You are on page 1of 36

Knowledge Representation

**Why do we need knowledge


Why do we need knowledge
• Before KR we have to collect first the fragment of knowledge
• This collection is composed of some steps such as the formulation of the
specific problem in spoken language and after that we represent it formal
language so that computer can understand
• The basic framework of KR: first informal formalism of the problem takes
place
• Then it is represented formally and the computer constructs an output
• This output can then be represented in a informal manner which acts like a
description of the solution and this description is understood by the user
• Finally, consistency of the solution is checked
Why do we need knowledge
• By using the concept of knowledge engineering knowledge can be gathered
from the knowledge base and the knowledge is based on various knowledge
sources like human experts, the knowledge engineer and database
• Separation of knowledge and inference, can allow inference systems to be
implemented in an application-independent way
• With this separation it becomes easier to substitute the knowledge base of a
medical expert system than to program a whole new system
• When we attempt to convert any English statement into some other
representation ,for instance logical proposition then it is mandatory to decide
first what facts the sentence is representing and then convert the fact into new
depiction
Why do we need knowledge

• Agent is a program that maps perception to actions


• This functionality of agent is application specific as for each mapping an agent
needs some information
• If the application is complex one, large amount of information need to be
processed by the agent
• The first step here is to separate information or knowledge from the system or
from inference program
• Knowledge is used to reach some conclusions, answers
• Knowledge is deposited in knowledge base
** Need to represent knowledge

• The main objective is to express knowledge in computer-tractable form so


that it can be used to help intelligent system to enhance its performance
• Also for solving complex problems
• Decoupling of knowledge and storing it declaratively helps in inference
• The clear separation of knowledge and processing of inference steps are
cost saving
• Formal language is a suitable interface between man and machine which
helps in representing knowledge in knowledge base
**Framework of knowledge representation (Poole 1998)
Knowledge representation with mapping
scheme
o AI comprises of many complex problems which requires huge
knowledge to resolve them along with mechanism for manipulation on
this knowledge
o Representing appropriate truth about world is KR and so facts can be
used as a weapon in representing knowledge
o We have to structure the entities in two layers for better understanding:
 knowledge level at which facts are represented
 symbol level at which representation of objects in terms of symbols
are done
Mapping between fact and representation
**Knowledge Representation

13
Knowledge representation schemes
**Properties of good knowledge based
system
• KR is the method which is used to encode knowledge in an intelligent system’s
knowledge base
• A high quality KR enables rapid and precise access to knowledge and also good
representation gives better understanding of the frame work
Representational adequacy :
• To represent any kind of statement domain knowledge is required
• This implies the ability to represent all kinds of knowledge that are required in
that domain for representation
Properties of good knowledge based system
Properties of good knowledge based system

• Inferential adequacy :
– Exploitation facility is one of the significant aspects in any KR as new structure
can be derived with the help of this property

– This is the ability to manipulate or explore the representational structure that


in turns helps in deriving new structure consequent to new knowledge which
is inferred from the old structure
Properties of good knowledge based system
Inferential efficiency :
• In inference a set of procedures by using representational language infers
new facts from new known ones
• Integration of additional information is required for inference
• It helps in improving deduction mechanism in KR
• If any knowledge-base has a provision to add new information for inference
then that knowledge-base is said to have inferential efficiency
• Inferences require search and so good knowledge-base should have the
ability to incorporate extra knowledge into the knowledge structure to
redirect the inference mechanism in a promising direction
Properties of good knowledge based system
Acquisitional efficiency :
• In most of the cases, KR is dependent on human involvement
• If any knowledge base has an ability to gain new knowledge whenever
possible, using automatic methods rather than depending only on human
intervention is said to have this property
**Knowledge representation issues
• An AI system is able to store and manipulate data, along with that it is able to
acquire, manipulate and can represent knowledge
• The main aim of KR is to achieve conclusion or inference
Knowledge representation issues
Knowledge Acquisition :
Knowledge representation issues
Knowledge Acquisition :

