Sl. No.

Questions
1. Which technique is used in Unsupervised Learning to reduce dimensionality?
(a) Linear Regression (b) clustering
(c) Principal Component Analysis (d)Decision trees
2. Which method is primarily used in Unsupervised Learning?
(a) Support Vector Machine (b) K-means Clustering
(c) Decision Trees (d) Linear Regression
3. Early machine learning models were limited by ________.
(a) computational power (b) training datasets
(c) statistical methods (d) neural networks
4. Which algorithm is used to predict the value of a dependent variable based on one or more
independent variables?
(a)Support Vector Machine (b) K-means Clustering
(c) Decision Trees (d) Linear Regression
5. What is one of the key features of static learning?
(a) Adapts to new data automatically
(b) Requires real-time data processing
(c) Assumes patterns in data do not change over time
(d) Operates with evolving datasets
6. What is the main objective of reinforcement learning in machine learning?
(a) Identifying clusters in data (b) Minimizing prediction errors
(c) Maximizing cumulative rewards (d) Creating decision boundaries
7. Hyper parameter tuning involves optimizing parameters that are not directly learned from
the data during _____.
(a) data pre processing (b) feature engineering
(c) model training (d) model evaluation
8. Which algorithm is best suited for binary classification tasks?
(a) Support Vector Machine (b) K-means Clustering
(c) Decision Trees (d) Linear Regression
9. K-Means Clustering is an example of a _____ learning algorithm.
(a) supervised (b) semi supervised
(c) unsupervised (d) Reinforcement Learning
10. What component of a general machine learning architecture is responsible for learning
complex patterns from data?
(a) data pre processing (b) feature engineering
(c) model training (d) model evaluation
11. In supervised learning, the algorithm learns from a labelled _____ set.
(a) testing (b)validation
( c) training (d) evaluation
12. A self-driving car needs to navigate through traffic and make decisions in real-time to reach
its destination safely. Which machine learning technique would be most appropriate for
training the car to learn optimal driving behaviours through trial and error?
(a) supervised (b) semi supervised
(c) unsupervised (d) Reinforcement Learning
13. A company is developing a recommendation system for its e-commerce platform. Which
machine learning technique would be most suitable for training the recommendation system
to predict user preferences based on past behaviour?
(a) supervised (b) semi supervised
(c) unsupervised (d) Reinforcement Learning
14 Your machine learning model performs exceptionally well on the training set but shows
poor performance on the testing set. What machine learning activity should be prioritized
to address this issue?
(a) Feature Engineering (b)Hyper parameter tuning
(c) Cross-validation (d) Regularization
15. Your dataset contains missing values in a crucial feature. How should you handle these
missing values during the data preparation phase?
(a) Ignore the missing values, as they won't impact model performance.
(b) Replace missing values with the mean of the feature.
(c) Remove instances with missing values from the dataset.
(d) Replace missing values with zeros.
16. The term "machine learning" was coined by _____ in the late 1950s.
(a) John McCarthy (b) Arthur Samuel
( c) Marvin Minsky (d) Alan Turing
17. Which learning algorithm is more prone to under fitting?
(a)Support Vector Machine (b) K-means Clustering
(c) Decision Trees (d) Linear Regression
18. What is the primary characteristic of nominal categorical data?
(a) It has a meaningful order among categories.
(b) It can take any numerical value within a range.
(c) It represents categories without any inherent order.
(d) It has a precise numerical relationship between categories.
19. You are building a model to predict the number of customer purchases. What type of
numerical data is this?
(a) Continuous numerical data (b) Discrete Numerical Data
(c) Ordinal data (d) Nominal data
20. Logistic regression uses _____ function or logistic function which is a complex cost
function.
(a) quadratic (b) sigmoid (c) linear (d) lasso

21. What is the primary goal of the Find-S algorithm?

(a) To find the most specific hypothesis
(b) To find the most general hypothesis
(c) To create a decision tree
(d) To optimize the hypothesis space
22. Which of the following is true about the Find-S algorithm?
(a) It starts with the most general hypothesis
(b) It works on both positive and negative examples
(c) It converges on the first hypothesis that matches the data
(d) It starts with the most specific hypothesis and generalizes it
23. Which of the following algorithms uses both positive and negative examples?
(a) Find-S
(b) Candidate Elimination
(c) List-Then-Eliminate
(d) Perceptron
24. In the Candidate Elimination algorithm, which type of example is used to refine the
specific hypothesis, S?
 (a) Positive examples
(b) Negative examples
(c) Mixed examples
(d) Neutral examples

