IMPORTANCE LEGEND:

• 🔥 Most Important: High frequency (3+ times) OR critical foundational topics essential for the course.

• ⭐ Very Important: Medium frequency (2 times) OR key operational concepts/extensions.

• 📌 Important: Standard frequency (1 time) OR specific/niche topics.

UNIT 1: Introduction to Artificial Intelligence

Q1. 🔥 Compare Artificial Narrow Intelligence, Artificial General Intelligence and Artificial Super Intelligence

[ Jun 2025 Q1(a), Jun 2024 Q1(a), Jun 2022 Q1(a) | Frequency: 3 ]

Artificial Intelligence can be categorized into three types based on capability and scope:

Diagram:

Q2. 📌 Give suitable example for Narrow AI

[ Jun 2022 Q1(a) | Frequency: 1 ]

Narrow AI (ANI) performs specific tasks with high efficiency but cannot generalize beyond its designed purpose.

Examples of Narrow AI:

Key Characteristic: Each system excels only at its designated task and cannot perform tasks outside its programming.

Q3. 📌 Give suitable example for General AI

[ Jun 2022 Q1(a) | Frequency: 1 ]

General AI (AGI) would possess human-like cognitive abilities to understand, learn, and apply knowledge across any domain.

Theoretical Examples & Representations:

Expected Capabilities of AGI:

• Reasoning and problem-solving across domains

• Understanding natural language contextually

• Learning new skills without explicit programming

• Emotional understanding and social intelligence

• Creative thinking and abstract reasoning

Current Status: AGI does not exist yet. It remains a goal of AI research with estimated timelines ranging from decades to centuries.

Q4. 📌 Give suitable example for Super AI

[ Jun 2022 Q1(a) | Frequency: 1 ]

Super AI (ASI) would surpass the brightest human minds in every conceivable domain including creativity, problem-solving, and social intelligence.

Hypothetical Examples & Representations:

Theoretical Capabilities of ASI:

• Self-improvement without human intervention

• Scientific discoveries beyond human comprehension

• Perfect decision-making capabilities

• Potential for technological singularity

• Complete mastery over all human knowledge

Current Status: Purely hypothetical. Raises significant ethical concerns about control, safety, and existential risks.

Q5. ⭐ Explain Turing test (with example/block diagram)

[ Dec 2023 Q1(b), Jun 2023 Q2(a) | Frequency: 2 ]

Definition: The Turing Test, proposed by Alan Turing in 1950 in his paper "Computing Machinery and Intelligence," is a test of a machine's ability to exhibit intelligent behavior indistinguishable from that of a human.

Original Name: Turing originally called it the "Imitation Game."

Test Setup:

Procedure:

1. A human interrogator (C) is separated from two respondents

2. One respondent is a human (B), the other is a machine (A)

3. Communication occurs only through text-based interface

4. Interrogator asks questions to determine which is human

5. If interrogator cannot reliably distinguish, machine passes the test

Example Conversation:

Key Points:

• Tests behavioral intelligence, not internal processes

• Machine must exhibit human-like responses including errors

• Text-only communication prevents physical appearance bias

• Benchmark for conversational AI systems

Q6. ⭐ Explain Chinese room test as Criticism to Turing test (with example)

[ Jun 2024 Q1(b), Jun 2023 Q2(a) | Frequency: 2 ]

Definition: The Chinese Room Argument, proposed by philosopher John Searle in 1980, is a thought experiment that challenges the idea that passing the Turing Test demonstrates true understanding or intelligence.

The Thought Experiment:

Scenario Explained:

How It Works:

1. Chinese speaker sends questions in Chinese characters

2. Person inside matches symbols using the rule book

3. Person produces appropriate Chinese symbol responses

4. From outside, responses appear perfectly fluent

The Argument:

• The person inside follows syntax (rules) perfectly

• BUT has zero semantic understanding (meaning)

• Similarly, computers manipulate symbols without understanding

• Passing Turing Test ≠ True Intelligence or Understanding

Example:

Searle's Conclusion:

• Strong AI Claim (Rejected): Computers can truly understand and have minds

• Weak AI Claim (Accepted): Computers can simulate intelligent behavior

Significance: Distinguishes between simulation of intelligence and actual understanding/consciousness.

Q7. 🔥 Compare Artificial Intelligence, Machine Learning and Deep Learning

[ Dec 2024 Q1(b), Dec 2023 Q2(a), Dec 2022 Q1(a) | Frequency: 3 ]

AI, ML, and DL are related but distinct concepts with a hierarchical relationship:

Diagram:

Comprehensive Comparison:

Relationship Flow:

Key Distinctions:

• AI: Goal is to create intelligent behavior (any method)

• ML: Method to achieve AI through learning from experience

• DL: Advanced ML using deep neural networks for complex patterns

Q8. 📌 What do you understand by the term "Agents" in Artificial Intelligence

[ Jun 2022 Q2(a) | Frequency: 1 ]

Definition: An Agent is anything that can perceive its environment through sensors and act upon that environment through actuators.

Formal Definition: An agent is a system that:

• Perceives its environment

• Makes decisions

• Takes actions to achieve goals

Diagram:

Components of an Agent:

Types of Agents:

1. Human Agent - Eyes, ears (sensors); Hands, legs (actuators)

2. Robotic Agent - Cameras, sensors; Motors, grippers

3. Software Agent - GUI inputs, APIs; Screen output, commands

Agent Function: The mathematical mapping from percept sequence to action:

• f: P* → A

• Where P* is the sequence of all percepts and A is the action

Q9. 📌 List the properties supposed to be possessed by agents

[ Jun 2022 Q2(a) | Frequency: 1 ]

An ideal AI agent should possess the following properties:

Diagram:

Q10. 📌 Compare the SR (simple reflex) agents with model based reflex agents

[ Jun 2022 Q2(a) | Frequency: 1 ]

Comparison Table:

Diagrams:

Simple Reflex Agent:

Model-Based Reflex Agent:

Key Difference Summary:

• Simple Reflex: "See → Act" (stateless)

• Model-Based: "See → Think (using memory & model) → Act" (stateful)

Q11. 📌 What are task environments in context of Intelligent Agents

[ Dec 2022 Q2(a) | Frequency: 1 ]

Definition: A Task Environment is the "problem" to which an intelligent agent is the "solution." It specifies the complete setting in which the agent operates.

Task Environment Components:

This is known as PEAS description (Performance, Environment, Actuators, Sensors)

Properties of Task Environments:

Q12. 📌 Explain the standard set of measures for specifying task environment under PEAS

[ Dec 2022 Q2(a) | Frequency: 1 ]

PEAS is a framework for describing the task environment of an intelligent agent:

Diagram:

PEAS Examples:

Importance of PEAS:

1. Provides structured approach to agent design

2. Clarifies what success means (Performance)

3. Identifies challenges (Environment)

4. Determines required capabilities (Actuators & Sensors)

5. Foundation for selecting appropriate agent architecture

UNIT 2: Problem Solving using Search

Q1. ⭐ What do you understand by state space in AI

[ Dec 2023 Q2(b), Jun 2023 Q3(a) | Frequency: 2 ]

Definition: A State Space is a mathematical representation of a problem that includes all possible states (configurations) that can be reached from the initial state through any sequence of actions.

Formal Definition: State Space is represented as a 4-tuple: S = {S, A, Action(s), Result(s,a)}

State Space Structure:

Diagram:

Key Elements:

1. Initial State: Starting point of the problem

2. Goal State: Desired end configuration

3. State: A unique configuration of the problem

4. Operators: Actions that transform one state to another

5. Path: Sequence of states from initial to goal

6. Path Cost: Sum of costs of individual actions

State Space as a Graph:

• Nodes represent states

• Edges represent actions/operators

• Root node is the initial state

• Goal nodes are target states

Q2. 📌 What is the utility of state space in AI

[ Dec 2023 Q2(b) | Frequency: 1 ]

State Space provides a structured framework for problem-solving in AI with the following utilities:

Diagram:

Practical Benefits:

1. Abstraction: Converts real-world problems into computational form

2. Generalization: Same search techniques work on different problems

3. Visualization: Graph representation aids understanding

4. Completeness Check: Ensures all possibilities are considered

5. Debugging: Easy to trace and verify solution paths

Q3. 📌 Discuss the formulation of state space search (with example)

[ Jun 2022 Q1(b) | Frequency: 1 ]

State Space Search Formulation involves defining a problem in terms of states, actions, and goals to enable systematic exploration.

Formulation Steps:

Example: 8-Puzzle Problem

Diagram:

Formulation for 8-Puzzle:

State Space Size: 9! = 362,880 possible states (only half are reachable from any given state)

Q4. 📌 What are the various factors for developing a state space representation

[ Dec 2024 Q5(b) | Frequency: 1 ]

When developing a state space representation, the following factors must be considered:

Diagram:

Key Principles:

1. Abstraction: Include only relevant details

2. Efficiency: Minimize state representation size

3. Completeness: Capture all necessary information

4. Uniqueness: Each state should have unique representation

5. Computability: States must be computationally manageable

Q5. ⭐ Write production rules for state space representation of water jug problem

[ Dec 2023 Q2(b), Jun 2023 Q3(a) | Frequency: 2 ]

Problem Statement: Given two jugs - one with capacity 4 gallons and another with capacity 3 gallons. Initially both are empty. Goal is to get exactly 2 gallons of water in the 4-gallon jug.

State Representation: (x, y) where:

• x = gallons in 4-gallon jug (0 ≤ x ≤ 4)

• y = gallons in 3-gallon jug (0 ≤ y ≤ 3)

Initial State: (0, 0) Goal State: (2, n) where n can be any value 0-3

Production Rules:

Solution Path:

Diagram:

Solution Trace:

Q6. 📌 Formulate the missionaries and cannibal problem

[ Jun 2025 Q5(b)(i) | Frequency: 1 ]

Problem Statement: Three missionaries and three cannibals are on one side of a river with a boat that can hold at most two people. Find a way to get everyone to the other side, without ever leaving a group of missionaries outnumbered by cannibals (on either side).

