Digital Signal Processing Reference
In-Depth Information
rate
η
0
and the maximum number of iterations
N
. Set the iteration
counter equal to zero,
ν
= 0. Randomly generate an initial codebook
L
=
{
L
1,0
,
···
,
L
c,0
}
.
2. Adaptation, part I:
Compute the updated learning rate
η
=
η
0
1
N
.
ν
−
Also set
ν
=
ν
+1.
3. Adaptation, part II:
For each input vector
x
find the winning proto-
type based on the equation
2
<
2
,
||
x
−
L
i,ν−1
||
||
x
−
L
j,ν−1
||
∀
j
=
i
(7.86)
Determine the membership functions
u
ir,ν
using
u
ir,ν
=
u
||
,
L
i,ν−1
||
2
x
−
∀
r
=
i.
(7.87)
||
x
−
L
r,ν−1
||
2
Determine
w
ir,ν
using
w
ir,ν
=
u
||
,
L
i,ν−1
||
2
x
−
∀
r
=
i.
(7.88)
||
x
−
L
r,ν−1
||
2
Determine
n
ir,ν
using
||
w
ir,ν
,
2
x
−
L
i,ν−1
||
n
ir,ν
=
u
ir,ν
−
∀
r
=
i.
(7.89)
||
x
−
L
r,ν−1
||
2
4. Adaptation part III:
Determine the update of the winning prototype
L
i
using
c
L
i,ν
=
L
i,ν−1
+
η
(
x
−
L
i,ν−1
)(1 +
w
ir,ν
)
(7.90)
r=i
Determine the update of the nonwinning prototype
L
j
=
L
i
using
L
j,ν
=
L
j,ν−1
+
η
(
x
−
L
j,ν−1
)
n
ij,ν
.
(7.91)
5. Continuation:
If
ν
=
N
, stop; else go to step 2.
Search WWH ::
Custom Search