Information Technology Reference
In-Depth Information
MCA-EXIN (Solid), MCA-FENG1 (Dashed)
0.13
0.12
0.11
0.1
0.09
0.08
8200 8400 8600 8800 9000 9200 9400 9600 9800
Iterations
Figure 2.11
Smallest eigenvalue computation by MCA EXIN and FENG (last iterations).
Equation (2.38) implies that for high
w (
t
)
2
, as during the divergence in the
MC direction,
w(
t
)
|
EXIN
≤
w(
t
)
|
OJAn
≤
w(
t
)
|
LUO
(2.136)
An analogous comparison can be made between MCA EXIN and the other neu-
rons by using the RQ minimal residual property (2.5), which can be written as
(
R
−
r
(w(
t
))
I
) w(
t
)
≤
(
R
−
µ
I
) w(
t
)
(2.137)
R
being, as usual, the autocorrelation matrix. Discretizing, it holds that
≤
y
y
2
(
t
)w(
t
)
w (
t
)
y
(
t
)
x
(
t
)
−
(
t
)
x
(
t
)
−
µw(
t
)
(2.138)
2
2
Under the same assumption of high
w (
t
)
2
, it follows that
w(
t
)
|
EXIN
≤
w(
t
)
|
OJA
(2.139)
by using (2.137) and
µ
=
y
2
(
t
)
,
w(
t
)
|
EXIN
≤
w(
t
)
|
OJA
+
(2.140)
Search WWH ::
Custom Search