About this previous year paper
SECTION A
(2 × 10 = 20 marks | Short Answers)
a) AI, ML and Deep Learning
Artificial Intelligence (AI): Broad field where machines perform tasks that require human intelligence.
Machine Learning (ML): Subset of AI where machines learn patterns from data.
Deep Learning: Subset of ML using multi-layer neural networks to learn complex patterns.
b) Types of Machine Learning
Supervised Learning: Uses labeled data (e.g., house price prediction).
Unsupervised Learning: Finds patterns in unlabeled data (e.g., clustering customers).
Reinforcement Learning: Learns by reward and penalty (e.g., game playing).
c) Single learning rate for all features
Using one learning rate may cause slow convergence for some features and overshooting for others, especially when features have different scales.
d) Training vs Testing data
Training data: Used to train the model.
Testing data: Used to evaluate model performance on unseen data.
e) Dimensionality reduction
It reduces the number of input features while retaining important information, improving efficiency and reducing overfitting (e.g., PCA).
f) Hierarchical clustering
A clustering method that builds a hierarchy of clusters using agglomerative or divisive approaches, represented by a dendrogram.
g) Loss function
A loss function measures the difference between predicted output and actual output, guiding model optimization (e.g., MSE, cross-entropy).
h) Overfitting
Overfitting occurs when a model performs well on training data but poorly on new data due to excessive complexity.
i) Genetic algorithm for scheduling
Steps: Initialize population
Evaluate fitness Selection
Crossover Mutation
Termination Used to find optimal schedules efficiently.
j) Kernel in Gaussian processes
A kernel defines the similarity between data points, controlling smoothness and structure of functions in Gaussian processes.
SECTION B
(Attempt any 3 | 10 marks each)
a) Artificial Intelligence and its industrial impact
AI improves automation, decision-making, quality control, predictive maintenance, robotics, and supply chain optimization across industries.
b) Support Vector Machines (SVM)
SVM finds an optimal hyperplane that maximizes the margin between classes. Kernels allow it to handle non-linear data.
c) Expectation-Maximization (EM) algorithm
EM is an iterative algorithm used to estimate parameters in probabilistic models with latent variables through:
Expectation step Maximization step
d) Challenges of Backpropagation
Vanishing gradients Slow convergence
Local minima High computational cost
e) Hidden Markov Models (HMM)
HMMs are probabilistic models for sequential data.
Applications: speech recognition, bioinformatics, time-series prediction.
SECTION C
Q3 (Attempt any one)
a) Steps in designing a learning system
Problem definition Data collection
Feature selection Model selection
Training Evaluation
Deployment
b) ML applications in mechanical engineering
Predictive maintenance Manufacturing defect detection
Energy optimization Robotics and automation
Q4 (Attempt any one)
a) Bayesian Decision Theory
It provides a probabilistic framework for classification by minimizing expected risk using prior probabilities and likelihoods.
b) Bias and variance
Bias: Error due to overly simple model Variance: Error due to overly complex model
Trade-off affects generalization.
Q5 (Attempt any one)
a) Principal Component Analysis (PCA)
PCA transforms correlated features into uncorrelated components, reducing dimensionality while preserving variance.
b) Clustering vs Classification Clustering: Unsupervised grouping
Classification: Supervised labeling
Q6 (Attempt any one)
a) Receptive field in CNN
The receptive field is the region of input influencing a neuron. It depends on kernel size, stride, and number of layers.
b) Kernel function in SVM
Kernel functions map data into higher-dimensional space for separation.
Examples: Linear kernel – text classification
Polynomial kernel – image recognition RBF kernel – non-linear problems
Q7 (Attempt any one)
a) Bayesian estimation vs MLE
MLE: Uses data only, simple but overfits
Bayesian: Uses prior knowledge, more robust but computationally heavy
b) Reinforcement learning vs Deep learning
Reinforcement Learning: Learns via rewards Deep Learning: Learns hierarchical features
RL focuses on decision-making, DL on representation learning.