State Representation: (M, C, B) where:

• M = Number of missionaries on left bank (0, 1, 2, or 3)

• C = Number of cannibals on left bank (0, 1, 2, or 3)

• B = Position of boat (L = Left, R = Right)

Diagram:

Formulation:

Possible Actions:

• Move 1 missionary

• Move 1 cannibal

• Move 2 missionaries

• Move 2 cannibals

• Move 1 missionary + 1 cannibal

Q7. 📌 Solve the missionaries and cannibal problem

[ Jun 2025 Q5(b)(i) | Frequency: 1 ]

Solution Steps:

Diagram:

Total Moves: 11 crossings

Q8. 📌 Draw the state-space search graph for solving missionaries and cannibal problem

[ Jun 2025 Q5(b)(ii) | Frequency: 1 ]

State Space Graph:

Diagram:

Legend:

• 🟡 Yellow: Initial State (3,3,L)

• 🟢 Green: Goal State (0,0,R)

• Blue edges: Solution path

• L states: Boat on left bank

• R states: Boat on right bank

Q9. ⭐ Discuss the Adversarial search briefly

[ Dec 2022 Q1(b), Jun 2022 Q3(a) | Frequency: 2 ]

Definition: Adversarial Search is a search strategy used in competitive environments where two or more agents have conflicting goals. It is primarily used in game-playing AI.

Characteristics:

Diagram:

Key Concepts:

Applications:

• Chess, Checkers, Tic-Tac-Toe

• Go, Backgammon

• Card games

• Real-time strategy games

Q10. 📌 How Adversarial Search is different from the normal search

[ Jun 2022 Q3(a) | Frequency: 1 ]

Comparison Table:

Diagram:

Key Differences:

1. Decision Making:
2. Normal: Choose best action for self

3. Adversarial: Choose best action considering opponent's response

4. Tree Structure:

5. Normal: Single type of nodes

6. Adversarial: Alternating MAX and MIN nodes

7. Evaluation:

8. Normal: Heuristic estimates distance to goal
9. Adversarial: Heuristic estimates game position quality

Q11. ⭐ Name/Discuss the techniques/types used for adversarial search

[ Dec 2022 Q1(b), Jun 2022 Q3(a) | Frequency: 2 ]

Major Adversarial Search Techniques:

Diagram:

Technique Details:

1. Minimax Algorithm:

• Assumes both players play optimally

• MAX player maximizes score, MIN player minimizes score

• Complete and optimal for finite games

2. Alpha-Beta Pruning:

• Enhancement to Minimax

• Prunes branches that won't affect final decision

• α = best value for MAX, β = best value for MIN

3. Expectiminimax:

• Extension for games with chance (dice, cards)

• Adds "chance nodes" for random events

4. Monte Carlo Tree Search (MCTS):

• Uses random simulations

• Balances exploration and exploitation

• Used in Go (AlphaGo)

Q12. 📌 What is Min-Max Search Strategy

[ Jun 2023 Q1(b) | Frequency: 1 ]

Definition: The Minimax Search Strategy is a recursive algorithm used in decision-making and game theory for finding the optimal move for a player, assuming that the opponent also plays optimally.

Core Principle:

• MAX Player: Tries to maximize the score (your move)

• MIN Player: Tries to minimize the score (opponent's move)

How It Works:

Diagram:

Value Propagation:

• Terminal: (3,1), (5,6), (0,2), (4,1)

• MAX level: max(3,1)=3, max(5,6)=6, max(0,2)=2, max(4,1)=4

• MIN level: min(3,6)=3, min(2,4)=2

• ROOT (MAX): max(3,2)=3 → Choose left branch

Q13. 📌 Discuss the Min-Max Search Strategy briefly

[ Dec 2022 Q3(a) | Frequency: 1 ]

Minimax is a backtracking algorithm used in game theory and AI for decision-making in two-player zero-sum games.

Key Components:

Algorithm Flow:

Diagram:

Properties:

• Complete: Yes, if tree is finite

• Optimal: Yes, against optimal opponent

• Time Complexity: O(b^m)

• Space Complexity: O(bm)

Where: b = branching factor, m = maximum depth

Assumptions:

1. Both players play optimally

2. Perfect information (both see full game state)

3. Zero-sum game (one's gain = other's loss)

Q14. 📌 Write MINIMAX algorithm

[ Jun 2023 Q1(b) | Frequency: 1 ]

MINIMAX Algorithm (Pseudocode):

function MINIMAX(node, depth, isMaximizingPlayer):

    if depth == 0 OR node is terminal:
        return evaluate(node)

    if isMaximizingPlayer:
        maxEval = -∞
        for each child of node:
            eval = MINIMAX(child, depth-1, FALSE)
            maxEval = max(maxEval, eval)
        return maxEval

    else:  // Minimizing player
        minEval = +∞
        for each child of node:
            eval = MINIMAX(child, depth-1, TRUE)
            minEval = min(minEval, eval)
        return minEval

// Initial call
bestMove = MINIMAX(rootNode, maxDepth, TRUE)

Alternative Representation:

function MINIMAX-VALUE(state):
    if TERMINAL-TEST(state):
        return UTILITY(state)

    if PLAYER(state) == MAX:
        return max(MINIMAX-VALUE(s) for s in SUCCESSORS(state))

    else:
        return min(MINIMAX-VALUE(s) for s in SUCCESSORS(state))

Function Descriptions:

Flow Diagram:

Q15. ⭐ Explain Min-Max search algorithm (with example, properties, advantages, disadvantages)

[ Jun 2024 Q2(b) | Frequency: 1 ]

Definition: Minimax is a recursive backtracking algorithm that finds the optimal move for a player in a two-player zero-sum game by minimizing the maximum possible loss.

Example: Simple Game Tree

Diagram:

Step-by-Step Evaluation:

Properties:

Where b = branching factor, m = maximum depth

Advantages:

Disadvantages:

Q16. ⭐ Give properties of MinMax search algorithm

[ Jun 2024 Q2(b), Dec 2022 Q3(a) | Frequency: 2 ]

Properties of Minimax Algorithm:

Where:

• b = Branching factor (average number of legal moves)

• m = Maximum depth of game tree

Diagram:

Complexity Examples:

Additional Properties:

• Deterministic: Same input always produces same output

• Recursive: Natural recursive implementation

• Zero-sum assumption: One player's gain equals other's loss

• Perfect information: Both players see complete game state

Q17. ⭐ Write/Give advantages of MinMax search algorithm

[ Jun 2024 Q2(b), Dec 2022 Q3(a) | Frequency: 2 ]

Advantages of Minimax Algorithm:

Diagram:

Q18. ⭐ Write/Give disadvantages of MinMax search algorithm

[ Jun 2024 Q2(b), Dec 2022 Q3(a) | Frequency: 2 ]

Disadvantages of Minimax Algorithm:

Diagram:

Solutions to Overcome:

UNIT 4: Predicate and Propositional Logic - Exam Answers

4.3-4.5 Introduction to Propositional Logic & Logical Connectives

Q1. 🔥 Differentiate between predicate and propositional logic.

[ Dec 2023 Q3(a) | Frequency: 1 ]

Propositional logic and predicate logic are two fundamental systems in mathematical logic used for knowledge representation in AI.

Examples:

Propositional Logic:

• P = "Socrates is a man" (True/False)

• Q = "Socrates is mortal" (True/False)

• Expression: P → Q

Predicate Logic:

• Man(Socrates) - Socrates is a man

• Mortal(x) - x is mortal

• Expression: ∀x (Man(x) → Mortal(x))

Key Difference: Propositional logic treats statements as atomic units, while predicate logic can analyze the internal structure of statements using predicates and quantifiers.

4.7 Propositional Rules of Inference

Q2. ⭐ Describe the Modus Ponens as propositional rule of inference.

[ Jun 2023 Q1(d) | Frequency: 1 ]

Modus Ponens (Latin: "mode of affirming") is one of the most fundamental and widely used rules of inference in propositional logic.

Formal Definition:

Premise 1:  P → Q    (If P then Q)
Premise 2:  P        (P is true)
─────────────────────
Conclusion: Q        (Therefore, Q is true)

Symbolic Representation:

Explanation:

• If we know that "P implies Q" is true

• And we also know that "P" is true

• Then we can validly conclude that "Q" must be true

Example 1:

• Premise 1: If it rains, the ground gets wet (Rain → WetGround)

• Premise 2: It is raining (Rain)

• Conclusion: The ground is wet (WetGround)

Example 2:

• Premise 1: If a student studies hard, they will pass (Study → Pass)

• Premise 2: John studies hard (Study)

• Conclusion: John will pass (Pass)

Truth Table Verification:

Importance in AI: Modus Ponens is the foundation of forward chaining in rule-based expert systems.

Q3. ⭐ Describe the Modus Tollens as propositional rule of inference.

[ Jun 2023 Q1(d) | Frequency: 1 ]

Modus Tollens (Latin: "mode of denying") is a rule of inference that allows us to derive a negation from a conditional statement and the negation of its consequent.

Formal Definition:

Premise 1:  P → Q    (If P then Q)
Premise 2:  ¬Q       (Q is false)
─────────────────────
Conclusion: ¬P       (Therefore, P is false)

Symbolic Representation:

Explanation:

• If we know that "P implies Q" is true

• And we also know that "Q" is false

• Then we can validly conclude that "P" must be false

Example 1:

• Premise 1: If it rains, the ground gets wet (Rain → WetGround)

• Premise 2: The ground is not wet (¬WetGround)

• Conclusion: It is not raining (¬Rain)

Example 2:

• Premise 1: If the battery is charged, the car will start (Charged → Start)

• Premise 2: The car does not start (¬Start)

• Conclusion: The battery is not charged (¬Charged)

Comparison of Modus Ponens and Modus Tollens:

Importance in AI: Modus Tollens is fundamental for backward chaining and goal-driven reasoning in expert systems.

4.9 Validity and Satisfiability

Q4. 📌 Obtain Conjunctive Normal Form (CNF) for given formula.

[ Dec 2022 Q1(d) | Frequency: 1 ]

Conjunctive Normal Form (CNF): A formula is in CNF when it is a conjunction (AND) of clauses, where each clause is a disjunction (OR) of literals.

General Form: (L₁ ∨ L₂ ∨ ...) ∧ (L₃ ∨ L₄ ∨ ...) ∧ ...

