Information Technology Reference
In-Depth Information
structural descriptor. Descriptor type is very important to determine operation of
applied descriptor.
In learning system Star, basic form of assertion is c-expression, which is
defined as a conjunction normal form:
< quantifier form><conjunction of relation statement> (6.5)
where < quantifier form > represent none or many quantifiers
。
<relation
statement> is specific form of predicate. Following is an example of
c-expression:
∃
P
1
)])
i.e. shape of object P
0
and P
1
is box, and object P
0
is more weighty than object
P
0
,P
1
([shape(
P
0
∧
P
1
) = box][weight(
P
0
) > weight(
1
.
An important specific form of c-expression is a-expression, i.e. atomic expression,
which does not include “inter disjunction”. Inter conjunction and disjunction
mean “and” and “or” of conjunction items respectively; outer conjunction and
disjunction mean “and” and “or” of conjunction predicate, i.e. “and” and “or” in
common sense.
P
7.2.4
Selective and constructive generalization rules
A generalization rule is to transform a description into a more general description.
A more general description tautologically implicates the initial description. Since
generalization rule is non- tautological, if
F
|<
H,
then for all facts that makes F
to be false, they make
H
to be false (~
F
¼
~
H
).
In concept acquisition, if a rule
E
::
>
K
is transformed into a more general rule
D
. So we can get generalization rules using tautological
implication in formal logic. For example, formal logic holds
::
>
K
, it must hold
E
¼
D
P
∧
Q
¼
P,
then it
can be transformed into generalization rule:
P
|
(7.6)
If using labeling predicate calculus to rep resent these generalization rule, we
should mainly consider transforming one or more statement into a single more
general generalization rule:
{
∧
Q
K
P
K
::
>
::
>
I
|
D
i
K
i
}
i
Ӥ
D
K
(7.7)
::
>
::
>
Equals to:
D
1
|
∧
D
2
∧
∧
D
n
K
D
K
(7.8)
┄
::
>
::
>
The rule represents that if an event meets all description
D
i
(
i
∈
I), then it must
meet a more general description
.
A basic characteristic of generalization transform is as following. What it gets
is only a hypothesis, and must be tested using new data. Furthermore,
generalization rule do not assure that the descriptor gotten from it is rational or
D