n l y

e o
Knowledge
o s
r p
Representation and
u
y P
Reasoning

t u d
S
Dr. S. S. OHOL,

r
Associate Professor,

o
Mechanical Engineering Department,

F
COEP Technological University, Pune.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 1
 n l y
o
Introduction to Knowledge Representation (KR) and Reasoning

s e
Definition:

r p o
u
• Knowledge Representation: A method for encoding knowledge about the world in a form that a machine can utilize to

P
solve complex problems.

y
• Reasoning: The process of deriving new information or drawing conclusions from known facts using logical rules.

d
Importance:

u
• Helps AI systems "understand" and "manipulate" knowledge.

t
• Essential for decision-making, automated reasoning, and understanding real-world scenarios.

S
Applications:

o r
• Natural language processing (NLP)

F
• Autonomous systems (robots, self-driving cars)
• Expert systems (medical diagnosis, legal reasoning)

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 2
 n l y
What is an Ontology
e o
o s
Definition:

u r p
P
• An ontology is a formal, explicit specification of a shared
conceptualization.

d y
• Provides a structured framework to model knowledge in terms of

u
entities, their attributes, and their relationships.

S
Examples:
t
o r
• Domain-specific ontologies in healthcare (diseases, symptoms,
treatments).

F
• Ontologies in e-commerce (products, categories, customer behaviors).

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 3
 n l y
Components of Ontologies
e o
o s
r p
• Classes: Represent concepts or categories (e.g., "Person," "Vehicle").

u
P
• Instances: Specific objects within a class (e.g., "John" is an instance of
"Person").

d y
• Attributes: Characteristics of classes or instances (e.g., "Age" of a

u
person, "Speed" of a vehicle).

S t
• Relations: Connections between entities (e.g., "hasPart," "isA").

o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 4
 n l y
Ontologies in AI
e o
o s
Applications:

u r p
• Natural Language Processing (NLP): Helps understand semantic relationships between

P
words (e.g., WordNet).

y
• Robotics: Used in perception and planning systems for reasoning about objects and their

d
relationships.

t u
• Semantic Web: Ontologies are used to organize and retrieve web data in a meaningful way.

S
Examples:

r
• NLP: Enhancing machine translation or question-answering systems by using ontologies to

o
resolve ambiguity.

F
• Robotics: Using an ontology to understand objects in the environment and plan
interactions (e.g., grasping objects).

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 5
 n l y
Example Ontologies
e o
o s
Medical Ontology

u r p
P
• A structured ontology for diseases, symptoms, treatments, and drugs.

y
• Shows how such an ontology can assist in diagnosis or decision-making.

t u d
E-commerce Ontology
• An ontology that maps out product categories, attributes, and customer

S
preferences.

r
o
• Demonstrates how e-commerce platforms personalize

F
recommendations.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 6
 n l y
Triple Stores: Introduction
e o
o s
• A triple store is a database for storing information modeled as

r p
subject-predicate-object (SPO) triples.

u
P
• Example of a triple: "Anna owns a Toyota"

y
o Subject (S): Anna

t u d
o Predicate (P): owns a
o Object (O): Toyota

r S
Fo
Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 7
Multiple Triples: Extending Relationships n l y
e o
o s
The same structure can be used to model various relationships. For
example:

u r p
P
• "Anna works at yWorks"

y
• "Anna uses a Yellow Umbrella"

d
S t u
o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 8
 n l y
o
Advantages and Challenges of Triple Stores
e
o s
Advantages:

u r p
• No schema needed beforehand.

P
• Every triple is an array of three strings (simple and fast to parse, index, and query).

y
d
• Easy to add new relationships (just add more triples).

u
Challenges:

S t
• No predefined structure leads to amorphic data (no clear form).

r
• Data integrity is difficult to maintain (same data can be stored in different ways).

F o
• Hard to maintain consistency across updates and deletions.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 9
 n l y
From Triples to Ontologies
e o
Problem with Simple Triples:

o s
r
• Ambiguity arises in real-world scenarios.

u p
• Example: "Anna owns a Toyota" vs. "Anna is driving a Toyota."

