Rules

About This Presentation

Title:

Rules

Description:

Rules( ) Representation of knowledge in the form of if-then-else Rule chaining Procedure applied to knowledge base of rules to solve a problem – PowerPoint PPT presentation

Number of Views:131

Avg rating:3.0/5.0

Slides: 93

Provided by: Kathy444

Category:

more less

Transcript and Presenter's Notes

Title: Rules

1
Rules Rule Chaining

Rules(??)
Representation of knowledge in the form of
if-then-else
Rule chaining
Procedure applied to knowledge base of rules to
solve a problem
Rule-based system(?? ?? ???)
combination of a rule base and inference
procedures to solve a particular problem

2
(No Transcript)
3
Rule-based Systems

Working memory contains assertions
Stretch is a giraffe
Long-term Memory
Rule base contains rules
Assertion adding
if giraffe(x) then long-legs(x)
system considered to be a deduction system
Action taking
if food(x) then eat(x)
system considered to be a reaction system

4
?? ??? Allan Newell (Production system)
5
??? ???? ??
6
Antecedent-Consequent Rules

Rn IF condition 1
condition 2
...
THEN consequent 1
consequent 2
...

7
Forward/Backward Chaining???/??? ??)

Remember logic proofs
Can work forward from initial state by applying
rules to assertions to generate new assertions
until goal is reached
match if-part to add then-part
Can work backward from goal state to work
backwards from goal to assertions
match then-part to goal, try to find support for
if-part

8
The Inference Process
Forward Chaining 1 Enter new data 2 Fire
forward chaining rules 3 Rule actions infer new
data values 4 Go to step 2 5 repeat until no
new data can be inferred 6 If no solution, rule
base is insufficient
9
The Inference Process

Forward Chaining

1
Add new data values to knowledge base
Get some new data
4
Fire forward chaining rules
2
Infer new data values from rule actions
3
10
The Inference Process

Cascading Rules
Only applicable in forward chaining
Rule executes causing inference of new data
New data is added to the knowledge base
New data added causes other rules to fire
Can be very time consuming and inefficient in
large systems

11
The Inference Process

Backward Chaining
1 State a specific goal (question)
2 Find rules which resolve the goal
3 At runtime, answer questions to satisfy the
antecedents of the rules as required
4 Obtain a result (goal resolved or not)

12
The Inference Process

Backward Chaining

1
State primary goal to source
Source sub goals
Fire backward chaining rules
2
3
Primary goal sourced
4
13
The Inference Process

Chaining in Action
A list of known facts - A, B, D, G, P, Q, R, S.

14
The Inference Process

Chaining in Action
A list of known facts - A, B, D, G, P, Q, R, S.

Raining outside?
15
The Inference Process

Chaining in Action
A list of known facts - A, B, D, G, P, Q, R, S.

Cold?
16
The Inference Process

Chaining in - X is the goal
A list of Rules
1 A I gt X
2 AB gt C
3 CD gt E
4 FG gt H
5 EH gt X
6 AC gt F
7 PQ gt R
8 RS gt T

17
The Inference Process

Chaining in Action (Backward)
A list of Rules
1 A I gt X
2 AB gt C
3 CD gt E
4 FG gt H
5 EH gt X
6 AC gt F
7 PQ gt R
8 RS gt T

Rule 1 never fired because either I was not
true or the user did not know it was true
Rules 78 never fired because they were not
relevant to proving our goal of X, though they
may have been used if we had another consultation
with a different goal
18
The Inference Process

Chaining in Action (Forward)
A list of known facts - A, B, D, G, P, Q, R, S,
are all true (we asked the user) and I is false
or not known

19
The Inference Process

Chaining in Action (Forward)
A list of Rules
1 A I gt X
2 AB gt C
3 CD gt E
4 FG gt H
5 EH gt X
6 AC gt F
7 PQ gt R
8 RS gt T

20
The Inference Process

Chaining in Action (Forward)
A list of Rules
1 A I gt X
2X AB gt C
3X CD gt E
4 FG gt H
5 EH gt X
6X AC gt F
7X PQ gt R
8X RS gt T

21
Terminology(??)

If giraffe(x) then long-legs(x)
x is a variable
x is bound when it becomes identified with a
particular value (e.g. giraffe(Geoffrey)
pattern is instantiated when variables are bound

22
ZOOKEEPER deduction system

Goal identify animals
Domain 7 kinds of animals (cheetah, tiger,
giraffe, zebra, ostrich, penguin, albatross)
Design intermediate conclusions are generated
(as opposed to having a rule for each kind of
animal)
advantages expansion of domain easier, simpler
rules to deal with, simpler rules to create (like
modular programming)

23
Rules

// ????(mammal)?? ??(bird)??
Z1 IF ((? animal) has hair) THEN ((? animal) is a
mammal))
?? P1. IF the animal has hair
THEN it is a mammal
Z2 IF ((? animal) gives milk) THEN ((? animal) is
a mammal))
?? P2. IF the animal gives milk
THEN it is a mammal
Z3 IF ((? animal) has feathers) THEN ((? animal)
is a bird))
Z4 IF ((? animal) flies) ((? animal) lays eggs)
THEN ((? animal) is a bird))

