Knowledge Representation and Expert Systems

1
Knowledge RepresentationandExpert Systems

• by
• Sujan Pakala Stephen Brown

2
What are they?

• Diagnose a problems in a given domain
• Capable of justifying its actions and conclusions
• Applications
• Medical diagnosis applications
• Locating equipment failures
• Interpreting measurement data

3
Structure of Expert System
User
Shell

• Major Components
• Knowledge Base
• Inference Engine
• User Interface
• The Shell contains the inference engine and
  the user interface.

User Interface
Inference Engine
Knowledge Base
4
Expert System Features

• Goal driven reasoning or backward chaining
• Data driven reasoning or forward chaining
• Coping with uncertainty
• Data representation
• User interface
• Generating Explanations

5
If then Rules

• Most popular formalism for knowledge
  representation
• Additional features
• - modularity
• - incrementability
• - modifiability
• - transparency
• categorical vs. probabilistic knowledge
• knowledge elicitation
• Example refer figure.

6
Knowledge base as an Inference Network
Kitchen_dry
Leak_in_bathroom
Hall_wet
Problem_in_kitchen
Bathroom_dry
Leak_in_kitchen
Window_closed
No_water_from_outside
No_rain
7
Backward chaining

• Follow a chain of rules backwards
• Stating rules into knowledgebase
• - as straightforward prolog rules
• hall_wet.
• bathroom_dry.
• leak_in_bathroom -
• hall_wet, kitchen_dry.

8
Contd..

• Disadvantage
• - not suited for normal user
• - hence, not syntactically distinct
• Better way
• - use if, then etc. as operators-
• -op(800, fx, if)
• - and write the rules as-
• if hall_wet and kitchen_dry then
  leak_in_bathroom.
• Major disadvantage
• - user has to state all relevant info.

9
Forward chaining
10
Generating Explanations

• Has the ability to explain its results.
• Two types of explanations
• how the system reached a given conclusion
• why the system is asking a question

11
Coping with Uncertainty

• Much of the time, the final answer is not known
  with complete certainty.
• We can model uncertainty by assigning some
  qualification or measure of belief factor.
• In our knowledge base, we can add a certainty
  factor to our conclusions
• if
• hall_wet and bathroom_dry
• then
• problem_in_kitchen 0.9.

12

• Example
• User specifies certainty estimates
• given(hall_wet, 1). Hall is wet
• given(bathroom_dry, 1). Bathroom is dry
• given(kitchen_dry, 0). Kitchen is not dry
• given(no_rain, 0.8). Probably no rain, but not
  sure
• given(window_closed, 0). Window not closed

13

• Continue Example
• An interpreter for rules with certainties
• certainty(P, Cert) -
• given(P, Cert).
• certainty(Cond1 and Cond2, Cert) -
• certainty(Cond1, Cert1),
• certainty(Cond2, Cert2),
• minimum(Cert1, Cert2, Cert).
• certainty(Cond1 or Cond2, Cert) -
• certainty(Cond1, Cert1),
• certainty(Cond2, Cert2),
• maximum(Cert1, Cert2, Cert).
• certainty(P, Cert) -
• if Cond then P C1,
• certainty(Cond, C2),
• Cert is C1 C2.

?- certainty( leak_in_kitchen, C). C
0.8 Obtained hall is wet bathroom is
dryproblem in the kitchen 0.9 Possibility of
some rainno water from outside 0.8 Leak in
kitchen is min(.8, .9) 0.8
14
Demo

• http//www.visual-prolog.com/vipexamples/geni/pdci
  ndex.htm

15
Questions?
Summary

• Expert Systems typical functions
• Solving problem in a given domain
• Explaining the problem-solving process
• Handling uncertainty and incomplete information