25. In Find-S, if the hypothesis is overly general, what can cause this?
(a) Inconsistent data
(b) Lack of negative examples
(c) Multiple generalizations
(d) Lack of positive examples
26. Which of the following algorithms produces a version space as an output?
(a) Decision Trees
(b) Find-S
(c) Candidate Elimination
(d) k-Nearest Neighbors
27. The process of narrowing down hypotheses until only those consistent with the training
examples remain is called:
(a) Hypothesis Generalization
(b) Hypothesis Restriction
(c) List-Then-Eliminate
(d) Cross-Validation
28. The inductive bias of the Find-S algorithm is that:
(a) The target concept is the most specific consistent hypothesis
(b) The target concept is the most general consistent hypothesis
(c) The target concept can be any consistent hypothesis
(d) It doesn’t make any assumptions about the target concept
29. Which algorithm assumes that the target concept lies within the hypothesis space?
(a) Find-S
(b) Candidate Elimination
(c) List-Then-Eliminate
(d) All of the above
30. When would the Candidate Elimination algorithm remove a hypothesis from the version
space?
(a) When it is inconsistent with a positive example
(b) When it is inconsistent with a negative example
(c) When it is consistent with a positive example
(d) Both (a) and (b)
31. What does inductive bias refer to in machine learning?
(a) The assumption about the solution space
(b) The assumption that allows learning algorithms to generalize
(c) The assumption about noise in the data
(d) The assumption about training time
32. In the Candidate Elimination algorithm, the specific hypothesis S is initialized as:
(a) The most specific hypothesis
(b) The most general hypothesis
(c) A random hypothesis
(d) The first hypothesis in the hypothesis space
33. What is the effect of using a highly restrictive inductive bias in learning algorithms?
(a) It leads to overfitting
 (b) It reduces the hypothesis space
(c) It improves training time
(d) It makes the model more general
34. In the Find-S algorithm, how is the hypothesis updated with each positive example?
(a) By removing inconsistent features
(b) By making it more specific
(c) By making it more general
(d) By completely rewriting the hypothesis
35. List-Then-Eliminate is a process that involves:
(a) Creating all possible hypotheses and filtering out inconsistent ones
(b) Starting with a general hypothesis and making it specific
(c) Choosing a hypothesis based on training time
(d) Randomly selecting a hypothesis from the list
36. The Candidate Elimination algorithm maintains a version space defined by:
(a) The boundary hypotheses in G and S
(b) Only the most specific hypothesis
(c) Only the most general hypothesis
(d) All hypotheses in the space
37. The primary limitation of the Find-S algorithm is:
(a) It is computationally expensive
(b) It only considers positive examples
(c) It uses both positive and negative examples
(d) It does not produce a specific hypothesis
38. What is a key difference between List-Then-Eliminate and Candidate Elimination?
(a) List-Then-Eliminate uses only positive examples
(b) Candidate Elimination uses both positive and negative examples to refine the
hypothesis space
(c) Candidate Elimination is less efficient than List-Then-Eliminate
(d) List-Then-Eliminate does not generate a version space
39. The inductive bias of the Candidate Elimination algorithm assumes that:
(a) The target concept is within the hypothesis space
(b) The hypothesis space is infinite
(c) Only positive examples define the target concept
(d) The target concept is not in the hypothesis space
40. Which of the following best describes deductive learning?
(a) Drawing conclusions based on specific examples to create general rules
(b) Using pre-existing rules and knowledge to make specific predictions
(c) Learning by adjusting hypotheses to fit observed data
(d) Generating new hypotheses without any prior knowledge
Important Questions (10 marks)

1. Explain in detail Supervised, Unsupervised and Reinforcement learning with example.

2. Explain the concept of supervised learning and provide an overview of popular algorithms
used in this paradigm. Discuss their applications in real-world problems.

3. Discuss the learning system in detail and discuss the perspectives and issues in the machine
learning system.

4. Explain the Find-S algorithm in detail. Describe its steps, use cases, and limitations in
machine learning. Provide an example to illustrate how it works.

5. Draw and solve the decision trees for the following set of training examples

6. Assess the Candidate-Elimination algorithm. (ii)Explain the candidate elimination

algorithm. Apply the algorithm to obtain the final version space for the training example.
Sky Humid Temp wind Water Forecas Label
t
1 Sunny Warm Normal Strong Warm Same Y

2 Sunny Warm High Strong Warm Same Y

3 Rainy Cold High Strong Warm Change N

4 Sunny Warm High Strong Cold Change Y

7. Illustrate the List-Then-Eliminate approach in machine learning. Explain how it refines

the hypothesis space and its applications in various learning tasks. Provide a step-by-step
example.
8. Discuss how inductive bias in the Find-S algorithm can lead to either overfitting or
underfitting. How can the choice of inductive bias influence the algorithm’s performance in
practical applications
9. Discuss the concept of version space in the Candidate Elimination algorithm. Explain
how the boundaries (S and G) are used to define the space and how they evolve with
positive and negative examples.
10. Discuss the concept of neural networks in machine learning. Describe the working of a
simple feedforward neural network and its application in tasks such as image recognition
and speech processing.
11. Discuss the unsupervised learning paradigm. Describe the k-means clustering algorithm
and its application in grouping similar data points. Analyze its advantages and limitations.
12. Analyze the trade-offs between inductive bias and deductive bias in machine learning
systems with an example.
 Part-A Course Outcome 4 (CO4)

