Representing knowledge with rules

1
Representing knowledge with rules

• Points
• Definitions
• Production systems
• Conflict resolution strategies
• Examples
• A financial advisor
• Bagger
• XCON
• Animal classification in Prolog
• A forward-chaining engine
• Animal classification revisited

2
Rules...

• A rule, in its simplest form, is a pair
• conditions ?? actions
• An action can be real (maybe software-simulated),
  or it can only add -- assert -- new conditions.
• A rule is not the same as an implication
• premises ? conclusions
• where false premises are allowed, but there are
  significant similarities.
• In rule-based systems, the basic rule form is
  enhanced with control information and,
  especially, with probabilities or confidence
  measures.

3
... and applications

• One class of applications are rule-based expert
  systems.
• Despite their apparent simplicity, rules are
  quite expressive. There are two types of uses.
• Synthesis (to construct or configure)
• XCON is an example of a deployed system
• our simple case study is Bagger.
• Analysis (to recognize or diagnose)
• MYCIN is the best-known (almost) success story
• our simple case study is animal classification.

4
Forward-chaining production systems

• Rule memory (rules are called productions).
• Working memory (literals that represent the
  current state of the problem) actions change
  working memory.
• Rule selection and application mechanisms (rules
  are indexed by situations).

Rule selection relies on pattern matching. A
conflict arises if more than one rule has been
selected to match the current situation. Conflict
resolution procedures select one rule. This rule
is fired if its conditions hold. Actions are
performed, working memory updated.
5
Some conflict resolution strategies

• Order rules (partially) by specificity -- the
  more conditions, the more specific the rule
  prefer more specific rules.
• Rank all rules prefer rules with a higher rank.
• Rank all facts prefer rules with more highly
  ranking facts.
• Record the time of changes to working memory
  prefer rules that depend of recent additions.

6
Implementing forward chaining

• An agenda is a handy data structure for
  implementing forward chaining. It is "a list of
  things to do, here -- candidate rules.
• Conditions are considered in some order any rule
  that has the currently examined condition in its
  left-hand side becomes partially activated. Its
  copy, with the matched condition removed, goes on
  the agenda.
• A rule is selected for firing when it has become
  fully activated, that is, when all its left-hand
  side conditions have been matched.
• Otherwise the rule is kept for future use, just
  in case the remaining conditions become asserted
  later.

7
A financial expert system

1. saved(low) ? investIn(savings)
2. saved(high) ? income(adequate) ??investIn(stocks)
3. saved(high) ? income(inadequate)
   ??investIn(mixed)
4. ?x amountSaved(x) ? ?y (nbrDependents(y) ?x gt
   y5000) ? saved(high)
5. ?x amountSaved(x) ? ?y (nbrDependents(y) ?x
   y5000) ? saved(low)
6. ?x earnings(x, steady) ? ?y (nbrDependents(y) ?x
   gt 15000 y4000) ? income(adequate)
7. ?x earnings(x, steady) ? ?y (nbrDependents(y) ?x
   15000 y4000) ? income(inadequate)
8. ?x earnings(x, unsteady) ? income(inadequate)

8
A financial expert system (2)

1. amountSaved(22000)
2. nbrDependents(3)
3. earnings(25000, steady)

Given facts 9-11 and rules 1-8, we perform
forward chaining. 9 and 10 match the
conditions of 4 and 5, but only 4 passes
the test. Add a new fact

1. saved(high)

10 and 11 match the conditions of 6 and
7, but only 7 passes the test. Add a new fact

1. income(inadequate)

12 and 13 match the conditions of 3. Add a
new fact

1. investIn(mixed)

Done.
9
Bagger a robot that bags groceries

• The main idea is simple.
• Bag the large items
• put large heavy items one per bag
• start a new bag when a large item does not fit in
  any partially filled bag.
• Bag the mid-sized items
• meat and frozen goods go in freezer bags
• start a new bag when a mid-sized item does not
  fit in any partially filled bag.
• Bag the small items
• try to avoid bagging them with heavy items
• start a new bag when a small item does not fit in
  any partially filled bag.
• What could be a good conflict resolution strategy
  for the following rules?

10
Bagger (2)

• (1) the step is bag-large there is a large
  unbagged item there is a large unbagged heavy
  item there is a bag with lt 6 large items?? bag
  the large heavy item
• (2) the step is bag-large there is a large
  unbagged item there is a bag with lt 6 large
  items?? bag the large item
• (3) the step is bag-large there is a large
  unbagged item?? start a new bag bag the large
  item
• (4) the step is bag-large?? end the step
  bag-large begin the step bag-medium

11
Bagger (3)

• (5) the step is bag-medium there is a medium
  unbagged item the medium item is a freezer-bag
  item there is an unfilled bag?? put the medium
  item in a freezer bag bag the medium item
• (6) the step is bag-medium there is a medium
  unbagged item there is an unfilled bag?? bag
  the medium item
• (7) the step is bag-medium there is a medium
  unbagged item?? start a new bag bag the
  medium item
• (8) the step is bag-medium?? end the step
  bag-medium begin the step bag-small

12
Bagger (4)

• (9) the step is bag-small there is a small
  unbagged item there is an unfilled bag without
  large heavy items?? bag the small item
• (10) the step is bag-small there is a small
  unbagged item there is an unfilled bag?? bag
  the small item
• (11) the step is bag-small there is a small
  unbagged item?? start a new bag bag the small
  item
• (12) the step is bag-small?? end the step
  bag-medium stop

13
XCON

• XCON configures DEC computing equipment. It is
  one of rather few (but well known!) examples of
  commercially successful expert systems.
• XCON is, in a sense, Bagger's big brother. Its
  general plan of action is quite simple a
  sequence of steps. (There are, however, dozens of
  rules for each step...)
• The configuring operation includes selecting
  components to match the order, laying out the
  spatial arrangement of cabinets, filling the
  cabinets, and so on.

14
XCON (2)

• Major steps
• Check the order, look for missing or incompatible
  items.
• Lay out the processor in cabinets.
• Put boxes in the input/output cabinets, put
  components in those boxes.
• Put panels in the input/output cabinets.
• Lay out the floor plan.
• Plan the cabling.

15
A classification system in Prolog

• http//www.site.uottawa.ca/nat/Courses/csi4106_20
  08/Material/animals.pl

16
Animal classificationin a rule-based system
The general rules

• http//www.site.uottawa.ca/nat/Courses/csi4106_20
  08/Material/animals_rules

17
Animal classification (2)

• A session with the forward chainer

http//www.site.uottawa.ca/nat/Courses/csi4106_20
08/Material/animals_session
18
Animal classification (3)
The forward chainer in Prolog
http//www.site.uottawa.ca/nat/Courses/csi4106_20
08/Material/forward_chainer.pl
19
A glimpse at expert systems

• Your reading for independent study ? sections
  8.1-8.2.

Architecture of a typical expert system
20
A glimpse at expert systems (2)

• Guidelines to determine whether a problem is
  appropriate for expert system solution
• The need for the solution justifies the cost and
  effort of building an expert system.
• Human expertise is not available in all
  situations where it is needed.
• The problem may be solved using symbolic
  reasoning.
• The problem domain is well structured and does
  not require commonsense reasoning.
• The problem may not be solved using traditional
  computing methods.
• Cooperative and articulate experts exist.
• The problem is of proper size and scope.

21
A glimpse at expert systems (3)
Exploratory development cycle
22
Later or much later

• More topics appear in the very useful chapter 8
• planning (section 8.4) will be discussed later
• case-based reasoning is left for another course ?.