P
• How to handle changes (e.g., updating "Yellow Umbrella" to "Red Umbrella")?

y
• Solution: Ontologies

d
• An ontology introduces a formal structure or schema to reduce ambiguity.

t u
• Defines classes (abstract concepts) and relationships between them.

Example:

r S
o
• Person - (owns a) → Car

F
• Person - (works at) → Company
• Person - (uses a) → Umbrella

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 10
 n l y
What is Knowledge Representation?
e o
o s
r p
• Human Understanding vs. Machine Understanding:

u
• Humans excel at interpreting and reasoning with knowledge.

y P
• KR allows machines to perform similar tasks, solving complex
problems (e.g., medical diagnosis).

u d
• Definition of KR:

t
S
• KR is a field of AI focused on how AI agents think and how it

r
contributes to intelligent behavior.

o
• It allows machines to use knowledge to act intelligently, rather

F
than merely storing data.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 11
 n l y
What to Represent in KR
e o
o s
•
r p
Objects: Facts about objects (e.g., "Guitars have strings").

u
P
• Events: Actions occurring in the world.

y
• Performance: How to perform tasks.
•
•
t u d
Meta-Knowledge: Knowledge about knowledge.
Facts: Truths about the real world.
•

r S
Knowledge Base (KB): The central repository of facts (sentences in a

o
technical sense).

F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 12
 n l y
Types of Knowledge
e o
• Declarative Knowledge: Knowing about

o s
facts, objects, and events.

u r
• Procedural Knowledge: Knowing how to do
p
P
tasks (rules, strategies).

y
• Meta-Knowledge: Knowledge about other

d
knowledge.

t u
• Heuristic Knowledge: Expert knowledge
based on experience (rules of thumb).

r S
• Structural Knowledge: Relationships

o
between concepts (e.g., "part of," "kind of").

F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 13
 n l y
Declarative Knowledge
e o
o s
•
r p
Definition: Knowledge about facts and objects.

u
P
• Examples:

y
• "Paris is the capital of France."
•
•
t u d
"Guitars have strings."
Expressed in: Declarative sentences (e.g., facts or statements).
•

r S
Importance: It is simpler than procedural knowledge and foundational

o
in AI.

F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 14
 n l y
AI Knowledge Cycle
e o
Components of the Cycle:
o s
• Perception: Gathering information
from the environment.
u r p
• Learning: Learning from data.

y P
d
• Knowledge Representation:

t u
Structuring the knowledge.

S
• Reasoning: Drawing conclusions from

r
knowledge.

Fo
• Planning and Execution: Acting on
knowledge.

Importance: Planning and execution rely on analysis from KR and reasoning

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 15
 Approaches to Knowledge Representationn l y
e o
o s
•
r p
Simple Relational Knowledge: Organizing facts in tables.

u
P
• Inheritable Knowledge: Storing data in a hierarchy of classes.

y
• Inferential Knowledge: Using formal logic to derive new facts.
•

t u
actions.
d
Procedural Knowledge: Using small programs or codes to perform

r S
F o
Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 16
 n l y
Simple Relational Knowledge
e o
• Definition: Storing facts in relational columns.
o s
• Example:

u r p
P
• Use: Common in database systems.

d y
S t u
o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 17
 n l y
Inheritable Knowledge
e o
• Definition: Storing data in
o s
r p
hierarchies where classes inherit

u
properties from their parent
classes.

y P
d
• Example: Mammals inherit traits

t
from Animals.

S u
o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 18
 n l y
Inferential Knowledge
e o
Definition: Representing knowledge in formal logic.
o s
Example:

u r p
P
• Premise 1: Marcus is a man.

y
• Premise 2: All men are mortal.

d
S t u
• Conclusion: Marcus is mortal.
Use: Helps derive new knowledge based on facts.

o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 19
 Techniques of Knowledge Representation n l y
e o
o s
r p
• Logical Representation: Using concrete rules and propositions.

u
P
• Semantic Networks: Graphical representation of objects and their
relationships.

d y
• Frame Representation: Storing knowledge in structures called frames.

t u
• Production Rules: "If-Then" rules for decision-making.

