T
E
L
K
O
M
N
I
K
A
T
elec
o
m
m
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
,
p
p
.
1
6
3
~
1
7
2
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Kem
en
r
is
tek
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
:
1
0
.
1
2
9
2
8
/TE
L
KO
MN
I
K
A.
v
1
9
i1
.
1
6
3
9
8
163
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
H
a
lf
G
a
uss
ia
n
-
ba
sed wa
v
elet tra
nsf
o
rm for poo
ling
l
a
y
er f
o
r
co
nv
o
lution neur
a
l net
wo
rk
Aqeel
M
.
H
a
m
a
d
Alhu
s
s
a
iny
,
Am
ma
r
D.
J
a
s
im
De
p
a
rtme
n
t
o
f
In
fo
rm
a
ti
o
n
E
n
g
i
n
e
e
rin
g
,
Al
-
Na
h
ra
i
n
Un
i
v
e
rsity
,
Ira
q
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ap
r
1
5
,
2
0
2
0
R
ev
is
ed
J
u
l 3
,
2
0
2
0
Acc
ep
ted
Au
g
2
9
,
2
0
2
0
P
o
o
l
in
g
m
e
th
o
d
s
a
re
u
se
d
to
se
le
c
t
m
o
st
sig
n
ifi
c
a
n
t
fe
a
tu
re
s
t
o
b
e
a
g
g
re
g
a
ted
to
sm
a
ll
re
g
io
n
.
In
t
h
is
p
a
p
e
r,
a
n
e
w
p
o
o
li
n
g
m
e
th
o
d
is
p
ro
p
o
se
d
b
a
se
d
o
n
p
ro
b
a
b
il
it
y
f
u
n
c
t
io
n
.
De
p
e
n
d
i
n
g
o
n
th
e
fa
c
t
t
h
a
t,
m
o
st
i
n
fo
rm
a
ti
o
n
is
c
o
n
c
e
n
trate
d
fro
m
m
e
a
n
o
f
th
e
s
ig
n
a
l
t
o
it
s
m
a
x
imu
m
v
a
lu
e
s,
u
p
p
e
r
h
a
lf
o
f
G
a
u
ss
ian
fu
n
c
ti
o
n
is
u
se
d
to
d
e
te
rm
in
e
we
ig
h
ts
o
f
th
e
b
a
sic
sig
n
a
l
sta
ti
stics
,
wh
ich
is
u
se
d
to
d
e
term
in
e
th
e
tran
sfo
rm
o
f
t
h
e
o
ri
g
i
n
a
l
sig
n
a
l
in
to
m
o
re
c
o
n
c
ise
fo
rm
u
la,
w
h
ich
c
a
n
re
p
re
se
n
t
sig
n
a
l
fe
a
tu
re
s,
t
h
is
m
e
th
o
d
n
a
m
e
d
h
a
lf
G
a
u
ss
ian
tran
sfo
rm
(HG
T).
Ba
se
d
o
n
stra
te
g
y
o
f
tran
sf
o
rm
c
o
m
p
u
tatio
n
,
Th
re
e
m
e
th
o
d
s
a
re
p
ro
p
o
se
d
,
t
h
e
first
m
e
th
o
d
(HG
T1
)
is
u
se
d
b
a
si
c
sta
ti
stics
a
fter
n
o
rm
a
li
z
e
d
it
a
s
we
i
g
h
ts
t
o
b
e
m
u
lt
i
p
li
e
d
b
y
o
ri
g
i
n
a
l
sig
n
a
l,
se
c
o
n
d
m
e
th
o
d
(HG
T2
)
is
u
se
d
d
e
term
in
e
d
sta
ti
stics
a
s
fe
a
tu
re
s
o
f
th
e
o
rig
in
a
l
sig
n
a
l
a
n
d
m
u
lt
i
p
ly
it
wit
h
c
o
n
sta
n
t
we
ig
h
ts
b
a
se
d
o
n
h
a
lf
G
a
u
ss
ian
,
wh
il
e
th
e
th
ird
m
e
th
o
d
(HG
T3
)
is
wo
r
k
e
d
in
si
m
il
a
r
to
(HG
T1
)
e
x
c
e
p
t,
it
d
e
p
e
n
d
o
n
e
n
ti
re
sig
n
a
l.
T
h
e
p
r
o
p
o
se
d
m
e
th
o
d
s
a
re
a
p
p
li
e
d
o
n
th
re
e
d
a
tab
a
se
s,
wh
ich
a
re
(M
NIST
,
CIF
AR1
0
a
n
d
M
IT
-
BI
H
ECG
)
d
a
tab
a
se
.
Th
e
e
x
p
e
rime
n
tal
re
su
l
ts
sh
o
w
t
h
a
t,
o
u
r
m
e
th
o
d
s
a
re
a
c
h
iev
e
d
g
o
o
d
imp
ro
v
e
m
e
n
t,
wh
i
c
h
is
o
u
t
p
e
rfo
rm
e
d
sta
n
d
a
rd
p
o
o
li
n
g
m
e
th
o
d
s
su
c
h
a
s
m
a
x
p
o
o
li
n
g
a
n
d
a
v
e
ra
g
e
p
o
o
li
n
g
.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
n
eu
r
al
n
etwo
r
k
Gau
s
s
ian
HGT
1
HGT
2
HGT
3
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r
th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Aq
ee
l M
.
Ham
ad
A
lh
u
s
s
ain
y
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
E
n
g
in
ee
r
in
g
Al
-
Nah
r
a
in
Un
iv
er
s
it
y
Al
J
ad
r
iy
ah
B
r
id
g
e,
B
ag
h
d
ad
6
4
0
7
4
,
I
r
a
q
E
m
ail: a
q
ee
l_
alh
u
s
s
ain
y
@
u
tq
.
ed
u
.
iq
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
r
ec
en
t
d
ev
elo
p
m
en
t
is
u
s
ed
n
eu
r
al
n
etwo
r
k
in
s
p
ir
ed
s
y
s
tem
s
u
ch
as
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
(
C
NN)
,
De
-
n
o
is
in
g
au
to
en
co
d
er
an
d
o
th
er
d
ee
p
lear
n
in
g
n
eu
r
al
n
etwo
r
k
h
as
d
er
iv
ed
s
ig
n
if
ican
t
d
ev
elo
p
m
en
t
f
r
o
m
b
u
ild
in
g
m
o
r
e
im
p
o
r
tan
t
an
d
co
m
p
licated
n
etw
o
r
k
s
tr
u
ctu
r
e,
wh
ich
lead
to
m
o
r
e
n
o
n
-
lin
ea
r
ac
tiv
atio
n
s
[
1
-
3
]
.
Desp
ite
th
e
d
ev
elo
p
m
en
t
p
r
o
g
r
ess
in
C
NNs,
th
er
e
ar
e
s
till
s
ev
er
al
ch
allen
g
es
en
co
u
n
ter
ed
b
y
th
i
s
n
etwo
r
k
s
u
ch
as
p
r
o
b
lem
o
f
h
ig
h
ca
p
ac
ity
b
ec
au
s
e
o
f
h
u
g
e
p
r
o
ce
s
s
in
g
d
ata,
wh
ich
m
ay
r
esu
lt
in
o
v
er
f
itti
n
g
p
r
o
b
lem
d
u
e
to
h
ig
h
ca
p
ac
ity
o
f
C
NN
[4
-
6
]
.
I
n
o
r
d
er
to
s
o
lv
e
th
ese
p
r
o
b
lem
s
,
d
if
f
er
en
t
r
eg
u
lar
izatio
n
m
eth
o
d
s
wer
e
p
r
o
p
o
s
ed
s
u
ch
as
weig
h
t d
ec
ay
,
weig
h
t
ty
in
g
an
d
p
o
o
lin
g
tech
n
iq
u
es.
T
h
e
ce
n
tr
al
r
o
le
f
o
r
C
NN
n
etwo
r
k
is
th
e
f
ea
tu
r
es
p
o
o
lin
g
o
p
er
atio
n
,
h
o
wev
er
,
p
o
o
lin
g
h
av
e
b
ee
n
litt
le
r
ev
is
ed
b
ey
o
n
d
s
tan
d
ar
d
m
eth
o
d
s
o
f
av
er
ag
e
an
d
m
ax
p
o
o
lin
g
[
7
-
9
]
.
I
n
th
is
p
ap
er
,
an
ew
p
o
o
lin
g
o
f
f
ea
tu
r
es
m
eth
o
d
is
p
r
o
p
o
s
ed
b
ased
o
n
p
r
o
b
ab
ilit
y
f
u
n
ctio
n
,
th
e
p
r
o
p
o
s
ed
m
eth
o
d
is
r
ep
lace
d
th
e
o
u
tp
u
t
o
f
co
n
v
o
lu
tio
n
al
lay
er
with
d
eter
m
in
is
ts
f
ea
tu
r
es
b
y
u
s
in
g
p
o
o
lin
g
o
p
er
atio
n
,
wh
ich
is
ev
alu
ated
b
ased
o
n
d
is
tr
ib
u
tio
n
s
tatis
tics
f
o
r
ea
ch
p
o
o
lin
g
win
d
o
w,
th
e
weig
h
t
o
f
t
h
e
s
e
s
t
a
t
i
s
t
i
c
s
a
r
e
c
o
m
p
u
t
e
d
d
e
p
e
n
d
i
n
g
o
n
n
o
r
m
a
l
d
i
s
t
r
i
b
u
t
i
o
n
o
f
s
t
a
t
i
s
t
i
c
s
[
1
0
-
1
3
]
.
t
h
e
m
a
i
n
c
o
n
t
r
i
b
u
t
i
o
n
s
o
f
t
h
i
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
1
6
3
-
17
2
164
w
o
r
k
i
s
t
h
a
t
,
t
h
e
b
a
s
i
c
p
r
o
p
e
r
t
i
e
s
o
f
t
h
e
s
i
g
n
a
l
a
r
e
f
i
l
t
e
r
e
d
b
y
s
e
l
e
c
t
t
h
e
m
o
s
t
s
i
g
n
i
f
i
c
a
n
t
i
n
f
o
r
m
a
t
i
o
n
,
w
h
i
l
e
t
h
e
d
e
t
a
i
l
o
f
t
h
e
s
i
g
n
a
l
w
i
l
l
h
a
v
e
l
i
t
t
l
e
e
f
f
e
c
t
,
s
o
t
h
e
e
l
i
m
i
n
a
t
i
o
n
o
f
t
h
e
s
i
g
n
a
l
w
i
l
l
b
e
s
a
t
i
s
f
i
e
d
b
y
d
i
s
c
a
r
d
l
e
s
s
s
i
g
n
i
f
i
c
a
n
t
i
n
f
o
r
m
a
t
i
o
n
t
h
r
o
u
g
h
t
h
e
C
N
N
s
a
n
d
t
h
i
s
i
s
e
l
i
m
i
n
a
t
e
d
s
h
o
r
t
c
o
m
i
n
g
o
f
m
a
x
a
n
d
a
v
e
r
a
g
e
p
o
o
l
i
n
g
m
e
t
h
o
d
s
[
1
4
,
1
5
]
.