1. Which of the following best describes a perceptron?

(a) A linear classifier used in supervised learning
(b) A neural network with multiple hidden layers
(c) An unsupervised learning algorithm
(d) A genetic algorithm for optimization
2. What is the primary purpose of the activation function in a neural network?
(a) To initialize weights
(b) To introduce non-linearity into the model
(c) To prevent overfitting
(d) To standardize input data
3. In the backpropagation algorithm, what is the primary goal?
(a) To increase the model's complexity
(b) To minimize the loss function by adjusting weights
(c) To prevent the model from generalizing
(d) To increase the number of layers
4. Which of the following functions is commonly used as an activation function in
neural networks?
(a) Mean Squared Error
(b) Sigmoid
(c) Euclidean Distance
(d) Manhattan Distance
5. In the context of neural networks, what does “gradient descent” refer to?
(a) The process of adding noise to the data
(b) The optimization technique to find the minimum of a function
(c) A method for increasing learning rates
(d) A way to enhance the number of nodes in a layer
6. What is the purpose of the learning rate in gradient descent?
(a) To control the update frequency of weights
(b) To prevent overfitting
(c) To regularize the model
(d) To control the size of weight adjustments
7. Which of the following statements is true about genetic algorithms?
(a) They rely on gradient-based optimization
(b) They mimic natural selection and evolution
(c) They are used only for classification tasks
(d) They cannot be used for continuous data
8. What is the role of mutation in genetic algorithms?
(a) To speed up convergence
(b) To introduce diversity into the population
(c) To minimize the fitness function
(d) To reinforce the current solution
9. In evolutionary models, which of the following is NOT typically a key process?
(a) Selection
(b) Mutation
(c) Crossover
(d) Normalization
10. Which algorithm uses a population of candidate solutions to evolve toward an
optimal solution?
(a) Perceptron
(b) Gradient Descent
(c) Genetic Algorithm
(d) Backpropagation
11. Which of the following neural network architectures is most commonly used for
image recognition?
(a) Recurrent Neural Network (RNN)
(b) Feedforward Neural Network
(c) Convolutional Neural Network (CNN)
(d) Radial Basis Function Network
12. Which type of gradient descent calculates the gradient for the entire dataset
before updating weights?
(a) Stochastic Gradient Descent
(b) Batch Gradient Descent
(c) Mini-Batch Gradient Descent
(d) Accelerated Gradient Descent
13. In the context of neural networks, what is the purpose of dropout
regularization?
(a) To increase the model complexity
(b) To prevent overfitting by randomly ignoring neurons during training
(c) To add layers to the network
(d) To improve backpropagation efficiency
14. What does a sigmoid activation function output?
(a) Values between -1 and 1
(b) Values between 0 and 1
(c) Only binary outputs
(d) Only integer outputs
15. Which of the following is NOT a common application of genetic algorithms?
(a) Optimizing neural network architectures
(b) Feature selection
(c) Linear regression
(d) Game development and AI
16. In backpropagation, which metric is commonly minimized to improve model
accuracy?
(a) Accuracy
(b) Loss function
(c) Activation function
(d) Regularization parameter
17. Which of the following methods is used to prevent gradient vanishing in deep
networks?
(a) Decreasing the learning rate
(b) Using activation functions like ReLU
(c) Removing layers
(d) Increasing weight initialization values
18. What is the primary difference between a single-layer perceptron and a multi-
layer perceptron?
(a) Single-layer perceptrons cannot solve non-linear problems, whereas multi-layer
perceptrons can
 (b) Multi-layer perceptrons are only used for unsupervised learning
(c) Single-layer perceptrons have multiple hidden layers
(d) Single-layer perceptrons use backpropagation
19. Which evolutionary algorithm technique selects the best individuals to create
the next generation?
(a) Crossover
(b) Mutation
(c) Selection
(d) Replacement
20. Which type of neural network is most suitable for processing sequential data,
such as text or time series?
(a) Convolutional Neural Network
(b) Radial Basis Function Network
(c) Feedforward Neural Network
(d) Recurrent Neural Network
21. In machine learning, the process of hypothesis space search helps in finding a
model that _____ well to unseen data.
(a)overfits
(b)underfits
(c)memorizes
(d)generalizes
22. Which technique is commonly used to search through the hypothesis space
efficiently?

(a)Grid search
(b)Random search
(c)Evolutionary algorithms
(d) Gradient descent
23. How does the size of the hypothesis space affect machine learning algorithms?

(a)Larger hypothesis spaces always lead to better performance.

(b)Smaller hypothesis spaces are preferable as they are easier to search.
(c) The size of the hypothesis space does not necessarily correlate with
performance.
(d) Smaller hypothesis spaces always result in overfitting.
24. What is hypothesis space in machine learning?

(a)The space where data points reside

(b)The set of all possible hypotheses that can be learned by a learning algorithm.
(c) The space where training data is generated
(d)The set of features used for training a model
25. The Travelling Salesman Problem is categorized under which class of problems?

(a)Pseudo-polynomial
(b)NP-complete
(c) NP-hard
(d)P-class
26. In genetic algorithms, a sequence of city indices represents a _____ for the TSP.

(a)fitness function
(b)chromosome
(c) gene
(d)crossover point
27. A numerical representation of solutions in a genetic algorithm is often called a
_____?

(a)gene
(b)chromosome
(c) genome
(d)character
28. When utilizing Genetic Algorithms, what is the role of the fitness function in
solving TSP?

(a)It introduces randomness in candidate solutions

(b)It selects initial configurations for routes
(c) It evaluates the quality of solutions based on tour length
(d)It primarily acts as the termination condition
29. What is the main purpose of the input layer in a Multi-Layer Perceptron?

(a)Perform classification
(b)Extract features
(c) Receive raw input data
(d)Calculate gradients
30. Multi-Layer Perceptron is a type of _____ neural network consisting of multiple
layers.