S
o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 20
 n l y
Logical Representation
e o
o s
r p
• Definition: Logical representation is a method of representing knowledge

u
using formal logic systems, where knowledge is expressed in terms of rules

P
and propositions. It is one of the most precise and structured ways of

y
representing information because it is based on mathematical logic.

d
• Key Components:
• Syntax:

S t u
• The rules that define how symbols and sentences are formed.

o r
• It involves defining which symbols can be used and how to construct
sentences from them.

F • Example: In propositional logic, we might use symbols like P for

propositions and ¬, ∧, ∨ for logical connectives (NOT, AND, OR).

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 21
 n l y
o
Key Components:
Semantics:

s e
• Deals with the meaning of sentences.

r p o
• It assigns truth values to sentences based on the conditions of the
real world or a model.

Pu
y
Inference Rules:

t
propositions.
u d
• Allows the system to draw conclusions from known facts or

r S
• Example: Inference rules like Modus Ponens can help infer new

o
knowledge from existing facts.

F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 22
 Example: Suppose we have two propositions:
n l y
• P: "It is raining."

e o
s
• Q: "The ground will be wet."

o
• If we know the logical rule that "If it rains, the ground will be wet" (P →

p
Q), and we know it’s raining (P is true), then using Modus Ponens, we

r
can infer that Q (the ground will be wet).

u
Advantages:

y P
d
• Enables precise reasoning.

t u
• Based on established mathematical principles.

S
Disadvantages:

r
• It can be computationally expensive, especially for complex problems.

o
F
• Some types of knowledge, such as uncertainty or ambiguous
information, are difficult to represent using logic alone.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 23
 n l y
Introduction to Predicate Logic
e o
o s
• Propositional Logic:

u r p
• Deals with simple, atomic statements (e.g., "It is raining").

y P
• No internal structure within statements; focuses on true/false values.

t u d
• Example: P → Q (If it rains, the ground will be wet).

S
• Predicate Logic:

o r
• Extends propositional logic by introducing variables and quantifiers.

F
• Allows representing relationships and more complex statements.
• Example : ∀x (Human(x) → Mortal(x)) ("All humans are mortal").

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 24
 n l y
o
Syntax and Semantics of Predicate Logic (Quantifiers, Variables,

e
Connectives)

o s
• Syntax:

u r p
P
• Predicates: Represent properties or relations (e.g., Loves(John, Mary)).

y
• Variables: Stand for elements in the domain (e.g x, y ).
• Quantifiers:

t u d
• Universal Quantifier: ∀ (For all)

r S
• Existential Quantifier: ∃ (There exists)
• Connectives:

Fo
• AND (∧), OR (∨), NOT (¬), IMPLIES (→).

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 25
 n l y
o
Predicate Logic for Knowledge Representation
e
o s
r p
• Representing facts, relationships, and rules in knowledge bases.

u
P
• Expert Systems: Encode rules such as "If a patient has a fever, they
might have an infection."

d y
• Natural Language Processing (NLP): Sentence meanings can be

u
expressed using logical forms (e.g., "Every cat is an animal" as

S t
• ∀ x (Cat(x)) → Animal (x))

o r
• Robotics: Describing environments and constraints (e.g., "If an object is
fragile, handle it carefully").

F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 26
 n l y
o
Limitations of Predicate Logic (Undecidability, Expressiveness)

s e
Undecidability:

r p o
Pu
• Not all logical problems have a computational solution. Some statements
may require infinite reasoning to determine their truth.
Expressiveness:

d y
u
• Predicate logic cannot fully capture uncertainty (e.g., probabilistic
reasoning).

S t
r
• Struggles to handle incomplete information or vague concepts.

o
Computational Complexity:

F
• Reasoning with predicate logic can be time-consuming for large systems.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 27
 n l y
Semantic Networks
e o
o s
r p
Definition: A Semantic Network is a knowledge representation technique

u
that uses a graphical structure to represent objects (or concepts) and

y P
their relationships. Objects are represented as nodes, while relationships
between them are depicted as labeled arcs.
Key Concepts:

t u d
• Nodes: Represent objects or concepts (e.g., "Dog," "Animal").

r S
• Arcs: Represent relationships between objects (e.g., "is-a," "has-part").