2.
L
I
T
E
R
AT
U
RE
SU
RVE
Y
T
av
is
W
illi
am
an
d
R
o
b
er
li
i
s
in
tr
o
d
u
ce
d
a
p
o
o
lin
g
m
eth
o
d
b
ased
o
n
wav
elet
tr
an
s
f
o
r
m
,
th
is
m
eth
o
d
was
b
ased
o
n
d
ec
o
m
p
o
s
e
th
e
o
r
ig
in
al
im
ag
e
in
to
s
ec
o
n
d
lev
el
tr
an
s
f
o
r
m
o
f
wav
elet,
th
en
d
elete
all
th
e
s
u
b
-
b
an
d
d
etail
s
o
f
f
ir
s
t
lev
el
d
ep
en
d
in
g
o
n
th
e
f
ac
t
th
at
,
ap
p
r
o
x
im
atio
n
co
ef
f
icie
n
ts
r
ep
r
esen
t
th
e
b
asic
in
f
o
r
m
atio
n
o
f
th
e
o
r
ig
in
al
d
ata,
th
is
ca
n
r
ed
u
ce
th
e
f
ea
tu
r
es
o
f
th
e
o
r
ig
in
al
s
ig
n
al
b
y
d
is
ca
r
d
in
g
less
s
ig
n
if
ican
t
in
f
o
r
m
atio
n
[
1
6
]
.
C
h
en
-
Yu
L
ee
et
a
l.
,
th
ey
ar
e
s
tu
d
ied
th
e
p
er
f
o
r
m
an
ce
o
f
co
m
b
in
in
g
av
er
ag
e
p
o
o
lin
g
with
m
ax
p
o
o
lin
g
an
d
th
e
s
tr
ateg
y
o
f
tr
ee
s
tr
u
ctu
r
ed
f
u
s
io
n
o
f
f
ilter
s
.
T
h
e
b
asic
id
ea
o
f
th
is
wo
r
k
is
u
s
ed
lear
n
in
g
p
r
o
ce
s
s
o
f
m
ix
ed
r
ate
b
etwe
en
m
ax
an
d
av
er
ag
e
p
o
o
lin
g
m
eth
o
d
,
th
ey
ar
e
r
ef
er
r
ed
to
th
is
m
eth
o
d
as
m
ix
ed
m
eth
o
d
,
wh
ile
th
e
s
ec
o
n
d
u
s
ed
m
eth
o
d
in
th
is
wo
r
k
was
b
ased
o
n
g
ated
m
ask
,
wh
ich
is
u
s
ed
to
f
in
d
m
ix
o
f
m
ax
an
d
av
er
ag
e
p
o
o
lin
g
,
th
ey
ar
e
r
ef
er
ed
to
th
is
m
eth
o
d
as g
ated
m
ax
-
av
er
ag
e
m
eth
o
d
p
o
o
lin
g
[
1
7
]
.
Din
g
ju
n
Yu
et
al
.
th
ey
ar
e
p
r
o
p
o
s
e
d
a
m
eth
o
d
f
o
r
f
ea
tu
r
e
p
o
o
lin
g
b
ased
o
n
r
ep
lacin
g
d
eter
m
in
is
ts
with
s
to
ch
asti
c
o
p
er
atio
n
,
th
is
is
ac
co
m
p
lis
h
ed
b
y
ch
o
s
e
r
an
d
o
m
v
alu
e
to
s
elec
t
th
e
m
ax
o
r
av
er
ag
e
p
o
o
lin
g
m
eth
o
d
,
th
e
b
asic
b
en
ef
it
f
r
o
m
th
is
m
eth
o
d
is
to
av
o
id
o
v
er
f
itti
n
g
p
r
o
b
lem
.
T
h
ey
ar
e
ap
p
lied
m
ix
ed
p
o
o
lin
g
b
y
th
r
ee
d
if
f
er
en
t
ap
p
r
o
ac
h
es,
wh
ich
ar
e
ap
p
ly
p
o
o
lin
g
f
o
r
all
f
ea
tu
r
es in
a
lay
er
,
o
r
b
y
u
s
in
g
m
ix
ed
with
in
lay
er
,
o
r
b
y
u
s
in
g
m
ix
in
g
b
etwe
en
r
eg
io
n
s
with
in
lay
er
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
s
ar
e
ap
p
lied
o
n
d
if
f
er
en
t
ty
p
es
o
f
d
atab
ase
[
1
8
]
.
M.
D.
Z
eiler
an
d
R
o
b
Feg
u
s
ar
e
p
r
o
p
o
s
ed
to
s
elec
t
ac
tiv
atio
n
,
th
at
is
d
r
iv
en
f
r
o
m
a
m
u
ltid
im
en
s
io
n
al
d
is
tr
ib
u
tio
n
b
y
ac
tiv
atio
n
in
th
e
r
eg
io
n
o
f
p
o
o
lin
g
(
p
o
o
l
s
ize)
,
th
is
is
p
er
f
o
r
m
ed
b
y
f
ir
s
t
co
m
p
u
tin
g
th
e
p
r
o
b
ab
ilit
y
f
o
r
ea
ch
r
eg
io
n
,
th
en
th
is
p
r
o
b
ab
ili
ty
is
n
o
r
m
alize
d
,
th
en
s
am
p
le
f
r
o
m
th
ese
d
is
tr
ib
u
tio
n
b
ased
o
n
th
e
p
r
o
b
ab
ilit
y
is
s
elec
ted
to
b
e
th
e
p
o
o
lin
g
f
ea
tu
r
es,
d
if
f
er
en
t
m
eth
o
d
s
ar
e
u
s
ed
to
s
elec
t
th
ese
p
r
o
b
ab
ilit
ies,
wh
ich
m
ea
n
s
th
at,
th
e
s
elec
ted
elem
en
ts
f
o
r
p
o
o
lin
g
f
ea
tu
r
e
m
ay
n
o
t
b
e
th
e
lar
g
est
v
alu
e
[
1
9
]
.
T
ak
u
m
i
Ko
b
ay
ash
i
is
p
r
o
p
o
s
ed
f
ea
tu
r
e
p
o
o
lin
g
lay
er
b
ased
o
n
d
is
tr
ib
u
tio
n
o
f
p
r
o
b
ab
ilit
ies
o
v
e
r
ac
tiv
atio
n
,
th
is
is
p
er
f
o
r
m
ed
b
y
d
eter
m
in
e
th
e
s
tatis
tics
o
f
s
tan
d
ar
d
d
ev
iatio
n
an
d
m
ea
n
d
ep
en
d
in
g
o
n
d
is
tr
ib
u
tio
n
o
f
Gau
s
s
ian
f
u
n
ctio
n
,
th
e
b
as
ic
id
ea
o
f
th
is
wo
r
k
is
to
s
u
m
m
ar
ize
th
e
d
is
tr
ib
u
tio
n
o
f
Gau
s
s
ian
an
d
ag
g
r
eg
ate
th
e
ac
tiv
atio
n
in
to
two
b
asic
v
alu
es,
wh
ich
ar
e
s
tan
d
ar
d
d
ev
iatio
n
an
d
av
er
ag
e,
th
is
m
eth
o
d
is
ap
p
lied
later
to
s
to
ch
asti
c
p
o
o
lin
g
m
eth
o
d
[
1
3
]
.
3.
M
E
T
H
O
DO
L
O
G
Y
I
n
t
h
i
s
p
a
p
e
r
,
w
e
h
a
v
e
p
r
o
p
o
s
e
d
a
n
e
w
p
o
o
l
i
n
g
m
et
h
o
d
s
b
as
e
d
o
n
p
r
o
b
a
b
i
l
i
t
y
f
u
n
c
ti
o
n
,
Fi
g
u
r
e
1
d
e
s
c
r
i
b
e
s
t
h
e
b
l
o
c
k
d
i
a
g
r
a
m
o
f
t
h
e
p
o
o
l
i
n
g
l
a
y
e
r
,
t
h
e
b
a
s
i
c
c
o
m
p
o
n
e
n
t
o
f
t
h
i
s
l
a
y
e
r
is
f
e
a
t
u
r
e
c
o
m
p
u
t
a
t
i
o
n
,
w
h
i
c
h
i
s
e
x
t
r
a
c
t
e
d
d
e
p
e
n
d
i
n
g
o
n
a
l
g
o
r
i
t
h
m
(
1
)
b
y
c
a
l
c
u
l
a
ti
n
g
t
h
e
b
a
s
i
c
s
t
a
ti
s
t
i
cs
,
w
h
i
c
h
c
a
n
b
e
u
s
e
d
t
o
c
o
m
p
u
t
e
t
h
e
w
e
i
g
h
t
s
o
f
e
a
c
h
e
l
e
m
e
n
t
a
c
c
o
r
d
i
n
g
t
o
(
1
)
a
n
d
(
2
)
,
w
h
i
c
h
a
r
e
r
e
p
r
e
s
e
n
t
e
d
a
v
e
r
a
g
e
a
n
d
s
t
a
n
d
a
r
d
d
e
v
i
a
ti
o
n
r
e
s
p
e
ct
i
v
e
l
y
[
1
3
,
2
0
]
.
=
1
│
│
∑
(
)
(
1
)
=
1
│
│
∑
(
−
(
)
)
2
(
2
)
Fig
u
r
e
1
.
T
h
e
p
r
o
p
o
s
ed
p
o
o
lin
g
lay
er
b
l
o
ck
d
ia
g
r
am
T
h
e
s
ec
o
n
d
h
alf
o
f
Gau
s
s
ian
f
u
n
ctio
n
r
e
p
r
esen
t
s
th
e
s
tatis
ti
cs
b
etwe
en
m
ea
n
an
d
m
ax
im
u
m
v
alu
e,
wh
ich
r
ep
r
esen
t
s
th
e
m
o
s
t
im
p
o
r
tan
t
ch
a
r
ac
ter
is
tics
o
f
th
e
s
ig
n
al.