(a)convolutional
(b)recurrent
(c) artificial
(d)modular

Sl.No. Part B Questions

1. Define a neural network and state its basic components.
2. List the primary characteristics of a perceptron as a linear classifier.
3. Compare and contrast the gradient descent and Delta rule
4. Brief Baldwin effect.
5. Distinguish between crossover and mutation.
6. Illustrate the Lamarckian Evolution.
7. Illustrate with a simple example how genetic algorithms use crossover and
mutation.
8. List two differences between gradient descent and genetic algorithms as
optimization methods.
9. Define the fitness function in the context of genetic algorithms and its purpose in
selection.
 10. Distinguish between single-layer and multi-layer perceptrons in terms of problem-
solving capabilities.
11. Identify the purpose of using a bias term in a perceptron model.
12. Name two types of evolutionary operations in genetic algorithms and briefly state
their functions.
13. Outline the main steps in the gradient descent optimization process.
14. Define two common activation functions used in neural networks and their typical
output ranges.
15. State the purpose of the learning rate in gradient descent and its effect on
convergence.

Sl.
Part C Questions
No.
1. Discuss the architecture of a basic neural network. Explain how each layer (input,
hidden, output) functions, and describe the role of weights and biases in the model’s
computations.
2. Explain the working of the perceptron algorithm as a linear classifier with its
limitations.

3. Describe the backpropagation algorithm in neural networks. Outline the steps

involved in calculating gradients and updating weights.
4. Evaluate the impact of learning rate on the performance of gradient descent. Discuss
what can happen if the learning rate is too high or too low, and suggest possible
solutions for adapting the learning rate effectively.
5. Examine the importance of activation functions in neural networks. Compare
different types of activation functions like Sigmoid, ReLU, Tanh and their
respective roles in improving model performance.
6. Illustrate the key operations in a genetic algorithm, including selection, crossover,
and mutation. Explain how these operations contribute to the evolution of solutions
in the search space.
7. Discuss evolutionary models and their application in machine learning.
8. Explain the concept of genetic programming (GP) and how it differs from traditional
genetic algorithms.
9. Discuss the limitations of a single-layer perceptron. Explain why it cannot solve non-
linearly separable problems and how the multi-layer perceptron addresses this
limitation
10. Evaluate the types of problems for which Artificial Neural Network (ANN) learning
is most effective, and outline its key characteristics.

Course Outcome 4 (CO4)

Sl.
Questions
No.
1. Bayes' theorem is used to calculate
(a) Conditional probability
(b) Marginal probability
(c) Joint probability
(d) All of the above
2. In concept learning, Bayes' theorem helps in
(a) Identifying the most likely concept
(b) Predicting future outcomes
(c) Estimating probabilities of observations
(d) Reducing the number of hypotheses
3. The Maximum Likelihood estimation method is used to
(a) Maximize the accuracy of a model
(b) Estimate the parameters that maximize the likelihood of observed data
(c) Minimize the error in a model
(d) Estimate the prior probability
4. The Minimum Description Length (MDL) principle suggests that the best model is
the one that
(a) Maximizes the complexity of the hypothesis
(b) Minimizes the description length of both data and model
(c) Maximizes the likelihood of the data
(d) Minimizes the error rate
5. The MDL principle combines
(a) A trade-off between data fitting and model simplicity
(b) A trade-off between complexity and bias
(c) A trade-off between model complexity and overfitting
(d) A trade-off between model interpretability and performance
6. The Bayes optimal classifier is based on
(a) A decision tree algorithm
(b) A nearest-neighbor approach
(c) The likelihood of different hypotheses
(d) A probabilistic approach to classification
7. Gibbs sampling is used to
(a) Estimate model parameters via sampling
(b) Compute the Bayes optimal classifier
(c) Find the most likely hypothesis in a Bayesian network
(d) Generate random variables for the likelihood computation
8. The Naïve Bayes classifier assumes
(a) The features are dependent on each other
(b) The features are independent of each other
(c) The features are not relevant to the classification task
(d) The classes are linearly separable
9. In Naïve Bayes, the class with the highest probability is
(a) Chosen using a decision tree
(b) Selected using maximum likelihood
(c) The one with the maximum posterior probability
(d) Chosen randomly from the set of possible classes
10. A Bayesian belief network represents
(a) Probabilistic relationships among variables
(b) Linear relationships among variables
(c) A deterministic set of rules
(d) Direct cause-and-effect relationships without uncertainty
11. In a Bayesian belief network, nodes represent
(a) Random variables
(b) Hypotheses
 (c) Observations
(d) Deterministic outputs
12. The Expectation-Maximization (EM) algorithm is used for
(a) Estimating parameters in models with incomplete data
(b) Finding exact solutions in Bayesian networks
(c) Maximizing a model’s likelihood directly
(d) Learning from labeled data only
13. In the EM algorithm, the E-step involves
(a) Maximizing the expected likelihood
(b) Estimating the parameters of the model
(c) Updating the hidden variables given the observed data
(d) Estimating the data distribution
14. The sample complexity in probability learning refers to
(a) The number of iterations required to train the model
(b) The minimum number of samples needed to learn a target concept with high
confidence
(c) The amount of data required for the model to generalize
(d) The number of hidden layers in a neural network
15. Probability learning becomes more efficient when
(a) More training data is available
(b) Fewer hypotheses are used
(c) The model is overfitted
(d) Sample complexity is reduced
16. A finite hypothesis space in machine learning refers to
(a) A limited number of possible models or hypotheses
(b) An infinite number of possible models
(c) A single hypothesis used throughout the learning process
(d) A continuous function that can be adapted
17. The mistake bound space for a hypothesis refers to
(a) The maximum number of errors a classifier can make during training
(b) The minimum number of errors a classifier must make
(c) The total number of possible mistakes a model can make in testing
(d) The average number of mistakes made over multiple trials
18. In the context of Bayes' theorem, the prior probability represents
(a) The probability of the hypothesis before observing the data
(b) The probability of the data given the hypothesis
(c) The probability of observing the data in future experiments
(d) The probability of data and hypothesis being true simultaneously
19. The Naïve Bayes classifier is called "naïve" because it assumes that
(a) Features are dependent on the class label
(b) Features are conditionally independent given the class label
(c) Classes are independent of each other
(d) All features are equally important in determining the class
20. In Maximum Likelihood estimation, the goal is to
(a) Minimize the variance of the model
(b) Find the parameter values that maximize the likelihood of observing the given
data
 (c) Maximize the prior probability
(d) Minimize the bias of the model

