Unit-I: Introduction to Artificial Intelligence

1. Define Artificial Intelligence (AI). Explain its importance and role in real-world applications.

Definition of AI:

Artificial Intelligence (AI) is the field of computer science that focuses on creating machines capable of mimicking human intelligence. It involves programming computers to perform tasks that typically require human cognition, such as learning, reasoning, problem-solving, perception, and language understanding.

Importance of AI:

AI is important because it enhances efficiency, reduces human error, and enables automation in various sectors. Some key benefits include:

Faster decision-making: AI systems can analyze large datasets quickly.
Automation of repetitive tasks: AI reduces manual effort in industries like manufacturing and customer service.
Enhanced accuracy: AI-powered systems improve precision in medical diagnostics and financial forecasting.

Real-World Applications of AI:

Healthcare: AI-powered diagnostic tools, robotic surgery, and drug discovery.
Finance: Fraud detection, algorithmic trading, and risk assessment.
Manufacturing: Predictive maintenance, quality control, and automation.
Education: AI-driven personalized learning, automated grading, and virtual tutors.
Transportation: Autonomous vehicles, smart traffic management, and logistics optimization.

2. Describe the evolution of AI. How has AI developed over the years?

Early History of AI (1950s-1970s):

1950: Alan Turing introduced the Turing Test to evaluate machine intelligence.
1956: John McCarthy coined the term Artificial Intelligence at the Dartmouth Conference.
1960s-1970s: AI research focused on symbolic reasoning and early expert systems.

AI Winters (1970s & 1980s):

Due to lack of computing power and funding cuts, AI research slowed down.

Revival and Machine Learning Era (1990s-2000s):

AI improved with neural networks, genetic algorithms, and expert systems.
IBM's Deep Blue defeated world chess champion Garry Kasparov in 1997.

Modern AI (2010-Present):

Deep learning breakthroughs (e.g., Google’s AlphaGo defeating human Go players).
NLP advancements (e.g., ChatGPT, Google Assistant).
Self-driving cars, robotics, and AI-powered automation in industries.

Future of AI:

AI is expected to revolutionize healthcare, finance, defense, and space exploration.

3. What is an intelligent agent? Explain its components and different types of agents with examples.

Definition of an Intelligent Agent:

An intelligent agent (IA) is an entity that perceives its environment and takes actions to achieve a goal. It operates autonomously, adapts to changing environments, and optimizes its performance.

Components of an Intelligent Agent:

Perception: Sensors collect data from the environment (e.g., cameras, microphones).
Reasoning: AI algorithms process and analyze the data.
Action: The agent makes decisions and takes actions using actuators.
Learning: The agent improves performance over time through experience.

Types of Intelligent Agents:

Simple Reflex Agents: Respond based on predefined rules (e.g., thermostat).
Model-Based Agents: Maintain internal memory to track past actions (e.g., chess programs).
Goal-Based Agents: Act to achieve a specific goal (e.g., GPS navigation systems).
Utility-Based Agents: Optimize performance based on utility values (e.g., AI in stock trading).
Learning Agents: Improve over time using machine learning (e.g., self-driving cars).

4. Explain the role of AI in mechanical engineering. How can AI be applied in design, manufacturing, and maintenance?

Role of AI in Mechanical Engineering:

AI plays a crucial role in automating tasks, optimizing designs, and improving efficiency in mechanical engineering fields.

Applications in Design:

AI-powered CAD (Computer-Aided Design): AI enhances 3D modeling and simulation.
Generative Design: AI generates optimized design structures based on constraints.
Finite Element Analysis (FEA): AI improves accuracy in stress and thermal simulations.

Applications in Manufacturing:

Smart Robotics: AI-driven robots handle assembly lines and welding.
Quality Control: AI detects defects in production lines using image processing.
Predictive Maintenance: AI predicts machine failures before breakdowns.

Applications in Maintenance:

Fault Detection: AI monitors vibration, heat, and noise patterns for equipment health.
Automated Repairs: AI-powered maintenance scheduling reduces downtime.

Impact of AI in Mechanical Engineering:

Reduces cost and improves efficiency.
Minimizes human errors and optimizes production.

