Information Technology Reference
In-Depth Information
consolidating final fuzzy ARB population()
Antigenic presentation
for each
pair of ARB's
do
if
affinity between two current ARB's is less than
ε
then
merge two ARB's in a single fuzzy ARB
Figure 5.12
A subfunction of fuzzy AINE.
fuzzy ARBs from dominating the least-stimulated ARBs, because their stimula-
tion cannot go beyond certain limit. h is process can be thought of as a niching
mechanism to maintain diversity in the population of fuzzy ARBs.
5.2.3.3.1 The Dynamic Weight B Cell Model
Nasraoui et al. (2003b) proposed a modifi cation of the fuzzy IN model to perform
dynamic unsupervised learning. h is model was proposed as an attempt to solve
the scalability problems present in most IN problems, which in general need to
manipulate a large number of antibodies (or B cells) and links that represent the
interactions among them. h is model is also intended to deal with dynamic envi-
ronments. h ereby, antigens are presented to the IN one at a time. Accordingly,
the stimulation levels and radius of infl uence of each ARB are updated after the
presentation of each antigen.
In this model, the concept of a dynamic weighted B cell (D-W-B cell) is
introduced. h is is based on the concept of a fuzzy ARB, but instead of only
considering a function to model the infl uence zone of an antigen, it also introduces
a temporal aspect in the model. h en, t he ma i n a s su mpt ion here i s t hat more c u rrent
data will have a higher infl uence on the network dynamics as compared to less
current or older data. h us, the membership function defi ned in Equation 5.22,
after
J
antigens have been presented, becomes
dx c
2
(
)
(
J
j
)
j
i
fx
()
e
(5.23)
i
j
2
2
i
where
d
(
x,c
i
) is the distance between the center of the
i
th D-W-B cell and antigen
x
j
,
which is the
j
th antigen encountered by the IN. Accordingly, the stimulation level
of the IN, after
J
antigens have been presented, is defi ned as
∑
fx
()
i
j
(5.24)
sx
()
i
j
2
i
Search WWH ::
Custom Search