21. What is the purpose of applying regularization techniques in the EM algorithm?

(a)To increase the likelihood of overfitting.
(b)To speed up convergence.
(c)To avoid singularity issues and improve stability.
(d)To decrease the likelihood of finding the global minimum.
22. Scenario-based: Given a Bayesian Network for a traffic system, how might new
sensor data be integrated? (a) Update probability
distributions using new data
(b) Ignore new data
(c) Build a new network from
scratch (d) Use
data as supplementary information

23. How does MLE(Maximum Likelihood Estimation) handle overfitting in small

datasets?

(a)By using fewer parameters.

(b)By incorporating regularization.
(c) By ignoring outliers.
(d) It generally tends to overfit due to large variance.
24. Which of the following is an advantage of using Bayesian Networks?

(a)Increased computational complexity

(b)Inefficient handling of incomplete data
(c)Provide interpretable probabilistic reasoning
(d)Unsuitable for probabilistic inference
25. What kind of model would a neural network with real-valued weights be classified
under in terms of hypothesis space?

(a)Finite model
(b)Deterministic model
(c)Infinite model
(d)Discrete model
26. A Bayesian Belief Network utilizes a _____ to define the probability distributions
for each node.

(a)Linear Regression Table

(b)Conditional Probability Table
(c) Decision Tree
(d)Markov Matrix
27. What is inferred by using the Variable Elimination method in Bayesian Networks?

(a)Network Topology
(b)Conditional Probability Tables
(c) Probabilistic Queries
 (d) Directed Edges
28. Which of the following is a limitation of finite hypothesis spaces?

(a)Increased risk of overfitting

(b)May not contain the true function
(c) Requires regularization for effective training
(d)Often requires gradient-based optimization
29. What principle helps the Halving Algorithm reduce its mistake bound during
learning?

(a)Reducing features
(b)Increasing data size
(c) Eliminating inconsistent hypotheses
(d) Random sampling of the data
30. Which of these search algorithms is an example of optimizing a scoring function in
Score-Based Learning for Bayesian Networks?

(a)Simulated Annealing
(b)Variable Elimination
(c) Hill Climbing
(d)Decision Tree

Sl.No. Part B Questions

1. Define Bayes' theorem and explain its importance in concept learning.
2. Identify the difference between prior and posterior probability in Bayes' theorem.
3. State the objective of Maximum Likelihood estimation.
4. Explain the concept of likelihood function in the Maximum Likelihood estimation.
5. Define the Minimum Description Length (MDL) principle and its role in model
selection.
6. State the relationship between MDL and overfitting.
7. List the steps involved in the Bayes optimal classifier.
8. Describe the key assumption made by the Naïve Bayes classifier.
9. Define a Bayesian belief network and its role in representing probabilistic
relationships.
10. List the two main steps of the Expectation-Maximization (EM) algorithm.
11. Identify the purpose of the E-step in the EM algorithm.
12. State the relationship between sample complexity and the amount of training data
in probability learning.
13. Define finite hypothesis space and give an example.
14. Explain the concept of mistake bounds in learning algorithms.
15. State the advantage of using a finite hypothesis space over an infinite one in terms
of learning efficiency.

Sl.
Part C Questions
No.
1. Explain Bayes theorem with an Example.
2. Describe how Bayes Theorem can be used to update the probability of a hypothesis
based on new evidence in a concept learning context.

3. Explain how Maximum Likelihood Estimation apply to parameter estimation in

probabilistic models with an example.
4. Discuss how Minimum Description Length (MDL) Principle is applied in model
selection.
5. Describe the Gibbs Algorithm and its significance in relation to the Bayes Optimal
Classifier.
6. Demonstrate the working of the Naïve Bayes Classifier, including its assumptions,
using a simple example
7. Explain the process of inference in Bayesian Belief Networks, including how
conditional dependencies are represented and utilized.
8. Describe the Expectation-Maximization (EM) Algorithm, including its steps, and
explain its application in clustering
9. Discuss the factors that influence sample complexity and provide examples of how
sample size impacts model accuracy
10. Illustrate the mistake-bound model in learning theory and explain its application in
a finite hypothesis space with an example.
 Course Outcome 6 (CO5)

