Knowledge is mans first frontier

1
Knowledge Representation

• Knowledge is mans first frontier

We have always sought to obtain knowledge, to
apply that knowledge to solve everyday problems,
and to expand on it to improve the world around
us and our environment
2
Knowledge Representation

• Many problems tackled by experts are difficult
  and poorly understood
• Rules of thumb are called heuristics and this
  type of knowledge is called heuristic knowledge

Experts often have no formal basis for problem
solving and often use rules of thumb developed
on the basis of their experience
3
Knowledge Representation

• Deep and Shallow knowledge
• Shallow knowledge is when a system has been built
  with a working knowledge of the problem domain
  only
• Deep Knowledge is when the system has a clearer
  understanding of the fundamental laws on which it
  is based

4
Knowledge Representation

• Rules
• Rules are usually represented as an IF-THEN
  statement
• IF ltsituationgt THEN ltactiongt
• A simple example rule could be -
• IF Christmas day falls on a Monday
• OR Christmas day falls on a Tuesday
• THEN Many factories will close for 1 week

5
Knowledge Representation

• Rule Based Systems
• Represent practical human reasoning in the form
  of if-then rules
• Their knowledge grows as more rules are added
• Can solve complex problems by combining groups of
  rules
• Can dynamically select the best set of rules to
  execute
• Use an inferencing technique (either Forward or
  Backward Chaining) to manipulate expertise to
  solve a problem

6
Knowledge Representation

• Declarative v. Procedural Programming
• A program written in Basic, C, Assembler etc.
  consists of a set of procedures that must be
  performed in a strict sequence to accomplish a
  purpose. This is Procedural Programming
• Rules in a KBS stand alone as statements of truth
  or fact, and can be used by an inference engine
  in any order to reach other true conclusions.
  This is Declarative Programming

7
Knowledge Representation

• Certainty Factors
• Experts may use some form of judgmental or
  inexact reasoning
• Certainty factors range from 0 (false) to 1 (true)

When conditions X and Y occur, Z is likely to be
the case
8
Knowledge Representation
Rule 1 IF the patient has hypothermia THEN he
may have respiratory problems (CF 0.5) AND he may
suffer cardiac arrest (CF 0.4)
Rule 2 IF there is a reduction in alveolar
minute volume THEN the respiratory centre may be
depressed (CF 0.6)
Rule 3 IF the arterial pCo2 rises above 12 kPa
in response to oxygen therapy AND it remains at
that level for more than 5 minutes THEN there is
a substantial risk of carbon dioxnarcosis (CF 0.9)
9
Knowledge Representation

• Frames
• Are standard
• Provide a method of combining declarative and
  procedural programming within a single KBS
• Package both data and procedures into a single
  knowledge structure
• Contain slots that describe the accuracy of an
  object
• Slots can contain a range of data types
• Slots can contain procedural code (Demons)
• E.G. ( if -added, if - needed )

10
Knowledge Representation

• Example Frame

CUP FRAME
TYPE
Value Coffee Mug
COLOUR
SIZE
Range Small, Medium, Large
PURPOSE
Value Drinking Coffee
COST
Demon (If-Needed) ...
MATERIAL
Default Pottery
11
Knowledge Representation

• Advantages of Frames
• Can be graphically represented
• Allow the representation of different knowledge
  types
• Can be used to contain default values
• Allow easy classification of knowledge
• Can reduce complexity
• Can clearly document information
• Can inherit procedural or declarative knowledge
• Can constrain allowed values

12
Knowledge Representation

• Example of Inheritance

CUP FRAME
Stores value fluid
TEACUP FRAME
COFFEECUP FRAME
TYPE value teacup
TYPE value coffee mug
SIZE range S, M, L
SIZE range S, M
PURPOSE value drinking tea
PURPOSE value drinking coffee
COST demon (if-needed)...
COST demon (if-needed)...
MATERIAL value bone china
MATERIAL default pottery
MYCUP FRAME
owner value simon
size value L
13
Knowledge Representation

• Frames Levels

Highest Level

• Frame
• The name of the frame
• Slot
• Attributes of the frame
• Facet
• Aspects of an attribute
• Values, Range , Default, Demons
• Data
• Information about the attributes

