Unit 1:

1. Define machine learning.

Machine learning is a field of artificial intelligence that focuses on the development of
algorithms and statistical models that enable computers to learn and make predictions or
decisions without being explicitly programmed. It involves the study of patterns and data,
allowing machines to automatically improve their performance and accuracy over time
through experience and exposure to new information.

2. List some applications of machine learning.

1. Autonomous Driving
2. Personalized Recommendations
3. Medical Diagnosis
4. Fraud Detection
5. Language Translation
6. Virtual Assistants

3. What are major components involved in each machine algorithm?

1. Data
2. Features
3. Model
4. Training
5. Evaluation
6. Optimization
7. Prediction/Inference
8. Deployment

4. List the limitations of Decision tree.

1. Overfitting
2. High variance
3. Instability
4. Lack of global optimization
5. Handling missing data
6. Lack of robustness

5. Summarize algorithm technique in ML.

1. Decision Trees
2. K-Nearest Neighbors (KNN)
3. Naive Bayes
4. Neural Networks
5. Clustering Algorithms
6. Dimensionality Reduction
6. Compare learning vs programming.

Learning Programming
Learning refers to the process of acquiring Programming, on the other hand, is the
knowledge, skills, or understanding through process of creating a set of instructions or
study, experience, or teaching. code that a computer can execute to perform
specific tasks.
It involves gathering information, processing It involves writing algorithms, coding, and
it, and gaining insights. developing software applications.
Learning is focused on gaining knowledge Programming is focused on solving
and understanding a particular subject or problems and creating software solutions
topic. using programming languages.
It can encompass various domains such as It involves logical thinking, problem-solving
academic subjects, practical skills, skills, and attention to detail.
languages, etc.

7. Write the limitation of Find-S algorithm.

The limitation of the Find-S algorithm:
 Limited to hypotheses with a single most specific hypothesis.
 Inability to handle noisy or inconsistent data
 Lack of generalization
 Dependency on the order of training examples
 Limited representation power
 Absence of negative examples

8. List different forms of learning.

1. Supervised Learning
2. Unsupervised Learning
3. Semi-Supervised Learning
4. Reinforcement Learning

9. How can you avoid overfitting?

1. Increase Training Data
2. Feature Selection
3. Cross-Validation
4. Early Stopping
5. Simplify the Model
6. Dropout
 10. Define Entropy.
Entropy is a measure of the impurity or uncertainty in a set of data. It quantifies the
unpredictability or disorder of information. Mathematically, entropy is calculated using the
formula:
Entropy(S) = -Σ (p(i) * log2(p(i)))

Unit 2:
1. State perceptron training rule.
The perceptron training rule is a simple algorithm used to train a perceptron, a type of
artificial neural network. The perceptron training rule adjusts the weights and bias to
minimize the error between the perceptron's output and the desired output, allowing it to
learn and make accurate predictions.

2. Why XOR problem could not be solved by simple perceptron?

The XOR problem cannot be solved by a simple perceptron because it is not linearly
separable. A simple perceptron can only learn linearly separable patterns, which means it
can only classify data points that can be separated by a single straight line or hyperplane.

3. Define Confidence Interval.

A confidence interval is a range of values that is used to estimate an unknown population
parameter, such as the mean or proportion. It provides a level of uncertainty or margin of
error associated with the estimate. The confidence interval is typically expressed as a
range with an associated confidence level, which represents the probability that the true
parameter value falls within the interval.

4. Define Variance.
Variance is the measure of spread in data from its mean position. In machine learning
variance is the amount by which the performance of a predictive model changes when it
is trained on different subsets of the training data.

Variance= E[(Y-E[Y])^2]
5. What is sample error?
Sample error is the difference between the results obtained from a sample and the true
value that would be obtained if the entire population were measured.
 6. What is the inductive bias of Candidate Elimination algorithm?
The inductive bias of the Candidate Elimination algorithm is to prefer the most specific
hypothesis consistent with the training data while still allowing for a potential
generalization to new, unseen examples.

7. State delta rule.

The delta rule, also known as the Widrow-Hoff rule or the incremental gradient descent
rule, is a learning algorithm used to update the weights of a neural network. It adjusts the
weights based on the difference between the predicted output and the desired output,
aiming to minimize the error.
8. What are the factors affecting the performance of machine learning algorithm?

1. Data Quality
2. Feature Selection
3. Model Complexity
4. Training Data Size
5. Computational Resources
6. Noise and Outliers
7. Algorithm Selection

9. What is estimated value?

The estimated value is used to represent an unknown or unobserved quantity of interest
based on the information available from a sample or population. It provides an estimate
or best guess of the value, taking into account the available data and the assumptions
made during the estimation process.