5. Differentiate between strong AI and weak AI. Provide real-world examples for both.

Aspect	Strong AI	Weak AI
Definition	AI that possesses human-like cognition and consciousness.	AI that performs specific tasks without real understanding.
Capabilities	Can reason, think, and make independent decisions.	Works based on predefined rules and algorithms.
Examples	- Hypothetical AI (e.g., AI with self-awareness).	- Virtual Assistants (e.g., Alexa, Siri).
Real-World Use	- Future AI with emotions and consciousness.	- AI used in recommendation systems, image recognition, chatbots.

6. What are the different types of AI techniques? Explain machine learning, expert systems, and deep learning.

1. Machine Learning (ML):

Uses algorithms to analyze data and improve performance.
Example: Spam email detection, self-driving cars, medical diagnosis.

2. Expert Systems:

AI that mimics human experts in a field.
Uses a knowledge base and inference engine.
Example: MYCIN (medical diagnosis), DENDRAL (chemical analysis).

3. Deep Learning:

A subset of machine learning using neural networks.
Example: Image recognition, speech processing (Google Translate, Face ID).

7. Define rationality in AI. How does rational decision-making apply to AI agents?

Definition of Rationality in AI:

Rationality in AI refers to making the best possible decision based on available data.
A rational agent takes actions that maximize performance and minimize risks.

Application in AI Agents:

Autonomous Vehicles: AI makes real-time driving decisions based on road conditions.
Medical Diagnosis AI: AI selects the best treatment by analyzing patient data.
Stock Market Prediction: AI optimizes trading decisions based on financial patterns.

8. Explain the concept of an agent environment. What are the different types of environments in AI?

Definition of Agent Environment:

The agent environment is where an AI interacts, perceives, and takes actions.

Types of Environments:

Fully Observable vs. Partially Observable:
- Fully Observable: AI gets complete data (e.g., chess game).
- Partially Observable: AI gets limited data (e.g., poker).
Deterministic vs. Stochastic:
- Deterministic: Predictable outcomes (e.g., tic-tac-toe).
- Stochastic: Random events affect decisions (e.g., weather forecasting).
Episodic vs. Sequential:
- Episodic: Past actions don’t affect the future (e.g., spam filtering).
- Sequential: Past decisions influence outcomes (e.g., driving a car).

Unit-II: Knowledge Representation & Reasoning questions.

1. What are Knowledge-Based Agents? How do they use logic to represent knowledge?

Definition of Knowledge-Based Agents:

A knowledge-based agent (KBA) is an AI system that stores, retrieves, and applies knowledge to make decisions. These agents rely on logic and reasoning to solve problems instead of just reacting to stimuli.

How KBAs Use Logic to Represent Knowledge:

Knowledge Base (KB): A repository of facts, rules, and heuristics.
Inference Engine: Applies logic to deduce new information.
Decision-Making: Uses logical reasoning to select the best action.

Example:

A medical diagnosis system (like MYCIN) stores symptoms and diseases in a knowledge base and reasons using inference rules to diagnose patients.

2. Explain Propositional Logic. What are its advantages and limitations?

Definition of Propositional Logic:

Propositional logic is a formal system of logic where statements (propositions) are either true or false. It uses logical operators such as:

AND (∧) → (e.g., "It is raining AND it is cold").
OR (∨) → (e.g., "It is raining OR it is sunny").
NOT (¬) → (e.g., "It is NOT raining").

Advantages:

✔ Simple to understand and implement.
✔ Works well for small-scale logical problems.
✔ Foundation for First-Order Logic (FOL).

Limitations:

✖ Cannot represent relationships between objects.
✖ Limited in handling complex real-world scenarios.
✖ Cannot infer new facts unless explicitly stated.

3. Describe First-Order Logic (FOL). How does it improve upon Propositional Logic?

Definition of First-Order Logic (FOL):

First-Order Logic (FOL) extends Propositional Logic by introducing variables, functions, and quantifiers, allowing for more expressive reasoning.

Components of FOL:

Constants: Represent specific objects (e.g., "John", "Apple").
Variables: Represent general objects (e.g., x, y).
Predicates: Describe relationships (e.g., "Loves(John, Mary)").
Quantifiers:
- Universal (∀): "For all" (e.g., ∀x Loves(x, Chocolate)).
- Existential (∃): "There exists" (e.g., ∃x Likes(x, Pizza)).