Fo
• IS-A Relationship: One of the most important types of relationships in
semantic networks. It represents inheritance, meaning that an object
belongs to a certain category or class.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 28
Example: A simple semantic network might show the following relationships:
• Representation of the Example:
n l y
e
• Car → Pick-up Truck: Pick-up truck is a type of car (IS-A relationship).
o
• Car → Audi: Audi is a type of car (IS-A relationship).

o s
p
• Car → BMW: BMW is a type of car (IS-A relationship).

u r
• Audi → Headlight: An Audi has headlights (HAS-PART relationship).

P
• BMW → Headlight: A BMW has headlights (HAS-PART relationship).

d y
S t u
o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 29
 n l y
Advantages:

e o
s
• Intuitive and Easy to Understand: The graphical nature makes it

o
easier for humans to visualize relationships.

r p
• Inheritance: Inheritance of properties from general to specific

u
concepts can reduce redundancy in knowledge representation.
Disadvantages:
y P
t u d
• Scalability: As the size of the network grows, traversing the entire
network can become slow and computationally expensive.

S
• Lack of Quantifiers: Does not easily handle quantifiers like "for

r
o
all" or "some."

F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 30
 n l y
Frame Representation
e o
o s
r p
Definition: Frame representation is a technique where knowledge is

u
stored in data structures called frames. A frame represents an entity or

y P
concept in the world by breaking it down into a set of attributes (or "slots")
and their corresponding values.
Key Concepts:

t u d
• Frames: Structured representation of an object or concept.

r S
• Slots: These are attributes or properties of the frame.

Fo
• Facets: Provide additional details about slots, like default values,
ranges, or conditions.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 31
 n l y
• Example: Consider a frame representing a car:
• Frame: Car
e o
• Slots:
o s
•
•
Make: Toyota
Model: Corolla

u r p
P
• Year: 2020

y
• Color: Red

d
• Owner: Sarah

S t u
• Each attribute (slot) in the frame holds a value. The frame allows the system to encapsulate a

r
complete set of related data about the car entity in a structured way.

o
• Facets Example: A slot for Fuel Efficiency might have a facet that includes a default value,

F
such as "30 MPG," and a condition like "IF electric, THEN 100 MPG."

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 32
 n l y
o
Advantages:

e
• Modular: Frames allow easy modification and extension of data by

s
adding new slots.

r p
making it easier to access and reason with.
o
• Encapsulation: Combines related knowledge in one structured entity,

u
• Default and Inferred Values: Facets allow for more intelligent behavior

P
in filling in gaps when information is missing.

Disadvantages:
d y
t u
• Lack of Inference Mechanism: While frames can store data efficiently,

S
they often rely on external systems for reasoning or inference.

r
• Not Ideal for Dynamic Knowledge: Frames represent static

o
knowledge and are not well-suited for representing rapidly changing

F
information.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 33
 n l y
Production Rules
e o
o s
r p
Definition: Production rules are a type of knowledge representation that consists of a set

u
of condition-action pairs, commonly referred to as IF-THEN rules. These rules define

P
how a system should behave based on certain conditions in its environment.

Key Concepts:

d y
t u
• IF-THEN Structure:

S
• Condition (IF): Specifies when the rule should be applied.

r
• Action (THEN): Defines the action to be taken when the condition is true.

Fo
• Working Memory: Stores the current state of knowledge or facts in the system.
• Conflict Resolution: When multiple rules are triggered at the same time, the system
must decide which rule to apply using a conflict resolution strategy.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 34
 n l y
• Example: In a traffic light control system:

e o
s
• Rule 1:

o
• IF cars are waiting and light is red,
• THEN change light to green.
• Rule 2:
u r p
P
• IF no cars are waiting and light is green,

y
d
• THEN change light to red after 30 seconds.

u
• The system applies these rules based on the state of the environment

S t
(whether cars are waiting or not) and determines the appropriate action
(whether to change the light or not).

o r
F Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 35
 n l y
o
Advantages:

e
• Modular and Scalable: Production rules can be easily added,

s
removed, or modified without affecting the rest of the system.

o
p
• Easy to Understand: Rules are expressed in natural language

u r
terms, making them intuitive for non-programmers to understand.

y P
d
Disadvantages:

t u
• Efficiency Issues: If many rules are active simultaneously, the

S
system may take a long time to resolve which rule to apply.