So
,
th
e
Ga
u
s
s
ian
is
r
ec
o
n
s
tr
u
cted
f
o
r
u
p
p
e
r
h
alf
o
f
its
f
u
n
ctio
n
as sh
o
wn
i
n
F
ig
u
r
e
2
,
th
en
m
o
s
t s
ig
n
if
ican
t statis
tic
s
ar
e
ca
lcu
lated
.
T
h
ese
s
tati
s
tics
w
ill b
e
u
s
ed
later
to
d
eter
m
in
e
th
e
f
ea
t
u
r
es
an
d
th
eir
weig
h
t
s
ac
co
r
d
i
n
g
to
th
e
s
ig
n
if
ican
t
o
f
ea
ch
o
f
th
em
.
T
h
e
s
elec
ted
f
ea
tu
r
es will b
e
d
eter
m
in
e
d
as sh
o
wn
in
(
3
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
H
a
lf
Ga
u
s
s
ia
n
-
b
a
s
ed
w
a
va
let
tr
a
n
s
fo
r
m
fo
r
p
o
o
lin
g
la
ye
r
fo
r
c
o
n
vo
lu
tio
n
.
.
.
(
A
q
ee
l M.
Ha
ma
d
A
lh
u
s
s
a
in
y
)
165
=
∑
(
4
=
0
µ
+
∗
2
)
∗
(
)
(
3
)
wh
er
e
(
)
r
ep
r
esen
t
th
e
weig
h
t
o
f
ea
ch
elem
en
t,
wh
ile
µ
,
ar
e
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
o
f
th
e
s
ig
n
al
r
esp
ec
tiv
ely
[
2
1
,
22]
.
Fig
u
r
e
2
.
Gau
s
s
ian
an
d
h
alf
G
au
s
s
ian
f
u
n
ctio
n
3
.
1
.
P
r
o
po
s
ed
a
lg
o
rit
hm
B
ased
o
n
th
e
s
tr
ateg
y
o
f
t
r
an
s
f
o
r
m
co
m
p
u
tatio
n
,
w
e
h
av
e
p
r
o
p
o
s
ed
th
r
ee
alg
o
r
ith
m
s
.
T
h
ese
m
eth
o
d
s
ar
e
h
alf
Gau
s
s
ian
t
r
an
s
f
o
r
m
1
(
HGT
1
)
,
h
alf
Gau
s
s
ian
t
r
an
s
f
o
r
m
2
(
HGT
2
)
a
n
d
h
al
f
Gau
s
s
ian
t
r
an
s
f
o
r
m
3
(
HGT
3
)
.
T
h
e
alg
o
r
ith
m
s
ar
e
d
if
f
er
en
ce
d
in
th
e
s
tr
ateg
y
o
f
tr
an
s
f
o
r
m
co
m
p
u
tatio
n
.
Als
o
,
th
e
s
tati
s
tic
s
ar
e
d
eter
m
in
ed
in
d
if
f
e
r
en
t w
in
d
o
w
s
ize.
T
h
e
d
etails ar
e
d
escr
ib
ed
in
n
ex
t sectio
n
s
.
3
.
1
.
1
.
H
G
T
1
T
h
is
alg
o
r
ith
m
is
u
s
ed
th
e
b
asic
f
ea
tu
r
es
o
f
th
e
s
ig
n
al
(
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
)
,
wh
ich
ar
e
d
eter
m
in
ed
as
s
h
o
wn
in
(
1
)
an
d
(
2
)
r
esp
ec
tiv
ely
.
At
f
ir
s
t,
th
e
s
ize
an
d
s
tr
id
e
f
o
r
ea
ch
p
o
o
l
s
ize
win
d
o
w
an
d
o
th
er
s
p
ar
am
eter
s
ar
e
in
itialized
,
th
en
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
ar
e
d
eter
m
in
ed
f
o
r
ea
ch
p
o
o
l
win
d
o
w,
th
en
th
e
u
p
p
er
h
alf
o
f
th
e
Gau
s
s
ian
d
is
tr
ib
u
tio
n
f
u
n
ctio
n
is
d
eter
m
in
ed
,
d
ep
en
d
in
g
o
n
th
ese
s
tatis
tics
,
b
asic
elem
en
ts
o
f
th
is
f
u
n
ctio
n
ar
e
co
m
p
u
ted
,
wh
ich
ar
e
(
µ
−
,
µ
−
1
2
∗
,
µ
,
µ
+
1
2
∗
µ
+
)
.
T
h
e
weig
h
ts
o
f
th
ese
elem
en
ts
ar
e
d
eter
m
in
ed
ac
co
r
d
in
g
to
Gau
s
s
ian
f
u
n
ctio
n
s
h
o
wn
in
(
4
)
,
th
ese
weig
h
ts
ar
e
m
u
ltip
lied
b
y
o
r
ig
in
al
s
ig
n
al
to
co
m
p
u
te
th
e
b
asics
f
ea
tu
r
es
(
p
o
o
led
s
ig
n
al)
.
T
h
e
d
etails
d
escr
ip
tio
n
o
f
th
is
m
eth
o
d
s
h
o
wn
in
Fig
u
r
e
3
,
wh
ich
s
h
o
ws alg
o
r
ith
m
HGT
1
.
(
)
=
1
√
2
∗
2
−
(
−
µ
)
2
2
2
(
4
)
3
.
1
.
2
.
H
G
T
2
I
n
th
is
alg
o
r
ith
m
,
th
e
weig
h
ts
ar
e
d
eter
m
in
ed
f
o
r
Ga
u
s
s
ian
f
u
n
ctio
n
,
th
en
f
o
r
ea
ch
p
o
o
l
s
ize
win
d
o
w,
th
e
m
ea
n
an
d
s
tan
d
ar
d
d
ev
iatio
n
ar
e
d
eter
m
in
ed
.
T
h
ese
s
tatis
tics
ar
e
u
s
ed
to
d
eter
m
in
e
th
e
b
asic
elem
en
ts
o
f
h
alf
Gau
s
s
ian
f
u
n
ctio
n
,
wh
ich
ar
e
(
(
µ
−
,
µ
−
1
2
∗
,
µ
,
µ
+
1
2
∗
µ
+
)
)
.
T
he
n
,
th
e
d
eter
m
in
ed
elem
en
ts
ar
e
m
u
ltip
lied
b
y
th
e
co
n
s
tan
t
weig
h
ts
,
wh
ich
ar
e
d
eter
m
in
ed
at
f
ir
s
t
s
tep
.
T
h
e
d
etails
d
escr
ip
tio
n
o
f
th
is
alg
o
r
ith
m
ar
e
s
h
o
wn
in
Fig
u
r
e
4
.
3
.
1
.
3
.
H
G
T
3
T
h
is
alg
o
r
ith
m
is
s
im
ilar
to
alg
o
r
ith
m
I
I
,
ac
ce
p
t
th
at,
it
is
d
ete
r
m
in
ed
t
h
e
m
ea
n
a
n
d
s
tan
d
ar
d
d
ev
iatio
n
f
o
r
e
n
tire
s
ig
n
al
i
n
s
tead
o
f
d
e
ter
m
in
ed
it
f
o
r
ea
ch
p
o
o
l
s
ize
.
At
f
ir
s
t,
th
ese
s
tatis
tics
ar
e
ca
lcu
lated
f
o
r
en
tir
e
s
ig
n
al,
th
en
th
e
b
asic
elem
en
ts
o
f
th
e
n
ew
Gau
s
s
ian
f
u
n
ctio
n
ar
e
d
ete
r
m
in
e,
wh
ic
h
ar
e
((
µ
−
,
µ
−
1
2
∗
,
µ
,
µ
+
1
2
∗
µ
+
)).
T
h
ese
v
alu
es a
r
e
u
s
ed
as in
p
u
ts
to
Gau
s
s
ian
f
u
n
cti
o
n
to
d
eter
m
in
e
th
e
f
ea
tu
r
es
,
wh
ich
a
r
e
m
u
ltip
lied
b
y
th
e
o
r
ig
i
n
al
s
ig
n
al
to
c
o
m
p
u
te
th
e
p
o
o
l
ed
s
ig
n
al
.
T
h
e
d
et
ails
ar
e
s
h
o
wn
in
Fig
u
r
e
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
1
6
3
-
17
2
166
Fig
u
r
e
3
.
HGT
1
al
g
o
r
ith
m
Fig
u
r
e
4
.
HGT
2
al
g
o
r
ith
m
Fig
u
r
e
5
.
HGT
3
a
l
g
o
r
ith
m
4.
E
XP
E
R
I
M
E
N
T
A
L
RE
SUL
T
S AN
D
D
I
SC
USS
I
O
N
T
h
e
p
r
o
p
o
s
ed
p
o
o
lin
g
m
eth
o
d
s
ar
e
u
s
ed
C
NN,
an
d
a
p
p
lied
o
n
d
i
f
f
er
en
t
ty
p
es
o
f
d
atab
ase
to
test
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
as
co
m
p
ar
e
d
with
o
th
er
m
eth
o
d
s
.
T
h
ese
d
atab
as
es
ar
e
MN
I
ST
an
d
C
I
FAR
1
0
,
wh
ich
ar
e
two
-
d
im
en
s
io
n
s
ig
n
al
(
im
a
g
e)
with
s
ize
(
2
8
*
2
8
)
a
n
d
(
3
2
*
3
2
)
r
es
p
ec
tiv
ely
.
T
h
e
o
th
e
r
d
atab
ase
was
MI
T
-
B
I
H
E
C
G
d
atab
ase,
wh
ich
is
one
-
d
im
e
n
s
io
n
s
ig
n
al.
T
h
e
ex
p
er
im
en
t
s
ar
e
ex
ec
u
ted
b
y
I
n
tel®co
r
e
™
i7
-
4
5
0
0
C
PU@
2
.
4
0
GHz
p
r
o
ce
s
s
o
r
,
with
8
GB
o
f
R
AM
,
64
-
b
it
win
d
o
ws
s
ev
en
o
p
er
atin
g
s
y
s
tem
,
o
n
Ma
tlab
(
2
0
1
9
a)
.
T
h
e
r
esu
lt
s
ar
e
co
m
p
ar
e
d
with
r
esu
lts
o
f
s
tan
d
ar
d
m
eth
o
d
s
.
4
.
1
.
M
NIS
T
da
t
a
ba
s
e
re
s
ult
s
T
h
is
d
atab
a
s
e
is
co
n
tain
ed
6
0
0
0
0
im
ag
e
o
f
g
r
ay
s
ca
le
im
ag
e
with
s
ize
(
2
8
*
2
8
)
,
it
is
d
iv
id
ed
in
to
(
5
0
0
0
0
)
im
ag
e,
wh
ich
ar
e
u
s
ed
f
o
r
tr
ain
in
g
,
wh
ile
th
e
r
em
ain
in
g
1
0
0
0
0
im
ag
es
ar
e
u
s
ed
f
o
r
test
th
e
p
r
o
p
o
s
e
d
m
o
d
el
[
2
3
]
.