How FOL Improves Upon Propositional Logic:

Can represent relationships between objects.
Allows for generalization using quantifiers.
Enables reasoning and inference beyond explicit facts.

4. What is the Wumpus World Problem? How does it demonstrate logical reasoning?

Wumpus World Problem:

The Wumpus World is a grid-based AI environment where an agent navigates a cave to find gold while avoiding dangers (like the Wumpus monster and pits).

Logical Reasoning in Wumpus World:

Percepts (inputs): The agent perceives stench (Wumpus nearby), breeze (pit nearby), glitter (gold).
Inference: Based on logic, the agent deduces safe or dangerous cells.
Decision-making: The agent chooses an optimal path to reach gold.

Example:

If (Breeze at (2,2)), then pit might be at (1,2) or (2,3).
If (Stench at (3,1)), then Wumpus is at (4,1) or (3,2).

This demonstrates how AI uses logical reasoning to solve problems in an uncertain environment.

5. Compare Propositional Logic and First-Order Logic. Which one is more useful for AI applications?

Feature	Propositional Logic (PL)	First-Order Logic (FOL)
Representation	Statements are true/false	Includes objects, predicates, and quantifiers
Expressiveness	Limited	More expressive
Inference	Simple logical rules	Complex reasoning possible
Scalability	Works for small problems	Better for real-world AI
AI Usage	Basic rule-based systems	Used in expert systems, NLP, robotics

Which is More Useful?

FOL is more useful for AI applications as it enables complex reasoning and generalization beyond specific facts.

6. Explain the process of inference in First-Order Logic. How does unification work?

Inference in FOL:

Inference in FOL involves deducing new facts from known facts using logical rules such as:

Modus Ponens: If A → B and A is true, then B is also true.
Generalization & Specialization: Using universal (∀) and existential (∃) quantifiers.

What is Unification?

Unification is the process of matching variables in logical expressions to find commonality.

Example of Unification:

Given:

Loves(John, x) → "John loves someone".
Loves(John, Mary) → "John loves Mary".

The unifier is x = Mary, meaning John loves Mary.

7. What is the significance of pattern representation in AI? How is it used for decision-making?

Definition:

Pattern representation in AI refers to structuring data in a meaningful way to recognize relationships and make predictions.

Importance in AI:

Helps in image recognition (face detection, handwriting recognition).
Used in speech processing (Google Assistant, Siri).
Enables fraud detection in banking.

Pattern-Based Decision-Making Example:

AI in spam detection analyzes email patterns (subject, sender, words).
If patterns match known spam emails, the AI classifies the email as spam.

8. What is a logical agent? How does it represent knowledge and make inferences?

Definition:

A logical agent is an AI system that uses formal logic to reason about its environment and take decisions.

How It Represents Knowledge:

Uses propositional and first-order logic to encode facts.
Stores rules in a knowledge base.

How It Makes Inferences:

Deduction: If "All birds can fly" and "Sparrow is a bird," it infers "Sparrow can fly".
Resolution: Uses contradictions to derive new facts.
Backward Chaining: Starts from a goal and works backward to find supporting facts.

Example:

If AI knows "If it rains, roads are wet" and it detects "roads are wet", it infers that it rained.

9. Explain forward chaining and backward chaining in reasoning. Provide real-world examples.

Feature	Forward Chaining	Backward Chaining
Approach	Starts with known facts and applies rules to reach conclusions	Starts with a goal and works backward to find supporting evidence
Usage	Used in diagnostic systems, expert systems	Used in goal-based AI, planning
Example	AI in medical diagnosis starts with symptoms → finds possible diseases	AI in robotics starts with a goal (pick object) → finds required actions

Unit-III: Bayesian & Computational Learning questions.

1. What is Bayes' Theorem? Explain its significance in AI and Machine Learning.

Bayes' Theorem Definition:

Bayes' Theorem is a mathematical formula used to determine conditional probability, i.e., the probability of an event occurring based on prior knowledge of related events.

Formula:

$P(A | B) = \frac{P(B | A) P(A)}{P(B)}$

Where:

$P(A | B)$ = Probability of event A happening given that B has occurred.
$P(B | A)$ = Probability of event B happening given A has occurred.
$P(A)$ = Prior probability of event A.
$P(B)$ = Total probability of event B.

