AL-405 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
   aHow does supervised learning differ from unsupervised learning? Discuss.
   How to answer

   • Define supervised learning with labelled data and unsupervised learning with unlabelled data
   • Compare on objective, feedback, algorithms and output
   • Give examples e.g. classification/regression vs clustering/association
   • Discuss applications and when to use each
2. 7 Marks
   bBriefly explain the need of Inductive Bias in decision Tree Learning.
   How to answer

   • Define inductive bias in decision tree learning
   • Explain need to generalize beyond training data and prefer simpler trees
   • Mention Occam's razor, preference for shorter trees, handling unseen instances
   • Briefly link to overfitting avoidance
3. 7 Marks
   aWhat is Principal Component Analysis (PCA) in machine learning? How does PCA help in reducing the dimensionality of data?
   How to answer

   • Define PCA as unsupervised dimensionality reduction technique
   • Outline steps: standardization, covariance, eigenvectors/eigenvalues, projection
   • Explain variance retention and selection of principal components
   • Mention benefits: visualization, noise reduction, efficiency
4. 7 Marks
   bEvaluate the effectiveness of biologically inspired neural network architectures in solving complex machine learning tasks.
   How to answer

   • Define biologically inspired architectures e.g. ANN, CNN, SNN
   • Discuss strengths in pattern recognition, vision, speech, complex tasks
   • Evaluate limitations: data, compute, interpretability
   • Conclude on effectiveness with examples
5. 7 Marks
   aWhat is the back propagation algorithm in neural networks? How does it compute gradients to update the weights of a neural network?
   How to answer

   • Define backpropagation as gradient-based weight update algorithm
   • Explain forward pass, loss computation, chain rule backward pass
   • Describe gradient descent weight update rule
   • Mention role of learning rate and activation derivatives
6. 7 Marks
   bDiscuss how the Multilayer Perceptron algorithm compute outputs from inputs through its interconnected layers?
   How to answer

   • Define MLP structure: input, hidden, output layers
   • Explain weighted sum plus activation at each neuron
   • Describe forward propagation layer by layer to output
   • Mention activation functions and fully connected nature
7. 7 Marks
   aWhat is Naive Bayes classification in machine learning? Describe the fundamental principle behind Naive Bayes and how it calculates probabilities.
   How to answer

   • State Bayes theorem and conditional independence assumption
   • Explain fundamental Naive Bayes principle
   • Show probability calculation: prior, likelihood, posterior
   • Give example of classification decision rule
8. 7 Marks
   bDiscuss about linear regression and derive the Individual error and Minimization functions.
   How to answer

   • Define linear regression model $y = w^T x + b$
   • Define individual squared error and overall cost function
   • Derive minimization via least squares / gradient descent
   • State normal equation or update rule
9. 7 Marks
   aDescribe the basic concept of a Random Forest and how it combines multiple decision trees for classification.
   How to answer

   • Define Random Forest as ensemble of decision trees
   • Explain bagging and random feature selection
   • Describe majority voting / averaging for classification
   • Mention reduction of overfitting and robustness
10. 7 Marks
    bList the applications of clustering and identify advantages and disadvantages of clustering algorithm.
    How to answer

    • List clustering applications e.g. segmentation, anomaly detection
    • Explain advantages: discovery of structure, no labels needed
    • Explain disadvantages: sensitivity to k, scaling, interpretation
    • Conclude with suitability conditions
11. 7 Marks
    aExplain the Expectation-Maximization (EM) algorithm used to train Gaussian Mixture Models.
    How to answer

    • Define GMM and latent variables
    • Explain E-step: compute responsibilities
    • Explain M-step: update means, covariances, mixing coefficients
    • Mention iteration to convergence
12. 7 Marks
    bWhat are evolutionary optimization techniques in machine learning? Provide an overview of common evolutionary algorithms used for optimization tasks.
    How to answer

    • Define evolutionary optimization inspired by natural evolution
    • Overview common algorithms: GA, PSO, DE, ES
    • Explain selection, crossover, mutation operators
    • Mention use in hyperparameter and non-differentiable optimization
13. 7 Marks
    aDescribe key elements of experimental design such as hypothesis formulation, variable manipulation, and control.
    How to answer

    • Define experimental design and hypothesis formulation
    • Explain variable manipulation: independent/dependent variables
    • Explain control, randomization, replication
    • Mention validity and bias control
14. 7 Marks
    bWhat is the significance of comparing machine learning models over multiple datasets?
    How to answer

    • Define need for evaluation over multiple datasets
    • Explain generalization, bias avoidance, statistical significance
    • Mention cross-dataset comparison and benchmarking
    • Conclude on robust model selection
15. 14 Marks
    Write a short note on any two of the following:

    1. a)Shrinkage Methods
    2. b)Vanishing and Exploding Gradients
    3. c)Issues in decision trees
    4. d)Hypothesis testing
    How to answer

    • Select any two topics with definition and key points
    • Explain core concept, formulas or steps for each
    • Give advantages, issues and applications
    • Conclude with suitable diagrams/examples if needed