Lowest Level
14
Knowledge Representation

• Disadvantages of Frames
• Can be inefficient at runtime
• Can lead to procedural fever
• Require care in the design stage to create
  suitable taxonomies

15
KnowledgeRepresentation

• Semantic Nets

Example semantic net, depicting relationships
Nodes which represent objects
KEY
Links show the relationship between objects
16
Knowledge Representation

• Advantages of Semantic Nets
• A powerful, flexible and graphical way of
  representing knowledge
• Often used as a communication tool between
  knowledge engineer and expert during the
  knowledge acquisition phase

17
Knowledge Representation

• Disadvantages of Semantic Nets
• Can be very difficult to inference
• Less reliable than Rules as inferring becomes a
  process of searching across the diagram
• Diagrams can become very complex

18
Knowledge Representation

• Blackboard Representation

So named because they organise and process
knowledge in a fashion similar to a group of
people working around a blackboard
The blackboard is used a repository for
knowledge
The group leader provides a control function,
guiding and focusing the activities of the
knowledge sources
Each person represents a knowledge source
19
Knowledge Representation

• Object Based Representations
• The major concepts in object oriented technology
  are -
• The Objects
• The Classes
• Messages and Methods
• Inheritance

20
Knowledge Representation

• Object Oriented Technology (The Object)

An entity which has a state and a defined set
of operations to access and modify that state
All the relevant information is held within the
object. This is referred to as encapsulation
Objects are independent and may be readily changed
21
Knowledge Representation

• Object Oriented Technology (The Class)

The basic modular unit
Represents a group of objects based on
similarities
Class
Also referred to as an abstract data type
Permits reuse of program code
22
Knowledge Representation

• Object Oriented Technology (Messages and
  Methods)

M Software Module
Object A
Object B
Communicate by Message
A procedure or method is a specification of
action on data contained in the object
When a message is sent to an object, the object
responds by locating an attached procedure
associated with the message and executes it
Thus, since different procedures are defined as
methods for the objects, different objects might
respond to the same message in different ways
23
Knowledge Representation

• Object Oriented Technology (Inheritance)

A class may inherit its structure and methods
from others - e.g.
24
Knowledge Representation

• Case Based Reasoning (CBR)
• Can help when the experience domain is rich but
  the knowledge domain is poor
• Utilises pervious experiences of problems in
  order to solve new ones
• Does not try to express knowledge in terms of
  rules, but builds up a set of cases to be
  referred to
• If exact matches are not found then the closest
  previous match is found and the answers adapted
  to solve the problem

25
Knowledge Representation

• Case Based Reasoning (CBR)
• CBR attempts to simulate human experts
• Is best applied -
• In areas where problems cannot be easily
  decomposed
• In cases where general principles involved are
  not completely understood, but a library of past
  experience can be generated
• Where complex problems exist for which there are
  no precise answers

26
Knowledge Representation

• Differences between RBR and CBR

Rule-Based Reasoning
Case-Based Reasoning
Rules are small, independent but consistent
Cases are large chunks of domain knowledge and
stored by describing the way things work
Knowledge content
Domain
Well enough understood
Well and not well understood
Matching
Exactly
Partially
Heuristic
Explicit
Implicit
Interpretation
Single
Multiple
27
Knowledge Representation

• Selecting a KR Scheme

Select the most natural representation
28
Knowledge Representation

• Select the most natural representation

A combination of different techniques can provide
all the advantages of these representation
schemes without the disadvantages associated
with some of them
Cost vs. complexity and flexibility is an issue!
29
Knowledge Representation

• Break down the problem

Complexity tends to increase with problem size
Complexity
Size of the Problem
Decomposition techniques can produce an efficient
KBSs
A 2000-rule KBS base will be far too much for
many computers to cope with.
However, a KBS of 20-rule sets having 100 rules
each can be processed quite rapidly
30
Knowledge Representation

• Plan for the representation
• Define the appropriate types of KRS that will be
  used to organise the knowledge
• Identify the best tools that support the
  different representations
• If none provide exactly what is needed then
  review the next best choices