Significance in AI & Machine Learning:

Used in Naïve Bayes Classifier for spam detection, sentiment analysis, and medical diagnosis.
Improves decision-making by updating predictions as new data arrives.
Used in Bayesian Networks for probabilistic reasoning in AI.
Helps in fraud detection by identifying unusual patterns in financial transactions.

Example:

If a person has flu-like symptoms, Bayes' Theorem can determine the probability that they have the flu given prior flu statistics and the frequency of symptoms in non-flu cases.

2. Explain the concept of Maximum Likelihood Estimation (MLE). How is it used in AI?

Definition:

MLE is a statistical method used to estimate the parameters of a probability distribution by maximizing the likelihood function.

Formula:

$\theta_{MLE} = \arg\max_{\theta} P(D | \theta)$

Where:

$\theta$ = Model parameters
$P(D | \theta)$ = Probability of data given the parameters

Application in AI & Machine Learning:

Used in Supervised Learning to optimize model parameters (e.g., weights in neural networks).
Speech recognition systems use MLE to model phoneme probabilities.
Used in Logistic Regression to find the best classification boundary.
Image processing algorithms use MLE for object detection and recognition.

3. What is the Naïve Bayes Classifier? Explain how it is applied in spam detection.

Definition of Naïve Bayes Classifier:

Naïve Bayes is a probabilistic machine learning algorithm that assumes all features are independent given the class label.

Formula:

$P(C | X) = \frac{P(X | C) P(C)}{P(X)}$

Where:

$P(C | X)$ = Probability of class C given feature X.
$P(X | C)$ = Probability of feature X given class C.
$P(C)$ = Prior probability of class C.
$P(X)$ = Probability of feature X.

Application in Spam Detection:

Feature Extraction: Identifies words like "free," "win," "offer" in emails.
Training Model: Uses labeled emails (spam & non-spam) to calculate probabilities.
Prediction: If $P(spam | words) > P(non-spam | words)$ , the email is classified as spam.

Example:

If an email contains "Congratulations! You won a free iPhone," the classifier calculates spam probability.
If the probability is high, the email is marked as spam.

4. Differentiate between generative and discriminative models. Provide examples.

Feature	Generative Model	Discriminative Model
Definition	Models the distribution of input features and generates new data points.	Models the decision boundary between classes.
Objective	Learns **P(X	Y)**, i.e., how data is generated given a class label.
Examples	Naïve Bayes, Hidden Markov Models, GANs.	Logistic Regression, SVM, Neural Networks.
Use Case	Text generation, image synthesis, anomaly detection.	Classification, spam filtering, object detection.

Example:

Generative Model: GANs (Generative Adversarial Networks) can create realistic human faces.
Discriminative Model: Logistic Regression classifies an email as spam or non-spam.

5. What is Instance-Based Learning? Explain the working of K-Nearest Neighbor (KNN) Algorithm.

Definition:

Instance-based learning stores training examples and classifies new instances based on similarity to past cases.

Working of K-Nearest Neighbor (KNN) Algorithm:

Store Training Data → KNN keeps all labeled examples in memory.
Calculate Distance → Measures similarity between a new input and stored examples using Euclidean distance: $d = \sqrt{(x_2 - x_1)^2 + (y_2 - y_1)^2}$
Find Nearest Neighbors → Selects k nearest points to the new instance.
Assign Class → Majority voting determines the class of the new data point.

Example:

If we want to classify a new flower based on petal length and width, KNN checks the closest k flowers and assigns the majority class.

6. Compare and contrast supervised learning and unsupervised learning. Provide use cases.

Feature	Supervised Learning	Unsupervised Learning
Definition	Learns from labeled data.	Learns from unlabeled data.
Objective	Predicts an output (classification/regression).	Finds hidden patterns and structures.
Examples	Decision Trees, SVM, Neural Networks.	K-Means Clustering, PCA, GANs.
Use Case	Spam filtering, medical diagnosis, sentiment analysis.	Customer segmentation, anomaly detection.

Example:

Supervised: Predicting house prices using past sales data.
Unsupervised: Grouping customers based on purchasing behavior.

7. What is the Minimum Description Length (MDL) Principle? How does it help in model selection?

Definition:

MDL is a principle in information theory that selects the best model by minimizing the total description length (model complexity + data encoding).

