CY-701 Machine Learning - Dec 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
   aDiscuss in detail the types of data visualization in ML.
   How to answer

   • Define data visualization purpose in ML workflow
   • List types: univariate, bivariate, multivariate plots
   • Examples: histogram, box plot, scatter, heatmap, pair plot
   • Use for EDA, missing values, outliers and feature relations
2. 7 Marks
   bWhat is the need for data augmentation? Discuss about data augmentation.
   How to answer

   • Define data augmentation and need: limited data, overfitting
   • Techniques for image/text/tabular data
   • Benefits: generalization, robustness, class balance
   • Limitations / care in label preservation
3. 7 Marks
   aDiscuss in detail the limitations of machine learning.
   How to answer

   • Define ML dependence on data and assumptions
   • Limitations: data hunger, bias, interpretability, generalization
   • Computational cost, overfitting, adversarial vulnerability
   • Domain adaptation and ethical issues
4. 7 Marks
   bWhat is an activation function? Explain ReLU function and Compute the output of ReLU function for $x = -3, 0, 5$.
   How to answer

   • Define activation function and need for non-linearity
   • Define ReLU formula $f(x)=\max(0,x)$ and graph shape
   • Compute: ReLU(-3)=0, ReLU(0)=0, ReLU(5)=5
   • Advantages: sparsity, vanishing gradient mitigation
5. 7 Marks
   aExplain the role of back propagation in training a multilayer network with weight adjustments in layers.
   How to answer

   • Define forward pass and loss in multilayer network
   • Explain gradient computation via chain rule backward
   • Weight update with gradient descent across layers
   • Role in credit assignment and deep learning
6. 7 Marks
   bExplain the working of batch normalization. Write its benefits and usage areas.
   How to answer

   • Define batch normalization: normalize mini-batch activations
   • Steps: mean, variance, normalize, scale-shift with gamma-beta
   • Benefits: faster training, stability, regularization
   • Usage: CNNs, deep networks before activation
7. 7 Marks
   aExplain how max pooling reduces dimensionality. Given a $4\times4$ feature map. $$\begin{bmatrix} 1 & 2 & 3 & 0 \\ 4 & 5 & 6 & 1 \\ 7 & 8 & 9 & 2 \\ 0 & 1 & 2 & 3 \end{bmatrix}$$ Apply $2\times2$ max pooling with stride 2 and show the resulting matrix.
   How to answer

   • Define max pooling: window, stride, max operation
   • Show 4x4 to 2x2 reduction with 2x2 stride 2
   • Compute resulting matrix [[5,6],[8,9]]
   • Explain dimensionality, translation invariance, computation saving
8. 7 Marks
   bWhat is the inception network architecture and Explain how inception modules use multiple kernel sizes simultaneously.
   How to answer

   • Define Inception / GoogLeNet architecture
   • Describe inception module parallel branches
   • Explain 1x1, 3x3, 5x5 and pooling with 1x1 bottleneck
   • Benefits: multi-scale features, parameter efficiency
9. 7 Marks
   aExplain the implementation of CNN in Tensor flow.
   How to answer

   • Outline TensorFlow workflow: data pipeline, layers
   • Define CNN layers: Conv2D, pooling, dense
   • Training: compile, loss, optimizer, fit
   • Evaluation and example code structure
10. 7 Marks
    bExplain in detail about Recurrent Neural Network.
    How to answer

    • Define RNN and recurrent connections, unfolding in time
    • Explain hidden state, BPTT, sequence handling
    • Variants: LSTM, GRU for long dependencies
    • Applications in text, speech, time-series
11. 7 Marks
    aExplain the role of beam width in beam search and how it affects performance.
    How to answer

    • Define beam search and beam width k
    • Effect of small vs large width on accuracy and cost
    • Trade-off: greedy vs exhaustive search
    • Role in seq2seq decoding
12. 7 Marks
    bExplain Q-learning. Write the differences between Q-learning and SARSA.
    How to answer

    • Define Q-learning: off-policy TD update formula
    • Explain Q-table, epsilon-greedy, Bellman update
    • Compare with SARSA: off-policy vs on-policy
    • Tabulate differences in update, exploration, convergence
13. 7 Marks
    aDiscuss in detail about Bayesian learning.
    How to answer

    • Define Bayesian learning: prior, likelihood, posterior
    • Explain Bayes theorem and MAP/MLE
    • Parameter estimation under uncertainty
    • Advantages and applications
14. 7 Marks
    bWrite the applications of machine learning in speech processing.
    How to answer

    • List speech tasks: recognition, speaker ID, emotion
    • ML models: HMM, DNN, RNN, transformers
    • Feature extraction: MFCC, spectrograms
    • Examples: assistants, transcription
15. 14 Marks
    Write short notes on any two of the following

    1. a)Natural language processing
    2. b)Actor - critic model
    3. c)Tuning hyper parameters.
    4. d)Supervised and unsupervised learning
    How to answer

    • Select any two topics for short notes
    • Define each with key principles and diagram/flow
    • Mention algorithms, advantages and use-cases
    • Conclude with applications