AD-601 Deep Learning - May 2023 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. 8 Marks
   aWhat is Deep Learning? Discuss the structure and units of Biological Neuron.
   How to answer

   • Define Deep Learning as multi-layer neural networks for hierarchical feature learning
   • Describe biological neuron structure: dendrites, soma, axon, synapse
   • Explain functional units: input reception, integration, activation, output transmission
   • Relate biological neuron to artificial neuron model
2. 6 Marks
   bExplain the difference between Machine Learning and Deep Learning.
   How to answer

   • Define Machine Learning with manual feature engineering
   • Define Deep Learning with automatic hierarchical feature learning
   • Compare on data requirement, hardware, interpretability, performance
   • Give examples of suitable applications for each
3. 7 Marks
   aExplain the importance of Gradient Descent algorithm in Deep Learning.
   How to answer

   • Define Gradient Descent and cost function minimization
   • Explain iterative weight update using gradient and learning rate
   • Discuss importance: optimization, convergence, handling large data
   • Mention variants: SGD, mini-batch, Adam
4. 7 Marks
   bWhat is Multilayer Perceptron? Explain Backpropagation algorithm in detail.
   How to answer

   • Define Multilayer Perceptron: input, hidden, output layers with non-linear activation
   • Explain forward pass to compute loss
   • Explain backward pass: chain rule for error backpropagation and weight update
   • Mention advantages and training issues
5. 7 Marks
   aBriefly discuss the role of LeNet in Convolutional Networks.
   How to answer

   • Introduce LeNet-5 by LeCun for handwritten digit recognition
   • Describe architecture: conv-pool layers, fully connected layers
   • Discuss role as pioneering CNN demonstrating end-to-end learning
6. 7 Marks
   bExplain the variants of the basic convolution function in detail.
   How to answer

   • Define basic discrete convolution operation
   • Explain variants: full, same, valid convolution, strided, dilated, separable
   • Discuss locally connected, tiled convolution
   • Give use-cases for each variant
7. 7 Marks
   aWhat are Bidirectional RNNs? Explain in detail.
   How to answer

   • Define Bidirectional RNN with forward and backward hidden states
   • Explain architecture and processing of past and future context
   • Discuss training with BPTT and output combination
   • Mention applications in NLP, speech
8. 7 Marks
   bDefine Recurrent Network? Discuss about Deep Recurrent Networks.
   How to answer

   • Define Recurrent Network with cyclic connections and memory
   • Explain unfolding in time and hidden state recurrence
   • Define Deep Recurrent Networks with stacked recurrent layers
   • Discuss advantages for hierarchical temporal features
9. 14 Marks
   Give an overview of the following:

   1. a)Deep Boltzmann Machines
   2. b)Deep Belief Network
   3. c)Supervised Learning
   How to answer

   • Define Deep Boltzmann Machine: undirected stacked RBMs
   • Define Deep Belief Network: directed/undirected hybrid generative model
   • Define Supervised Learning: labelled data mapping
   • Compare generative pre-training vs discriminative learning
10. 9 Marks
    aDiscuss the role of Deep Learning in Data Science.
    How to answer

    • Define Data Science workflow: data collection to decision
    • Explain deep learning for feature extraction from big unstructured data
    • Give applications: image, text, recommendation, prediction
    • Discuss advantages and challenges
11. 5 Marks
    bWrite a short note on neural network.
    How to answer

    • Define neural network as interconnected neurons
    • Describe basic components: weights, bias, activation
    • Mention types: feedforward, recurrent, convolutional
    • State applications briefly
12. 8 Marks
    aWhat is Convolution Operation? Explain in detail.
    How to answer

    • Define convolution operation mathematically for continuous and discrete cases
    • Explain 2D convolution for images with kernel/filter
    • Discuss stride, padding, feature maps
    • Give example of edge detection
13. 6 Marks
    bWrite about Gated RNNs.
    How to answer

    • Define Gated RNNs to solve vanishing gradient
    • Explain LSTM gates: input, forget, output
    • Explain GRU as simplified gating
    • Mention long-term dependency applications
14. 14 Marks
    Explain any two of the following:

    1. a)Deep Generic Models
    2. b)Gibbs Sampling
    3. c)Feed forward Network
    4. d)Artificial Intelligence verses Deep Learning
    How to answer

    • Select any two topics from the four options
    • Give definition, working principle and key features for each
    • Include diagram/formula/example where relevant
    • Conclude with applications