Formula:

$L(H) + L(D | H)$

Where:

$L(H)$ = Complexity of the model.
$L(D | H)$ = Complexity of the data given the model.

Use in AI & ML:

Prevents Overfitting: Avoids too complex models that fit noise.
Feature Selection: Chooses the best subset of features.
Used in Decision Trees & Bayesian Networks to optimize models.

8. Explain the concept of Gibbs Sampling. How is it used in AI models?

Definition:

Gibbs Sampling is a Monte Carlo Markov Chain (MCMC) algorithm used to generate samples from a probability distribution when direct sampling is difficult.

Steps in Gibbs Sampling:

Initialize a random starting point.
Sample one variable at a time while keeping others fixed.
Repeat the process until convergence.

Applications in AI:

Natural Language Processing (NLP): Topic modeling (e.g., Latent Dirichlet Allocation).
Bayesian Networks: Probabilistic reasoning in AI.
Image Restoration: Removing noise from images.

Unit-IV: Supervised & Unsupervised Learning questions.

1. Explain the working of a Decision Tree Algorithm. How is it used for classification?

Definition of Decision Tree:

A Decision Tree is a machine learning algorithm used for classification and regression. It splits data into branches based on feature values, forming a tree-like structure where each internal node represents a decision rule, and leaves represent final outcomes (classes).

Steps in Decision Tree Algorithm:

Select the Best Feature: The feature that provides the most information gain is chosen first.
Split Data Based on the Feature: The dataset is divided into subsets based on feature values.
Repeat the Process: Further split each subset until a stopping criterion is met (e.g., pure class labels).
Make Predictions: For a new data point, traverse the tree from root to leaf.

Example:

If we classify whether a person will buy a car based on income and age, a decision tree might look like:

            Income?
           /       \
        High       Low
       /    \      /   \
   Age<30  Age>30 Buy  No Buy

Use in Classification:

Spam Detection: Classifies emails as spam or not spam.
Medical Diagnosis: Determines if a patient has a disease based on symptoms.
Loan Approval: Classifies applicants as approved or rejected.

2. What is Support Vector Machine (SVM)? Explain soft margin and hard margin classification.

Definition of SVM:

SVM is a supervised learning algorithm that finds the optimal decision boundary (hyperplane) that best separates different classes in a dataset.

Soft Margin vs. Hard Margin Classification:

Feature	Hard Margin SVM	Soft Margin SVM
Definition	A strict classifier that perfectly separates classes.	Allows some misclassification to handle noisy data.
Use Case	When data is perfectly separable.	When data has some overlap or noise.
Regularization Parameter (C)	Not used (strict rule).	Used to control misclassification.
Example	Handwritten digit recognition (ideal cases).	Email classification (spam vs. non-spam).

Example:

Hard Margin: Used when classes are perfectly separable, like classifying apples and oranges.
Soft Margin: Used when there is some overlap, like customer segmentation based on spending patterns.

3. Explain distance-based learning. How is it different from rule-based learning?

Definition of Distance-Based Learning:

Distance-based learning uses mathematical distances between data points to classify new data. K-Nearest Neighbors (KNN) is the most common algorithm.

Difference Between Distance-Based and Rule-Based Learning:

Feature	Distance-Based Learning	Rule-Based Learning
Working	Classifies based on similarity to existing data points.	Uses explicit rules defined by humans.
Algorithm Examples	KNN, K-Means, SVM.	Decision Trees, Expert Systems.
Flexibility	Can handle new patterns dynamically.	Requires manual updating of rules.
Use Case	Image recognition, recommender systems.	Fraud detection, medical diagnosis.

Example:

Distance-Based: KNN classifies a flower based on closest flower species.
Rule-Based: If-Else rules decide whether a person gets a loan approval.

4. How does K-Means Clustering work? Explain its steps with a real-world example.

Definition:

K-Means is an unsupervised learning algorithm that groups data into K clusters based on feature similarity.

Steps in K-Means Algorithm:

Select K (number of clusters).
Randomly initialize K cluster centers.
Assign each data point to the nearest cluster.
Recalculate cluster centers (centroids).
Repeat until cluster centers no longer change.

Example - Customer Segmentation:

A bank uses K-Means to group customers based on spending behavior:

Cluster 1: Low spenders
Cluster 2: Medium spenders
Cluster 3: High spenders

By analyzing these clusters, the bank creates personalized marketing strategies.