T
h
e
C
NN
is
tr
ain
ed
with
in
itial
lear
n
in
g
r
ate
0
.
0
1
,
1
0
ep
o
ch
s
an
d
5
8
iter
atio
n
p
er
ep
o
ch
.
T
ab
le
1
d
escr
ib
es
th
e
r
esu
lts
as
co
m
p
ar
ed
with
s
tan
d
ar
d
m
ax
an
d
av
er
ag
e
p
o
o
lin
g
m
eth
o
d
s
,
it
is
clea
r
th
at
th
e
p
r
o
p
o
s
e
d
m
eth
o
d
ar
e
o
u
tp
er
f
o
r
m
ed
th
ese
m
eth
o
d
,
th
e
b
est ac
cu
r
ac
y
is
s
atis
f
ied
with
(
HGT
1
+a
v
er
ag
e
)
m
eth
o
d
,
w
h
ich
is
ac
h
iev
ed
ac
cu
r
ac
y
(
9
9
.
7
2
%)
v
er
s
es
(
9
9
.
4
8
%)
an
d
(
9
9
.
4
2
%)
f
o
r
m
ax
.
an
d
av
er
ag
e
p
o
o
lin
g
m
eth
o
d
s
r
esp
ec
tiv
ely
,
also
th
is
m
eth
o
d
is
ac
h
iev
ed
lo
west
FP
R
(
0
.
2
8
%)
co
m
p
ar
ed
with
(
0
.
3
4
%)
f
o
r
Ma
x
m
eth
o
d
as
s
h
o
wn
in
T
ab
le
2
,
wh
ich
s
h
o
ws th
e
d
if
f
er
en
t p
er
f
o
r
m
an
ce
m
etr
ics f
o
r
(
HGT
1
)
m
eth
o
d
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
H
a
lf
Ga
u
s
s
ia
n
-
b
a
s
ed
w
a
va
let
tr
a
n
s
fo
r
m
fo
r
p
o
o
lin
g
la
ye
r
fo
r
c
o
n
vo
lu
tio
n
.
.
.
(
A
q
ee
l M.
Ha
ma
d
A
lh
u
s
s
a
in
y
)
167
T
ab
le
1
.
R
esu
lts
o
f
(
HGT
1
)
m
eth
o
d
f
o
r
MN
I
ST
class
if
icatio
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T1
H
G
T1
+
M
a
x
H
G
T1
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
9
.
4
8
9
9
.
4
2
9
9
.
6
8
9
9
.
7
2
9
9
.
9
6
T
ab
le
2
. P
er
f
o
r
m
an
ce
m
etr
ics o
f
(
HGT
1
)
m
eth
o
d
s
f
o
r
h
an
d
w
r
ite
d
ig
it c
lass
if
icatio
n
M
e
t
h
o
d
H
G
T1
H
G
T1
+
M
a
x
H
G
T1
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
9
.
6
8
9
9
.
7
2
9
9
.
9
6
S
e
n
s
i
t
i
v
i
t
y
(
S
N
%)
9
9
.
6
6
9
9
.
6
8
9
9
.
7
2
F
a
l
se
p
o
si
t
i
v
e
R
a
t
e
F
P
R
(
%)
0
.
3
4
0
.
2
8
0
.
2
8
S
p
e
c
i
f
i
c
i
t
y
(
%)
9
9
.
6
6
9
9
.
6
8
9
9
.
7
2
ER
R
(
%)
0
.
3
2
0
.
2
8
0
.
0
4
T
h
e
r
esu
lts
o
f
s
ec
o
n
d
m
eth
o
d
is
d
escr
ib
ed
in
T
ab
le
3
,
wh
ich
g
iv
es
th
e
b
est
r
esu
lts
b
y
(
HG
T
2
+M
ax
)
,
an
d
o
th
e
r
m
etr
ics
o
f
p
er
f
o
r
m
a
n
ce
ar
e
ex
p
lain
ed
in
T
a
b
le
4
,
f
r
o
m
T
a
b
le
4
it
is
clea
r
th
at
(
HGT
2
+M
ax
)
g
iv
es
lo
west
FP
R
(
0
.
2
8
)
with
th
e
h
ig
h
est
ac
cu
r
ac
y
(
9
9
.
7
2
)
.
T
h
e
tab
l
es
ar
e
d
escr
ib
ed
th
e
im
p
r
o
v
e
m
en
ts
o
f
o
u
r
m
eth
o
d
s
in
ter
m
s
o
f
ac
cu
r
ac
y
,
s
en
s
itiv
ity
an
d
p
r
ec
is
io
n
with
m
in
im
u
m
f
alse p
o
s
itiv
e
r
ate
(
FP
R
)
.
T
h
e
ac
cu
r
ac
y
an
d
lo
s
s
tr
ain
in
g
p
r
o
g
r
ess
f
o
r
(
HGT
2
+
Ma
x
)
m
eth
o
d
ar
e
s
h
o
wn
in
F
ig
u
r
e
s
6
an
d
7
r
esp
ec
tiv
ely
,
i
t
is
c
lear
th
at
,
th
e
ac
cu
r
ac
y
is
r
ea
c
h
ed
to
m
o
r
e
th
an
9
8
.
5
with
less
th
an
2
ep
o
ch
s
,
th
is
is
d
u
e
to
e
x
tr
ac
tin
g
b
asic
f
ea
tu
r
es
o
f
th
e
im
ag
e
with
less
elim
in
atio
n
as
co
m
p
a
r
ed
with
m
ax
a
n
d
av
er
ag
e
p
o
o
lin
g
m
eth
o
d
s
,
also
th
e
lo
s
s
is
atten
u
ated
to
less
th
an
0
.
1
5
.
T
h
e
co
n
f
u
s
io
n
m
atr
ix
d
etails
f
o
r
th
is
m
e
th
o
d
is
d
escr
ib
ed
in
F
ig
u
r
e
8
,
wh
i
ch
is
d
escr
ib
ed
th
e
h
ig
h
m
atch
in
g
b
etwe
en
th
e
p
r
ed
icted
an
d
ac
t
u
al
v
alu
es,
s
in
ce
m
o
s
t
o
f
th
e
class
es
ar
e
m
atch
ed
p
er
f
ec
tly
.
T
ab
le
5
s
h
o
ws
th
e
r
esu
lt
o
f
th
i
r
d
m
eth
o
d
(
HGT
3
)
,
wh
ich
is
ac
h
iev
ed
less
r
esu
lts
as
co
m
p
ar
ed
with
HGT
1
an
d
HGT
2
m
eth
o
d
s
,
th
e
d
etail
d
escr
ip
tio
n
o
f
th
ese
m
eth
o
d
s
f
o
r
a
ll p
er
f
o
r
m
a
n
ce
m
etr
ics ar
e
d
escr
ib
ed
in
T
a
b
le
6
.
T
ab
le
3
.
R
esu
lts
o
f
(
HGT
2
)
m
eth
o
d
f
o
r
MN
I
ST
class
if
icatio
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T2
H
G
T2
+
M
a
x
H
G
T2
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
9
.
4
8
9
9
.
4
2
9
9
.
5
2
9
9
.
7
2
9
9
.
5
2
T
ab
le
4
. P
er
f
o
r
m
an
ce
m
etr
ics o
f
(
HGT
2
)
m
eth
o
d
s
f
o
r
h
an
d
w
r
ite
d
ig
it c
lass
if
icatio
n
M
e
t
h
o
d
H
G
T2
H
G
T2
+
M
a
x
H
G
T2
+
A
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
9
.
5
2
9
9
.
7
2
9
9
.
5
2
S
e
n
s
i
t
i
v
i
t
y
(
S
N
%)
9
9
.
5
2
9
9
.
7
2
9
9
.
5
8
F
a
l
se
Er
r
o
r
R
a
t
e
F
ER
(
%)
0
.
4
8
0
.
2
8
0
.
4
2
S
p
e
c
i
f
i
c
i
t
y
(
%)
9
9
.
5
2
9
9
.
7
2
9
9
.
5
6
ER
R
(
%)
0
.
4
8
0
.
2
8
0
.
4
8
Fig
u
r
e
6
.
A
cc
u
r
ac
y
tr
ain
i
n
g
p
r
o
g
r
ess
f
o
r
(
HGT
2
+M
ax
)
m
et
h
o
d
Fig
u
r
e
7
.
L
o
s
s
tr
ain
in
g
p
r
o
g
r
e
s
s
f
o
r
(
HGT
2
+M
ax
)
m
et
h
o
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
1
6
3
-
17
2
168
T
ab
le
5
.
R
esu
lts
o
f
(
HGT
3
)
m
eth
o
d
f
o
r
MN
I
ST
class
if
icatio
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T3
HG
T3
+
M
a
x
H
G
T3
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
9
.
4
8
9
9
.
4
2
9
9
.
0
4
9
9
.
5
2
9
9
.
9
6
T
ab
le
6
.
Per
f
o
r
m
an
ce
m
ea
s
u
r
e
s
f
o
r
HGT
3
f
o
r
h
a
n
d
wr
ite
d
ig
i
t c
lass
if
icatio
n
M
e
t
h
o
d
H
G
T3
H
G
T3
+
M
a
x
H
G
T3
+
A
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
9
.
0
4
9
9
.
5
2
9
9
.
9
6
S
e
n
s
i
t
i
v
i
t
y
(
S
N
%)
9
9
.
3
0
9
9
.
5
2
9
9
.
9
6
F
a
l
se
Er
r
o
r
R
a
t
e
F
ER
(
%)
0
.
7
0
.
4
8
0
.
0
4
S
p
e
c
i
f
i
c
i
t
y
(
%)
9
9
.
1
2
9
9
.
5
2
9
9
.
9
6
ER
R
(
%)
0
.
9
6
0
.
4
8
0
.
0
4
Fig
u
r
e
8
.
c
o
n
f
u
s
io
n
m
atr
ix
o
f
(
HGT
3
+M
ax
)
m
eth
o
d
4
.
2
.
Resul
t
s
CIFAR
1
0
da
t
a
s
et
T
h
is
d
ataset
is
co
n
s
tr
u
cted
f
r
o
m
6
0
0
0
0
im
a
g
e,
ea
c
h
im
ag
e
with
s
ize
(
3
2
*
3
2
)
R
GB
co
lo
r
im
ag
e,
th
e
m
o
d
el
is
tr
ain
ed
o
n
(
5
0
0
0
0
)
,
wh
ile
th
e
test
d
ataset
was
1
0
0
0
0
im
ag
es
[
2
4
]
.
