KnowledgeBased Problem Solving

1
Knowledge-Based Problem Solving

• Chap. 6, Chap. 8

2
Introduction

• Problem solving
• Knowledge Control
• Knowledge
• Logical sentence
• ltIfthengt rules
• Past experience
• Control (search)
• Recursive search with backward chaining Prolog
• Production systems with forward chaining Jess
• Find similar cases - CBR

3
Prolog

• Logic programming
• Program
• Sequence of FOPC sentences
• All sentences are Horn clauses
• p1 ? p2 ? ? pn ? q
• Inferencing mechanism
• Backward-chaining, DFS (recursive search)

4
Production Systems

• Production System
• A computational model of problem solving process
• Used for implementation of expert systems

5
Production Systems

• Production rules
• Problem solving knowledge
• ltconditiongt ? ltactiongt
• Exgt if earning(X, unsteady)
• then income(inadequate)
• if clear(X)
• then pickup(X)
• Working memory
• Current states of the world
• Exgt earning(1500, unsteady)
• clear(a)

6
Production Systems

• Recognize-act cycle
• Match
• Patterns in working memory are matched against
  rules
• ? Conflict set
• Conflict resolution
• One of the rules in the conflict set is selected
• Fire
• Action is performed. Working memory is updated

7
Production Systems

• Example

Match
Production Rules
Working Memory
R1 A ? B R2 A ? C R3 B ? D
A B
Fire R1
Conflict set R1, R2
8
Forward Chaining
9
Advantages

• Separation of knowledge and control
• Knowledge Production rules
• Control Recognize-act cycle
• ? Modification / update of KB is easy
• Modularity
• Production rules are independent each other
• ? Incremental development is possible

10
JESS

• CLIPS
• C Language Integrated Production System
• An expert system shell developed 1985 in NASA
• Solve the problem of portability and
  extensibility of the LISP based systems
• JESS
• Java Expert System Shell
• Developed 1997 in Sandia National Lab.
• Simplified version of CLIPS written in Java
• Easily integrated into Java codes
• Inference engine
• Forward-chaining, production system

11
JESS

• How to use?
• Download the User Manual and Jess package
• http//herzberg.ca.sandia.gov or,
  http//www.jessrules.com/jess/index.shtml
• Run in the command-line interface
• Run Jess.bat (java jess.Main)
• JESSgt (batch myfile.clp) (run)
• Integrating with Java
• Embed Jess program into Java program
• Extend Jess with commands written in Java

12
Jess Language

• Constants
• Atoms kim, car1, 100, Good Bye
• Lists (a b c), ( 1 2), (like kim lee)
• Variables
• Variables ?x, ?name
• Binding (bind ?x kim)
• Built-in functions
• , -, , /, , lt, lt, eq, clear,
• Usage ( ( 2 3) ( 1 2)), (clear)
• Function definitions
• (deffunction max (?a ?b)
• (if (gt ?a ?b) then ?a
• else ?b))

13
Jess Language

• Facts
• (like kim lee), (temperature 20), (start)
• Add facts at runtime
• (assert (like kim lee))
• (assert (car (id 6433) (make benz)))
• Add facts when Jess start (or reset)
• (deffacts myfacts
• (parent kim lee)
• (parent lee park) .. )

14
Jess Language

• Rules
• (defrule alert
• (temperature 100)
• gt
• (printout t Finished !!!))
• " x,y Female(x) Ù Parent(x,y) Þ Mother(x,y)
• (defrule rule-1
• (female ?x)
• (parent ?x ?y)
• gt
• (assert (mother ?x ?y)))

15
Jess Language

• Connectives
• ? multiple patterns in LHS
• ? not supported (make separate rules)
• (not ltpatterngt)
• Test
• Specify match condition
• (defrule r1
• (man (age ?a))
• (test (gt ?a 20))
• gt
• ...

16
Jess Language

• Pattern bindings
• A variable that is binded to a pattern
• (defrule r1
• ?fact lt- (step ?n)
• gt
• (bind ?m ( ?n 1))
• (retract ?fact)
• (assert (step ?m)))
• Other functions
• Program control clear, reset, run, exit, ...
• I/O read, printout, open, ...
• Debug facts, rules, watch, unwatch, ...

