AL-405 Machine Learning - Jun 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
   aDefine Machine Learning. Explain how Machine Learning differs from traditional programming paradigms. Illustrate your answer with examples.
   How to answer

   • Definition of Machine Learning
   • Difference from traditional programming with data vs rules flow
   • Examples e.g., spam filter, prediction vs explicit code
   • Illustration of learning from data
2. 7 Marks
   bDefine unsupervised learning. Explain its significance and challenges. Discuss different approaches to clustering and dimensionality reduction, citing examples of common algorithms used.
   How to answer

   • Definition of unsupervised learning
   • Significance and challenges like no labels, evaluation
   • Clustering approaches e.g., k-means, hierarchical
   • Dimensionality reduction e.g., PCA with examples
3. 7 Marks
   aWhat is a neural network? Explain how a neural network is represented mathematically. Describe the role of weights, biases, and activation functions in the working of a neural network.
   How to answer

   • Definition of neural network and neuron model
   • Mathematical representation $y = f(Wx+b)$
   • Role of weights, biases
   • Role of activation functions
4. 7 Marks
   bWhat is a Multilayer Perceptron (MLP)? Describe its architecture and how it differs from a single-layer perceptron. Explain how hidden layers contribute to solving complex problems.
   How to answer

   • Definition and architecture of MLP
   • Difference from single-layer perceptron
   • Role of hidden layers and non-linearity
   • Ability to solve complex non-linear problems e.g., XOR
5. 7 Marks
   aExplain the concept of Support Vector Machines (SVM) for classification. What is the role of margin maximization? Discuss the importance of kernel tricks and explain different types of kernels used in SVM.
   How to answer

   • Concept of SVM and maximum margin hyperplane
   • Role of margin maximization and support vectors
   • Kernel trick for non-linear data
   • Types of kernels: linear, polynomial, RBF, sigmoid
6. 7 Marks
   bWhat is Logistic Regression? Explain how logistic regression is used for binary classification problems. Derive the logistic function and explain how the cost function is optimized during training.
   How to answer

   • Definition of logistic regression for binary classification
   • Derivation of sigmoid/logistic function $\sigma(z)=1/(1+e^{-z})$
   • Cost function: binary cross-entropy
   • Optimization via gradient descent during training
7. 7 Marks
   aExplain the k-means clustering algorithm in detail. Describe the steps involved, the objective function it minimizes, and the limitations of k-means. How does the choice of k affect the outcome?
   How to answer

   • k-means algorithm steps: initialize, assign, update
   • Objective function: within-cluster sum of squares
   • Effect of choice of k
   • Limitations: initialization, spherical clusters, outliers
8. 7 Marks
   bExplain the difference between hard clustering and soft clustering. Which clustering algorithms produce hard clusters and which produce soft clusters? Provide examples and use cases for each.
   How to answer

   • Definition of hard vs soft clustering
   • Hard clustering algorithms e.g., k-means; soft e.g., GMM, fuzzy c-means
   • Examples and use cases for each
   • Comparison of membership assignment
9. 7 Marks
   aWhat are the key guidelines to follow when designing machine learning experiments? Discuss the importance of reproducibility, dataset partitioning, and the control of confounding variables. How do these guidelines ensure the validity and generalizability of experimental results?
   How to answer

   • Guidelines for ML experiments: reproducibility, partitioning
   • Importance of dataset splits train/validation/test
   • Control of confounding variables
   • How guidelines ensure validity and generalizability
10. 7 Marks
    bWhy is it important to compare machine learning models over multiple datasets? Discuss the benefits and challenges of cross-dataset validation. How can the results of comparisons over multiple datasets provide more robust insights into model performance?
    How to answer

    • Need for comparison over multiple datasets
    • Benefits: robustness, generalizability; challenges: heterogeneity, cost
    • Cross-dataset validation concept
    • How multi-dataset results give robust insights
11. 7 Marks
    aExplain the concept of Principal Components Analysis (PCA). Derive the mathematical formulation of PCA. How does PCA help in dimensionality reduction and visualization?
    How to answer

    • Concept of PCA and variance maximization
    • Mathematical formulation: covariance, eigenvectors/eigenvalues
    • Dimensionality reduction procedure
    • Use in visualization and feature compression
12. 7 Marks
    bDescribe the gradient descent algorithm. Compare its different variants like Stochastic Gradient Descent (SGD), Mini-batch Gradient Descent, Momentum, RMSProp, and Adam optimizers.
    How to answer

    • Gradient descent update rule
    • Variants: SGD, Mini-batch, Momentum, RMSProp, Adam
    • Comparison of convergence, stability, computation
    • Use cases for each optimizer
13. 7 Marks
    aExplain the bias-variance tradeoff in supervised learning. How do different supervised learning techniques like decision trees and random forests handle bias and variance?
    How to answer

    • Definition of bias and variance
    • Bias-variance tradeoff explanation
    • Decision trees: low bias high variance
    • Random forests: bagging reduces variance
14. 7 Marks
    bHow can Genetic Algorithms be applied to optimize clustering solutions? Discuss chromosome representation, fitness function design, and genetic operations in the context of clustering.
    How to answer

    • Use of genetic algorithms for clustering optimization
    • Chromosome representation of cluster centers
    • Fitness function design e.g., intra-cluster distance
    • Genetic operators: selection, crossover, mutation
15. 7 Marks
    aDiscuss how to evaluate the stability of a machine learning model. What metrics or methods can be used to assess how consistent a model's performance is across different datasets, environments, or perturbations? How does model stability impact its generalization ability?
    How to answer

    • Concept of model stability
    • Metrics: variance across folds, datasets, perturbations
    • Consistency across environments assessment
    • Link between stability and generalization ability
16. 7 Marks
    bDiscuss the trade-off between bias and variance when designing machine learning experiments. How can experimental design mitigate the effects of high bias or high variance? Explain how this trade-off is related to overfitting and underfitting in model evaluation.
    How to answer

    • Bias vs variance tradeoff in experimental design
    • Mitigating high bias and high variance
    • Relation to overfitting and underfitting
    • Design choices: complexity, regularization, data size