5. What is Principal Component Analysis (PCA)? How does it help in dimensionality reduction?

Definition of PCA:

PCA is a mathematical technique that reduces high-dimensional data into fewer dimensions while retaining the most important information.

How PCA Works:

Standardize Data: Normalize feature values.
Compute Covariance Matrix: Understand feature relationships.
Calculate Eigenvalues & Eigenvectors: Identify principal components.
Select Top Components: Keep only the most significant ones.

How PCA Helps in Dimensionality Reduction:

Reduces computational cost in ML models.
Removes redundancy and noise in data.
Helps visualize high-dimensional data in 2D/3D graphs.

Example - Image Compression:

PCA reduces image size by removing less important pixel information while keeping main details.

6. Compare Decision Trees and Random Forest Algorithms. Which one is better?

Feature	Decision Tree	Random Forest
Definition	Single tree that splits data based on rules.	Uses multiple decision trees (ensemble).
Accuracy	Prone to overfitting.	More accurate and robust.
Computational Complexity	Faster for small datasets.	Slower but handles large datasets better.
Best For	Simple problems (loan approval).	Complex problems (fraud detection).

Which is Better?

Random Forest is better for complex problems as it reduces overfitting and improves accuracy.

7. What is Kernel Trick in SVM? How does it help in handling non-linearly separable data?

Definition:

The Kernel Trick allows SVM to map input data to a higher-dimensional space where a linear decision boundary can separate classes.

Why is it Needed?

Some data is not linearly separable in its original form. The Kernel Trick helps transform such data into a higher-dimensional space where separation becomes possible.

Types of Kernel Functions:

Linear Kernel: Used for simple linearly separable data.
Polynomial Kernel: Allows curved decision boundaries.
Radial Basis Function (RBF) Kernel: Best for complex, non-linear problems.

Example:

SVM with an RBF kernel can classify images of cats vs. dogs by mapping pixel values to a higher-dimensional space.

8. Explain hierarchical clustering. How does it differ from K-Means Clustering?

Definition of Hierarchical Clustering:

A clustering algorithm that builds a tree-like structure (dendrogram) to represent data relationships.

Types:

Agglomerative: Starts with individual points and merges them.
Divisive: Starts with one cluster and splits it into smaller clusters.

Difference from K-Means:

Feature	Hierarchical Clustering	K-Means Clustering
Cluster Structure	Builds a hierarchy (tree).	Creates fixed K clusters.
Scalability	Slower for large datasets.	Faster and works well for large data.
Flexibility	Can generate clusters at different levels.	Must predefine K value.

Example:

Used in biological taxonomy to classify species based on similarities.

9. What is feature engineering? How does it improve machine learning models?

Definition:

Feature engineering is the process of creating, selecting, and transforming features to improve machine learning model performance.

Techniques:

Feature Selection: Removing irrelevant features.
Feature Extraction: Converting raw data into useful features (e.g., PCA).
Feature Scaling: Standardizing values (e.g., Min-Max Scaling).

Example - Predicting House Prices:

Adding "rooms per floor" as a new feature can improve model accuracy.

Unit-V: Advanced Machine Learning & Deep Learning questions.

1. What is ensemble learning? Explain the concepts of Bagging and Boosting.

Definition of Ensemble Learning:

Ensemble learning is a technique in machine learning where multiple models (weak learners) are combined to improve overall prediction accuracy.

Types of Ensemble Learning:

Bagging (Bootstrap Aggregating):
- Trains multiple models independently using random subsets of data.
- Final prediction is made by majority voting (classification) or averaging (regression).
- Example: Random Forest is a bagging-based algorithm.
Boosting:
- Models are trained sequentially, with each model correcting the errors of the previous one.
- Example: AdaBoost (Adaptive Boosting), Gradient Boosting, XGBoost.

Comparison of Bagging vs. Boosting:

Feature	Bagging	Boosting
Model Training	Independent models	Sequential models
Focus	Reduces variance	Reduces bias
Example	Random Forest	AdaBoost, XGBoost

Applications:

Fraud detection
Stock price prediction
Medical diagnosis

2. Describe the working of the Random Forest Algorithm. How does it improve accuracy?

Definition of Random Forest:

Random Forest is an ensemble learning algorithm that builds multiple decision trees and aggregates their outputs to improve accuracy.

