CY-701 Machine Learning - Dec 2024 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 machine learning concept by taking an example. Describe the perspective and issues in machine learning.
   How to answer

   • Define machine learning with example e.g. spam classification / house price prediction
   • Describe perspectives - statistical, computational, AI viewpoint
   • List issues - overfitting, bias-variance, data quality, generalization, ethics
2. 7 Marks
   bWhat is the role of preprocessing of data in machine learning? Why it is needed?
   How to answer

   • Define data preprocessing and its goal
   • Steps - cleaning, integration, transformation, reduction
   • Need - improve quality, handle missing/noisy data, improve model accuracy
3. 7 Marks
   aWhat is activation function? Discuss its type with suitable formula.
   How to answer

   • Define activation function and its need for non-linearity
   • Types - sigmoid, tanh, ReLU, softmax with formulas
   • Comparison of properties and use cases
4. 7 Marks
   bExplain the role of back propagation in training a multilayer network and how it adjusts weights through successive layers?
   How to answer

   • Define backpropagation and chain rule basis
   • Forward pass, loss computation, backward error propagation
   • Gradient descent weight update through successive layers
5. 7 Marks
   aWhat is the purpose of using a $1\times 1$ convolution in CNNs, and how does it contribute to feature reduction and model efficiency?
   How to answer

   • Define 1x1 convolution operation
   • Role in channel-wise dimensionality reduction
   • Benefits - fewer parameters, efficiency, added non-linearity
6. 7 Marks
   bWhat are the different types of padding (e.g., same, valid) and how does each influence the preservation of spatial features?
   How to answer

   • Define padding and types - same, valid
   • Effect of each on output size formula
   • Influence on preservation of border/spatial features
7. 7 Marks
   aHow does reinforcement learning differ from supervised and unsupervised learning?
   How to answer

   • Define reinforcement, supervised, unsupervised learning
   • Compare on feedback, labels, objective
   • Examples of each paradigm
8. 7 Marks
   bHow does SARSA differ from Q-learning in terms of updating Q-values?
   How to answer

   • Define SARSA (on-policy) and Q-learning (off-policy)
   • Write Q-value update formulas for both
   • Explain difference - actual next action vs max next action
9. 7 Marks
   aHow do SVMs find the optimal hyperplane for classification?
   How to answer

   • Define hyperplane, margin, support vectors
   • Objective - maximize margin optimization formulation
   • Role of support vectors in optimal hyperplane
10. 7 Marks
    bHow is machine learning applied in speech recognition and synthesis?
    How to answer

    • Pipeline - feature extraction, acoustic modeling
    • Models - HMM, DNN, RNN/Transformer for recognition
    • Application to synthesis - TTS, vocoders
11. 7 Marks
    aWhat is the structure of an Inception module and how does it help improve the efficiency of model.
    How to answer

    • Describe Inception module parallel branches 1x1, 3x3, 5x5, pooling
    • Role of 1x1 bottleneck for dimension reduction
    • Efficiency - multi-scale features with low computation
12. 7 Marks
    bWhat is sub sampling and how does it contribute to feature extraction?
    How to answer

    • Define subsampling / pooling
    • Types - max, average pooling
    • Contribution - dimensionality reduction, translation invariance, feature extraction
13. 7 Marks
    aDiscuss in detail about Actor-critic model.
    How to answer

    • Define actor and critic roles
    • Policy gradient and value estimation interaction
    • Advantages - reduced variance, continuous actions
14. 7 Marks
    bDifferentiate between Value Iteration and Policy Iteration.
    How to answer

    • Define value iteration and policy iteration steps
    • Compare - evaluation + improvement vs Bellman optimality update
    • Trade-offs in convergence and computation
15. 14 Marks
    Write short notes on :

    1. a)NLP
    2. b)Computer Vision
    How to answer

    • NLP - definition, tasks like sentiment, translation, models
    • Computer Vision - definition, tasks like classification, detection
    • ML role and example applications in each