(5th sem) AIML Chapter 2: Complete Concepts Guide
Chapter 2: Knowledge Representation and Concept Learning -> Generated and Prepared By Thiruselvan (ThiruXD)
Part A: Introduction to Knowledge Representation (Unit 6)
1. What is Knowledge Representation?
Knowledge Representation (KR) is the method of storing information, facts, rules and relationships in a form that a computer can understand, process and use for reasoning.
Simple meaning:
It is the way we give knowledge to an AI system so that it can think and make decisions.
Why is it important?
Without proper knowledge representation, an AI system cannot reason, answer questions or solve problems.
2. Goals of Knowledge Representation
- Represent knowledge accurately
- Allow easy inference (drawing conclusions)
- Support efficient reasoning
- Make knowledge easy to acquire and update
3. Knowledge Representation Issues
- What to represent? (Which facts are important?)
- How to represent? (Which format to use?)
- Knowledge Acquisition (How to get knowledge into the system?)
- Updating and Maintenance (How to change knowledge when new information comes?)
4. Requirements of a Good Knowledge Representation
| Requirement | Meaning |
|---|---|
| Representational Adequacy | Can represent all kinds of required knowledge |
| Inferential Adequacy | Can draw new conclusions from existing knowledge |
| Inferential Efficiency | Can draw conclusions quickly |
| Acquisitional Efficiency | Easy to add new knowledge |
5. Types of Knowledge
| Type | Meaning | Example |
|---|---|---|
| Declarative Knowledge | Facts and information | “Earth is round” |
| Procedural Knowledge | How to do something | Steps to diagnose a disease |
| Heuristic Knowledge | Rules of thumb / expert experience | “If fever + cough → high chance of flu” |
| Meta Knowledge | Knowledge about knowledge | “This rule is more reliable than that rule” |
Part B: Predicate Logic (Unit 7)
1. Why Predicate Logic?
Propositional logic is too limited (it can only say True/False for whole sentences).
Predicate Logic (also called First-Order Logic) is more powerful because it can talk about objects, properties and relationships.
2. Components of Predicate Logic
| Component | Meaning | Example |
|---|---|---|
| Constants | Specific objects | Ram, 5, Delhi |
| Variables | Placeholders for objects | x, y, z |
| Predicates | Properties or relations | Student(x), Loves(x,y) |
| Functions | Mapping from objects to objects | FatherOf(x), Age(x) |
3. Quantifiers
-
Universal Quantifier (∀) → “For all”
Example: ∀x Student(x) → Intelligent(x)
(All students are intelligent)
-
Existential Quantifier (∃) → “There exists”
Example: ∃x Student(x) ∧ Intelligent(x)
(There exists at least one intelligent student)
4. Representing Knowledge using Predicate Logic
Facts:
- Ram is a student → Student(Ram)
- Ram loves Sita → Loves(Ram, Sita)
Rules:
- All humans are mortal → ∀x Human(x) → Mortal(x)
5. Inference in Predicate Logic
Important techniques:
- Unification – Making two expressions look the same by substituting values
- Resolution – A powerful method to prove theorems
- Logical reasoning using quantifiers and predicates
Part C: Rule-Based Knowledge Representation (Unit 8)
1. What are Rule-Based Systems?
A rule-based system stores knowledge in the form of IF–THEN rules.
Structure of a Rule:
IF condition THEN action/conclusion
Example:
IF fever AND cough THEN diagnosis = flu
2. Components of a Rule-Based System
| Component | Role |
|---|---|
| Knowledge Base | Stores all the IF–THEN rules |
| Inference Engine | Applies the rules to draw conclusions |
| Working Memory | Stores current facts and temporary data |
3. Inference Mechanisms
A. Forward Chaining (Data-driven)
- Starts from known facts
- Applies rules
- Reaches conclusion
- Example: Given symptoms → find disease
B. Backward Chaining (Goal-driven)
- Starts from a goal
- Works backwards to find supporting facts
- Example: Suspect flu → check if symptoms match
4. Applications
- Expert Systems
- Medical Diagnosis
- Decision Support Systems
Part D: Concept Learning Fundamentals (Unit 9)
1. What is Concept Learning?
Concept Learning is the process of learning a general concept (category) from a set of positive and negative training examples.