I
n
th
is
ex
p
er
im
en
t,
th
e
s
am
e
p
ar
am
eter
s
ar
e
u
s
ed
f
o
r
all
p
o
o
lin
g
m
eth
o
d
s
(
th
e
p
r
o
p
o
s
ed
a
n
d
s
tan
d
ar
d
)
,
w
h
ich
ar
e
1
0
ep
o
ch
,
1
2
8
b
atch
s
ize
with
0
.
0
1
lear
n
i
n
g
r
ate
.
T
h
e
r
e
s
u
lts
o
f
HGT
1
m
eth
o
d
ar
e
d
es
cr
ib
ed
in
T
ab
le
7
,
it
clea
r
th
at
,
o
u
r
m
eth
o
d
(
HGT
)
g
iv
es
th
e
b
est
r
esu
lts
,
b
ec
a
u
s
e
co
m
b
i
n
in
g
th
is
m
eth
o
d
with
m
ax
an
d
a
v
er
ag
e
ca
n
elim
in
at
e
s
o
m
e
s
ig
n
if
ica
n
t
in
f
o
r
m
atio
n
f
r
o
m
th
e
im
ag
e,
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
eth
o
d
s
ar
e
s
h
o
wn
i
n
T
ab
l
e
8
,
th
e
lo
west FPE
is
s
at
is
f
ied
in
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
(
HGT
1
)
,
wh
ich
is
ac
h
iev
ed
(
2
6
.
3
%).
T
h
e
co
n
f
u
s
io
n
m
atr
ix
o
f
th
is
m
eth
o
d
is
s
h
o
wn
in
F
ig
u
r
e
9
,
wh
ic
h
s
h
o
ws g
o
o
d
m
a
tch
in
g
b
etwe
en
p
r
ed
icte
d
a
n
d
ac
tu
al
class
es.
T
h
e
p
r
o
g
r
ess
o
f
th
e
ac
cu
r
ac
y
an
d
lo
s
s
tr
ain
in
g
ar
e
s
h
o
wn
in
F
ig
u
r
e
s
1
0
an
d
1
1
r
esp
ec
tiv
ely
,
th
e
ac
cu
r
ac
y
is
r
ea
ch
ed
to
(
6
0
%)
in
less
th
an
2
ep
o
ch
s
,
th
en
in
cr
ea
s
ed
g
r
ad
u
ally
,
wh
ile
th
e
lo
s
s
is
atten
u
ated
to
less
th
an
1
in
2
ep
o
ch
s
an
d
th
en
,
it
is
d
ec
r
ea
s
ed
s
lo
wly
.
T
h
e
r
esu
lts
o
f
HGT
2
ar
e
p
r
e
s
en
ted
in
T
ab
le
s
9
an
d
1
0
r
esp
ec
tiv
ely
,
th
er
e
is
s
m
all
im
p
r
o
v
em
en
t
co
m
p
ar
ed
with
m
ax
an
d
av
er
ag
e
p
o
o
lin
g
m
eth
o
d
s
,
b
ec
au
s
e
th
is
m
eth
o
d
is
d
ep
en
d
ed
o
n
f
ea
tu
r
e
o
f
th
e
im
ag
e
in
s
tead
o
f
th
e
im
ag
e
its
elf
f
o
r
ex
tr
ac
tio
n
th
e
p
o
o
led
s
ig
n
al.
T
ab
les
1
1
an
d
1
2
r
ep
r
es
en
t r
esu
lts
o
f
H
GT
3
m
eth
o
d
,
wh
ich
is
less
in
m
o
s
t
p
er
f
o
r
m
an
ce
m
etr
ics
(
ac
c
7
2
.
4
2
%
)
an
d
(
FP
E
2
7
.
5
8
%),
th
is
d
u
e
to
th
at,
th
is
m
eth
o
d
is
d
ep
en
d
ed
o
n
th
e
s
tatis
tics
o
f
en
tire
s
ig
n
al
in
s
tead
o
f
ea
ch
p
o
o
l
s
ize
win
d
o
w
f
r
o
m
th
e
s
ig
n
al,
wh
ich
n
o
t
g
iv
es
th
e
m
eth
o
d
h
ig
h
d
y
n
am
ic
in
d
ea
lin
g
with
th
e
s
ig
n
al,
an
d
th
is
is
h
ap
p
en
ed
in
th
e
f
ir
s
t m
eth
o
d
(
HGW1
)
.
T
ab
le
7
.
R
esu
lts
o
f
d
if
f
e
r
en
t p
r
o
p
o
s
ed
p
o
o
lin
g
m
eth
o
d
f
o
r
C
I
FAR
1
0
class
if
icatio
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T1
H
G
T1
+
M
a
x
H
G
T1
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
7
2
.
5
9
7
2
.
4
1
7
3
.
6
7
7
2
.
2
7
2
.
7
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
H
a
lf
Ga
u
s
s
ia
n
-
b
a
s
ed
w
a
va
let
tr
a
n
s
fo
r
m
fo
r
p
o
o
lin
g
la
ye
r
fo
r
c
o
n
vo
lu
tio
n
.
.
.
(
A
q
ee
l M.
Ha
ma
d
A
lh
u
s
s
a
in
y
)
169
T
ab
le
8
.
Per
f
o
r
m
an
ce
m
ea
s
u
r
e
s
o
f
HGT
m
eth
o
d
f
o
r
C
I
FAR
1
0
d
atab
ase.
M
e
t
h
o
d
H
G
T1
H
G
T1
+
M
a
x
H
G
T1
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
7
3
.
6
7
7
2
.
2
7
2
.
7
S
e
n
si
t
i
v
i
t
y
(
S
N
%)
7
3
.
6
7
7
2
.
2
7
7
2
.
7
F
a
l
se
p
o
si
t
i
v
e
R
a
t
e
F
P
R
(
%)
2
6
.
3
2
6
.
6
2
7
.
2
7
S
p
e
c
i
f
i
c
i
t
y
(
%)
7
3
.
3
7
2
.
2
9
7
2
.
7
3
ER
R
(
%)
2
6
.
3
3
2
7
.
8
2
7
.
3
Fig
u
r
e
9
.
C
o
n
f
u
s
io
n
m
atr
i
x
o
f
HGT
1
m
eth
o
d
Fig
u
r
e
1
0
.
ac
cu
r
ac
y
p
r
o
g
r
ess
f
o
r
tr
ain
in
g
HGT
1
m
eth
o
d
Fig
u
r
e
1
1
.
L
o
s
s
p
r
o
g
r
ess
f
o
r
tr
ain
in
g
HGT
1
m
eth
o
d
T
ab
le
9
.
R
esu
lts
o
f
HGT
2
m
et
h
o
d
f
o
r
C
I
FAR
1
0
class
if
icatio
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T2
H
G
T2
+
M
a
x
H
G
T2
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
7
2
.
5
9
7
2
.
4
1
7
2
.
2
1
7
2
.
4
2
7
2
.
7
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
1
6
3
-
17
2
170
T
ab
le
1
0
.
Per
f
o
r
m
a
n
ce
m
ea
s
u
r
es o
f
HGT
2
m
eth
o
d
f
o
r
C
I
FAR
1
0
d
atab
ase
M
e
t
h
o
d
H
G
T2
H
G
T2
+
M
a
x
H
G
T2
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
7
2
.
2
1
7
2
.
4
2
7
2
.
7
S
e
n
si
t
i
v
i
t
y
(
S
N
%)
7
2
.
2
3
7
2
.
4
2
7
2
.
7
F
a
l
se
p
o
si
t
i
v
e
R
a
t
e
F
P
R
(
%)
2
7
.
7
7
2
7
.
5
8
2
7
.
3
S
p
e
c
i
f
i
c
i
t
y
(
%)
7
2
.
2
1
7
2
.
3
8
7
2
.
6
8
ER
R
(
%)
2
7
.
7
9
2
7
.
5
8
2
7
.
3
T
ab
le
1
1
.
R
esu
lts
o
f
HGT
3
m
e
th
o
d
f
o
r
C
I
FAR
1
0
class
if
icati
o
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T3
H
G
T3
+
M
a
x
H
G
T3
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
7
2
.
5
9
7
2
.
4
1
7
2
.
4
1
7
2
.
3
3
7
2
.
3
8
Ta
b
le
1
2
.
Per
f
o
r
m
a
n
ce
m
ea
s
u
r
es o
f
HGT
3
m
eth
o
d
f
o
r
C
I
FAR
1
0
d
atab
ase
M
e
t
h
o
d
H
G
T3
H
G
T3
+
M
a
x
H
G
T3
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
7
2
.
4
1
7
2
.
3
3
7
2
.
3
8
S
e
n
si
t
i
v
i
t
y
(
S
N
%)
7
2
.
5
1
7
2
.
3
6
7
2
.
3
9
F
a
l
se
p
o
si
t
i
v
e
R
a
t
e
F
P
R
(
%)
2
7
.
4
9
2
7
.
6
4
2
7
.
6
2
S
p
e
c
i
f
i
c
i
t
y
(
%)
7
2
.
4
0
7
2
.
3
0
7
2
.
3
5
ER
R
(
%)
2
8
.
5
9
2
7
.
6
7
2
7
.
6
2
4
.
3
.
Resul
t
o
f
E
CG
s
ig
na
l
T
h
i
s
d
at
a
s
e
t
i
s
c
o
n
t
a
i
n
e
d
d
a
t
a
w
i
t
h
s
i
ze
(
1
0
9
4
4
6
*
1
8
8
)
,
w
h
i
c
h
r
e
p
r
e
s
e
n
t
(
1
0
9
4
4
6
)
s
i
g
n
a
l
,
e
a
ch
o
n
e
w
it
h
o
n
e
d
i
m
e
n
s
i
o
n
,
w
it
h
1
8
8
s
a
m
p
le
s
,
t
h
e
t
r
a
i
n
i
n
g
s
et
w
it
h
s
i
z
e
(
8
7
5
5
4
)
,
w
h
i
l
e
t
es
t
s
i
z
e
is
(
2
1
8
9
2
)
[
2
5
,
26
].
T
h
e
m
o
d
e
l
i
s
t
r
ai
n
e
d
w
i
t
h
s
a
m
e
p
a
r
a
m
et
e
r
s
f
o
r
a
l
l
m
e
t
h
o
d
s
o
f
p
o
o
l
i
n
g
l
ay
e
r
s
,
w
h
i
c
h
a
r
e
1
0
e
p
o
c
h
s
,
b
atc
h
s
i
z
e
1
2
8
a
n
d
0
.