Steps to Convert to CNF:

1. Eliminate implications (→): P → Q ≡ ¬P ∨ Q

2. Eliminate biconditionals (↔): P ↔ Q ≡ (P → Q) ∧ (Q → P)

3. Move negation inward (De Morgan's Laws):

4. ¬(P ∧ Q) ≡ ¬P ∨ ¬Q
5. ¬(P ∨ Q) ≡ ¬P ∧ ¬Q

6. ¬¬P ≡ P

7. Distribute OR over AND: P ∨ (Q ∧ R) ≡ (P ∨ Q) ∧ (P ∨ R)

Example: Convert (P → Q) ∧ (¬P → R) to CNF

Step 1: Eliminate implications

• P → Q ≡ ¬P ∨ Q

• ¬P → R ≡ ¬(¬P) ∨ R ≡ P ∨ R

Step 2: Combine with conjunction

• (¬P ∨ Q) ∧ (P ∨ R)

Result: (¬P ∨ Q) ∧ (P ∨ R) — Already in CNF

Example 2: Convert ¬(P ∧ Q) → R to CNF

Step 1: Eliminate implication

• ¬(P ∧ Q) → R ≡ ¬(¬(P ∧ Q)) ∨ R ≡ (P ∧ Q) ∨ R

Step 2: Distribute OR over AND

• (P ∧ Q) ∨ R ≡ (P ∨ R) ∧ (Q ∨ R)

Result: (P ∨ R) ∧ (Q ∨ R) — CNF achieved

Q5. 📌 Obtain Disjunctive Normal Form for given well form formula.

[ Jun 2022 Q1(d) | Frequency: 1 ]

Disjunctive Normal Form (DNF): A formula is in DNF when it is a disjunction (OR) of clauses, where each clause is a conjunction (AND) of literals.

General Form: (L₁ ∧ L₂ ∧ ...) ∨ (L₃ ∧ L₄ ∧ ...) ∨ ...

Steps to Convert to DNF:

1. Eliminate implications (→): P → Q ≡ ¬P ∨ Q

2. Eliminate biconditionals (↔): P ↔ Q ≡ (P ∧ Q) ∨ (¬P ∧ ¬Q)

3. Move negation inward (De Morgan's Laws)

4. Distribute AND over OR: P ∧ (Q ∨ R) ≡ (P ∧ Q) ∨ (P ∧ R)

Example: Convert (P ∨ Q) ∧ (R ∨ S) to DNF

Step 1: Apply distribution of AND over OR

• (P ∨ Q) ∧ (R ∨ S)

• = (P ∧ (R ∨ S)) ∨ (Q ∧ (R ∨ S))

• = (P ∧ R) ∨ (P ∧ S) ∨ (Q ∧ R) ∨ (Q ∧ S)

Result: (P ∧ R) ∨ (P ∧ S) ∨ (Q ∧ R) ∨ (Q ∧ S) — DNF achieved

Example 2: Convert ¬(P → Q) to DNF

Step 1: Eliminate implication

• ¬(P → Q) ≡ ¬(¬P ∨ Q)

Step 2: Apply De Morgan's Law

• ¬(¬P ∨ Q) ≡ P ∧ ¬Q

Result: P ∧ ¬Q — Already in DNF (single term)

Comparison CNF vs DNF:

4.14 Propositional Resolution

Q6. 🔥 What do you understand by the term 'Resolution' in AI / Explain the concept.

[ Jun 2024 Q3(b), Dec 2023 Q3(b) | Frequency: 2 ]

Resolution is a powerful inference rule and proof technique used in automated theorem proving and AI systems for logical reasoning.

Definition: Resolution is a single inference rule that can be applied repeatedly to derive new clauses from existing ones, ultimately proving whether a statement is satisfiable or a consequence of given premises.

The Resolution Rule:

Clause 1:  A ∨ B
Clause 2:  ¬A ∨ C
─────────────────
Resolvent: B ∨ C

Formal Definition:

Where L is a literal, and α, β are disjunctions of literals.

Key Concepts:

1. Complementary Literals: Two literals are complementary if one is the negation of the other (e.g., P and ¬P)

2. Resolvent: The new clause produced by resolving two clauses

3. Empty Clause (□): When resolution produces an empty clause, it indicates a contradiction

Diagram:

Resolution Procedure:

1. Convert all statements to CNF

2. Negate the conclusion and add to clause set

3. Repeatedly apply resolution rule

4. If empty clause (□) is derived → Original statement is proved

5. If no new clauses can be derived → Statement cannot be proved

Q7. 📌 Discuss the utility of Resolution mechanism in AI.

[ Jun 2024 Q3(b) | Frequency: 1 ]

Utility of Resolution Mechanism in AI:

Key Advantages:

1. Completeness: Resolution is refutation-complete — if a contradiction exists, resolution will find it

2. Uniform Procedure: Single rule applicable to all logical problems

3. Mechanical Nature: Can be fully automated by computers

4. Sound: Never produces incorrect conclusions

5. Efficient for Certain Problems: Works well with Horn clauses

Practical Applications:

• Prolog: Uses SLD-resolution for query answering

• SAT Solvers: Use resolution-based techniques

• Verification Systems: Prove program correctness

• Database Query Optimization: Logical query processing

Q8. 📌 Discuss the relevance of resolution mechanism in AI briefly.

[ Jun 2022 Q1(e) | Frequency: 1 ]

Relevance of Resolution in AI:

1. Foundation of Logic Programming:

• Prolog and similar languages are built on resolution

• Enables declarative programming paradigm

2. Automated Reasoning:

• Provides systematic method for machine reasoning

• Eliminates need for human intervention in proofs

3. Knowledge Base Validation:

• Checks consistency of expert system knowledge bases

• Identifies contradictory rules

4. Query Answering Systems:

• Derives answers from database facts and rules

• Powers intelligent information retrieval

5. AI Planning:

• Proves whether goals are achievable

• Validates plan correctness

Why Resolution is Preferred:

Q9. 🔥 Apply Resolution to conclude statement from knowledge / with example.

[ Jun 2024 Q3(b), Dec 2023 Q3(b) | Frequency: 2 ]

Example Problem: Prove "Socrates is mortal" from the following knowledge base:

1. Socrates is a man

2. All men are mortal

Step 1: Represent in Predicate Logic

• Man(Socrates)

• ∀x (Man(x) → Mortal(x))

Step 2: Convert to Propositional Form (for simplicity)

• P: Socrates is a man

• Q: Socrates is mortal

• Knowledge: P, P → Q

• Goal: Prove Q

Step 3: Convert to CNF

• Clause 1: P (given)

• Clause 2: ¬P ∨ Q (from P → Q)

• Clause 3: ¬Q (negation of goal for refutation)

Step 4: Apply Resolution

Clause 1: P
Clause 2: ¬P ∨ Q
Clause 3: ¬Q

Resolution Step 1: Resolve Clause 1 and Clause 2
    P
    ¬P ∨ Q
    ─────────
    Q (Resolvent 1)

Resolution Step 2: Resolve Resolvent 1 and Clause 3
    Q
    ¬Q
    ─────────
    □ (Empty Clause - Contradiction!)

Diagram:

Conclusion: Since we derived the empty clause (□), the negation of our goal leads to a contradiction. Therefore, the original goal "Socrates is mortal" is PROVED.

Another Example: Prove R from {P, P→Q, Q→R}

Result: Goal R is proved by deriving empty clause.

Summary Table: Unit 4 Key Concepts

UNIT 5: First Order Logic - Exam Answers

5.2 Syntax of First Order Predicate Logic (FOPL)

Q1. 🔥 Symbolize "Every rational number is a real number" using predicate logic.

[ Dec 2024 Q1(d)(i), Dec 2023 Q3(a)(i) | Frequency: 2 ]

Given Statement: "Every rational number is a real number"

Step 1: Identify Predicates

• Let R(x) = "x is a rational number"

• Let E(x) = "x is a real number"

Step 2: Analyze the Statement

• "Every" indicates Universal Quantifier (∀)

• The statement means: For all x, if x is rational, then x is real

Step 3: Symbolic Representation

Interpretation: For all x, if x is a rational number, then x is a real number.

Alternative Notation: ∀x (Rational(x) → Real(x))

Q2. 🔥 Symbolize "Some real numbers are rational numbers" using predicate logic.

[ Dec 2024 Q1(d)(ii), Dec 2023 Q3(a)(ii) | Frequency: 2 ]

Given Statement: "Some real numbers are rational numbers"

Step 1: Identify Predicates

• Let R(x) = "x is a rational number"

• Let E(x) = "x is a real number"

Step 2: Analyze the Statement

• "Some" indicates Existential Quantifier (∃)

• The statement means: There exists at least one x that is both real AND rational

Step 3: Symbolic Representation

Interpretation: There exists an x such that x is a real number AND x is a rational number.

Important Note: With existential quantifier, we use conjunction (∧) not implication (→).

Q3. 🔥 Symbolize "Not every real number is a rational number" using predicate logic.

[ Dec 2024 Q1(d)(iii), Dec 2023 Q3(a)(iii) | Frequency: 2 ]

Given Statement: "Not every real number is a rational number"

Step 1: Identify Predicates

• Let R(x) = "x is a rational number"

• Let E(x) = "x is a real number"

Step 2: Analyze the Statement

• "Not every" = negation of universal quantifier

• Equivalent to: "There exists some real number that is not rational"

Step 3: Symbolic Representation

Method 1 (Direct Negation):

Method 2 (Equivalent Form using De Morgan's for Quantifiers):

Equivalence Explanation:

• ¬∀x P(x) ≡ ∃x ¬P(x)

• ¬(E(x) → R(x)) ≡ E(x) ∧ ¬R(x)

Interpretation: There exists an x such that x is a real number AND x is NOT a rational number (e.g., √2, π).

5.4 Semantics of Quantifiers

Q4. 📌 Translate "∃ₓC(x)" into English (car dealer context).

[ Jun 2024 Q3(a)(i) | Frequency: 1 ]

Given: ∃x C(x)

Where: C(x) = "x is a car dealer"

Translation:

Alternative Translations:

• "Some person is a car dealer"

• "At least one car dealer exists"

• "There is a car dealer"

Q5. 📌 Translate "∃ₓH(x)" into English (honest context).

[ Jun 2024 Q3(a)(ii) | Frequency: 1 ]

Given: ∃x H(x)

Where: H(x) = "x is honest"

Translation:

Alternative Translations:

• "Some person is honest"

• "At least one honest person exists"

• "There is an honest person"

Q6. 🔥 Translate "∀x C(x)→∼H(x)" into English (car dealer/honest context).

[ Jun 2025 Q1(d)(i), Jun 2024 Q3(a)(iii) | Frequency: 2 ]

Given: ∀x (C(x) → ¬H(x))

Where:

• C(x) = "x is a car dealer"

• H(x) = "x is honest"

Translation:

Alternative Translations:

• "Every car dealer is dishonest"

• "For every person, if they are a car dealer, then they are not honest"

• "No car dealer is honest"

Logical Breakdown:

Q7. 🔥 Translate "∃x (C(x)∧H(x))" into English (car dealer/honest context).