24
More Rules

// ????(carnivore)??
Z5 IF ((? animal) is a mammal) ((? animal) eats
meat) THEN ((?animal) is a carnivore))
Z6 IF ((? animal) is a mammal) ((? animal) has
pointed teeth) ((? animal) has claws) ((?
animal) has forward-pointing eyes) THEN
((?animal) is a carnivore))
//???(ungulate)? ??(hoof)? ?? ????.
Z7 IF ((? animal) is a mammal) ((? animal) has
hoofs) THEN ((?animal) is a ungulate))
// ???? ?? ??? ???(cud)? ?? ???.
Z8 IF ((? animal) is a mammal) ((? animal)
chews cud) THEN ((? animal) is a ungulate))

// ????? ??? ? ?? ??? ??.
Z9 IF ((? animal) is a carnivore) ((? animal)
has tawny color) ((? animal) has dark spots)
THEN ((? animal) is a cheetah))
Z10 IF ((? animal) is a carnivore) ((? animal)
has tawny color) ((? animal) has black
stripes) THEN ((? animal) is a tiger))
// ?????? ??? ?? ? ?? ??? ??.
Z11 IF ((? animal) is a ungulate) ((? animal)
has long neck) ((? animal) has long legs)
((? animal) has dark spots)THEN ((? animal) is
a giraffe))
Z12 IF ((? animal) is a ungulate) ((? animal)
has black stripes) THEN ((? animal) is a zebra))

26
More Rules

// ?? ? ??? ??? ?? ???.
Z13 IF ((? animal) is a bird) ((? animal) does
not fly) ((? animal) has long legs) ((?
animal) has long neck) ((? animal) is black
and white) THEN ((? animal) is a ostrich))
Z14 IF ((? animal) is a bird) ((? animal) does
not fly) ((? animal) swims) ((? animal) is
black and white) THEN ((? animal) is a penguin))
Z15 IF ((? animal) is a bird) ((? animal) flies
well) THEN ((? animal) is a albatross))

27
Example with forward chaining

Assertions (p.124)
Stretch has hair
Stretch chews cud
Stretch has long legs
Stretch has a long neck
Stretch has tawny color
Stretch has dark spots

28
(No Transcript)
29
Example with forward chaining

Process (p.126 Fig. 7.2)
Z1 IF ((? animal) has hair) THEN ((? animal) is a
mammal))
(? animal) bound to Stretch when matching with
first assertions Stretch has hair
Stretch is a mammal
Z8 IF ((? animal) is a mammal) ((? animal)
chews cud) THEN ((? animal) is a ungulate))
Stretch is a ungulate
Z11 IF ((? animal) is a ungulate) ((? animal)
has long neck) ((? animal) has long legs)
((? animal) has dark spots)THEN ((? animal) is
a giraffe))
gt Stretch is a giraffe

? ??? ??? ???? ?? ?(dark spot)? ???.
?? Z9? Z11? ??? ? ??? ????? ? ????? ? ??? ????
??? ? ??.
? ?? ???? Z1? ?? ??? ?? ????.

? ??? ????? ???? ??. Z8? ??? ?????, ???? ? ??
?(even-toed)??? ????.
? ??? ? ??? ? ?? ?? ??. ??? Z11? ????, ??? ????
?? ??.

32
Zookeepers forward chaining

Until no rule produces a new assertion or the
animal is identified
For each rule,
Try to support each of the rules antecedents
by matching it to known facts
If all the rules antecedents are supported,
assert the consequent unless there is an
identical assertion already
Repeat for all matching and instantiation
alternatives

33
Backward Chaining

Given a hypothesis determine if it is true
Until all hypotheses have been tried and none
have been supported or the animal is identified
For each hypothesis,
Try to support each of the rules antecedents
by matching it to known facts or by
backward chaining through another rule,
creating new hypotheses. Be sure to check all
matching and instantiation alternatives.
If all the rules antecedents are supported,
announce success and conclude that the
hypothesis is true

34
Backward Chaining Example

Assertions
Swifty has forward-pointing eyes
Swifty has claws
Swifty has pointed teeth
Swify has hair
Swifty has a tawny color
Swifty has dark spots
Hypothesis Swifty is a cheetah

35
(No Transcript)
36
Process

Z9 IF ((? animal) is a carnivore) ((? animal)
has tawny color) ((? animal) has dark spots)
THEN ((? animal) is a cheetah))
Check first antecedent (Swifty is a carnivore)?
Rules Z5 Z6 possible
Z5 IF ((? animal) is a mammal) ((? animal)
eats meat) THEN ((?animal) is a carnivore))
Check first antecedent (Swifty is a mammal)?
Rules Z1 and Z2 possible
Z1 IF ((? animal) has hair) THEN ((?
animal) is a mammal))
Swifty has hair so conclude Swifty is a
mammal
Check second antecedent (Swifty eats meat)?
unknown so rule cant be used
Z6 IF ((? animal) is a mammal) ((? animal)
has pointed teeth) ((? animal) has claws)
((? animal) has forward-pointing eyes) THEN
((?animal) is a carnivore))
Check first antecedent (Swifty is a mammal)?
YES, established earlier
Check other antecedents all true from
assertions
Conclude (Swifty is a carnivore)
Check other antecedents true from assertions
Conclude (Swifty is a cheetah)