10. How to compute expected value and variance of a random variable?

Expected Value:
 Multiply each possible outcome of the random variable by its corresponding
probability.
 Sum up all the products obtained in the previous step.

Variance:
 Calculate the expected value of the random variable.
 Subtract the expected value from each possible outcome, square the differences.
 Multiply each squared difference by its corresponding probability.
 Sum up all the products obtained in the previous step.

Unit 3:
1. State Bayes theorem.
Bayes' theorem is a fundamental principle in machine learning that describes the
relationship between conditional probabilities. In a concise form, Bayes' theorem
states:
P(A|B) = (P(B|A) * P(A)) / P(B)
Here, P(A|B) represents the probability of event A occurring given that event B has
already occurred. P(B|A) is the probability of event B occurring given that event A
has already occurred. P(A) and P(B) are the probabilities of events A and B occurring
independently of each other.

2. What is prior probability?

Prior probability refers to the initial or existing belief in the likelihood of an event
occurring, before any specific evidence or data is taken into account. It represents our
initial subjective assessment or knowledge about the probability of an event. The
prior probability serves as a starting point or baseline for making predictions or
updating our beliefs using methods such as Bayes' theorem.

3. Define PAC learnability.

PAC (Probably Approximately Correct) learnability refers to the ability of a machine
learning algorithm to learn a concept or hypothesis that is probably correct and
approximately accurate, with controlled errors.

4. State the limitations of KNN algorithm.

Limitations of KNN Algorithm:
1. Computational Complexity
2. Memory Usage
3. Optimal K Selection
4. Imbalanced Data
5. Feature Scaling
6. Curse of Dimensionality
5. What is perspective in machine learning?
In machine learning, perspective refers to a specific viewpoint or approach taken
when designing and analyzing algorithms, models, or datasets. It involves considering
different angles, contexts, or considerations based on the goals, requirements, or
biases involved in the machine learning process.

6. Write the approaches for instance based learning.

1. K-Nearest Neighbors (KNN)
2. Case-Based Reasoning (CBR)
3. Locally Weighted Learning (LWL)
4. Memory-Based Reasoning (MBR)

7. Define maximum likelihood estimation (MLE).

 Maximum Likelihood Estimation (MLE) is a statistical method used to estimate the
parameters of a probability distribution or statistical model. The goal of MLE is to
find the values of the model parameters that maximize the likelihood function, which
measures the probability of observing the given data.
8. Distinguish Joint probability vs conditional probability.
s.no Joint probability Conditional probability
1 Joint probability refers to Conditional probability
the probability of two or refers to the probability of
more events occurring an event occurring given
simultaneously or in that another event has
conjunction with each already occurred.
other.
2 It represents the likelihood It represents the likelihood
of the intersection of of one event happening
multiple events happening given the information
together. about another event.
3 Joint probability is Conditional probability is
calculated by multiplying calculated by dividing the
the probabilities of the joint probability of two
individual events. events by the probability of
the condition.
4 For example, if events A For example, the
and B are independent, the conditional probability of
joint probability of A and B event A given event B is
occurring is P(A and B) = P(A|B) = P(A and B) /
P(A) * P(B). P(B).

9. What are benefits of K- NN algorithm?

1. Simplicity
2. Versatility
3. Non-parametric
4. Instance-Based Learning
5. Robust to Outliers
6. Interpretability
10. What is meant by Bayesian belief network (BBN)?
A Bayesian Belief Network (BBN), also known as a Bayesian Network or
Probabilistic Graphical Model, is a graphical representation of probabilistic
relationships among variables. It uses Bayesian probability theory to model and infer
dependencies and uncertainties in a system.
Unit 4:
1. What is Genetic Algorithm?
 Genetic Algorithm (GA) is a search and optimization algorithm inspired by the
principles of natural evolution and genetics. It is used to find optimal or near-
optimal solutions to complex problems where traditional methods may be
impractical.

2. Define fitness function in GA

A fitness function in Genetic Algorithms (GA) is a measure used to evaluate the
quality of an individual solution within a population. It quantifies how well a
solution performs in solving the problem and guides the selection process by
assigning higher fitness values to more desirable solutions.

3. Give the limitations of GA.

1. Premature Convergence
2. Computational Complexity
3. Tuning Parameters
4. Representation Dependency
5. Scalability
6. Difficulty with Continuous Search Spaces

4. Give list of ensemble methods.

1. Bagging
2. Boosting
3. Random Forest
4. Stacking
5. Gradient Boosting

5. How we can do crossover between multiple parents in genetic algorithm?

Crossover between multiple parents in a genetic algorithm can be achieved
through methods such as
 Simulated Binary Crossover (SBX),
 Weighted Crossover,
 Arithmetic Crossover,
 Heuristic-Based Crossover, or
 Multi-parent Crossover.
These techniques enable the combination of genetic information from
multiple parents to create new offspring.

