Transcript Sistem Pakar - Gunadarma University

SISTEM INFORMASI MANAJEMEN 2*
D3 MANAJEMEN INFORMATIKA
ATA 05/06
Sistem Pakar
Ati Harmoni
Artificial Intelligence
• Definition:
The activity of providing such machines as
computers with the ability to display behavior
that would be regarded as intelligent if it were
observed in humans.
History
• 1956, Dartmouth College. John McCarthy
coined term. Same year, Logic Theorist
(first AI program. Herbert Simon played a
part)
• Past 20 or so years, DOD and NSF have
funded AI research at top schools
(Stanford, Carnegie Mellon, etc.)
• Desert Storm opened up new funding
(smart bombs, night vision)
Areas of Artificial Intelligence
Expert
systems
AI
hardware
Robotics
Natural
language
Learning
Neural
networks
Artificial Intelligence
Perceptive
systems
(vision,
hearing)
The Appeal of Expert Systems
A computer program that attempts to code
the knowledge of human experts in the
form of heuristics (i.e. a rule of thumb)
Two distinctions from DSS
1. has the potential to extend the manager’s
problem-solving ability beyond his or her
normal capabilities
2. the ability to explain how the solution was
reached
User
Instructions &
information
Solutions &
explanations
Knowledge
User
interface
Inference
engine
Expert
system
Development
engine
Expert and
knowledge engineer
Knowledge
base
Problem
Domain
An Expert
System Model
Expert system model - main
parts:
•
•
•
•
User interface
Knowledge base
Inference engine
Development engine
User Interface
• User enters:
– Instructions
– Information
}
Menus,
Menus,commands,
commands,natural
naturallanguage,
language,GUI
GUI
• Expert system provides:
– Solutions
– Explanations of
• Questions
• Problem solutions
Knowledge Base
• Description of problem domain
• Rules: A knowledge representation
technique
– such as ‘IF:THEN’ logic
– networks of rules
• Lowest levels provide evidence
• Top levels produce 1 or more conclusions
• A conclusion is called a Goal variable.
A Rule Set That
Produces One Final
Conclusion
Conclusion
Conclusion
Evidence
Evidence
Conclusion
Evidence
Evidence
Evidence
Evidence
Evidence
Evidence
Cheetah
Tiger
R9
And
And
Tawny
color
Giraffe
R10
Zebra
R11
And
R12
And
Dark
spots
Long
legs
Ostrich
R13
And
Black
strips
Or
Carnivore
Or
Or
R1
Hair
R2
Gives
milk
A Rule Set That
Can Produce More
Than One Final
Conclusion
Black&
White
R8
And
And
Swims
Flies
Well
Bird
Hoofs
Chews
cud
R3
R4
Feathers
And
R6
And
Flies
Claws
Lays
eggs
LEGEND:
Rules
Pointed
teeth
And
Or
R7
R5
milk
Eats
milk
R15
And
Can’t
fly
Ungulate
Mammal
R14
And
Long
neck
Albatross
Penguin
Forward
Eyes
Conditions
Action
(conclusions)
Rule Selection
• Selecting rules to efficiently solve a
problem is difficult
• Some goals can be reached with only a
few rules; rules 3 and 4 identify bird
Inference Engine
• Two basic approaches to using rules
1. Forward reasoning (data driven)
2. Reverse reasoning (goal driven)
Forward Reasoning
(forward chaining)
• Rule is evaluated as:
– (1) true, (2) false, (3) unknown
• Rule evaluation is an iterative process
• When no more rules can fire, the
reasoning process stops even if a goal
has not been reached
Rule 1
IF A
THEN B
Rule 2
T
Rule 7
F
IF B OR D
THEN K
IF C
THEN D
Rule 3
T
IF K
THEN F
Rule 5
IF G
THEN H
Rule 6
IF I
THEN J
IF K AND
L THEN N
T
T
IF M
THEN E
Rule 4
Rule 10
The Forward
Reasoning
Process
Rule 8
Rule 12
T
IF N OR O
THEN P
IF E
THEN L
T
T
Legend:
Rule 9
T
F
IF (F AND H)
OR J
THEN M
T
First pass
Rule 11
IF M
THEN O
T
Second pass
Third pass
Reverse Reasoning
(backward chaining)
• Divide problem into subproblems
• Try to solve one subproblem
• Then try another
A Problem and Its Subproblems
Rule 10
IF K AND L
THEN N
Rule 12
Rule 11
IF N OR O
THEN P
Legend:
Problem
IF M
THEN O
Subproblem
