Information Technology Reference
In-Depth Information
the best choice, because it has the least number of deleted patterns. When
m
<2
r
, the
filtration process can be described as following.
(1)
Get the matching start point
p
and the maximum continuous matching length
m
between the detector
d
and the matched self string.
(2)
If
lp
, stuff the (
p
+
m
-
r
)-th bit of
d
with '#', or else stuff the
(
p
+
r-
1)-th bit of
d
with '#'.
(3)
If the maximum length of contiguous 0/1 bit in
d
is shorter than
r
, delete
d
and
try to generate a new detector.
≤
−
p
−
m
+
2
When
m
≥
2
r
, set all bits between (
p
+
r
-1,
p
+
m
-
r
) as '#'. Obviously, it is very
simple when
.
Notable, only one segment that the number of the maximum continuous matching
bits is no smaller than
r
, has been considered above. However, when
r
is small and
l
is
relative large, the number of such segments is possibly more than one. If there are two
or more segments of the maximum continuous matching bits not shorter than
r
,
methods for finding the bits to be stuffed by '#' can be designed according to the same
idea described above.
Note that if the maximum length of contiguous 0/1 bit in
d
is shorter than
r
, the bit
replacing operation has made detector
d
useless because it can not match any
self/non-self string in the string space, then
d
must be deleted and a new detector should
be try to be generated.
The following is the process of trying to generate a new detector when a detector is
deleted in the process of strategy II. This algorithm is described below.
m
≥
2
r
(1)
Generate a new detector
d'
randomly.
(2)
If
d'
is already included in the current set
R
, delete it and go back to (1).
(3)
Perform partial matching between
d'
and strings in
S
one by one, if
d'
matches
any self string, perform the above filtration process of strategy II on
d'
. If
d'
is
deleted in the filtration process, go to (1).
(4)
Add
d'
to
R
.
To limit the time cost for generating a new detector, when an old detector is deleted
in the updating process, step (1)-(3) in this process will be performed for only a small
constant number of cycles.
4 Simulation Experiments
To show the improvements of these two novel detector set updating strategies, they are
compared with the strategy used by ASTA-CED algorithm [8] which just delete a
detector matching self and try to generate a new one randomly. Binary strings in form
of “previous state/current state” are used for representing state transitions and
detectors. The framework of the simulation experiment system used in [8, 9] is adopted
with some modifications here, as shown in Fig 3.