o r
• No Learning Capability: Production rule systems do not

F
inherently learn from their actions; they only follow predefined
rules.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 36
 n l y
Representing Objects and Relations
e o
o s
r p
• Object Representation: Objects are represented as entities in a

u
knowledge system with attributes and values. Each object has distinct

y P
characteristics that help define its properties.
• Attribute-Value Pairs: This approach involves associating objects with

u d
their attributes (characteristics) and their respective values.
• Example:
t
r S
• Object: Car

o
• Attributes: Color = Red, Make = Toyota, Model = Corolla

F
• Relations between objects can be depicted, such as Car has a
Headlight.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 37
 n l y
Representing Events and Actions
e o
o s
r p
• Event-based Systems: Events describe occurrences at specific times,

u
like "a car started" or "a user clicked a button." Systems based on events

y P
are triggered by actions or changes in the environment.
• Temporal Logic: Used to represent and reason about events and their

u d
sequence over time, ensuring time-dependent relations (e.g., before,

t
after) are accounted for in systems.

S
• Example:

r
o
• Action: Pressing the start button

F
• Event: Engine starts after button press

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 38
 n l y
Representing Time
e o
o s
• Time Points vs. Intervals:

u r p
• Time points refer to specific moments (e.g., 3:00 PM), while time intervals refer to

P
durations (e.g., from 3:00 PM to 4:00 PM).

actions.

d y
• Time representations are crucial for understanding sequences and durations of

t u
• Temporal Reasoning: Involves reasoning about the relationships between time points

S
or intervals, allowing the system to predict or infer future states based on past events.

r
• Example:

o
• Before: Order received

F • During: Processing
• After: Delivery

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 39
 n l y
Representing Space
e o
o s
r p
• Spatial Logic: Deals with reasoning about space, distances, and locations of objects.

u
It plays a critical role in domains like robotics, geographic information systems (GIS),

P
and mapping.

y
• Examples in Robotics and Mapping:

d
• Robots use spatial representation to navigate environments (e.g., calculating routes or

t u
avoiding obstacles).

S
• Mapping involves representing the location of objects in a two- or three-dimensional

r
space, such as in self-driving cars or robotic systems mapping rooms.

o
• Example:

F
• Robot Location: At point (x, y)
• Obstacle: 10 meters ahead

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 40
 n l y
o
Example of Representation for a Real-world Problem

s e
•
r
Case Study: Autonomous Driving
p o
•

Pu
Objects: Vehicles, pedestrians, traffic signals

y
• Relations: Vehicle stops for pedestrian, vehicle follows road
•
•
t u d
Events: Traffic light turns red, pedestrian crosses
Time: Traffic light remains red for 30 seconds
•

r S
Space: Car is at (lat, long), 100 meters from intersection

o
• The system must integrate knowledge of objects, time, space, and

F events to make decisions.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 41
 n l y
o
Limitations and Complexity of Representation
e
o s
r p
• Scalability: Managing large sets of data (e.g., in dynamic environments
like cities).
u
y P
• Ambiguity: Unclear or imprecise data (e.g., unexpected events like a
sudden pedestrian crossing).

u d
• Incompleteness: Representations might not include all relevant

t
information, leading to limited system capabilities.

r S
• Complexity: As the number of objects, events, and relationships grows,

o
so does the computational complexity of reasoning and decision-

F
making.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 42
 n l y
What is Situation Calculus?
e o
o s
Definition:

u r p
• Situation Calculus is a formal language used for representing and reasoning about dynamic systems, particularly to

P
describe how the world evolves based on actions. It allows us to model changes in the world as a result of actions
taken by agents.

y
Examples:

d
• Scenario 1: A robot is moving from room A to room B.

t u
• Initial Situation: Robot is in Room A.

S
• Action: Move to Room B.
• Resulting Situation: Robot is now in Room B.

o r
• Scenario 2: A light is turned on or off.