Sl.
Questions
No.
1. The K-NN algorithm is classified under which type of learning?
(a) Supervised learning
(b) Unsupervised learning
(c) Reinforcement learning
(d) Deep learning
How does K-NN determine the class of a test instance?
2. (a) By averaging all data points
(b) By the majority class among the K-nearest neighbours
(c) By selecting the first instance only
(d) By randomly choosing a class
3. Which of the following metrics is commonly used in K-NN to measure similarity?
(a) Manhattan distance
(b) Euclidean distance
(c) Jaccard index
(d) Cross-entropy
What is the primary disadvantage of using K-NN in high-dimensional data?
4. (a) It requires a labeled dataset
(b) It has a long training phase
(c) Distance measurements become less meaningful
(d) It has low interpretability
5. In weighted regression, weights are applied to data points based on their
__________.
(a) Closeness to the origin
(b) Importance or influence
(c) Assigned class labels
(d) Dimension values
6. Which of the following is often used to define weights in weighted regression?
(a) Mean of each feature
(b) Data distribution density
(c) Distance from a point of interest
(d) Sum of squared errors
7. A radial basis function is primarily used in which type of machine learning model?
(a) Support Vector Machine
(b) Decision Tree
(c) Neural Network
(d) Naïve Bayes Classifier
8. Radial basis functions are based on the distance between:
(a) Data points and a central point
(b) All pairs of data points
(c) Randomly chosen points
(d) Only categorical data
9. Radial basis functions have an advantage over linear functions in that they:
(a) Are faster to compute
(b) Perform well with linear data
 (c) Model non-linear relationships effectively
(d) Reduce data points to clusters
10. In case-based learning, knowledge is represented as:
(a) Rules
(b) Predefined categories
(c) Individual cases or experiences
(d) Decision boundaries
11. Case-based learning is also known as:
(a) Feature learning
(b) Example-based learning
(c) Unsupervised learning
(d) Structured learning
12. Case-based learning is particularly useful when:
(a) There are a few examples available
(b) Data is non-sequential
(c) Data involves frequently recurring scenarios
(d) Supervision is limited
13 The process of solving a problem in case-based learning involves:
(a) Building a new model from scratch
(b) Learning from past experiences or similar cases
(c) Running a deep learning algorithm
(d) Preprocessing data extensively
In a machine learning case study, the primary goal is to:
14. (a) Develop a universally applicable model
(b) Understand a specific problem in detail
(c) Test a new hypothesis
(d) Conduct a random sampling
15. A major advantage of case studies is that they:
(a) Provide quantitative data only
(b) Allow extensive exploration of a specific case
(c) Focus on generalized outcomes
(d) Avoid in-depth analysis
16. A machine learning case study often involves:
(a) Testing a hypothesis without context
(b) Comparing multiple techniques in a real-world scenario
(c) Training on limited data
(d) Avoiding any domain-specific applications
17 The outcomes of a case study are most useful for:
(a) Predicting broad trends
(b) Refining specific approaches for similar problems
(c) Developing large-scale models
(d) Generating large datasets
18 A commonly used radial basis function is:
(a) Linear kernel
(b) Polynomial kernel
(c) Gaussian kernel
(d) Sigmoid function
19 Weighted regression is most suitable for data that exhibits:
(a) Constant variance
(b) Linear relationships only
 (c) Heteroscedasticity
(d) Categorical distributions
20 Which technique can reduce the computational burden in K-NN when working with
large datasets?
(a) Random sampling
(b) Dimensionality reduction
(c) Increasing K value
(d) Increasing dataset size
21. Which of the following should be considered before applying the KNN
algorithm?
(a)Parameter selection
(b)Feature selection and scaling
(c)Curse of Dimensionality
(d)All of the mentioned
22. Parametric machine learning algorithms are often referred to as _____ machine
learning algorithms because they assume a linear combination of input
variables.
(a)combination
(b)linear
(c)simplification
(d)supervised
23. Algorithms that do not make strong assumptions about the form of the
mapping function are called _____ machine learning algorithms.
(a)parametric
(b)nonparametric
(c)functional
(d)structural
24. A Radial Basis Function Network (RBFN) is a type of neural network that
performs classification by measuring the input’s similarity to examples from
the _______.
(a)output layer
(b)training set
(c)hidden layer
(d)test set
25. The radial basis function computes the _____ between the input data and the
center of each radial basis function.
(a)mean
(b)euclidean distance
(c)median
(d)standard deviation
26. Which regression technique builds a model at each query point without a
global assumption about the function form?
(a)Linear Regression
(b)Polynomial Regression
(c)Support Vector Regression
(d)Locally weighted regression
27. Which of the following best describes the cost function minimized in Locally
weighted regression?
(a)Regular Sum of Squared Errors
 (b)Weighted sum of squared errors
(c)Absolute Error
(d)Modified Squared Error
28. The process called _____ in Case based reasoning involves incorporating new
experiences into the case base.
(a)revise
(b)retain
(c)reuse
(d)retrieve
29. Which challenge in Case based reasoning involves changing a past solution to
fit a new scenario?
(a)Retrieval complexity
(b)Adaptation difficulty
(c)Storage limitation
(d)Similarity assessment
30. The process of searching for previously encountered cases in case based
reasoning is known as _____.
(a)retrieve
(b)reuse
(c)revise
(d)retain