[ Jun 2025 Q1(d)(ii), Jun 2024 Q3(a)(iv) | Frequency: 2 ]

Given: ∃x (C(x) ∧ H(x))

Where:

• C(x) = "x is a car dealer"

• H(x) = "x is honest"

Translation:

Alternative Translations:

• "There exists a car dealer who is honest"

• "At least one person is both a car dealer and honest"

• "There is an honest car dealer"

Note: The conjunction (∧) indicates that both properties must hold simultaneously for the same individual.

Q8. 🔥 Translate "∃x (H(x)→C(x))" into English (car dealer/honest context).

[ Jun 2025 Q1(d)(iii), Jun 2024 Q3(a)(v) | Frequency: 2 ]

Given: ∃x (H(x) → C(x))

Where:

• C(x) = "x is a car dealer"

• H(x) = "x is honest"

Translation:

Logical Analysis:

Since P → Q ≡ ¬P ∨ Q, we have:

• H(x) → C(x) ≡ ¬H(x) ∨ C(x)

Equivalent Translations:

• "There exists someone who is either not honest or is a car dealer"

• "Some person being honest implies they are a car dealer"

Important Observation: This statement is TRUE if there exists anyone who is:

• Dishonest (regardless of being a car dealer), OR

• A car dealer (regardless of being honest)

5.6 Conversion to Clausal Form

Q9. ⭐ What is Prenex Normal Form (PNF)?

[ Jun 2023 Q1(e) | Frequency: 1 ]

Prenex Normal Form (PNF) is a standardized way of writing First Order Logic formulas where all quantifiers appear at the beginning (prefix) of the formula.

Definition: A formula is in PNF if it has the form:

Where:

• Q₁, Q₂, ..., Qₙ are quantifiers (∀ or ∃)

• x₁, x₂, ..., xₙ are variables

• M is a quantifier-free formula (called the matrix)

Structure:

Example:

• Not in PNF: ∀x P(x) → ∃y Q(y)

• In PNF: ∃x ∀y (¬P(x) ∨ Q(y))

Importance in AI:

• Required for resolution-based theorem proving

• Standardizes formulas for automated processing

• Simplifies unification algorithms

Q10. ⭐ Write steps to transform FOPL to PNF.

[ Jun 2024 Q1(c) | Frequency: 1 ]

Steps to Convert FOPL Formula to Prenex Normal Form:

Detailed Rules for Step 4:

¬∀x P(x) ≡ ∃x ¬P(x)
¬∃x P(x) ≡ ∀x ¬P(x)

(∀x P(x)) ∧ Q ≡ ∀x (P(x) ∧ Q)    [if x not free in Q]
(∀x P(x)) ∨ Q ≡ ∀x (P(x) ∨ Q)    [if x not free in Q]
(∃x P(x)) ∧ Q ≡ ∃x (P(x) ∧ Q)    [if x not free in Q]
(∃x P(x)) ∨ Q ≡ ∃x (P(x) ∨ Q)    [if x not free in Q]

Diagram:

Q11. ⭐ Apply the steps to transform given formula to PNF.

[ Jun 2024 Q1(c), Jun 2023 Q1(e) | Frequency: 2 ]

Example: Convert ∀x P(x) → ∃y Q(y) to PNF

Step 1: Eliminate Implication

Step 2: Move Negation Inward

Step 3: Standardize Variables

• Variables x and y are already distinct ✓

Step 4: Move Quantifiers to Front

Final PNF:

Another Example: Convert ∀x(P(x) → ∀y Q(x,y)) to PNF

Final PNF: ∀x ∀y (¬P(x) ∨ Q(x,y))

Q12. 🔥 What is Skolemization?

[ Dec 2023 Q1(c), Dec 2022 Q1(e) | Frequency: 2 ]

Skolemization is a technique for eliminating existential quantifiers (∃) from a formula while preserving satisfiability.

Definition: The process of replacing existentially quantified variables with Skolem constants or Skolem functions.

Why Skolemization?

• Resolution requires formulas in clausal form

• Clausal form cannot have existential quantifiers

• Skolemization removes ∃ while maintaining logical equivalence for satisfiability

Rules of Skolemization:

Key Points:

• Skolem constants are new unique constants (a, b, c...)

• Skolem functions are new unique functions (f, g, h...)

• The resulting formula is satisfiability-equivalent (not logically equivalent)

Q13. ⭐ What is the utility of Skolemization?

[ Dec 2023 Q1(c) | Frequency: 1 ]

Utility of Skolemization in AI:

Importance in Resolution:

1. Resolution works only with universally quantified clauses

2. Skolemization removes all ∃ quantifiers

3. After Skolemization, all variables are implicitly universally quantified

4. Enables unification algorithm to work properly

Workflow:

Original Formula → PNF → Skolemization → CNF → Resolution

Q14. 🔥 Skolemize the given expression.

[ Dec 2023 Q1(c), Dec 2022 Q1(e) | Frequency: 2 ]

Example 1: Skolemize ∃x ∀y P(x, y)

Analysis:

• ∃x is NOT in scope of any ∀

• Replace x with Skolem constant 'a'

Result:

Example 2: Skolemize ∀x ∃y P(x, y)

Analysis:

• ∃y IS in scope of ∀x

• Replace y with Skolem function f(x)

Result:

Example 3: Skolemize ∀x ∀y ∃z P(x, y, z)

Analysis:

• ∃z is in scope of both ∀x and ∀y

• Replace z with Skolem function f(x, y)

Result:

Summary Table:

5.7 Resolution & Unification

Q15. 🔥 Explain the concept of unification in AI.

[ Dec 2023 Q3(b), Jun 2022 Q1(e) | Frequency: 2 ]

Unification is a fundamental process in AI that finds substitutions to make two logical expressions identical.

Definition: Unification is the process of finding a substitution (called the Most General Unifier or MGU) that makes two expressions syntactically identical.

Key Concepts:

Unification Algorithm Steps:

1. If both expressions are constants, they unify if identical

2. If one is a variable, substitute it with the other term

3. If both are compound terms, unify functor and arguments recursively

4. Apply occurs check to prevent infinite substitutions

Diagram:

Relevance in AI:

• Essential for resolution-based theorem proving

• Used in Prolog for pattern matching

• Enables logical inference in expert systems

Q16. ⭐ Explain unification in AI with suitable example.

[ Dec 2023 Q3(b) | Frequency: 1 ]

Unification Examples:

Example 1: Simple Unification

Unify: P(x, a) and P(b, a)

MGU: {x/b} Unified Expression: P(b, a)

Example 2: Complex Unification

Unify: P(x, f(y)) and P(a, f(b))

MGU: {x/a, y/b} Unified Expression: P(a, f(b))

Example 3: Unification Failure

Unify: P(x, x) and P(a, b)

Result: Cannot unify (no MGU exists)

Example 4: Occurs Check

Unify: P(x) and P(f(x))

Result: Cannot unify (would create infinite term)

Summary of Unification Rules:

Quick Reference: FOPL Symbolization Patterns

UNIT 6: Rule Based Systems and Other Formalism

6.2 Rule Based Systems

Q1. 🔥 Explain Rule Based Systems in AI and its types.

[ Dec 2023 Q3(c), Jun 2022 Q5(a) | Frequency: 2 ]

Rule Based Systems (RBS): A Rule Based System is an AI system that uses a set of IF-THEN rules to represent knowledge and make inferences. It consists of a knowledge base containing rules and facts, and an inference engine that applies rules to derive new information or make decisions.

Components of Rule Based System:

Diagram:

Types of Rule Based Systems:

Rule Format:

IF <condition/antecedent> THEN <action/consequent>

Example:

Rule 1: IF fever AND cough THEN suspect_flu
Rule 2: IF suspect_flu AND body_ache THEN diagnose_flu
Rule 3: IF diagnose_flu THEN prescribe_rest_and_fluids

Q2. ⭐ Give advantages of Rule Based Systems.

[ Dec 2023 Q3(c) | Frequency: 1 ]

Advantages of Rule Based Systems:

Q3. ⭐ Give disadvantages of Rule Based Systems.

[ Dec 2023 Q3(c) | Frequency: 1 ]

Disadvantages of Rule Based Systems:

Q4. 📌 Give the two important sources of uncertainty in Rule Based Systems.

[ Dec 2023 Q3(c) | Frequency: 1 ]

Two Important Sources of Uncertainty in Rule Based Systems:

1. Uncertainty in Rules (Rule Uncertainty):

• Rules themselves may not be absolutely certain

• Expert knowledge is often approximate or probabilistic

• The relationship between conditions and conclusions may not be 100% reliable

Example:

IF patient_has_fever AND patient_has_cough 
THEN patient_has_flu (Certainty Factor: 0.7)

The rule only suggests 70% confidence in the conclusion.

2. Uncertainty in Data/Facts (Data Uncertainty):

• Input data may be incomplete, imprecise, or unreliable

• Sensor readings may have errors

• User-provided information may be uncertain or vague

Example:

Fact: patient_has_fever (Confidence: 0.8)

The observation itself has only 80% reliability.

Diagram:

Handling Uncertainty:

• Certainty Factors (CF)

• Fuzzy Logic

• Bayesian Probability

• Dempster-Shafer Theory

6.2.1 Forward Chaining

Q5. 🔥 Explain Forward Chaining Systems with example.

[ Jun 2024 Q1(d), Dec 2022 Q5(a) | Frequency: 2 ]

Forward Chaining: Forward chaining is a data-driven reasoning strategy that starts with available facts and applies inference rules to extract more data until a goal is reached. It works from antecedent (IF part) to consequent (THEN part).

Characteristics:

• Also called bottom-up reasoning or data-driven reasoning

• Begins with known facts in working memory

• Applies all applicable rules to generate new facts

• Continues until goal is achieved or no more rules can fire

Algorithm:

1. Start with initial facts in working memory

2. Find all rules whose conditions (IF part) match the facts

3. Fire the matched rules and add conclusions to working memory

4. Repeat steps 2-3 until goal is reached or no rules fire

Diagram:

Example:

Knowledge Base (Rules):

R1: IF animal_has_feathers THEN animal_is_bird
R2: IF animal_is_bird AND animal_cannot_fly THEN animal_is_penguin
R3: IF animal_is_penguin THEN animal_lives_in_cold_regions

Initial Facts:

• animal_has_feathers

• animal_cannot_fly

Forward Chaining Process:

Applications:

• Planning systems

• Monitoring and control systems

• Configuration systems

• CLIPS, OPS5 expert systems

6.2.2 Backward Chaining

Q6. 🔥 Explain Backward Chaining System with example.

