolding RNN over time steps
    • Forward pass and backward gradient flow
    • Example sequence calculation
    • Issues like vanishing gradient
14. 7 Marks
    bExplain speech recognition techniques in detail.
    How to answer

    • Speech signal preprocessing and features
    • Acoustic modelling with deep networks
    • Language modelling and decoding
    • End-to-end deep speech systems
15. 14 Marks
    Write a short note on the following:

    1. a)Image recognition
    2. b)Video analytics
    3. c)Weight decay
    How to answer

    • Image recognition: CNN-based classification concept
    • Video analytics: detection and tracking over frames
    • Weight decay: L2 regularization to prevent overfitting
    • Applications and formulas for each