Simple meaning:
The system learns “what makes something belong to a category”.
Example: Learning the concept of “bird” from examples of sparrows, eagles (positive) and bats, airplanes (negative).
2. Concept Learning Task
- Training Examples – Positive and negative instances
- Target Concept – The true concept we want to learn
- Hypothesis Space – All possible concepts the system can consider
3. Concept Learning as Search
The system searches through the hypothesis space to find the correct concept.
Two common strategies:
- General-to-Specific Search
- Specific-to-General Search
Version Space = Set of all hypotheses that are consistent with the training examples seen so far.
Part E: Find-S Algorithm (Unit 10)
1. Purpose of Find-S
Find-S finds the most specific hypothesis that fits all positive training examples.
2. How Find-S Works
- Start with the most specific hypothesis (everything is null)
- For each positive example:
- Generalize the hypothesis just enough to cover the example
- Ignore negative examples
- Final hypothesis is the most specific one that covers all positive examples
3. Characteristics
- Very simple
- Finds the most specific hypothesis
- Does not consider negative examples
- Can miss better hypotheses
4. Advantages & Limitations
Advantages: Simple, fast, easy to understand
Limitations: Ignores negative examples, finds only one hypothesis, sensitive to noise
Part F: Candidate Elimination Algorithm (Unit 11)
1. Why Candidate Elimination?
Find-S gives only one hypothesis.
Candidate Elimination maintains the entire Version Space (all possible consistent hypotheses).
2. Version Space Representation
Version Space is represented by two boundaries:
- S-boundary (Specific Boundary) → Most specific hypotheses
- G-boundary (General Boundary) → Most general hypotheses
All hypotheses between S and G are possible.
3. How the Algorithm Works
- Positive example → Generalize S-boundary
- Negative example → Specialize G-boundary
- Remove any hypothesis that becomes inconsistent
4. Advantages & Limitations
Advantages:
- Finds all possible consistent hypotheses
- Handles both positive and negative examples
- More complete than Find-S
Limitations:
- Computationally expensive for large hypothesis spaces
- Sensitive to noisy data
Part G: Inductive Bias (Unit 12)
1. What is Inductive Bias?
Inductive Bias is the set of extra assumptions a learning algorithm makes to generalize beyond the training examples.
Simple meaning:
Without bias, a system cannot learn anything useful from limited examples. Bias helps it make smart guesses.
2. Types of Inductive Bias
| Type | Meaning |
|---|---|
| Restriction Bias | Limits the hypothesis space (considers only some types of hypotheses) |
| Preference Bias | Prefers some hypotheses over others (e.g., simpler ones) |
3. Inductive Bias in Algorithms
- Find-S → Prefers the most specific hypothesis (Preference Bias)
- Candidate Elimination → Considers only hypotheses in the version space (Restriction Bias)
4. Importance of Inductive Bias
- Enables generalization
- Improves learning efficiency
- Helps in selecting better models
- Essential for real-world learning
5. Applications
- Machine Learning systems
- Pattern Recognition
- Intelligent decision-making systems
Complete Chapter 2 Summary (Must Remember)
- Knowledge Representation = Storing knowledge so AI can reason.
- Good KR needs: Representational Adequacy, Inferential Adequacy, Efficiency, Acquisitional Efficiency.
- Predicate Logic is more powerful than Propositional Logic (uses predicates, quantifiers, variables).
- Rule-Based Systems use IF–THEN rules + Inference Engine (Forward & Backward Chaining).
- Concept Learning = Learning a general concept from examples.
- Find-S → Finds most specific hypothesis (ignores negatives).
- Candidate Elimination → Maintains Version Space (S and G boundaries).
- Inductive Bias is necessary for generalization.