[ Dec 2023 Q1(d), Jun 2023 Q5(a) | Frequency: 2 ]

Backward Chaining: Backward chaining is a goal-driven reasoning strategy that starts with a goal (hypothesis) and works backward to find facts that support the goal. It works from consequent (THEN part) to antecedent (IF part).

Characteristics:

• Also called top-down reasoning or goal-driven reasoning

• Begins with a goal to prove

• Finds rules whose conclusions match the goal

• Sets rule conditions as new subgoals

• Continues until all subgoals are proven by known facts

Algorithm:

1. Start with the goal to be proven

2. Find rules whose THEN part matches the goal

3. Set the IF conditions of matching rules as subgoals

4. Recursively try to prove each subgoal

5. If subgoal matches a known fact, it is proven

6. Goal is proven when all subgoals are proven

Diagram:

Example:

Knowledge Base (Rules):

R1: IF has_feathers THEN is_bird
R2: IF is_bird AND cannot_fly THEN is_penguin
R3: IF is_penguin THEN lives_in_cold

Known Facts: has_feathers, cannot_fly

Goal: Prove "lives_in_cold"

Backward Chaining Process:

Result: Goal "lives_in_cold" is PROVEN

Applications:

• Diagnostic systems (MYCIN)

• Query answering systems

• Prolog programming language

• Theorem proving

Comparison:

6.3 Semantic Nets

Q7. 🔥 Explain Semantic Nets with suitable example.

[ Jun 2024 Q3(c)(iii), Dec 2022 Q5(b) | Frequency: 2 ]

Semantic Networks: Semantic networks (Semantic Nets) are knowledge representation schemes that use a graph structure consisting of nodes and labeled edges (arcs) to represent objects, concepts, and relationships between them.

Components:

Common Relationship Types:

Example: Representing Animal Kingdom

Diagram:

Advantages of Semantic Networks:

1. Visual and intuitive representation

2. Supports inheritance through IS-A links

3. Easy to understand relationships

4. Natural for representing taxonomies

5. Efficient for certain types of queries

Disadvantages:

1. No standard definition of link types

2. Difficult to represent complex relationships

3. Limited expressiveness for logic

4. Inheritance can be ambiguous

Applications:

• Natural Language Processing

• Knowledge bases

• Information retrieval

• Conceptual modeling

Q8. ⭐ Draw a semantic network for given English statement.

[ Dec 2024 Q5(a) | Frequency: 1 ]

Statement Example: "John owns a red car. The car has four wheels. John is a person who works at IBM."

Diagram:

General Steps to Draw Semantic Network:

6.4 Frames

Q9. 🔥 Explain Frames with suitable example.

[ Jun 2024 Q3(c)(i), Jun 2022 Q5(b) | Frequency: 2 ]

Frames: Frames are a structured knowledge representation scheme proposed by Marvin Minsky (1975). A frame represents a stereotyped situation or object as a collection of slots (attributes) and their values (fillers). Frames capture both declarative and procedural knowledge.

Structure of a Frame:

Slot Facets:

Example: University Frame System

Frame: PERSON
  Slots:
    Name: [Type: String]
    Age: [Type: Integer, Range: 0-150]
    Gender: [Type: {Male, Female, Other}]

Frame: STUDENT (IS-A: PERSON)
  Slots:
    Student_ID: [Type: String]
    Department: [Type: String]
    GPA: [Type: Float, Range: 0.0-10.0, Default: 0.0]
    Courses: [Type: List]
    Advisor: [Type: PROFESSOR]

Frame: JOHN (INSTANCE-OF: STUDENT)
  Slots:
    Name: "John Smith"
    Age: 21
    Gender: Male
    Student_ID: "CS2024001"
    Department: "Computer Science"
    GPA: 8.5
    Courses: [AI, DBMS, Networks]

Diagram:

Features of Frames:

1. Inheritance - Child frames inherit slots from parent frames

2. Default Values - Provide typical values for slots

3. Procedural Attachment - Demons/triggers for slot operations

4. Multiple Inheritance - Can inherit from multiple parent frames

Advantages:

• Natural representation of objects and classes

• Supports inheritance and default reasoning

• Combines data and procedures

• Easy to understand and organize

Disadvantages:

• Complex inheritance can be confusing

• Difficult to handle exceptions

• No standard frame language

6.5 Scripts

Q10. 🔥 Explain Scripts with suitable example.

[ Jun 2024 Q3(c)(ii), Jun 2023 Q5(b) | Frequency: 2 ]

Scripts: Scripts are structured knowledge representation schemes proposed by Schank and Abelson (1977) for representing stereotyped sequences of events. A script captures the typical sequence of actions that occur in familiar situations, enabling understanding and prediction of events.

Components of a Script:

Example: Restaurant Script

SCRIPT: RESTAURANT
Track: Sit-down Restaurant

ENTRY CONDITIONS:
  - Customer is hungry
  - Customer has money
  - Restaurant is open

ROLES:
  - Customer (C)
  - Waiter (W)
  - Chef (CH)
  - Cashier (CA)

PROPS:
  - Tables, Chairs, Menu
  - Food, Plates, Utensils
  - Bill, Money/Card

SCENES:

Scene 1: ENTERING
  C enters restaurant
  C waits to be seated
  W leads C to table
  C sits down

Scene 2: ORDERING
  W brings menu
  C reads menu
  C decides on food
  W takes order
  W gives order to CH

Scene 3: EATING
  CH prepares food
  W brings food to C
  C eats food

Scene 4: PAYING
  C asks for bill
  W brings bill
  C pays CA
  C leaves tip
  C exits restaurant

RESULTS:
  - Customer is not hungry
  - Customer has less money
  - Restaurant has more money

Diagram:

Applications of Scripts:

1. Natural Language Understanding - Understanding stories and conversations

2. Question Answering - Inferring unstated information

3. Prediction - Anticipating what happens next

4. Planning - Generating action sequences

Example Usage: Statement: "John went to a restaurant and ordered pizza."

Inferences from Restaurant Script:

• John was probably hungry

• John sat at a table

• A waiter took his order

• John will receive pizza

• John will pay a bill

Advantages:

• Captures temporal sequences naturally

• Enables inference of unstated events

• Supports expectation-based understanding

• Good for stereotyped situations

Disadvantages:

• Limited to familiar, routine situations

• Cannot handle novel situations

• Difficult to modify or combine scripts

• Scripts may be culture-specific

Summary: Comparison of Knowledge Representation Schemes

7.3 Review of Probability Theory

Q1. 📌 Describe the event "The maximum score is 6" when a six-faced die is thrown twice.

[ Jun 2025 Q5(a)(i) | Frequency: 1 ]

Sample Space: When a six-faced die is thrown twice, the sample space S consists of all ordered pairs (i, j) where i, j ∈ {1, 2, 3, 4, 5, 6}.

Total outcomes = 6 × 6 = 36

Event Description: The event "Maximum score is 6" means at least one of the two throws must show 6, and no throw shows more than 6 (which is impossible anyway).

Event A = {(i, j) : max(i, j) = 6}

This includes:

• First throw is 6, second is anything: (6,1), (6,2), (6,3), (6,4), (6,5), (6,6)

• Second throw is 6, first is anything except 6: (1,6), (2,6), (3,6), (4,6), (5,6)

Event Set:

A = {(6,1), (6,2), (6,3), (6,4), (6,5), (6,6), 
     (1,6), (2,6), (3,6), (4,6), (5,6)}

Number of favorable outcomes = 11

Probability = 11/36

Diagram:

Q2. 📌 Describe the event "The total score is 9" when a six-faced die is thrown twice.

[ Jun 2025 Q5(a)(ii) | Frequency: 1 ]

Sample Space: When a six-faced die is thrown twice, total outcomes = 6 × 6 = 36

Event Description: The event "Total score is 9" means the sum of both throws equals 9.

Event B = {(i, j) : i + j = 9}

Finding pairs where i + j = 9:

Event Set:

B = {(3,6), (4,5), (5,4), (6,3)}

Number of favorable outcomes = 4

Probability = 4/36 = 1/9

Diagram:

Q3. 📌 Describe the event "Each throw results in an even score larger than 2" when die thrown twice.

[ Jun 2025 Q5(a)(iii) | Frequency: 1 ]

Sample Space: When a six-faced die is thrown twice, total outcomes = 6 × 6 = 36

Event Description: The event requires BOTH throws to satisfy:

• Even number (2, 4, or 6)

• Larger than 2 (3, 4, 5, or 6)

Intersection of conditions: Even AND > 2 → {4, 6}

Event C = {(i, j) : i ∈ {4, 6} and j ∈ {4, 6}}

Event Set:

C = {(4,4), (4,6), (6,4), (6,6)}

Number of favorable outcomes = 4

Probability = 4/36 = 1/9

Diagram:

7.4 Introduction to Bayesian Theory

Q4. 🔥 Write Bayes' Theorem and explain each component.

[ Jun 2024 Q4(b), Jun 2022 Q1(g) | Frequency: 2 ]

Bayes' Theorem:

Expanded Form (using Total Probability):

Components Explained:

Diagram:

Interpretation:

• Prior P(H): What we believed before seeing evidence

• Likelihood P(E|H): How well evidence supports hypothesis

• Posterior P(H|E): Updated belief after seeing evidence

Example:

Medical Diagnosis:

• H = Patient has disease

• E = Test result is positive

• P(H) = 0.01 (1% of population has disease)

• P(E|H) = 0.95 (test is 95% accurate for diseased)

• P(E|¬H) = 0.05 (5% false positive rate)

Even with positive test, only 16.1% chance of actually having the disease!

Applications:

• Medical diagnosis

• Spam filtering

• Machine learning classifiers

• Decision making under uncertainty

7.5 Bayes' Networks

Q5. ⭐ Explain Bayes' Networks.

[ Dec 2022 Q5(c) | Frequency: 1 ]

Bayesian Networks (Belief Networks): A Bayesian Network is a probabilistic graphical model that represents a set of random variables and their conditional dependencies using a Directed Acyclic Graph (DAG). It combines graph theory with probability theory for reasoning under uncertainty.

Components:

Key Concepts:

Example: Burglary Alarm Network

Diagram:

Conditional Probability Tables:

P(Burglary):

P(Alarm | Burglary, Earthquake):

P(JohnCalls | Alarm):

Joint Probability Calculation:

Properties:

1. Conditional Independence: A node is independent of non-descendants given its parents

2. Compact Representation: Requires fewer parameters than full joint distribution

3. Inference: Can compute any probability query from the network

Advantages:

• Visual representation of dependencies

• Handles uncertainty naturally

• Supports both prediction and diagnosis

• Compact knowledge representation

Applications:

• Medical diagnosis systems

• Fault detection

• Risk assessment

• Decision support systems

7.9 Dempster-Shafer Theory

Q6. ⭐ What are the different types of evidences in D-S (Dempster-Shafer) theory? Explain them briefly.