6. Mention the terminology used in first order rules

Predicate, Constants, Variables, Functions, Quantifiers, Connectives, Atomic
Formula, First-Order Rule, Knowledge Base, Inference.
 7. List some attractive features to formulating the learning task.
Clear Objective, Measurable Metrics, Relevant Features, Representative Data,
Well-Defined Input/Output, Scalability, Generalizability, Interpretability,
Real-World Relevance, Ethical Considerations
8. Contrast of Practical Difficulties in Formulating the Learning Task:
 Lack of data
 Ambiguity in objectives
 Incomplete or noisy data
 Feature selection and representation
 Complexity and dimensionality
 Trade-off between model complexity and interpretability
 Computational resources
9. Define reinforcement learning.
Reinforcement learning is a type of machine learning where an agent learns by
trial and error to make decisions that maximize rewards in a given environment.
Reinforcement learning is commonly used in scenarios where explicit training
data or a pre-defined set of actions is not available, and the agent must learn
through continuous interaction with the environment.
10. What is Q-learning.
Q-learning is a popular reinforcement learning algorithm used for learning
optimal actions in a Markov decision process (MDP) setting. It involves an agent
that learns through trial and error to make decisions based on the maximum
expected cumulative reward, represented by a Q-value. The agent iteratively
updates its Q-values based on the rewards received and uses these values to guide
its actions in different states, gradually converging towards an optimal policy.

Unit 5:
1. Define analytical learning.
Analytical learning, also known as analytical modeling or analytical machine learning,
refers to a type of learning approach that focuses on deriving insights and making
predictions from data through the application of statistical and mathematical models. It
 involves analyzing the available data, identifying patterns, and building mathematical
models to understand relationships and make predictions or decisions.

2. What do you mean by explanation based learning?

Explanation-based learning is a machine learning approach where new knowledge is
acquired by using pre-existing knowledge to explain and generalize from specific
instances. It involves analyzing and understanding the explanations or justifications for
certain examples or observations, extracting general rules or concepts, and applying them
to new situations.

3. List the learning rules of prolog-EBG.

1. general-to-specific
2. specific-to-general
3. dual incremental
4. selective dual incremental

4. Mention the different perspectives on EBL.

1. Cognitive Perspective
2. Knowledge Representation Perspective
3. Problem-Solving Perspective
4. Inductive Inference Perspective
5. Automated Reasoning Perspective
6. Knowledge Transfer Perspective

5. Define SOAR.
SOAR, which stands for "State, Operator, And Result," is a cognitive architecture and
problem-solving framework designed to model human intelligence and decision-making
processes. It combines aspects of rule-based systems, artificial intelligence, and cognitive
psychology. SOAR represents knowledge in the form of rules and employs a production
system to guide its problem-solving and decision-making activities.

6. Give the motivation of Inductive learning.

The motivation behind inductive learning is to develop a learning approach that can
generalize from specific examples and make accurate predictions or decisions on unseen
data. The goal is to extract patterns, regularities, and underlying relationships from the
given data to form generalized rules, models, or hypotheses.

7. Distinguish between inductive and analytical learning methods.

s.no Inductive learning Analytical learning
1 These rely on training data these use prior knowledge to
to learn patterns and do not guide learning, allowing for
require explicit prior more accurate
knowledge. generalization from less
 data.

2. They may fail with They can be misled if the

insufficient training data prior knowledge is incorrect
or insufficient.
3. Inductive learning involves Analytical learning involves
generalizing from specific examining and analyzing
instances or examples to existing knowledge or
form broader conclusions or information to extract
general rules. insights or draw
conclusions.
4. Inductive learning is data- Analytical learning is
driven, as it relies heavily knowledge-driven, as it
on observed data or relies on existing knowledge
examples to induce general or information to gain
rules or hypotheses. insights or draw
conclusions.

8. What is deductive learning?

Deductive learning is a learning approach where knowledge or conclusions are derived
from general principles or rules and then applied to specific instances. Deductive learning
typically involves using deductive logic or formal reasoning methods to make inferences
and reach logical conclusions based on the given premises or rules.

9. Define inductive bias in EBL.

Inductive bias in Explanation-Based Learning (EBL) refers to the inherent assumptions
or preferences that guide the learning process and influence the types of generalizations
or hypotheses that are considered. It represents the prior knowledge or assumptions that
shape the learning algorithm's behavior and restrict the space of possible generalizations.

10. Which algorithm is used for knowledge level learning?

The algorithm commonly used for knowledge level learning is the Explanation-Based
Generalization (EBG) algorithm. EBG is specifically designed to learn high-level,
abstract knowledge or rules from individual instances by using explanations. It focuses
on generalizing from specific examples and extracting general rules or concepts that can
be applied to new situations.