Information Technology Reference
In-Depth Information
Algorithm 1
Evolving GMACA
Input : Pattern Size
(
n
)
, Pattern Set to be learnt,
Initial temperature (Temp
initial
)
Output : GMACA Rule.
Temp
point
= Temp
initial
Initialize CS, BS as zero rules.
while Temp
point
>
0
{
if Temp
point
>
0
.
50
× Temp
initial
Randomly generate graph as guess soln.
else
Generate graph by Scheme 1 or 2.
Generate state transition table and rule table.
NS = GMACA - Rule
δ
cost
= cost value(NS) - cost value(CS)
if δ
cost
<
0
{
CS=NS
if cost value(NS) < cost value(BS)
{
BS=NS
}
}
else
accept CS = NS with prob. e
−|δ
cost
|/Temp
point
.
Reduce Temp
point
exponentially.
}
The time required to find out an
n
-cell
GMACA
increases with the value of
r
max
which is the
maximum permissible noise
. The number of nodes in a graph
p
increases with
r
max
. So, we investigate the minimum value of
r
max
for which
cost function attains minimum value close to zero.
3.4 Minimum Value of Maximum Permissible Noise (
r
max
)
r
max
specifies the noise level at the training/synthesis phase. To find out the
minimum value of
r
max
, we carry out extensive experiments to evolve pattern
recognizable
n
-cell
GMACA
for different values of
n
. The results are reported
in
Table 1
.
Column I
of
Table 1
represents noise present in training phase; whereas
Col-
umn II
represents the percentage of convergence for different noise value per
bit in identification/recognition phase. The results of the
Table 1
clearly estab-
lish that in the training (synthesis) phase if we consider that the patterns are
corrupted with single bit noise, then the percentage of convergence at the recog-
nition is better irrespective of noise level.
So, the minimum value of
r
max
is
set to 1 for which
GMACA
based associative memory performs better
.
Then,
Equation 1
reduces to
p
=1+
n
.