• It implies the transformation of potential problem solving, from a knowledge


source to a program
• This strategy comprises of some basic steps like: extracting knowledge,
interpretation of knowledge by accumulating sufficient detail to make it clear,
then documented it in a symbolic form, finally verify the symbolic form against
the original statement
• This is part of accessing problem and is done by KE
• Responsible for building knowledge-based system
• Final step is to developing a model of the reasoning like flow chart , decision
tree or protocol analysis
Knowledge representation issues
Knowledge Representation :
• The main objective is to allow users to represent English language in a predicate
logic format
• KR is generally concerned with fact representation – which is a truth about real
world
• It is also termed as ’base knowledge level’
• General KR includes logical representation, semantic or associative network,
production rules and frames
• When users has to represent knowledge in natural language there may be a
problem as natural language is very ambiguous, syntax and semantics used in
natural language are not understood properly
• There exists a little inconsistency in the sentence structure
Knowledge representation issues
Knowledge manipulation :
• Forward chaining and backward chaining mechanism are used
•The basic structure of forward chaining : IF antecedent THEN consequents
• Backward chaining authenticates hypothesis matching ‘THEN’ pattern, by verifying
‘IF’ patterns
• Forward chaining is a process of moving from the ‘if’ patterns to ‘then’ patterns,i.e.
condition-action rules
Knowledge representation issues
• Whenever an ‘if’’ pattern is observed to match an assertion; then the
antecedent is satisfied

• And when all the ‘if’ patterns of a rule are satisfied, then the rule is
triggered

• A triggered rule establishes a new assertion and then it is fired

• It is the objective of search methods to decide on the order in which the


rules to be fired

• Conflict resolution in forward chaining inference engine decides which of


the several rules could be fired and incorporates some decision making
into matching process
Knowledge representation issues
Forward chaining
Knowledge representation issues
Backward chaining :
• The process starts with the hypothesis and uses the antecedent-consequent rules to work
backward towards hypothesis supporting assertions
**AND-OR Graphs
• This AND-OR graphs are structured representation of knowledge and is used to

put into the expert system’s knowledge-base


• For large and complex problems it need to be decomposed using the concept of
divide and conquer and decomposition continues till a smaller sub-problem which
is easy to solve is reached
• After solving all small decomposed parts then all the solutions are combined to
get the unique solution of the large problem provided initially
• This will generate AND-OR graphs as these contain both the AND node and OR
node
• All successors must be solved in order for and AND node but alternative solution
paths can also be generated because of OR node
AND-OR Graphs
Problem Decomposition using AND-OR Graphs

• In this case first the principal goal is identified ;goal can be defined as

something that has to be achieved ; sub-goal is something that must be


achieved to reach the goal
• This goal is split into two or more sub-goals and these two are further split
up
• Each goal is written in a box, termed as ‘node’ and sub-goals are situated
below it with a ‘link’
• The boxes at the bottom of the graphs are leaf-nodes ; they don’t have any
link and is considered as piece of data needed to solve the problem
Problem Decomposition using AND-OR Graphs

Different situations in AND-OR Graphs


Problem Decomposition using AND-OR Graphs

Step 1: Decide what is the ultimate advise that the system should provide:
It is a statement along the lines of “the investment” should be x”

Step 2: Split the goal into sub-goals. X can be ‘saving', 'stock’ or ‘mixture’
Problem Decomposition using AND-OR Graphs
Step 3: It is only true that “X is saving” if “saving is inadequate” that provides sub-goal
under “X is saving”

If savings are adequate and income is adequate , then it is only true “X is stocks”. So two
sub-goals under “X is stocks”

Similarly there is “mixture”


Problem Decomposition using AND-OR Graphs
Step 4: Decision if something is really big enough. Such as savings are only inadequate ig=f
they are smaller than a certain figure ‘Y’, saving are only adequate if they are bigger than
‘Y’ and income is adequate if it is bigger than certain amount ‘W’
**Types of knowledge
Relational knowledge: Offers a structure of comparing two objects based on
their comparable attributes and it associates elements of one domain with the
other elements of another domain

Inheritable knowledge: It represents a framework in which new objects are


created and this newly created object may inherit all or some properties from
existing object class
A hierarchical format of representation is used to express inheritable
knowledge
Types of knowledge
Inferential knowledge: Is derived or inferred from objects through relations
among the objects given.

Declarative knowledge: It is a passive knowledge in the form of facts about


the world

Procedural knowledge: It describes how a problem is solved. It defines “what


to do when". It is a processed form of information and is associated to the
performance of some task. A sequence of steps to solve a problem is a
procedural knowledge.

You might also like