AD-502 Machine Learning - Nov 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. 7 Marks
   aExplain the importance of continuous model monitoring and retraining in the life cycle.
   How to answer

   • Definition of model monitoring and retraining in ML life cycle
   • Need due to data drift, concept drift and performance degradation
   • Monitoring metrics and triggers for retraining
   • Benefits: sustained accuracy, reliability and adaptability
2. 7 Marks
   bExplain the concept of clustering in machine learning. How does it differ from other tasks like classification and regression?
   How to answer

   • Definition of clustering as unsupervised learning
   • Objective: grouping similar instances without labels
   • Classification vs clustering: supervised vs unsupervised, labeled data
   • Regression vs clustering: prediction of continuous value vs grouping
3. 7 Marks
   aExplain the concept of logistic regression in classification. How does it model the probability of class membership?
   How to answer

   • Logistic regression for binary/multiclass classification
   • Linear combination followed by sigmoid/logistic function
   • Output as probability of class membership between 0 and 1
   • Decision threshold and cost function
4. 7 Marks
   bExplain the concept of ensemble learning in machine learning. What is the underlying idea behind ensemble methods?
   How to answer

   • Definition of ensemble learning
   • Combining multiple weak learners to improve performance
   • Reduction of variance, bias and overfitting
   • Examples: bagging, boosting, stacking, voting
5. 7 Marks
   aDefine principal components in the context of PCA. How are they calculated from the original data?
   How to answer

   • Definition of principal components in PCA
   • Linear combinations capturing maximum variance, orthogonal
   • Computation: centering, covariance matrix, eigenvectors/eigenvalues
   • Selection of top-k components for dimensionality reduction
6. 7 Marks
   bDescribe the stages of the machine learning life cycle, from data collection to model deployment.
   How to answer

   • Stages: data collection, preprocessing, feature engineering
   • Model training, evaluation and validation
   • Deployment, monitoring and maintenance
   • Iterative nature of life cycle
7. 7 Marks
   aExplain the concept of distribution model-based clustering. What are Gaussian Mixture Models (GMMs), and how are they used in this context?
   How to answer

   • Distribution-based clustering: clusters as probability distributions
   • Gaussian Mixture Models as mixture of Gaussians with weights
   • Parameters: mean, covariance, mixing coefficients
   • EM algorithm for fitting and soft assignment
8. 7 Marks
   bDescribe decision tree classification and how it constructs a tree-based model for classification tasks.
   How to answer

   • Decision tree structure: nodes, branches, leaves
   • Recursive splitting using impurity measures like Gini, entropy
   • Tree construction top-down for classification
   • Prediction by traversing from root to leaf
9. 7 Marks
   aExplain the averaging technique in ensemble learning. How does it combine predictions from multiple models?
   How to answer

   • Averaging as ensemble combination for regression/soft voting
   • Simple vs weighted averaging of predictions
   • Reduction of variance and improved stability
   • Difference from voting for classification
10. 7 Marks
    bDifferentiate between projection-based dimensionality reduction and manifold learning. What are their primary objectives?
    How to answer

    • Projection-based: linear/nonlinear projection preserving variance/structure
    • Manifold learning: preserving nonlinear manifold geometry
    • Objectives: PCA vs Isomap, t-SNE, LLE
    • Comparison of assumptions and applications
11. 7 Marks
    aDefine supervised learning and give examples of tasks that can be solved using this approach.
    How to answer

    • Definition: learning from labeled input-output pairs
    • Training to map inputs to outputs, loss minimization
    • Examples: classification, regression, spam detection
    • Contrast with unsupervised learning
12. 7 Marks
    bDescribe hierarchical clustering and its ability to create hierarchical structures of clusters. How does agglomerative hierarchical clustering work?
    How to answer

    • Hierarchical clustering builds tree of nested clusters/dendrogram
    • Agglomerative bottom-up: each point as cluster then merge
    • Linkage criteria and distance measures
    • Divisive vs agglomerative approach
13. 7 Marks
    aDiscuss the advantages and disadvantages of decision trees in handling complex decision boundaries.
    How to answer

    • Advantages: interpretable, handles nonlinear boundaries, mixed data
    • Disadvantages: overfitting, instability, greedy splits
    • Effect of depth on complex boundaries
    • Mitigation via pruning and ensembles
14. 7 Marks
    bDescribe the max voting technique in ensemble learning. How does it work in the context of classification?
    How to answer

    • Max voting / majority voting in classification ensembles
    • Each classifier votes for a class, majority wins
    • Use with diverse classifiers for robustness
    • Example and handling of ties
15. 7 Marks
    aDiscuss the significance of the VC dimension in understanding the generalization ability of learning algorithms.
    How to answer

    • VC dimension as capacity measure of hypothesis class
    • Relation to shattering and generalization bounds
    • Higher VC dimension: complex model, overfitting risk
    • Use in model selection and SRM
16. 7 Marks
    bExplain bagging and pasting as ensemble techniques. What are the key differences between them?
    How to answer

    • Bagging: bootstrap sampling with replacement
    • Pasting: sampling without replacement
    • Effect on diversity and variance reduction
    • Key differences in sampling and performance