17
Example

• (deffacts initial-facts
• (start))
• (defrule ask
• (start)
• gt
• (printout t "How many legs? (2 or 4)")
• (assert (legs (read))))
• (defrule human
• (legs 2)
• gt
• (assert (answer human)))
• (defrule animal
• (legs 4)
• gt
• (assert (answer animal)))

18
Example

• (defrule outout
• (answer ?x)
• gt
• (printout t "It is " ?x "!" crlf))
• (reset)
• (run)
• Example run
• JESSgt ( 1 2)
• 3
• JESSgt (batch test.clp)
• How many legs? (2 or 4)
• 2
• It is human!

19
Expert Systems

• Expert system
• Knowledge-based program
• Provides answers to problems in specific domain
• Problem Categories
• Interpretation, prediction, diagnosis, design,
• Separation of knowledge and control
• Easy to represent knowledge
• Easy to build and change KB
• Same control and interface can be used for
  various systems
• Expert System Shell

20
Expert Systems

• Expert system architecture

KB
Shell
21
Knowledge Engineering

• Knowledge acquisition
• Building KB by acquiring knowledge from experts
• ? Major bottleneck in building expert system
• Knowledge engineer
• Select tools
• Extract knowledge (interview)
• Build efficient KB

22
Rule-Based Reasoning

• Represent knowledge as explicit rules
• ltIf ... Then gt
• Expert system as a production system
• Inference engine - Production system
• Knowledge base - Set of rules
• Example Diagnosing automotive problems
• Production system with backward chaining
• Knowledge
• Engine gets gas ? Light on (Battery is OK)
• ? Turn over (Start motor is OK) ? Start (Spark
  plug OK)

23
Rule-Based Reasoning

• Rules
• Rule 1 if the engine is getting gas, and
• the engine will turn over
• then the problem is spark plugs
• Rule 2 if the engine does not turn over, and
• the lights do not come on
• then the problem is battery
• Rule 3 if the engine does not turn over, and
• the lights do come on
• then the problem is starter motor
• Rule 4 if there is gas in fuel tank, and
• there is gas in the carburetor
• then the engine is getting gas

24
Rule-Based Reasoning
25
Rule-Based Reasoning

• Backward chaining
• Rule 1 fire, Rule 4 fire
• Dialog with user
• gas in fuel tank? yes
• gas in carburetor? yes
• engine turn over? why ?
• It has been established that
• 1. engine is getting gas
• therefore if
• 2. engine turn over
• then the problem is spark plugs how 1 ?

26
Rule-Based Reasoning

• This follows from rule 4
• if
• gas in fuel tank, and
• gas in carburetor
• then
• engine is getting gas
• gas in fuel tank was given by the user
• gas in carburetor was given by the user
• engine turn over? yes
• The problem is spark plugs

27
Rule-Based Reasoning

• Advantages
• Use of expert knowledge
• ? Good performance in limited domains
• Separation of knowledge and control
• Explanation is possible
• Disadvantages
• Need knowledge acquisition
• Rules are brittle so can not handle noisy data

28
Case-Based Reasoning

• Reasoning from cases
• Case examples from past
• Exgt Law cases
• Procedure
• 1. For a new problem, retrieve similar case
• (based on common feature)
• 2. Modify the retrieved case
• (make it suitable for the current situation)
• 3. Apply the modified case to get the solution
• 4. Save the new case

29
Case-Based Reasoning

• Example
• customer sex age action?(f)
• c1 M 40 N
• c2 M 20 Y
• c3 F 30 N
• c4 M 30 Y
• c5 F, 20 ? Y or N?
• Rule-based
• Rule if (sexF) then N. Therefore N
• Case-based
• C3 is most similar to c5. Therefore N

30
Case-Based Reasoning

• k-Nearest Neighbor method
• Define distance functions
• dsex(A, B) A B (female0, male1)
• dage(A, B) A B / max difference
• d dsex dage
• Example
• dsum(c5, c1) 0 1 20 40 / 20 2.0
• dsum(c5, c2) 0 1 20 20 / 20 1.0
• dsum(c5, c3) 0 0 20 30 / 20 0.5
• dsum(c5, c4) 0 1 20 30 / 20 1.5

31
Case-Based Reasoning

• Prediction of value
• Weighted average of neighbors
• Example
• 3-NN ? c3, c2, c4
• f3 -1(N), f2 1(Y), f4 1(Y)
• d35 0.5, d25 1.0, d45 1.5
• f5 -11/0.5 11/1.0 11/1.5
• -0.33
• ? N

32
Case-Based Reasoning

• Issues
• Selection of features
• Computing similarity
• Advantages
• Knowledge acquisition can be simplified
• Provide learning capability to expert systems
• Disadvantages
• Lack of explanation capability
• Store/compute trade-offs