F
• Initial Situation: Light is off.
• Action: Turn light on.
• Resulting Situation: Light is now on.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 43
 n l y
Key Concepts in Situation Calculus
e o
o s
Situations:

u r p
• Situations are snapshots of the world at a given point in time. They capture the state of

P
the environment after a series of actions. Each situation reflects the result of

y
performing actions in a particular sequence.

d
Actions:

t u
• Actions are events that cause transitions between situations. They describe how the

S
world can change, such as "Move to Room B" or "Turn on the light."

r
Fluents:

Fo
• Fluents are properties that can change over time. They describe aspects of the world
that may vary between situations, like "the robot is in Room A" or "the light is on."
Fluents can be true in one situation and false in another.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 44
 n l y
Applications of Situation Calculus
e o
o s
Planning:

u r p
• Situation Calculus is widely used in AI planning, where agents need to decide the

P
sequence of actions to achieve a goal. By reasoning about actions and their effects, an

y
agent can plan a path from the initial situation to the desired outcome.

d
• Example: A robot planning a route to navigate through a building.

t u
Reasoning about Actions:

S
• It helps in determining the consequences of actions, predicting future situations, and

r
handling uncertainties in dynamic environments. Situation Calculus can also be used

o
to verify that certain conditions will always hold after executing a series of actions.

F
• Example: Ensuring safety constraints in robotic systems, such as never moving
through restricted areas.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 45
 n l y
Introduction to Description Logics
e o
o s
Definition:

u r p
• Description Logics (DL) are a family of formal knowledge representation languages

P
designed for representing the knowledge of an application domain in a structured and

y
unambiguous manner. They are used to describe concepts (classes), roles (properties), and

d
individuals (instances) in a domain.

u
Importance:

S t
• Description Logics provide the foundation for:

r
• Ontology Engineering: Structuring knowledge in a way that allows machines to interpret it.

o
• Semantic Web: Enabling intelligent data search, reasoning, and retrieval on the web.

F
• Knowledge-based Systems: Allowing reasoning about entities, their attributes, and
relationships.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 46
 Concepts and Roles in Description Logics n l y
e o
o s
• Concepts (Classes):

u r p
Concepts represent sets of objects or entities. They are used to categorize individuals. For

P
example, "Person" or "Vehicle" are concepts in a domain.

y
• Roles (Relationships):

t u d
Roles define relationships between individuals or between individuals and data values. For
example, "hasChild" or "ownsCar" are roles.

S
• Hierarchical Structure:

r
Description Logics support the organization of concepts and roles into a hierarchical

o
taxonomy, where more general concepts (e.g., Vehicle) are subdivided into more specific

F
concepts (e.g., Car, Truck).
Subsumption: One concept may be a subset of another, capturing inheritance relationships.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 47
 n l y
Example of Description Logics in Use
e o
o s
Ontology Engineering:

u r p
• Description Logics are used to define ontologies, which are structured frameworks that

P
define the relationships between concepts in a specific domain. For example, in a

y
medical ontology, you can define concepts like "Doctor", "Patient", and relationships

d
like "treats".

t u
Semantic Web:

S
• On the Semantic Web, Description Logics provide a framework for machines to

r
understand and reason about the meaning of information. For instance, they allow

o
intelligent systems to answer complex queries by reasoning about the relationships

F
between data.
• Example: A search engine can infer that a “pediatrician” is a “doctor” and thus provide
relevant results for queries involving both concepts

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 48
 n l y
What is Default Reasoning?
e o
o s
Default Reasoning:

u r p
• It is a type of non-monotonic reasoning, where conclusions can be drawn based on

P
default rules or assumptions that hold in the absence of contrary information. Unlike

y
classical logic, where the truth of statements is permanent, default reasoning allows

d
for the possibility of retracting conclusions if new information arises.

t u
Key Concepts:

S
• Non-monotonic Reasoning: As knowledge is updated, conclusions can be withdrawn.

r
• Defaults: Rules or assumptions used to fill in gaps in incomplete knowledge.

