Information Technology Reference
In-Depth Information
p
p
1
2
(a)
3
0
0
-1
3
0
0
-1
1
-1
(b)
3
0
0
-1
3
-1
0
1
-1
(c)
3
0
0
-1
1
-1
3
0
0
-1
1
2
3
4
5
6
7
8
9
10
Fig. 6.
An Example of Mutation Technique
3.5
Fitness Function
The fitness
F
(
s
) of a particular
MACA
in a population is determined by the
weighted mean of two factors -
F
1
and
F
2
. The fitness criteria
F
1
of the
MACA
is determined by the percentage of patterns satisfying the
relation 1
.
F
2
has been
defined as -
F
2
=1
−
[(
m −
1)
/n
]
l
(2)
where 2
m
denotes the number of attractor basins for the
n
cell
CA
, and
l
is
equal to 1
·
8. The value of
l
is set empirically.
Subsequent to extensive experimentation, we have fixed up the relative
weighage of
F
1
and
F
2
to arrive at the following empirical relation for the fitness
function
F
(
s
)=0
.
8
· F
1
+0
.
2
· F
2
(3)
4 Characterization of Attractor Basin - Its Capacity to
Accommodate Noise
A classification machine is supposed to identify the zone, the trained patterns
occupy. That is, let
P
i
be a representative pattern of
n
-bit learnt by the classifier
as (say) Class A. Then a new
n
-bit pattern
P
i
also belongs to the same class, if
the hamming distance (
r
) between
P
i
and
P
i
is small (
r<<n
).
The hamming
distance
r
is termed as noise while the pattern
P
i
is termed as noisy
pattern
.
This section provides a comprehensive study on the capacity of the classifier
to accommodate noise. The study is developed in four phases
-
Phase I:
Establishes the fact that probing into the nature of pattern dis-
tribution of zero basin is equivalent to studying the noise accommodating
capacity of the classifier.
-
Phase II:
Establishes the fact that characterization of the zero basin of
MACA
with two attractor basins can be easily extended to
MACA
with
more than two attractors.
-
Phase III:
Characterizes the zero basin of
MACA
with two attractor basins.
-
Phase IV:
Generalizes the study with 2
m
number of basins,
m
varying from
1to
n
.