[ Jun 2025 Q1(e) | Frequency: 1 ]

Dempster-Shafer Theory (also called Theory of Evidence) handles uncertainty by allowing belief to be assigned to sets of hypotheses rather than single hypotheses.

Types of Evidence in D-S Theory:

1. Vacuous Evidence:

• Provides no information about hypotheses

• Belief mass assigned entirely to the frame of discernment (Θ)

• m(Θ) = 1, m(A) = 0 for all proper subsets A

• Represents complete ignorance

2. Certain Evidence:

• Provides complete certainty about a single hypothesis

• All belief mass assigned to one singleton set

• m({h}) = 1 for some hypothesis h

• Represents absolute knowledge

3. Consonant Evidence:

• Focal elements are nested (one contains another)

• Forms a chain: A₁ ⊂ A₂ ⊂ ... ⊂ Aₙ

• Belief and Plausibility values are related simply

• Example: m({a}) = 0.3, m({a,b}) = 0.5, m({a,b,c}) = 0.2

4. Bayesian Evidence:

• All focal elements are singletons

• Equivalent to standard probability

• m assigns mass only to individual hypotheses

• Bel(A) = Pl(A) for all subsets A

5. Categorical/Dogmatic Evidence:

• Supports a specific proper subset with certainty

• m(A) = 1 for some A ⊂ Θ (where A ≠ Θ)

• No uncertainty about the supported set

Diagram:

Q7. 🔥 Explain Dempster-Shafer Theory with example.

[ Jun 2025 Q4(b), Jun 2022 Q5(c) | Frequency: 2 ]

Dempster-Shafer Theory: The Dempster-Shafer (D-S) theory is a mathematical framework for representing and combining uncertain evidence. It generalizes Bayesian probability by allowing belief to be assigned to sets of hypotheses rather than requiring probability for each individual hypothesis.

Key Concepts:

Mass Function Properties:

1. m(∅) = 0 (no belief assigned to empty set)

2. Σ m(A) = 1 for all A ⊆ Θ

Belief and Plausibility:

Relationship: Bel(A) ≤ P(A) ≤ Pl(A)

Dempster's Rule of Combination: For combining evidence from two independent sources:

Where conflict: $K = \sum_{B \cap C = \emptyset} m_1(B) \cdot m_2(C)$

Example: Medical Diagnosis

Frame of Discernment: Θ = {Flu, Cold, Allergy}

Evidence from Source 1 (Doctor 1):

• m₁({Flu}) = 0.6

• m₁({Cold}) = 0.3

• m₁(Θ) = 0.1 (uncertainty)

Evidence from Source 2 (Doctor 2):

• m₂({Flu}) = 0.4

• m₂({Allergy}) = 0.4

• m₂(Θ) = 0.2

Combining Evidence:

Diagram:

Combination Calculation Table:

Conflict: K = 0.24 + 0.12 + 0.12 = 0.48

Normalization factor: 1 - K = 0.52

Combined masses:

• m₁₂({Flu}) = (0.24 + 0.12 + 0.04) / 0.52 = 0.40/0.52 ≈ 0.77

• m₁₂({Cold}) = 0.06 / 0.52 ≈ 0.12

• m₁₂({Allergy}) = 0.04 / 0.52 ≈ 0.08

• m₁₂(Θ) = 0.02 / 0.52 ≈ 0.04

Belief and Plausibility for Flu:

• Bel({Flu}) = m({Flu}) = 0.77

• Pl({Flu}) = m({Flu}) + m(Θ) = 0.77 + 0.04 = 0.81

Advantages of D-S Theory:

1. Handles ignorance explicitly (mass to Θ)

2. Does not require complete probability assignments

3. Combines evidence from multiple sources

4. Distinguishes between uncertainty and disbelief

Disadvantages:

1. Computationally expensive (2^n subsets)

2. Conflict handling can be problematic

3. Requires independent evidence sources

4. May produce counterintuitive results with high conflict

Applications:

• Expert systems

• Sensor fusion

• Target identification

• Medical diagnosis

• Risk assessment

Summary: Comparison of Uncertainty Handling Methods

UNIT 8: Fuzzy and Rough Set

8.2-8.4 Fuzzy Systems & Fuzzy Set Representation

Q1. 📌 Find fuzzy set A∪B for given fuzzy sets A and B

[ Dec 2024 Q1(e)(i) | Frequency: 1 ]

Fuzzy Set Union (A∪B):

The union of two fuzzy sets A and B is defined by taking the maximum of the membership values for each element.

Formula:

Example:

Let Universal Set X = {1, 2, 3, 4, 5}

Result:

• A = {0.2/1, 0.5/2, 0.8/3, 0.4/4, 0.1/5}

• B = {0.6/1, 0.3/2, 0.7/3, 0.9/4, 0.2/5}

• A∪B = {0.6/1, 0.5/2, 0.8/3, 0.9/4, 0.2/5}

Q2. 📌 Find fuzzy set A∩B for given fuzzy sets A and B

[ Dec 2024 Q1(e)(ii) | Frequency: 1 ]

Fuzzy Set Intersection (A∩B):

The intersection of two fuzzy sets A and B is defined by taking the minimum of the membership values for each element.

Formula:

Example:

Using the same fuzzy sets A and B:

Result:

• A∩B = {0.2/1, 0.3/2, 0.7/3, 0.4/4, 0.1/5}

Q3. 📌 Find fuzzy set (A∩B)' for given fuzzy sets A and B

[ Dec 2024 Q1(e)(iii) | Frequency: 1 ]

Complement of Fuzzy Set Intersection (A∩B)':

First, find A∩B (intersection), then calculate its complement.

Complement Formula:

Step 1: Find A∩B

From Q2: A∩B = {0.2/1, 0.3/2, 0.7/3, 0.4/4, 0.1/5}

Step 2: Find (A∩B)'

Result:

• (A∩B)' = {0.8/1, 0.7/2, 0.3/3, 0.6/4, 0.9/5}

Q4. ⭐ Prove Commutative property of fuzzy sets with suitable example

[ Dec 2024 Q4(b)(i) | Frequency: 1 ]

Commutative Property:

For fuzzy sets A and B:

• A ∪ B = B ∪ A (Union is commutative)

• A ∩ B = B ∩ A (Intersection is commutative)

Proof:

For Union:

For Intersection:

Example:

Let X = {a, b, c}

• A = {0.3/a, 0.7/b, 0.5/c}

• B = {0.6/a, 0.4/b, 0.8/c}

Verified: A∪B = B∪A = {0.6/a, 0.7/b, 0.8/c} and A∩B = B∩A = {0.3/a, 0.4/b, 0.5/c}

Q5. ⭐ Prove Associative property of fuzzy sets with suitable example

[ Dec 2024 Q4(b)(ii) | Frequency: 1 ]

Associative Property:

For fuzzy sets A, B, and C:

• (A ∪ B) ∪ C = A ∪ (B ∪ C)

• (A ∩ B) ∩ C = A ∩ (B ∩ C)

Proof:

For Union:

Example:

Let X = {1, 2}

• A = {0.2/1, 0.5/2}

• B = {0.4/1, 0.3/2}

• C = {0.6/1, 0.8/2}

For Union:

Verified: (A∪B)∪C = A∪(B∪C) = {0.6/1, 0.8/2}

Q6. ⭐ Prove Distributive property of fuzzy sets with suitable example

[ Dec 2024 Q4(b)(iii) | Frequency: 1 ]

Distributive Property:

For fuzzy sets A, B, and C:

• A ∪ (B ∩ C) = (A ∪ B) ∩ (A ∪ C)

• A ∩ (B ∪ C) = (A ∩ B) ∪ (A ∩ C)

Proof:

By properties of max and min functions, both expressions are equal.

Example:

Let X = {p, q}

• A = {0.3/p, 0.6/q}

• B = {0.5/p, 0.4/q}

• C = {0.7/p, 0.2/q}

Verified: A∪(B∩C) = (A∪B)∩(A∪C) = {0.5/p, 0.6/q}

Q7. 🔥 Prove De-Morgan's laws for fuzzy sets with suitable example

[ Dec 2024 Q4(b)(iv) | Frequency: 1 ]

De-Morgan's Laws for Fuzzy Sets:

• (A ∪ B)' = A' ∩ B'

• (A ∩ B)' = A' ∪ B'

Proof of First Law:

LHS: $\mu_{(A \cup B)'}(x) = 1 - \max[\mu_A(x), \mu_B(x)]$

RHS: $\mu_{A' \cap B'}(x) = \min[1-\mu_A(x), 1-\mu_B(x)]$

Since $1 - \max[a, b] = \min[1-a, 1-b]$, LHS = RHS

Example:

Let X = {1, 2, 3}

• A = {0.2/1, 0.7/2, 0.4/3}

• B = {0.5/1, 0.3/2, 0.6/3}

Verified: (A∪B)' = A'∩B' = {0.5/1, 0.3/2, 0.4/3}

Diagram:

8.7 Rough Set Theory

Q8. 🔥 Explain Rough Set Theory

[ Dec 2022 Q5(d) | Frequency: 1 ]

Rough Set Theory:

Rough Set Theory, introduced by Zdzisław Pawlak in 1982, is a mathematical approach to deal with imprecision, vagueness, and uncertainty in data. Unlike fuzzy sets, rough sets work with indiscernibility based on available information.

Key Concepts:

Approximations:

1. Lower Approximation (R_X): Objects that certainly belong to set X

2. Contains objects whose equivalence class is completely contained in X

3. Upper Approximation (R̄_X): Objects that possibly belong to set X

4. Contains objects whose equivalence class has non-empty intersection with X

5. Boundary Region: Objects that cannot be classified with certainty

6. Boundary = Upper Approximation - Lower Approximation

Diagram:

Example:

Consider patients with symptoms and diagnosis:

For X = {Patients with Flu} = {P1, P2}

• Indiscernibility: P1 and P2 are indiscernible (same symptoms)

• Lower Approximation: {P1, P2} (certainly have Flu)

• Upper Approximation: {P1, P2} (possibly have Flu)

• Boundary: Empty (crisp classification possible)

Properties of Rough Sets:

Applications:

• Data Mining: Feature selection and reduction

• Medical Diagnosis: Handling incomplete patient data

• Pattern Recognition: Classification with uncertain data

• Decision Support Systems: Rule extraction from data

Comparison: Fuzzy Sets vs Rough Sets:

Diagram:

Summary: Fuzzy Set Operations Quick Reference

Properties Summary

UNIT 9: Introduction to Machine Learning Methods

Q1. 📌 State and discuss the basic steps of machine learning cycle.