Fo
Importance:
• Default reasoning is crucial in AI systems where full information isn’t always available,
enabling them to function effectively in uncertain environments.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 49
 n l y
Examples of Default Reasoning
e o
o s
• Example 1:
• Assumption: "Birds can fly."
u r p
y P
• Default Rule: Assume a bird can fly unless there is information that suggests

d
otherwise (e.g., if the bird is a penguin).

u
• Example 2:

S t
• Medical Diagnosis: A doctor may assume that a patient has a common illness

r
based on symptoms unless tests reveal an exception.

o
• Case Study:

F
• Default reasoning in autonomous vehicles: Vehicles assume that other cars
will follow traffic rules unless signs of erratic behavior are detected.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 50
 n l y
Epistemic Logic
e o
o s
Epistemic Logic:

u r p
• A formal system for reasoning about knowledge and belief. It provides a framework for

P
understanding what agents know, what they believe, and how knowledge is shared

y
among agents.

d
Key Features:

t u
• Knowledge Operators: Express what an agent knows (e.g., K_a(P) means "Agent A

S
knows P").

r
• Belief Operators: Represent what agents believe, which may or may not be true.

Fo
Use Cases:
• Used to model reasoning in situations where agents have incomplete or uncertain
information, such as multi-agent systems and distributed AI.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 51
 n l y
Multi-Agent Knowledge Reasoning
e o
o s
Common Knowledge:

u r p
• In multi-agent systems, common knowledge refers to facts that are known by all agents

P
and that all agents know that others know it as well (e.g., a shared protocol in a

y
distributed system).

d
Distributed Systems:

t u
• Knowledge reasoning is vital in distributed systems where multiple agents must

S
coordinate and share knowledge. Reasoning about what others know helps improve

r
decision-making in these systems.

o
Key Concept:

F
• S4 Logic: A system used to represent multi-agent knowledge reasoning, capturing
notions like individual knowledge and common knowledge.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 52
 n l y
Applications of Knowledge Reasoning
e o
o s
Collaborative AI:

u r p
• In systems where multiple AI agents work together (e.g., in robotics or complex

P
simulations), reasoning about each agent's knowledge and beliefs is crucial for

y
effective collaboration.

d
Example:

t u
• In collaborative healthcare systems, different AI agents may have access to different

S
pieces of patient data. Epistemic reasoning allows these systems to communicate

r
efficiently and share vital information to provide the best care.

o
Other Use Cases:

F
• Autonomous drones coordinating to achieve a mission while considering the
knowledge of other drones in the fleet.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 53
 n l y
o
KR and Reasoning in Expert Systems (Medical Diagnosis Example)

s e
r p o
Definition: Expert systems use knowledge-based reasoning to solve problems in

u
specific domains by simulating expert-level decision-making.
Example:

y P
d
• MYCIN System: Early medical diagnosis system for bacterial infections.

u
• How it Works: Uses rule-based reasoning (IF-THEN rules) to suggest

t
diagnoses and treatments.

S
r
• Reasoning Type: Deductive reasoning to infer conclusions from known facts

o
and rules.

F
Impact: Enhances decision-making in healthcare by reducing errors and
assisting doctors.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 54
 n l y
o
KR and Reasoning in Robotics (Autonomous Agents Example)

s e
r p o
Overview: Robotics systems rely heavily on knowledge representation for understanding

u
their environment and making decisions.

P
Example:

y
• Self-driving Cars: Use semantic maps to represent roads, obstacles, and signs.

d
u
• Reasoning Mechanism: Combines spatial reasoning (for positioning) and temporal

t
reasoning (for dynamic environments).

S
• Task: Detects obstacles, predicts pedestrian movement, and makes safe decisions in

r
real time.

Fo
Impact: Autonomous systems are safer, more efficient, and adaptable in various
conditions.

Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 55
 n l y
Conclusion
e o
o s
Summary of Key Concepts:

u r p
P
• Knowledge Representation (KR) enables systems to model complex
environments.

d y
• Reasoning allows systems to make logical decisions based on the

u
information they possess.

S t
• Different KR techniques, such as ontologies, rule-based systems, and

r
predicate logic, have unique advantages and limitations.

Fo
Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 56
 Future Trends
n l y
e o
o s
u r p
• Integration with Machine Learning: Hybrid systems combining symbolic KR with

P
neural networks.

y
• Semantic Web: Expanding the internet with more meaningful data through

d
ontologies.

u
• Collaborative Multi-Agent Systems: AI agents that reason together in shared

t
environments (e.g., smart homes, factories).

r S
• Explainable AI (XAI): Focus on creating systems that can reason and explain their
decisions.

Fo
Dr. S S Ohol, Mechanical Engg. Dept., COEP Technological University Pune (COEP Tech) 57