0
1
l
e
a
r
n
i
n
g
r
a
t
e
.
T
a
b
l
e
1
3
s
h
o
ws
th
e
r
e
s
u
l
ts
o
f
HG
T
1
c
o
m
p
a
r
e
d
w
i
t
h
o
t
h
e
r
m
o
s
t
c
o
m
m
o
n
m
e
t
h
o
d
s
,
t
h
e
b
e
s
t
r
es
u
l
ts
a
r
e
a
c
h
i
e
v
e
d
w
i
t
h
(
HG
T1
)
m
e
th
o
d
(
a
c
c
u
r
a
c
y
9
4
.
5
1
%
,
)
w
i
t
h
l
o
w
e
s
t
F
P
E
(
4
.
4
4
)
,
w
h
i
l
e
c
o
m
b
i
n
in
g
t
h
i
s
m
e
t
h
o
d
w
i
t
h
m
a
x
a
n
d
a
v
e
r
a
g
e
a
r
e
a
c
h
i
e
v
e
d
l
e
s
s
a
c
c
u
r
a
c
y
,
t
h
is
is
h
a
p
p
e
n
e
d
d
u
e
t
o
t
h
a
t
E
C
G
s
i
g
n
a
l
is
o
s
ci
l
l
a
te
d
s
i
g
n
a
l
,
M
a
x
o
r
a
v
e
r
a
g
e
c
a
n
p
r
o
d
u
c
e
e
l
i
m
i
n
a
t
i
o
n
o
f
m
o
r
e
s
i
g
n
i
f
i
c
a
n
t
i
n
fo
r
m
a
t
i
o
n
,
w
h
i
c
h
m
a
y
r
e
d
u
c
e
t
h
e
o
v
e
r
a
l
l
a
c
c
u
r
a
c
y
.
T
h
e
r
e
s
u
l
ts
o
f
H
G
T
2
is
s
h
o
w
n
i
n
T
a
b
l
e
1
4
,
i
t
g
i
v
es
h
i
g
h
e
s
t
a
c
c
u
r
ac
y
(
9
4
.
5
1
%
)
.
T
h
e
r
esu
lts
o
f
th
ir
d
p
r
o
p
o
s
ed
m
eth
o
d
(
HGT
3
)
ar
e
s
h
o
wn
in
T
ab
le
1
5
.
I
t
is
clea
r
th
at
,
th
is
m
eth
o
d
g
iv
es
th
e
lo
west
r
esu
lt
as
co
m
pa
r
ed
with
o
th
er
p
r
o
p
o
s
ed
m
eth
o
d
s
(
HGT
1
an
d
HGT
2
)
,
wh
ic
h
s
atis
f
ied
(
Acc
=9
2
.
3
5
%),
th
e
r
esu
lts
ar
e
d
r
o
p
p
ed
b
ec
au
s
e
t
h
is
m
eth
o
d
i
s
d
ep
en
d
ed
o
n
s
tatis
tics
o
f
th
e
en
tire
s
ig
n
al
in
s
tead
o
f
ev
er
y
p
o
o
l
s
ize,
wh
ich
is
v
er
y
d
if
f
e
r
en
t
b
ec
a
u
s
e
E
C
G
s
i
g
n
al
h
av
e
h
ig
h
d
if
f
er
e
n
ce
s
in
th
eir
s
am
p
les.
T
h
e
d
etail
p
er
f
o
r
m
an
ce
m
etr
ics
f
o
r
o
u
r
m
eth
o
d
s
ar
e
d
escr
ib
e
d
in
T
ab
le
1
6
,
wh
ic
h
is
co
n
clu
d
e
d
t
h
at,
th
e
b
est
r
esu
lts
ar
e
o
b
tain
e
d
with
(
HGT
2
)
m
et
h
o
d
,
w
h
ich
is
ac
h
iev
e
d
ac
cu
r
a
cy
(
9
4
.
9
4
%)
with
E
R
R
(
5
.
0
9
%
)
,
an
d
FP
R
(
4
.
4
4
%)
th
is
im
p
ro
v
em
e
n
t
is
ac
h
iev
ed
b
ec
au
s
e
th
e
p
o
o
led
s
ig
n
al
is
d
ep
en
d
ed
o
n
ex
tr
ac
tio
n
o
f
th
e
m
o
s
t
s
ig
n
if
ican
t
f
ea
tu
r
e
o
f
th
e
s
ig
n
al.
T
h
e
p
r
o
g
r
ess
o
f
t
r
ain
in
g
f
o
r
ac
cu
r
a
cy
an
d
lo
s
s
f
o
r
(
HGT
2
)
m
et
h
o
d
a
r
e
s
h
o
wn
in
F
ig
u
r
e
s
1
2
a
n
d
1
3
r
esp
ec
tiv
ely
,
af
ter
o
n
e
ep
o
c
h
,
th
e
tr
ai
ni
n
g
ac
cu
r
ac
y
is
r
ea
ch
e
d
to
ap
p
r
o
x
im
ately
(
9
0
%)
an
d
th
e
lo
s
s
is
d
ec
r
ea
s
ed
to
less
th
an
(
0
.
4
)
.
T
ab
le
1
3
.
R
esu
lts
o
f
d
if
f
er
en
t
p
r
o
p
o
s
ed
p
o
o
lin
g
m
et
h
o
d
f
o
r
C
I
FAR
1
0
clas
s
if
icatio
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T1
H
G
T1
+
M
a
x
H
G
T1
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
3
.
2
7
9
4
.
0
1
9
4
.
5
1
9
3
.
9
2
9
3
.
9
7
T
ab
le
1
4
.
R
esu
lts
o
f
HGT
2
m
eth
o
d
f
o
r
C
I
FAR
1
0
class
if
icat
io
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T2
H
G
T2
+
M
a
x
H
G
T2
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
3
.
2
7
9
4
.
0
1
9
4
.
9
4
9
4
.
5
4
9
4
.
3
0
T
ab
le
1
5
.
R
esu
lts
o
f
HGT
2
m
eth
o
d
f
o
r
C
I
FAR
1
0
class
if
icat
io
n
M
e
t
h
o
d
M
a
x
a
v
e
r
a
g
e
H
G
T3
H
G
T3
+
M
a
x
H
G
T3
+
a
v
e
r
a
g
e
A
c
c
u
r
a
c
y
(
%)
9
3
.
2
7
9
4
.
0
1
9
2
.
3
5
9
2
.
1
3
9
2
.
2
4
T
ab
le
1
6
.
P
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
M
e
t
h
o
d
H
G
T1
H
G
T2
H
G
T3
A
c
c
u
r
a
c
y
(
%)
9
4
.
5
1
9
4
.
9
4
9
2
.
3
5
S
e
n
si
t
i
v
i
t
y
(
S
N
%)
9
4
.
2
1
9
4
.
5
6
9
1
.
8
5
F
a
l
se
p
o
si
t
i
v
e
R
a
t
e
F
P
R
(
%)
5
.
7
9
4
.
4
4
8
.
1
5
S
p
e
c
i
f
i
c
i
t
y
(
%)
9
5
.
3
0
5
9
4
.
5
6
9
1
.
5
5
ER
R
(
%)
4
.
4
9
5
.
0
9
7
.
6
5
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
H
a
lf
Ga
u
s
s
ia
n
-
b
a
s
ed
w
a
va
let
tr
a
n
s
fo
r
m
fo
r
p
o
o
lin
g
la
ye
r
fo
r
c
o
n
vo
lu
tio
n
.
.
.
(
A
q
ee
l M.
Ha
ma
d
A
lh
u
s
s
a
in
y
)
171
Fig
u
r
e
1
2
.
A
cc
u
r
ac
y
p
r
o
g
r
ess
f
o
r
HGT
2
m
et
h
o
d
Fig
u
r
e
1
3
.
L
o
s
s
p
r
o
g
r
ess
f
o
r
H
GT
2
m
eth
o
d
5.
CO
NCLU
SI
O
N
T
h
e
m
o
s
t im
p
o
r
tan
t la
y
er
in
C
NNs is
co
n
v
o
lu
tio
n
al
la
y
er
,
b
u
t a
cc
o
r
d
in
g
to
th
e
s
ize
o
f
in
p
u
t
s
,
n
u
m
b
e
r
o
f
u
s
ed
f
ilter
s
an
d
k
er
n
el
s
ize
o
f
ea
ch
f
ilter
in
th
is
lay
er
,
t
h
e
o
u
tp
u
t
o
f
th
is
lay
e
r
will b
e
t
o
o
m
u
ch
an
d
th
is
m
ay
r
ed
u
ce
th
e
ef
f
icien
cy
o
f
t
h
e
n
e
two
r
k
a
n
d
in
c
r
ea
s
e
its
co
m
p
lex
ity
.
So
,
d
if
f
er
e
n
t
s
tu
d
ies
an
d
r
esear
ch
h
av
e
b
ee
n
p
er
f
o
r
m
ed
to
r
ed
u
ce
th
is
p
r
o
b
lem
.
I
n
th
is
p
a
p
er
,
t
h
r
ee
m
et
h
o
d
s
h
av
e
b
ee
n
p
r
o
p
o
s
ed
b
ased
o
n
th
e
p
r
in
ci
p
le
o
f
Gau
s
s
ian
f
u
n
ctio
n
,
b
y
u
s
in
g
th
e
f
ac
t
th
at
th
e
s
ec
o
n
d
h
alf
o
f
G
au
s
s
ian
f
u
n
ct
io
n
r
ep
r
esen
ts
th
e
s
tatis
tics
b
etwe
e
n
m
ea
n
an
d
m
ax
im
u
m
v
alu
e,
wh
ich
r
ep
r
esen
ts
th
e
m
o
s
t
im
p
o
r
tan
t
ch
ar
ac
te
r
is
tics
o
f
th
e
s
ig
n
al.
So
,
th
e
m
ain
co
n
ce
n
tr
atio
n
o
f
i
n
f
o
r
m
atio
n
is
f
r
o
m
m
ea
n
to
m
ax
,
a
n
d
d
ep
en
d
in
g
o
n
th
is
f
ac
t,
th
e
Gau
s
s
ian
is
r
ec
o
n
s
tr
u
cte
d
f
o
r
u
p
p
e
r
h
alf
o
f
its
f
u
n
ctio
n
,
an
d
d
ep
en
d
in
g
o
n
th
e
m
o
s
t
s
i
g
n
if
ican
t
f
ea
tu
r
es.