[ Dec 2024 Q1(f) | Frequency: 1 ]

The Machine Learning Cycle consists of systematic steps to build and deploy ML models:

Q2. 🔥 Draw block diagram for the machine learning cycle.

[ Jun 2024 Q1(e), Jun 2022 Q2(b) | Frequency: 2 ]

Diagram:

Key Components:

• Data Collection → Preparation: Raw data is collected and cleaned

• Feature Engineering → Model Selection: Features extracted and algorithm chosen

• Training → Evaluation → Tuning: Iterative process until optimal performance

• Deployment → Monitoring: Model goes live with continuous monitoring

Q3. 📌 List the steps involved in machine learning cycle.

[ Jun 2024 Q1(e) | Frequency: 1 ]

Steps in Machine Learning Cycle:

1. Problem Definition - Clearly define the objective and success criteria

2. Data Collection - Gather data from relevant sources

3. Data Preprocessing - Clean, normalize, and handle missing values

4. Exploratory Data Analysis (EDA) - Understand data patterns and distributions

5. Feature Engineering - Create and select important features

6. Data Splitting - Divide data into training, validation, and test sets

7. Model Selection - Choose appropriate ML algorithm

8. Model Training - Train model on training dataset

9. Model Evaluation - Assess performance using appropriate metrics

10. Hyperparameter Tuning - Optimize model parameters

11. Model Deployment - Deploy model to production

12. Monitoring & Maintenance - Track performance and update as needed

Q4. 📌 Discuss the role of each component of machine learning cycle briefly.

[ Jun 2022 Q2(b) | Frequency: 1 ]

Q5. 📌 Differentiate between Machine Learning and Data Mining.

[ Jun 2022 Q3(b) | Frequency: 1 ]

Q6. 📌 Give an example of Machine Learning.

[ Jun 2022 Q3(b) | Frequency: 1 ]

Example: Email Spam Detection

• Problem: Classify incoming emails as spam or not spam

• Data: Historical emails labeled as spam/not-spam

• Features: Words frequency, sender information, links, attachments

• Algorithm: Naive Bayes or Neural Network

• Training: Model learns patterns distinguishing spam from legitimate emails

• Prediction: New emails are automatically classified

Other Examples:

• Netflix movie recommendations

• Voice assistants (Siri, Alexa)

• Fraud detection in banking

• Self-driving cars

• Medical diagnosis systems

Q7. 📌 Give an example of Data Mining.

[ Jun 2022 Q3(b) | Frequency: 1 ]

Example: Market Basket Analysis in Retail

• Problem: Discover which products are frequently bought together

• Data: Transaction records from supermarket

• Technique: Association Rule Mining (Apriori Algorithm)

• Discovery: "Customers who buy bread also tend to buy butter (confidence: 85%)"

• Application: Product placement, cross-selling strategies

Other Examples:

• Customer segmentation in banking

• Detecting fraudulent insurance claims

• Identifying disease outbreaks from health records

• Social network analysis

• Web usage mining for website optimization

Q8. 🔥 Discuss Supervised Learning with suitable example.

[ Dec 2024 Q1(a), Dec 2022 Q2(b)(ii) | Frequency: 2 ]

Supervised Learning is a type of machine learning where the model is trained on labeled data - data with known input-output pairs. The algorithm learns to map inputs to correct outputs.

Characteristics:

• Requires labeled training data

• Goal is to learn a mapping function: Y = f(X)

• Used for classification and regression tasks

• Model is "supervised" by correct answers

Working Process:

1. Provide training data with labels

2. Algorithm learns patterns from input-output relationships

3. Model is evaluated on test data

4. Trained model predicts outputs for new inputs

Diagram:

Example: House Price Prediction

• Input Features: Size, location, bedrooms, age

• Label: Actual house price

• Training: Model learns relationship between features and prices

• Prediction: Estimate price for new houses

Algorithms: Linear Regression, Decision Trees, SVM, Neural Networks, Random Forest

Q9. 🔥 Discuss Unsupervised Learning with suitable example.

[ Dec 2024 Q1(a), Dec 2022 Q2(b)(iii) | Frequency: 2 ]

Unsupervised Learning is a type of machine learning where the model is trained on unlabeled data - data without predefined categories or outcomes. The algorithm discovers hidden patterns or structures.

Characteristics:

• No labeled data required

• Goal is to find hidden structure in data

• Used for clustering, association, dimensionality reduction

• Algorithm finds patterns on its own

Working Process:

1. Provide unlabeled data

2. Algorithm identifies patterns, similarities, or groupings

3. Output reveals data structure or clusters

4. Human interprets the discovered patterns

Diagram:

Example: Customer Segmentation

• Input: Customer purchase history, demographics, browsing behavior

• No Labels: Categories are unknown

• Process: Algorithm groups similar customers together

• Output: Customer segments (e.g., "budget shoppers," "premium customers")

• Application: Targeted marketing strategies

Algorithms: K-Means Clustering, Hierarchical Clustering, PCA, Apriori Algorithm, DBSCAN

Q10. ⭐ Compare Supervised Learning and Unsupervised Learning.

[ Jun 2022 Q1(f) | Frequency: 1 ]

Q11. 📌 List the algorithms for Supervised Learning.

[ Jun 2022 Q1(f) | Frequency: 1 ]

Classification Algorithms:

1. Logistic Regression

2. Decision Trees

3. Random Forest

4. Support Vector Machine (SVM)

5. K-Nearest Neighbors (KNN)

6. Naive Bayes

7. Neural Networks

8. Gradient Boosting (XGBoost, AdaBoost)

Regression Algorithms:

1. Linear Regression

2. Polynomial Regression

3. Ridge Regression

4. Lasso Regression

5. Decision Tree Regression

6. Random Forest Regression

7. Support Vector Regression (SVR)

8. Neural Network Regression

Q12. 📌 List the algorithms for Unsupervised Learning.

[ Jun 2022 Q1(f) | Frequency: 1 ]

Clustering Algorithms:

1. K-Means Clustering

2. Hierarchical Clustering

3. DBSCAN (Density-Based)

4. Gaussian Mixture Models

5. Mean Shift Clustering

Association Rule Algorithms:

1. Apriori Algorithm

2. FP-Growth Algorithm

3. Eclat Algorithm

Dimensionality Reduction:

1. Principal Component Analysis (PCA)

2. t-SNE (t-Distributed Stochastic Neighbor Embedding)

3. Linear Discriminant Analysis (LDA)

4. Singular Value Decomposition (SVD)

Anomaly Detection:

1. Isolation Forest

2. One-Class SVM

3. Local Outlier Factor (LOF)

Q13. 📌 Discuss Rote Learning with suitable example.

[ Dec 2022 Q2(b)(i) | Frequency: 1 ]

Rote Learning is the simplest form of machine learning where the system memorizes training examples exactly and retrieves them during inference without generalization.

Characteristics:

• Direct Memorization: Stores input-output pairs exactly

• No Generalization: Cannot handle new, unseen examples

• Fast Retrieval: Quick lookup from memory

• Memory Intensive: Requires large storage for all examples

• Exact Matching: Only works for previously seen inputs

Working Process:

1. Store all training examples in memory

2. When new input arrives, search for exact match

3. Return stored output for matched input

4. If no match found, cannot provide answer

Example: Lookup Table for Multiplication

• System memorizes all multiplication facts

• Query: 5 × 7 → Retrieves stored answer: 35

• Query: 11 × 13 → Cannot answer (not memorized)

Limitations:

• Cannot generalize to new situations

• Impractical for large problem spaces

• No learning of underlying concepts

Q14. 🔥 Compare/Discuss Descriptive, Predictive, and Prescriptive Analytics in Machine Learning.

[ Jun 2025 Q1(b), Jun 2024 Q2(a), Jun 2023 Q1(a) | Frequency: 3 ]

Diagram:

Examples:

Q15. 📌 Discuss Rule-based Machine Learning with suitable example.

[ Jun 2023 Q2(b)(i) | Frequency: 1 ]

Rule-based Machine Learning uses a set of IF-THEN rules to make predictions or decisions. These rules can be created by domain experts or learned from data.

Characteristics:

• Uses explicit IF-THEN rules

• Highly interpretable and explainable

• Easy to modify and maintain

• Works well with structured knowledge

• May not handle complex patterns

Structure of Rules:

IF condition1 AND condition2 THEN action/conclusion

Learning Approaches:

1. Expert-defined Rules: Domain experts create rules manually

2. Rule Induction: Algorithms learn rules from data (RIPPER, OneR)

3. Association Rule Mining: Discover rules from patterns (Apriori)

Example: Medical Diagnosis System

Rule 1: IF fever > 101°F AND cough = yes AND fatigue = yes 
        THEN diagnosis = "Flu"

Rule 2: IF fever > 101°F AND rash = yes AND joint_pain = yes 
        THEN diagnosis = "Dengue"

Rule 3: IF fever < 99°F AND runny_nose = yes AND sneezing = yes 
        THEN diagnosis = "Common Cold"

Advantages:

• Transparent decision-making

• Easy to audit and validate

• Domain knowledge integration

Disadvantages:

• Difficult to handle exceptions

• May conflict or overlap

• Doesn't generalize well

Q16. 📌 Discuss Bayesian Algorithms with suitable example.

[ Jun 2023 Q2(b)(ii) | Frequency: 1 ]

Bayesian Algorithms are based on Bayes' Theorem, which calculates the probability of a hypothesis given observed evidence. They provide probabilistic predictions with uncertainty quantification.

Bayes' Theorem:

Where:

• P(H|E): Posterior probability (probability of hypothesis given evidence)

• P(E|H): Likelihood (probability of evidence given hypothesis)

• P(H): Prior probability (initial belief)

• P(E): Evidence probability

Common Bayesian Algorithms:

1. Naive Bayes Classifier

2. Bayesian Networks

3. Bayesian Linear Regression

4. Gaussian Naive Bayes

Example: Email Spam Classification using Naive Bayes

Training Data Analysis:

• P(Spam) = 0.3 (30% of emails are spam)

• P("Free" | Spam) = 0.8

• P("Free" | Not Spam) = 0.1

Prediction for new email containing "Free":

P(Spam | "Free") = P("Free"|Spam) × P(Spam) / P("Free")
                 = (0.8 × 0.3) / P("Free")
                 = High probability → Classify as SPAM

Advantages:

• Handles uncertainty naturally

• Works well with small datasets

• Provides probability scores

• Fast training and prediction

Applications: Spam filtering, sentiment analysis, medical diagnosis, recommendation systems

Q17. 📌 What is Reinforcement Learning?