Sl.No. Questions
1. Define K-Nearest Neighbor (K-NN) learning.
2. Identify the purpose of the "K" value in the K-NN algorithm.
3. List two common distance metrics used in K-NN.
4. Summarize how Weighted Regression differs from Ordinary Least Squares (OLS)
regression
5. Recall what a radial basis function (RBF) is and its typical application in machine
learning.
6. Identify the primary factor used to assign weights in weighted regression
7. List two types of radial basis functions commonly used in neural networks.
8. Outline the meaning of a "lazy learning" algorithm and give an example.
9. State why K-NN can be computationally intensive for large datasets.
10. List two benefits of using case studies in machine learning applications.
11. Define "Sample Complexity" in probability learning.
12. Summarize how a Case-Based Learning system approaches solving a new
problem
13. Outline the difference between finite and infinite hypothesis spaces in learning
theory.
14. Define the concept of a Bayes optimal classifier.
15. List two factors that impact the sample complexity of a learning algorithm.
Sl.
Questions
No.
1. Explain the K-Nearest Neighbor (K-NN) algorithm, including its working
principles, advantages, and limitations.
2. Describe the process of Weighted Regression, outlining its applications, benefits,
and situations where it’s advantageous over Ordinary Least Squares regression.

3. Compare and contrast Radial Basis Function (RBF) networks with traditional neural
networks. Discuss the advantages of using RBF in specific problem types.
4. Illustrate the working of Case-Based Learning, and provide an example of a
problem where it would be a suitable choice. Discuss the advantages and challenges
of this learning approach.
5. Assess the detail about distance-weighted nearest neighbour algorithm.
6. Examine the Instance-based learning methods.
7. Describe the disadvantages and advantages of Lazy and Eager learning.
8. Explain about the Case-based reasoning (CBR).
9. Examine the role of weight assignment in Weighted Regression models. How does
the choice of weights affect the results of the model?
10. Evaluate the role of case studies in the machine learning process. Provide an example
where a case study helped improve the performance of a machine learning model.

Course Outcome 6 (CO6)

Sl.
Questions
No.
1. The Sequential Covering Algorithm is primarily used for:
(a) Supervised learning
(b) Unsupervised learning
(c) Reinforcement learning
(d) Neural network training
2. In the context of the Sequential Covering Algorithm, a rule is covered when:
(a) It is satisfied by all the training data points
(b) It correctly classifies a subset of data points
(c) It is the most complex rule
(d) It only applies to one data point
3. In the context of the Sequential Covering Algorithm, a rule is covered when:
(a) It is satisfied by all the training data points
(b) It correctly classifies a subset of data points
(c) It is the most complex rule
(d) It only applies to one data point
4. The Sequential Covering Algorithm is typically used for:
(a) Classification tasks
(b) Clustering tasks
 (c) Regression tasks
(d) Anomaly detection tasks
5. Which of the following is a key advantage of the Sequential Covering Algorithm?
(a) It produces a small number of rules
(b) It generates compact decision trees
(c) It can handle large datasets efficiently
(d) It can learn rules incrementally
6. First-order rules in machine learning involve:
(a) Unary predicates
(b) Quantifiers and variables
(c) Only numerical variables
(d) Binary relations
7. Which of the following best describes a first-order rule in logic-based learning?
(a) A rule that uses only constants
(b) A rule that is limited to boolean variables
(c) A rule that uses predicates, quantifiers, and variables
(d) A rule with no variables
8. What is the primary benefit of using sets of first-order rules in machine learning?
(a) They can generalize better to unseen data
(b) They are easier to interpret than other models
(c) They avoid overfitting by using fewer rules
(d) They only work with categorical data
9. Which type of logic is commonly used in first-order rule learning?
(a) Predicate logic
(b) Propositional logic
(c) Linear logic
(d) Fuzzy logic
10. First-order rules are mainly used in:
(a) Neural networks
(b) Decision trees
(c) Inductive logic programming (ILP)
(d) Genetic algorithms
11. Inverting resolution is primarily used to:
(a) Convert a set of rules into a decision tree
(b) Identify a solution by working backward from a goal
(c) Map a continuous function into discrete states
(d) Generate probabilistic models from data
12. Analytical learning is characterized by:
(a) Learning from data and human-provided explanations
(b) Using only data for learning
(c) Using unsupervised techniques for clustering
(d) A purely symbolic learning approach
13. In induction on inverted deduction, the process involves:
(a) Deductively generating rules from data
(b) Generating hypotheses by inverting known conclusions
(c) Using forward chaining to find solutions
(d) Using neural networks for regression tasks
14. The main goal of analytical learning is to:
(a) Improve the accuracy of learning by adding more data
(b) Incorporate human knowledge to refine the learning process
 (c) Minimize the computation cost
(d) Focus on supervised learning
15. The FOCL algorithm is associated with:
(a) Learning from positive and negative examples
(b) Combining logical reasoning with machine learning
(c) Training neural networks for optimization
(d) Genetic programming for rule generation
16. Perfect domain theories in machine learning refer to:
(a) Theories that accurately model every aspect of a problem
(b) Theories that have no exceptions
(c) Theories that can predict future events with high precision
(d) Theories that cover all potential edge cases in a dataset
17. Explanation-based learning involves:
(a) Understanding and explaining the reasoning behind the learning process
(b) Creating a model based purely on data
(c) Learning through direct interaction with an environment
(d) Using labeled data to train a classifier
18. The FOCL algorithm is designed to learn:
(a) Probabilistic models from noisy data
(b) First-order logic rules from examples
(c) Neural network weights using backpropagation
(d) Decision trees for classification tasks
19. Explanation-based learning (EBL) works by:
(a) Using a set of rules to generate new examples
(b) Creating explanations for observed data and refining learning
(c) Enhancing deep learning models through supervised signals
(d) Ignoring domain knowledge and learning purely from data
20. The primary goal of Explanation-Based Learning (EBL) is to:
(a) Generate rules from raw data
(b) Use domain-specific explanations to improve learning efficiency
(c) Build complex models with deep learning techniques
(d) Train a model without any prior knowledge of the domain
21. Consider a scenario where a machine learning model has learned the following
general rule: "If a fruit is red and round, then it is likely to be an apple." This is
an example of:
(a)Deduction
(b)Induction
(c)Abduction
(d)Reduction
22. In sequential covering algorithms, a new rule is generated by selecting the
__________ that maximally covers a subset of the remaining examples.
(a)best attribute
(b)highest support
(c)lowest confidence
(d)random attribute
23. What happens if a sequential covering algorithm encounters instances not
covered by any rule?
(a)The algorithm terminates with incomplete coverage
(b)New rules are generated to cover the remaining instances
 (c)The algorithm restarts from scratch
(d)The uncovered instances are assigned to the most similar rule
24. Learning sets of rules involves discovering _____ relationships between features
and outcomes in the data.
(a)linear
(b)causal
(c)probabilistic
(d)logical
25. How does EBL handle new instances not covered by existing explanations?
(a)It ignores new instances
(b)It adapts existing explanations to fit new instances
(c)It discards existing explanations
(d)It relies on reinforcement learning
26. Which algorithm is typically used for classifying emails into spam or not spam
in Inductive Learning?
(a)Naïve Bayes
(b)Explanation-Based Learning
(c)Theorem Proving
(d)Logic Programming
27. Which learning paradigm is best suited for handling real-world noisy data,
such as in image recognition?
(a)Analytical Learning
(b)Symbolic Reasoning
(c)Inductive Learning
(d)Hybrid Learning
28. Which algorithm in reinforcement learning is specifically used for robotic
control applications?
(a)Q-Learning
(b)Reinforce
(c)Deep Q-Network
(d)Temporal Difference Learning
29. Which key feature of FOCL allows it to adapt when domain knowledge is
incorrect?
(a)Deductive reasoning
(b)Data augmentation
(c)Inductive learning
(d)Feature elimination
30. Which feature allows Q-Learning to learn the optimal policy regardless of the
agent's actions?
(a)On-policy learning
(b)Off-policy learning
(c)Model-based learning
(d)Episodic learning