A Subproblem Becomes the New Problem
Rule 7
IF B OR D
THEN K
Rule 8
IF E
THEN L
Rule 10
IF K AND
LTHEN N
Legend:
Rule 12
IF N OR O
THEN P
Problem
Subproblem
Step 4
Rule 1
IF A THEN
B
T
Step 3
Rule 7
IF B OR D
THEN K
T
Rule 2
IF C
THEN D
The First Five Problems
Are Identified
Step 2
Rule 10
IF K AND L
THEN N
Step 5
Rule 3
IF M
THEN E
Step 1
Rule 12
IF N OR O
THEN P
IF E
THEN L
Rule 11
IF (F AND H)
Rule 9 OR J
THEN M
IF M
M
IF
THEN O
THEN
O
Legend:
Problems to
be solved
The Next Four Problems Are
Rule 12
Identified
Step 8
If N Or O
Then P T
Rule 4
If K
Then F
Rule 5
T
Step 7
Step 6
IF (F And H)
Or J
Then M T
If M
Then O
Step 9
If G
Then H
T
Rule 6
If I
Then J
Rule 9
T
Rule 11
Legend:
Problems to
be solved
Forward Versus Reverse
Reasoning
• Reverse reasoning is faster than
forward reasoning
• Reverse reasoning works best when
– there are multiple goal variables
– there are many rules
– all or most rules do not have to be
examined in the process of reaching a
solution
Handling Uncertainty
• Two types of uncertainty
– Rules
– Conditions
• Certainty factors (CFs) range from 0.00
to 1.00
Development Engine
• Programming languages Lisp, Prolog,
and recently C++
• Expert system shells
Role of the Systems Analyst
• Knowledge engineers work with the
expert in designing expert systems
• Beyond traditional analyst skills, the
following skills are needed
– understand how the expert applies his or
her knowledge
– be able to extract the description of the
knowledge (rules as well as facts)
System Development Process
•
•
•
•
Initiate the development process
Develop the expert system prototype
User participation
Expert system maintenance
Prototyping Approach
• A new player: the expert
• Delayed user involvement
• Need for maintenance
Prototyping Is Incorporated in the Development of an Expert System
Systems analyst
step 1
Expert
User
Studythe
the
Study
problem
Problem
domain
domain
step 2
Define the problem
step 3
Specify the rule set
step 4
Test the prototype system
step 5
Construct the interface
step 6
step 7
step 8
Maintain the system
Conduct
user tests
Use the
system
Example:
Financial Expert System
• Credit approval
• Knowledge base for the example
consists of rules and a mathematical
model
• User interface
• Five decision categories; credit amount
influences weightings
Expert System Advantages
• To managers
– Consider more alternatives
– Apply high level of logic
– Have more time to evaluate decision rules
– Consistent logic
• To the firm
– Better performance from management
team
– Retain firm’s knowledge resource
Expert System Disadvantages
• Can’t handle inconsistent knowledge
• Can’t apply judgment or intuition
Neural Networks
• Expert systems should be able to learn,
and improve their performance
• Neural net design -- a bottom-up
approach to modeling human intuition
The Human Brain
• Neuron -- the information processor
– Input -- dendrites
– Processing -- soma
– Output -- axon
• Neurons are connected by the synapse
Simple Biological Neurons
Soma
(processor)
Axonal Paths
(output)
Synapse
Axon
Dendrites
(input)
Artificial Neural Systems (ANS)
• McCulloch-Pitts mathematical neuron
function (late 1930s)
• Hebb’s learning law (early 1940s)
• Neurocomputers
– Marvin Minsky’s Snark (early 1950s)
– Rosenblatt’s Perceptron (mid 1950s)
Current Methodology
• Mathematical models
• Complex networks
• Repetitious training -- the ANS “learns”
by example. An ANS can learn; an
expert system cannot.
Single Artificial Neuron
y1
w1
y2
w2
y3
w3
wn-1
yn-1
y
OUT1
OUTn
The Multi-Layer
Perceptron
Input
Layer
Y1
Yn2
OutputL
ayer
IN1
INn
Prerequisite Activities for the
EIS
Information
needs
Information
technology
standards
Analysis of
organization
Information
systems plan
Corporate
data model
Production and
performance systems
EIS