De
p
en
d
in
g
o
n
th
e
n
ew
f
u
n
ctio
n
(
HG)
,
th
e
b
asic
s
tatis
tics
v
alu
es
ar
e
ca
lcu
lated
to
b
e
weig
h
t
s
f
o
r
th
e
o
r
ig
in
al
s
ig
n
al
to
ca
l
cu
late
th
e
f
ea
tu
r
es
(
s
elec
tin
g
f
ea
tu
r
e
)
.
T
h
r
ee
m
e
th
o
d
a
r
e
p
r
o
p
o
s
ed
HGT
1
,
wh
ich
is
u
s
ed
th
e
v
alu
es
o
f
b
a
s
ic
s
tatis
t
ics
af
ter
n
o
r
m
alize
d
it
as
weig
h
ts
to
b
e
m
u
ltip
lied
b
y
o
r
ig
in
al
s
ig
n
a
l,
th
e
HGT
2
is
u
s
ed
th
e
d
eter
m
in
ed
s
tatis
tics
as
f
ea
tu
r
es
o
f
th
e
o
r
ig
i
n
al
s
ig
n
al
an
d
m
u
ltip
ly
it
with
co
n
s
tan
t
weig
h
ts
b
ased
o
n
h
alf
Gau
s
s
ian
,
wh
ile
t
h
e
HGT
3
is
wo
r
k
ed
in
s
im
ilar
way
to
(
H
GT
1
)
ex
ce
p
t
th
at,
it
is
d
ep
en
d
ed
o
n
en
tir
e
s
ig
n
al
in
s
tead
o
f
ev
er
y
p
o
o
l
s
ize
f
o
r
ca
lcu
latio
n
th
e
b
asic
s
tatis
tics
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
s
ar
e
ap
p
lied
to
th
r
ee
ty
p
es
o
f
d
atasets
,
wh
ich
ar
e
(
MN
I
ST
an
d
C
I
FAR
1
0
)
,
wh
ich
ar
e
two
-
d
im
en
s
io
n
s
ig
n
al
a
n
d
MI
T
-
B
I
H
E
C
G
d
ataset,
wh
ich
is
one
-
d
im
en
s
io
n
s
ig
n
al.
Fo
r
MN
I
ST
d
ataset,
th
e
b
est
r
esu
lts
ar
e
ac
h
iev
ed
with
HGT
1
+a
v
er
ag
e
,
(
ac
cu
r
ac
y
9
9
.
9
6
%
an
d
FP
R
0
.
2
8
%),
wh
ile
f
o
r
C
I
FAR
1
0
d
ataset,
th
e
b
est
r
esu
lt
ar
e
s
at
is
f
ied
wi
th
HGT
1
m
eth
o
d
(
ac
cu
r
ac
y
7
3
.
6
7
%
an
d
FP
R
2
6
.
3
%).
Fo
r
E
C
G
d
ataset,
th
e
HGT
1
g
iv
es
th
e
g
o
o
d
r
esu
l
ts
(
ac
c=
9
4
.
5
1
%),
(
s
en
9
4
.
2
1
%)
an
d
(
FP
E
5
.
7
9
%),
an
d
HGT
2
g
iv
es
ap
p
r
o
x
im
atel
y
b
etter
r
esu
lts
,
wh
ich
ar
e
(
ac
c
=9
5
.
9
1
%),
(
s
en
.
9
4
.
5
6
%)
an
d
(
FP
E
4
.
4
4
%),
wh
ile
th
e
HGT
3
is
s
ati
s
f
ied
th
e
lo
w
est
r
esu
lts
(
ac
c=
9
2
.
3
5
%),
(
s
en
.
9
1
.
8
5
%)
an
d
(
FP
E
8
.
1
5
%),
t
h
e
r
esu
lt
is
d
r
o
p
p
e
d
b
ec
au
s
e
th
is
m
eth
o
d
is
d
ep
en
d
ed
o
n
th
e
s
tatis
tics
o
f
o
v
er
all
s
ig
n
al
in
s
tead
o
f
s
tatis
tic
s
o
f
ev
er
y
p
o
o
l
s
ize
as
in
HGT
1
,
wh
ich
is
v
er
y
d
if
f
e
r
en
t
b
ec
au
s
e
E
C
G
s
ig
n
al
h
av
e
h
i
g
h
d
if
f
er
e
n
ce
s
in
th
eir
s
am
p
les.
T
h
e
ex
p
e
r
im
en
tal
r
esu
lt sh
o
w
th
at,
o
u
r
m
eth
o
d
s
ar
e
ac
h
iev
e
d
g
o
o
d
im
p
r
o
v
em
e
n
ts
,
wh
ich
is
p
e
r
f
o
r
m
ed
o
r
o
u
t
p
er
f
o
r
m
ed
s
tan
d
a
r
d
p
o
o
lin
g
m
eth
o
d
s
s
u
ch
as m
ax
p
o
o
lin
g
an
d
a
v
er
ag
e
p
o
o
lin
g
,
an
d
ca
n
b
e
u
s
ed
in
class
if
ic
ati
o
n
p
r
o
b
lem
.
RE
F
E
R
E
NC
E
S
[1
]
Ag
o
stin
e
ll
i
L.
,
e
t
a
l.
,
“
Lea
rn
i
n
g
a
c
ti
v
a
ti
o
n
fu
n
c
ti
o
n
s t
o
imp
r
o
v
e
d
e
e
p
n
e
u
ra
l
n
e
two
r
k
s
,”
.
ICL
R
,
2
0
1
5
.
[2
]
Kriz
h
e
v
sk
y
,
S
.
,
e
t
a
l
.
,
“
Im
a
g
e
Ne
t
c
las
sifica
ti
o
n
with
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
,
”
Ad
v
a
n
c
e
s
in
n
e
u
ra
l
in
fo
rm
a
ti
o
n
p
ro
c
e
ss
in
g
sy
ste
ms
,
v
o
l.
2
5
,
n
o
.
2
,
Ja
n
u
a
ry
2
0
1
2
.
[3
]
M
a
ll
a
t
,
S.
,
“
A
th
e
o
r
y
f
o
r
m
u
lt
ires
o
lu
ti
o
n
sig
n
a
l
d
e
c
o
m
p
o
si
ti
o
n
:
th
e
wa
v
e
let
re
p
re
se
n
tati
o
n
,”
IEE
E
tr
a
n
sa
c
ti
o
n
s
o
n
p
a
tt
e
rn
a
n
a
lys
is a
n
d
ma
c
h
i
n
e
i
n
telli
g
e
n
c
e
,
v
o
l
.
1
1
,
n
o
.
7
,
p
p
.
6
7
4
-
6
9
3
,
J
u
ly
1
9
8
9
.
[4
]
Y.
Bo
u
re
a
u
,
P
.
,
e
t
a
l.
,
“
A
th
e
o
r
e
ti
c
a
l
a
n
a
ly
sis
o
f
fe
a
tu
re
p
o
o
li
n
g
in
v
isu
a
l
re
c
o
g
n
it
io
n
,
”
Pro
c
e
e
d
i
n
g
o
f
th
e
2
7
th
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
L
e
a
rn
i
n
g
(IC
M
L
-
10)
,
Ju
n
e
2
0
1
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6
9
3
0
T
E
L
KOM
NI
KA
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l Co
n
tr
o
l
,
Vo
l.
19
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
1
:
1
6
3
-
17
2
172
[5
]
K.
He
,
Zh
a
n
g
,
Re
n
,
a
n
d
J.
S
u
n
,
“
De
e
p
re
sid
u
a
l
lea
rn
in
g
f
o
r
ima
g
e
re
c
o
g
n
it
i
o
n
,”
IEE
E
Co
n
fer
e
n
c
e
o
n
Co
m
p
u
ter
Vi
sio
n
a
n
d
P
a
tt
e
rn
Rec
o
g
n
it
i
o
n
,
p
p
.
7
7
0
–
7
7
8
,
Ju
n
e
2
0
1
6
.
[6
]
Trav
is
Wi
ll
iam
s
a
n
d
Ro
b
e
rt
Li
,
“
Ad
v
a
n
c
e
d
ima
g
e
c
las
sifica
ti
o
n
u
s
in
g
wa
v
e
lets
a
n
d
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
,
”
1
5
t
h
IEE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
L
e
a
rn
i
n
g
a
n
d
A
p
p
li
c
a
ti
o
n
s
,
p
p
.
2
3
3
–
2
3
9
,
De
c
e
m
b
e
r
2
0
1
6
.
[7
]
J.
Kim
,
e
t
a
l
.
,
“
Ac
c
u
ra
te
ima
g
e
su
p
e
r
-
re
so
lu
t
io
n
u
si
n
g
v
e
ry
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
two
r
k
s
,”
IEE
E
Co
n
fer
e
n
c
e
o
n
Co
mp
u
ter
V
isio
n
a
n
d
Pa
t
ter
n
Rec
o
g
n
it
io
n
,
p
p
.
1
6
4
6
–
1
6
5
4
,
Ju
n
e
2
0
1
6
.
[8
]
C.
S
z
e
g
e
d
y
,
e
t
a
l
.,
“
G
o
in
g
d
e
e
p
e
r
with
c
o
n
v
o
l
u
ti
o
n
s
,
”
2
0
1
5
IEE
E
Co
n
fer
e
n
c
e
o
n
Co
m
p
u
ter
Vi
si
o
n
a
n
d
Pa
tt
e
rn
Rec
o
g
n
it
io
n
(CVP
R)
,
Ju
n
e
2
0
1
5
.
[9
]
I.
Io
f
fe
a
n
d
S
z
e
g
e
d
y
,
“
Ba
tch
n
o
r
m
a
li
z
a
ti
o
n
:
Ac
c
e
lera
ti
n
g
d
e
e
p
n
e
t
wo
rk
trai
n
in
g
b
y
re
d
u
c
in
g
i
n
tern
a
l
c
o
v
a
riate
sh
if
,”
a
rXiv p
re
p
ri
n
t
a
rX
iv:1
5
0
2
.
0
3
1
6
7
,
F
e
b
ru
a
ry
2
0
1
5
.
[1
0
]
A.
F
e
rr
a
,
e
t
a
l
.
,
“
Wav
e
let
p
o
o
li
n
g
f
o
r
CNN
s
,”
ECCV
,
v
ol
.
4
,
p
p
.
6
7
1
–
6
7
5
,
2
0
1
9
.
[1
1
]
P
.
Li
u
,
e
t
a
l
.
,
“
M
u
lt
i
-
lev
e
l
wa
v
e
le
t
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
,”
Co
RR
,
J
u
ly
2
0
1
9
.
[1
2
]
Alli
so
n
M
.