Sl.No. Questions
1. Define the Sequential Covering Algorithm and its purpose in rule learning.
2. Identify the process used by the Sequential Covering Algorithm to generate rules
from data.
3. Explain what a first-order rule is and how it differs from propositional logic rules.
 4. Compare the use of first-order rules in logic-based learning and propositional logic
in machine learning.
5. Describe the concept of inverted resolution in machine learning.
6. Differentiate between analytical learning and traditional inductive learning
methods.
7. Summarize what is meant by a perfect domain theory in the context of machine
learning.
8.
Clarify the role of Explanation-Based Learning (EBL) in improving learning
efficiency.

9. Distinguish between Q-Learning and Temporal Difference Learning in

reinforcement learning.
10. Define the key components of a reinforcement learning task.
11. Evaluate the advantages of the Sequential Covering Algorithm over other rule-
based learning methods.
12. Explain how rule coverage is determined in the Sequential Covering Algorithm.
13. Illustrate the purpose of using first-order rules in machine learning.

14. Discuss how sets of first-order rules contribute to generalization in machine

learning models.
15. Examine how inverting resolution helps in hypothesis generation in machine
learning.

Sl.
Questions
No.
1. Explain the Sequential Covering Algorithm and its significance in learning sets of
rules. Evaluate its advantages and limitations in rule generation with an example.
2. Analyze the process of rule generation in the Sequential Covering Algorithm.
Compare its performance with other rule-learning algorithms, such as decision trees.

3. Define first-order rules in the context of machine learning. Illustrate how they differ
from propositional rules and discuss their advantages in terms of generalization.
4. Explain the concept of induction on inverted deduction and its application in
learning tasks. Discuss the benefits of inverting resolution in hypothesis generation,
and compare it to traditional resolution methods.
5. Describe what is meant by a perfect domain theory in machine learning. Discuss its
significance in improving model accuracy and handling complex domains, with
examples.
6. Explain the Explanation-Based Learning (EBL) approach and its role in optimizing
the learning process.
7. Define the core components of a reinforcement learning task. Analyze how Q-
Learning and Temporal Difference Learning differ in terms of learning strategies,
and evaluate which approach is more effective for various types of problems.
8. Discuss the primary steps involved in the Sequential Covering Algorithm. Evaluate
how it handles rule conflicts and incomplete rule coverage. Provide examples to
illustrate its effectiveness in classification tasks.
9. Illustrate the process of learning first-order logic rules in an inductive logic
programming (ILP) framework. Discuss the trade-offs between expressiveness and
efficiency when using first-order rules in machine learning.
10. Examine the role of inverted deduction in hypothesis generation within the context
of machine learning. Analyze the application of this method in domains where
logical reasoning is crucial, and evaluate its effectiveness in improving model
accuracy.