Working of Random Forest:

Bootstrap Sampling: Random subsets of data are selected.
Decision Trees Training: Each tree is trained on a different subset.
Majority Voting: The final prediction is made by averaging the outputs from all trees.

How Random Forest Improves Accuracy:

Reduces Overfitting: By averaging multiple trees, it generalizes better than a single decision tree.
Handles Missing Data: Uses multiple predictions to reduce bias.
Works for Both Classification and Regression Tasks.

Example:

In medical diagnosis, Random Forest helps predict diseases based on multiple patient features.

3. What is an Autoencoder? How does it work in deep learning models?

Definition:

An autoencoder is a type of neural network used for unsupervised learning to compress and reconstruct data.

Working of an Autoencoder:

Encoder: Compresses input data into a lower-dimensional representation.
Bottleneck Layer: Stores the compressed information.
Decoder: Reconstructs the original input from the compressed data.

Applications:

Image Denoising: Removes noise from images.
Anomaly Detection: Identifies fraudulent transactions.
Dimensionality Reduction: Extracts important features from high-dimensional data.

4. Explain the role of Deep Learning in AI. How is it different from traditional ML techniques?

Role of Deep Learning in AI:

Deep learning is a subset of ML that uses neural networks to learn complex patterns.

How Deep Learning is Different from Traditional ML:

Feature	Traditional ML	Deep Learning
Feature Engineering	Requires manual feature extraction	Automatically extracts features
Performance on Large Data	Struggles with large datasets	Performs well on big data
Computational Power	Requires less computation	Needs GPUs for training
Example	Logistic Regression, SVM	CNNs, RNNs

Example Applications:

Facial Recognition (Face ID)
Autonomous Vehicles (Tesla)
Speech Processing (Google Assistant, Alexa)

5. What is a Deep Boltzmann Machine (DBM)? How is it used in AI?

Definition of DBM:

A Deep Boltzmann Machine (DBM) is a deep learning model with multiple hidden layers that learns data representations.

How DBM Works:

Uses restricted Boltzmann machines (RBMs) stacked together.
Learns hierarchical features from input data.
Uses probabilistic reasoning for complex pattern recognition.

Applications of DBM in AI:

Image recognition
Speech recognition
Recommendation systems

6. Explain how deep learning is applied in self-driving cars.

Deep Learning in Self-Driving Cars:

Self-driving cars use deep learning to process and interpret sensor data for navigation.

Key AI Components in Autonomous Vehicles:

Convolutional Neural Networks (CNNs): Analyze images from cameras for object detection (pedestrians, traffic signs).
Recurrent Neural Networks (RNNs): Predict future actions based on past driving patterns.
Reinforcement Learning: Optimizes driving strategies by learning from past experiences.

Example:

Tesla’s Autopilot uses deep learning to recognize roads and avoid collisions.

7. What is the difference between a Convolutional Neural Network (CNN) and a Recurrent Neural Network (RNN)?

Feature	CNN	RNN
Purpose	Used for image processing	Used for sequential data (time series, speech)
Architecture	Uses convolutional layers to extract spatial features	Uses feedback loops to store previous inputs
Use Case	Facial recognition, object detection	Language translation, speech recognition

Example:

CNN: Recognizing objects in an image.
RNN: Predicting the next word in a sentence.

8. Describe the different activation functions used in neural networks.

1. Sigmoid Function:

$f(x) = \frac{1}{1+e^{-x}}$

✔ Used in binary classification.
✖ Prone to vanishing gradient problem.

2. ReLU (Rectified Linear Unit):

$f(x) = \max(0, x)$

✔ Fast computation.
✖ Can suffer from dying neurons.

3. Tanh (Hyperbolic Tangent):

$f(x) = \frac{e^x - e^{-x}}{e^x + e^{-x}}$

✔ Used in hidden layers.
✖ Still has gradient issues.

9. How does dropout work in deep learning? Why is it important?

Definition of Dropout:

Dropout is a regularization technique that randomly removes neurons during training to prevent overfitting.

How Dropout Works:

Randomly deactivates neurons in each training iteration.
Helps models generalize better on new data.

Example:

Applied in CNNs for image classification to improve accuracy.