[ Dec 2022 Q1(f) | Frequency: 1 ]

Reinforcement Learning (RL) is a type of machine learning where an agent learns to make decisions by interacting with an environment. The agent receives rewards or penalties based on its actions and learns to maximize cumulative rewards over time.

Key Concepts:

• Agent: The learner/decision-maker

• Environment: The world the agent interacts with

• State: Current situation of the agent

• Action: Possible moves the agent can make

• Reward: Feedback signal (positive or negative)

• Policy: Strategy for choosing actions

Learning Process:

1. Agent observes current state

2. Agent takes an action based on policy

3. Environment provides reward and new state

4. Agent updates policy to maximize future rewards

5. Process repeats until goal is achieved

Examples:

• Game playing (Chess, Go, Video games)

• Robot navigation

• Autonomous vehicles

• Stock trading algorithms

Q18. ⭐ Discuss Reinforcement Learning.

[ Dec 2024 Q1(a) | Frequency: 1 ]

Reinforcement Learning (RL) is a learning paradigm where an agent learns optimal behavior through trial and error interactions with the environment, guided by reward signals.

Core Components:

Types of RL:

1. Positive Reinforcement: Reward for good actions

2. Negative Reinforcement: Penalty for bad actions

Key Characteristics:

• No labeled data required

• Learns from consequences of actions

• Balances exploration vs exploitation

• Goal is long-term reward maximization

Applications:

• AlphaGo (game playing)

• Self-driving cars

• Robotics control

• Resource management

• Personalized recommendations

Q19. 🔥 Explain Reinforcement Learning with block diagram and component roles.

[ Dec 2023 Q4(a) | Frequency: 1 ]

Reinforcement Learning is a goal-oriented learning approach where an agent learns to achieve goals by interacting with an environment.

Diagram:

Component Roles:

Learning Loop:

1. Agent observes state St

2. Agent selects action At using policy

3. Environment transitions to state St+1

4. Environment provides reward Rt+1

5. Agent updates policy based on experience

6. Repeat until convergence

Q20. 📌 Classify the various Reinforcement Learning algorithms.

[ Dec 2022 Q1(f) | Frequency: 1 ]

Diagram:

Classification Summary:

Q21. 📌 Discuss Delayed-Reinforcement Learning with suitable example.

[ Dec 2022 Q2(b)(iv) | Frequency: 1 ]

Delayed Reinforcement Learning refers to situations where rewards are not received immediately after an action but are delayed until much later, often after a sequence of actions.

Characteristics:

• Reward signal comes after many time steps

• Difficult to attribute credit to specific actions

• Requires temporal credit assignment

• Common in real-world problems

Challenges:

1. Credit Assignment Problem: Which actions led to the reward?

2. Long-term Dependencies: Early actions affect late rewards

3. Exploration: Must try sequences to discover rewards

4. Sparse Rewards: Few feedback signals to learn from

Example: Chess Game

• Agent plays hundreds of moves during a game

• Reward: +1 for winning, -1 for losing, 0 for draw

• Reward received only at game end

• Challenge: Which moves were responsible for the outcome?

Timeline:

Move 1 → Move 2 → Move 3 → ... → Move 50 → Win (+1 Reward)
  ↑
Which move was crucial for winning?

Solutions:

1. Temporal Difference (TD) Learning: Estimate value at each step

2. Eligibility Traces: Track which states were visited

3. Monte Carlo Methods: Average rewards over complete episodes

4. Reward Shaping: Add intermediate rewards to guide learning

Other Examples:

• Robot learning to walk (reward only when destination reached)

• Stock trading (profit/loss realized later)

• Drug treatment effects (outcomes visible after months)

Q22. ⭐ Compare Model-free Reinforcement Learning with Model-based Reinforcement Learning.

[ Jun 2024 Q4(a) | Frequency: 1 ]

Diagram:

When to Use:

• Model-Free: When environment is complex, hard to model

• Model-Based: When sample efficiency is critical, model is simple

Q23. ⭐ Discuss the sub-classes of Model-free Reinforcement Learning.

[ Jun 2024 Q4(a) | Frequency: 1 ]

Model-Free RL has three main sub-classes:

Diagram:

1. Value-Based Methods:

• Learn value function (V or Q)

• Policy derived from value function

• Algorithms: Q-Learning, SARSA, DQN

2. Policy-Based Methods:

• Learn policy directly (π: state → action)

• No value function required

• Algorithms: REINFORCE, Policy Gradient

3. Actor-Critic Methods:

• Combines value-based and policy-based

• Actor: Learns policy

• Critic: Evaluates actions using value function

• Algorithms: A2C, A3C, PPO, SAC

Q24. ⭐ Discuss Ensemble Learning / What is Ensemble Learning.

[ Dec 2024 Q1(g), Jun 2023 Q1(f) | Frequency: 2 ]

Ensemble Learning is a machine learning technique that combines multiple models (base learners) to produce a more accurate and robust prediction than any individual model.

Core Principle: "Wisdom of the crowd" - combining diverse opinions often yields better results.

Key Ensemble Methods:

Diagram:

Advantages:

• Higher accuracy than individual models

• Reduced overfitting

• More robust predictions

• Handles noise better

Example - Random Forest:

• Combines multiple decision trees

• Each tree trained on different data subset

• Final prediction: majority vote of all trees

Q25. 📌 Explain Ensemble Methods.

[ Dec 2022 Q5(g) | Frequency: 1 ]

Ensemble Methods combine predictions from multiple machine learning models to improve overall performance.

Types of Ensemble Methods:

1. Bagging (Bootstrap Aggregating):

• Creates multiple training sets by sampling with replacement

• Trains independent models on each set

• Aggregates predictions (voting/averaging)

• Example: Random Forest

2. Boosting:

• Trains models sequentially

• Each model focuses on errors of previous models

• Weighted combination of models

• Examples: AdaBoost, Gradient Boosting, XGBoost

3. Stacking:

• Trains diverse base models

• Uses another model (meta-learner) to combine predictions

• More complex but often more accurate

4. Voting:

• Hard Voting: Majority class prediction

• Soft Voting: Average of predicted probabilities

Diagram:

Comparison:

Summary Table: Unit 9 - Machine Learning Methods

UNIT 10: Classification

Q1. ⭐ Differentiate between Lazy Learners and Eager Learners in Classification.

[ Dec 2023 Q1(e), Dec 2022 Q3(b) | Frequency: 2 ]

Key Difference: Lazy learners wait until prediction time to generalize, while eager learners generalize during the training phase itself.

Q2. ⭐ List/Give examples of algorithms used for Lazy Learners.

[ Dec 2023 Q1(e), Dec 2022 Q3(b) | Frequency: 2 ]

Lazy Learner Algorithms:

1. K-Nearest Neighbors (KNN)
2. Classifies based on majority class of K nearest neighbors

3. Stores all training instances

4. Case-Based Reasoning (CBR)

5. Solves new problems using solutions from similar past cases

6. Maintains case library

7. Locally Weighted Learning (LWL)

8. Builds local models around query point

9. Weights instances by proximity

10. Radius Neighbors Classifier

11. Similar to KNN but uses fixed radius instead of K neighbors

12. Lazy Bayesian Rules

13. Constructs Bayesian rules only when needed

Q3. ⭐ List/Give examples of algorithms used for Eager Learners.

[ Dec 2023 Q1(e), Dec 2022 Q3(b) | Frequency: 2 ]

Eager Learner Algorithms:

1. Decision Trees (ID3, C4.5, CART)

2. Builds tree structure during training

3. Naive Bayes Classifier

4. Learns probability distributions from training data

5. Support Vector Machine (SVM)

6. Constructs optimal hyperplane during training

7. Neural Networks

8. Learns weights through backpropagation

9. Logistic Regression

10. Learns model parameters during training

11. Random Forest

12. Builds ensemble of decision trees

13. Linear Discriminant Analysis (LDA)

14. Computes discriminant functions during training

Q4. ⭐ Discuss the concept of Classification.

[ Jun 2025 Q1(f) | Frequency: 1 ]

Classification is a supervised machine learning technique that categorizes data into predefined classes or labels based on input features.

Key Characteristics:

• Supervised Learning: Requires labeled training data

• Discrete Output: Predicts categorical class labels

• Pattern Recognition: Learns decision boundaries from data

Working Process:

Diagram:

Components:

1. Training Phase: Algorithm learns from labeled examples

2. Model Building: Creates decision boundaries

3. Prediction Phase: Classifies new, unseen instances

Types:

• Binary Classification (2 classes)

• Multi-class Classification (>2 classes)

• Multi-label Classification (multiple labels per instance)

Applications:

• Spam detection, Medical diagnosis, Image recognition, Sentiment analysis

Q5. 🔥 Give/List the algorithms for Classification.

[ Jun 2025 Q1(f), Jun 2024 Q1(g), Dec 2023 Q1(a) | Frequency: 3 ]

Classification Algorithms:

Most Commonly Used:

1. Logistic Regression - Simple, interpretable

2. Decision Trees - Easy to visualize and understand

3. Random Forest - High accuracy, handles overfitting

4. SVM - Effective in high-dimensional spaces

5. KNN - Simple, no training phase

6. Naive Bayes - Fast, works well with text data

7. Neural Networks - Complex pattern recognition

Q6. 📌 Give suitable example for Classification.

[ Jun 2024 Q1(g) | Frequency: 1 ]

Example 1: Email Spam Classification

Example 2: Medical Diagnosis

• Input: Patient symptoms, test results, age, medical history

• Output: Disease category (Diabetic / Non-Diabetic)

Example 3: Loan Approval

• Features: Income, credit score, employment status, debt ratio

• Classes: Approved / Rejected

Example 4: Image Classification

• Input: Pixel values of an image

• Output: Cat / Dog / Bird

Example 5: Sentiment Analysis

• Input: Customer review text

• Output: Positive / Negative / Neutral

Q7. ⭐ Compare/Differentiate Clustering and Classification.

[ Jun 2024 Q1(g), Dec 2023 Q1(a) | Frequency: 2 ]