R
o
ss
e
tt
o
,
e
t
a
l
.,
“
Im
p
r
o
v
i
n
g
c
las
sifica
ti
o
n
wit
h
CNN
s u
sin
g
Wav
e
let
P
o
o
li
n
g
with
Ne
ste
ro
v
-
Ac
c
e
lera
ted
Ad
a
m
,
EP
iC
S
e
ries
in
Co
m
p
u
ti
n
g
,”
Pro
c
e
e
d
in
g
s
o
f
1
1
th
I
n
t
e
rn
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
B
io
i
n
fo
rm
a
t
ics
a
n
d
Co
mp
u
t
a
ti
o
n
a
l
B
io
l
o
g
y
,
v
o
l.
6
0
,
p
p
.
8
4
-
9
3
,
2
0
1
9
.
[1
3
]
Tak
u
m
i
K.
,
“
G
a
u
ss
ian
-
b
a
se
d
p
o
o
li
n
g
fo
r
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
u
ra
l
n
e
t
wo
rk
s
,
”
3
3
rd
C
o
n
fer
e
n
c
e
o
n
Ne
u
ra
l
In
f
o
rm
a
ti
o
n
Pro
c
e
ss
in
g
S
y
ste
ms
,”
Ad
v
a
n
c
e
s in
Ne
u
ra
l
In
f
o
rm
a
ti
o
n
P
r
o
c
e
ss
in
g
S
y
ste
m
s 3
2
,
2
0
1
9
.
[1
4
]
Da
n
il
o
P
.
,
e
t
a
l
,
“
Ac
o
u
stic
imp
u
l
siv
e
n
o
ise
b
a
se
d
o
n
n
o
n
-
g
a
u
ss
ian
m
o
d
e
ls:
a
n
e
x
p
e
rime
n
tal
e
v
a
lu
a
t
io
n
,
”
S
e
n
so
rs
,
v
o
l.
1
9
,
n
o
.
1
2
,
p
p
.
2
8
2
7
,
Ju
n
e
2
0
1
9
.
[1
5
]
Ja
wa
d
N.,
e
t
a
l
,
“
M
a
x
-
p
o
o
li
n
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
fo
r
v
i
sio
n
-
b
a
se
d
h
a
n
d
g
e
stu
re
re
c
o
g
n
i
ti
o
n
,
”
2
0
1
1
IEE
E
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
S
ig
n
a
l
a
n
d
Ima
g
e
Pro
c
e
ss
in
g
Ap
p
li
c
a
t
io
n
s
,
N
o
v
e
m
b
e
r
2
0
1
1
.
[1
6
]
T.
Wi
ll
iam
s a
n
d
R.
Li
.
,
“
Wa
v
e
let
p
o
o
li
n
g
f
o
r
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
s
,”
IC
L
R
,
2
0
1
8
.
[1
7
]
C.
-
Y.
Lee
,
e
t
a
l
,
“
G
e
n
e
ra
li
z
in
g
p
o
o
li
n
g
f
u
n
c
ti
o
n
s
in
c
o
n
v
o
lu
t
i
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s:
M
i
x
e
d
,
g
a
ted
,
a
n
d
tree
,”
AIS
T
AT
S
,
p
p
.
4
6
4
–
4
7
2
,
2
0
1
6
.
[1
8
]
D.
Yu
,
e
t
a
l
.
,
“
M
ix
e
d
p
o
o
l
in
g
fo
r
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
r
k
s
,”
RS
KT
,
p
p
.
3
6
4
–
3
7
5
,
Oc
to
b
e
r
2
0
1
4
[1
9
]
M
.
Zeiler,
e
t
a
l
.,
“
S
to
c
h
a
stic p
o
o
li
n
g
f
o
r
re
g
u
lariz
a
ti
o
n
o
f
d
e
e
p
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
r
k
,
”
ICL
R
,
Ja
n
u
a
ry
2
0
1
3
.
[2
0
]
S
wie
to
jan
sk
i
a
n
d
Re
n
a
ls.
"
Diffe
r
e
n
ti
a
b
le
p
o
o
l
in
g
fo
r
u
n
s
u
p
e
rv
ise
d
a
c
o
u
stic
m
o
d
e
l
a
d
a
p
tatio
n
,”
IEE
E
T
ra
n
s
a
c
ti
o
n
s
o
n
A
u
d
io
,
S
p
e
e
c
h
a
n
d
L
a
n
g
u
a
g
e
Pro
c
e
ss
in
g
,
v
o
l
.
2
4
,
n
o
.
1
0
,
p
p
.
1
7
7
3
–
1
7
8
4
,
M
a
rc
h
2
0
1
6
.
[2
1
]
A.
P
e
ws
e
y
,
“
Larg
e
-
sa
m
p
le
in
fe
re
n
c
e
fo
r
t
h
e
g
e
n
e
ra
l
h
a
lf
-
n
o
rm
a
l
d
i
strib
u
ti
o
n
,”
Co
mm
u
n
ica
t
io
n
s
in
S
t
a
ti
stics
-
T
h
e
o
ry
a
n
d
M
e
t
h
o
d
s
,
v
o
l.
3
1
,
n
o
.
7
,
p
p
.
1
0
4
5
–
1
0
5
4
,
Ju
l
y
2
0
0
2
.
[2
2
]
Wen
h
a
o
G
.
,
“
A
g
e
n
e
ra
li
z
a
ti
o
n
o
f
th
e
sla
sh
h
a
lf
n
o
rm
a
ld
istri
b
u
ti
o
n
:
p
ro
p
e
rti
e
s a
n
d
in
fe
re
n
c
e
s
,”
J
o
u
r
n
a
l
o
f
S
ta
ti
st
i
c
a
l
T
h
e
o
ry
a
n
d
Pra
c
ti
c
e
,
v
o
l.
8
,
n
o
.
2
,
p
p
.
2
8
3
–
2
9
6
,
M
a
rc
h
2
0
1
4
.
[
2
3
]
Y
a
n
n
L
e
.
C
.
,
e
t
a
l
,
"
T
h
e
M
I
N
S
T
d
a
t
a
b
a
s
e
f
o
r
h
a
n
d
w
r
i
g
h
t
d
i
g
i
t
s
,
”
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
:
/
/
y
a
n
n
.
l
e
c
u
n
.
c
o
m
/
e
x
d
b
/
m
n
i
s
t
/
.
A
c
c
e
s
s
e
d
:
2005
.
[2
4
]
Ale
x
K.,
e
t
a
l
.
,
“
Th
e
CIF
AR
-
1
0
d
a
tas
e
t”
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
s:
//
ww
w.cs
.
to
ro
n
to
.
e
d
u
/
~
k
riz/c
ifar.
h
tml
.
[2
5
]
G
e
o
rg
e
B.
,
e
t
a
l
,
“
T
h
e
i
m
p
a
c
t
o
f
th
e
M
IT
-
BIH
Arrh
y
t
h
m
ia
d
a
tab
a
se
,”
IEE
E
e
n
g
i
n
e
e
rin
g
in
me
d
ici
n
e
a
n
d
b
il
o
l
o
g
y
,
v
o
l.
2
0
,
n
o
.
3
,
Ju
n
e
2
0
0
1
.
[2
6
]
G
o
ld
b
e
rg
e
r,
A.
,
e
t
a
l.
,
“
P
h
y
sio
B
a
n
k
,
P
h
y
sio
To
o
lk
i
t,
a
n
d
P
h
y
si
o
N
e
t:
Co
m
p
o
n
e
n
ts
o
f
a
n
e
w
re
se
a
rc
h
re
so
u
rc
e
fo
r
c
o
m
p
lex
p
h
y
si
o
l
o
g
ic si
g
n
a
ls
,”
Cir
c
u
latio
n
, v
o
l.
1
0
1
,
n
o.
2
3
,
p
p
.
e
2
1
5
–
e
2
2
0
,
Ju
l
y
2
0
0
0
.
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
As
sist.
Pro
f.
Aqee
l
M.
H
a
m
a
d
Alh
u
ss
a
in
y
.
He
o
b
tain
e
d
h
is
BS
c
.
Tec
h
in
C
o
m
p
u
ter
a
n
d
C
o
n
tro
l
E
n
g
in
e
e
rin
g
fro
m
U
n
iv
e
rsity
o
f
T
e
c
h
n
o
lo
g
y
I
ra
q
2
0
0
3
a
n
d
M
.
S
c
.
fro
m
Ba
sra
h
Un
iv
e
rsity
,
Ira
q
in
2
0
1
0
a
n
d
h
e
is
P
h
.
D.
stu
d
e
n
t
n
o
w
a
t
Al
-
Na
h
ra
in
Un
iv
e
rsity
,
Ira
q
.
He
is
h
a
v
in
g
1
0
y
e
a
rs
o
f
tea
c
h
in
g
a
n
d
re
se
a
rc
h
e
x
p
e
rien
c
e
wo
rk
e
d
a
s
a
h
e
a
d
fo
r
c
o
m
p
u
ter
c
e
n
ter
i
n
Th
iQa
r
U
n
iv
e
rsity
fo
r
th
re
e
y
e
a
rs.
He
is
p
re
se
n
tl
y
wo
rk
in
g
a
s
As
sista
n
t
P
ro
fe
ss
o
r
in
T
h
iQa
r
U
n
iv
e
rsity
, C
o
ll
e
g
e
o
f
C
o
m
p
u
ter an
d
M
a
th
e
m
a
ti
c
s.
As
sist.
Pro
f
.
Dr
.
Am
m
a
r
D.
J
a
sim
.
He
o
b
tain
e
d
h
is
B
Sc
.
Tec
h
in
C
o
m
p
u
ter
fro
m
U
n
iv
e
rsi
t
y
o
f
Al
-
Na
h
ra
in
,
I
ra
q
1999
a
n
d
M
.
S
c
fro
m
Al
-
Na
h
ra
in
Un
iv
e
rsity
,
Ira
q
in
2
0
0
2
a
n
d
P
h
.
D
fro
m
Al
-
Na
h
ra
in
Un
iv
e
rsity
,
Ira
q
.
He
is
h
a
v
in
g
2
0
y
e
a
rs
o
f
tea
c
h
in
g
a
n
d
re
se
a
rc
h
e
x
p
e
rien
c
e
wo
rk
e
d
a
s a
lec
tu
re
r
in
C
o
ll
e
g
e
o
f
I
n
fo
rm
a
ti
o
n
E
n
g
in
e
e
rin
g
,
Al
-
Na
h
ra
in
Un
iv
e
rsity
.
Evaluation Warning : The document was created with Spire.PDF for Python.