37
Which direction to choose?

Depends on problem (p.129)
search space features
fan out
set of facts lead to many conclusions gt high fan
out
fan in
typical hypotheses result in many questions gt
high fan in
If more fan out than fan in gt backward
If more fan in than fan out gt forward
problem features
If interested in one of many possible conclusions
gt backward
Notice that this results in a focus on the
particular hypothesis being considered. Only
information relevant to that will be pursued.
If interested in everything that can be concluded
from a set of facts that wont change gt forward

38
MYCIN

Early rule-based system for identifying bacterial
infections
Background
antibiotic treatment should begin right away
antibiotics that are specific to a particular
organism are more effective than broad-spectrum
antibiotics
takes time to culture organism
System uses backward chaining to keep questions
focussed
also makes interface easier (specific questions
are asked)

gtWhat is the patients name?
John Doe
gtMale or female?
male
gtage?
He is 65
gtHave you obtained positive cultures indicating
general type?
Yes
gtWhat type of infection is it?
Primary bacteremia
gtWhen did symptoms first appear?
May 5
gtLets call the most recent positive culture C1.
From what site was C1 taken?
From the blood
gtWhen?
May 9

40
gtLets call the first significant organism from
this culture U1. Do you know the identity of
U1? No gtIs U1 a rod or a coccus or something
else? Rod gtWhat is the gram stain of
U1? Gram-negative gtHave there been positive
cultures from sites that are not associated with
the infections about which you are seeking
advice? No gtDid U1 grow aerobically? Yes gtMy
recommendation will be based on the opinion that
the identity of U1 may be1. Pseudomonas-aeruginos
a 2. Klebsiella-pneumoniae 3. E. coli 4.
Bacteroides-fragilis 5. Enterobacter 6.
Proteus-nonmirabilis
41
gtOn a scale of 0 to 4, where higher numbers
indicate greater severity, how would you rate the
patients degree of sickness? 3 gtDoes the patient
have a clinically significant allergic reaction
to any antimicrobial agent? No gtHow much does the
patient weigh? 70 kilograms gtMy first
recommendation is as follows to cover for items
1, 2, 3, 5, and 6, give gentamycin using a dose
of 119 mg (1.7 mg/kg) q8h IV or IM for 10 days.
Modify does in renal failure. Also, to cover for
item 4, give clindamycin using a dose of 595 mg
(8.5 mg/kg) q6h IV or IM for 14 days. If
diarrhea or other gastrointestinal symptoms
develop, check for pseudomembranous colitis.
42
Mycin

About 500 rules
Handles about 100 causes
Typical rule
If xs type is primary bacteremia the
suspected entry point of x is the
gastrointestinal tract the site if the
culture of x is one of the sterile sitesthen
there is evidence that x is bacteroides

43
MYCINs reasoning

Backward chaining (goal-directed)
Heuristics for efficiency and user-friendliness
(dont ask unnecessary questions)
Strategy
first ask a general set of questions to rule out
totally unlikely diagnoses.
focus on particular disorders, and backward chain
to try to prove each one.
Rule firing
first checks all the premises of the rule to see
if any are known to be false (dont pursue)
first look at rules that have more certain
conclusions, abandon a search once the
certainties involved get below 0.2

44
Constraints on MYCIN

system must be useful (implying competency,
consistency and high perfomance).
task area chosen partly because of a demonstrated
need
e.g., in early 70's 1/4 of the population of the
US were given penicillin 90 of these
prescriptions were unnecessary
emphasis of its supportive role as a tool for a
physician, rather than as a replacement for his
own reasoning process
accommodate a large and changing body of
technical knowledge
capable of handling an interactive dialog
supplying coherent explanation both its line of
reasoning and its knowledge.
Speed, access and ease of use.

45
MYCINs Success

MYCIN introduced several new features which have
become the hallmarks of the expert system.
knowledge base made of rules
rules are probabilistic - MYCIN is robust enough
to to arrive at correct conclusions even when
some of the evidence is incomplete or incorrect.
ability to explain its own reasoning process.
The user (a physician) can interrogate it in
various ways - by enquiring, why it asked a
particular question or how it reached an
intermediate conclusion, for example. It was one
of the first genuinely user-friendly systems.
it actually works - It does a job that takes a
human years of training.

46
MYCINs Success cont.

Commercial or routine use?
produced high-quality results - its performance
equalled that of specialists
never used routinely itself by physicians
spawned a whole series of medical-diagnostic
'clones', several of which are in routine
clinical use
Other systems resulting from work on MYCIN
TEIRESIAS (explanation facility concept) ,
EMYCIN (first shell),
PUFF,
CENTAUR,
VM,
GUIDON (intelligent tutoring),
SACON
ONCOCIN
ROGET