10. What is Transfer Learning? How does it help in AI model development?

Definition:

Transfer learning allows AI models trained on one task to be reused for another task.

How It Works:

A pretrained model (e.g., ResNet, VGG) is used.
The model is fine-tuned on a new dataset.
Requires less data and training time.

Example:

Google’s BERT model trained on text data is adapted for chatbots.

11. What is a Generative Adversarial Network (GAN)? Explain its applications.

Definition:

GANs consist of two networks:

Generator: Creates fake data.
Discriminator: Detects real vs. fake data.

Applications:

✔ Image generation (Deepfake)
✔ Art & design
✔ Data augmentation

12. Explain the process of evaluating machine learning models. What are the different performance metrics?

Performance Metrics:

Metric	Definition
Accuracy	Measures overall correctness.
Precision	Measures how many predicted positives are actually positive.
Recall (Sensitivity)	Measures how well actual positives are detected.
F1 Score	Harmonic mean of precision & recall.
ROC Curve	Shows trade-off between true positive rate and false positive rate.

No title

Unit-I: Introduction to Artificial Intelligence

1. Define Artificial Intelligence (AI). Explain its importance and role in real-world applications.

Definition of AI:

Importance of AI:

Real-World Applications of AI:

2. Describe the evolution of AI. How has AI developed over the years?

Early History of AI (1950s-1970s):

AI Winters (1970s & 1980s):

Revival and Machine Learning Era (1990s-2000s):

Modern AI (2010-Present):

Future of AI:

3. What is an intelligent agent? Explain its components and different types of agents with examples.

Definition of an Intelligent Agent:

Components of an Intelligent Agent:

Types of Intelligent Agents:

4. Explain the role of AI in mechanical engineering. How can AI be applied in design, manufacturing, and maintenance?

Role of AI in Mechanical Engineering:

Applications in Design:

Applications in Manufacturing:

Applications in Maintenance:

Impact of AI in Mechanical Engineering:

5. Differentiate between strong AI and weak AI. Provide real-world examples for both.

6. What are the different types of AI techniques? Explain machine learning, expert systems, and deep learning.

1. Machine Learning (ML):

2. Expert Systems:

3. Deep Learning:

7. Define rationality in AI. How does rational decision-making apply to AI agents?

Definition of Rationality in AI:

Application in AI Agents:

8. Explain the concept of an agent environment. What are the different types of environments in AI?

Definition of Agent Environment:

Types of Environments:

Unit-II: Knowledge Representation & Reasoning questions.

1. What are Knowledge-Based Agents? How do they use logic to represent knowledge?

Definition of Knowledge-Based Agents:

How KBAs Use Logic to Represent Knowledge:

Example:

2. Explain Propositional Logic. What are its advantages and limitations?

Definition of Propositional Logic:

Advantages:

Limitations:

3. Describe First-Order Logic (FOL). How does it improve upon Propositional Logic?

Definition of First-Order Logic (FOL):

Components of FOL:

How FOL Improves Upon Propositional Logic:

4. What is the Wumpus World Problem? How does it demonstrate logical reasoning?

Wumpus World Problem:

Logical Reasoning in Wumpus World:

Example:

5. Compare Propositional Logic and First-Order Logic. Which one is more useful for AI applications?

Which is More Useful?

6. Explain the process of inference in First-Order Logic. How does unification work?

Inference in FOL:

What is Unification?

Example of Unification:

7. What is the significance of pattern representation in AI? How is it used for decision-making?

Definition:

Importance in AI:

Pattern-Based Decision-Making Example:

8. What is a logical agent? How does it represent knowledge and make inferences?

Definition:

How It Represents Knowledge:

How It Makes Inferences:

Example:

9. Explain forward chaining and backward chaining in reasoning. Provide real-world examples.

Unit-III: Bayesian & Computational Learning questions.

1. What is Bayes' Theorem? Explain its significance in AI and Machine Learning.

Bayes' Theorem Definition:

Formula:

Significance in AI & Machine Learning:

Example:

2. Explain the concept of Maximum Likelihood Estimation (MLE). How is it used in AI?

Definition:

Formula:

Application in AI & Machine Learning:

3. What is the Naïve Bayes Classifier? Explain how it is applied in spam detection.

Definition of Naïve Bayes Classifier:

Formula:

Application in Spam Detection: