Information Technology Reference
In-Depth Information
the rule represents that if a concept description consists a part F
1
, F
1
implicate
another concept F
2
, a more general description can be gotten by using F
2
to
replace F
1
.
(2) Computing variable rule
Computing quantifier (CQ rule):
Ӡ
V1,V2,
,Vk
F
[V
1
,V
2
,
,V
k
]
CQ rule will generate a new descriptor “#
-COND”, representing number of v
i
meeting a certain condition COND. For example, “#
v
v
i
-length
─
2..4” represents
number of vi whose length is between 2 and 4.
Computing number of predicate (CA rule): in description, descriptor is a
relation which possesss several variables REL(v
1
,v
2
, ···), CA rule will compute
the number of predicate which meets condition COND in relation REL.
(3) Generating link attribute rule
In concept description, if variables in a concept description formulate a link
because of the difference of transfer relation, the rule can generate descriptors of
some particular objects in depicting link. This kind of object may be as follows:
LST-object: “minimum object”, or starting object of link
MST-object: ending object of link
MID-object: middle object of link
Nth-object: the N-th object of link
(4) Rule detecting dependence relation of descriptors
Assume that given an object set representing a concept, we use attribute
description to depict features of object. The description defines attribute values of
object, and do not depict structural features of object. Assume that in all fact
descriptions, values of linear descriptor
x
are ascending ordered, while values of
another linear descriptor
are ascending or descending ordered, then a
2-dimensional descriptor M(
y
x,y
) is generated, which represents that there is
monotonic relation between
x
and
y
. Descriptor has value
↑
when value
y
is
ascending, otherwise descriptor has value ↓.
7.3 Inductive Bias
Bias plays an important role in concept learning. Bias means all factors affecting
assume selection in concept learning except for positive and negative example.
These factors include: (a) language describing assumption. (b) assumption space
considered by program. (c) assumption procedure according to what order. (d)
admit definitive criterion, that is, research procedure with known assumption
