I
n
d
on
e
s
ian
Jou
r
n
al
o
f
E
lec
t
r
ica
l
E
n
gin
e
e
r
in
g
a
n
d
Com
p
u
t
e
r
S
c
ience
Vo
l
.
36
,
N
o
.
3
,
De
c
e
m
b
e
r
20
24
,
pp.
1
7
69
~1
7
77
I
S
S
N:
2
502
-
4
7
52
,
DO
I
:
10
.
11591/i
j
e
e
cs
.v
36.
i
3
.
pp1
7
69
-
1
7
77
1769
Jou
r
n
al
h
o
m
e
page
:
ht
tp:
//
ij
e
e
cs
.
iaes
c
or
e
.
c
om
A
n
e
f
f
ic
ie
n
t
c
on
v
ol
u
t
io
n
al
n
e
u
r
al
n
e
t
w
o
r
k
f
or
ad
v
e
r
sar
ia
l
t
r
ai
n
in
g again
st
a
d
v
e
r
sa
r
ia
l
at
t
ac
k
S
r
in
ivas
A
.
Vadd
ad
i
1
,
S
an
j
a
ik
an
t
h
E
.
Vadak
k
e
t
h
il
S
om
an
at
h
an
P
il
l
ai
2
,
S
an
t
os
h
Re
d
d
y
Add
u
l
a
1
,
Rohi
t
h
Val
l
ab
h
an
e
n
i
1
,
B
h
u
van
e
s
h
Anan
t
h
an
3
1
D
e
pa
r
tm
e
nt
of
I
n
f
or
ma
ti
o
n
T
e
c
hn
o
l
o
g
y
, U
ni
v
e
r
s
it
y
of
t
h
e
C
umbe
r
la
nds
,
W
il
li
a
ms
bur
g
,
U
ni
te
d
S
ta
te
s
2
S
c
h
oo
l
of
E
l
e
c
tr
ic
a
l
E
ngi
ne
e
r
in
g a
nd C
o
mput
e
r
S
c
i
e
n
c
e
, U
ni
ve
r
s
it
y
of
N
o
r
th
D
a
ko
ta
,
G
r
a
nd F
or
ks
,
U
ni
te
d S
ta
t
e
s
3
D
e
pa
r
tm
e
nt
of
E
l
e
c
tr
i
c
a
l
a
nd E
l
e
c
tr
o
ni
c
s
E
ngi
n
e
e
r
in
g, P
S
N
C
ol
le
g
e
of
E
ng
in
e
e
r
in
g a
nd
T
e
c
hn
o
l
o
g
y
,
T
ir
un
e
l
ve
li
, I
nd
ia
Ar
t
ic
l
e
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
i
ve
d
M
a
r
18
,
202
4
R
e
vi
s
e
d
A
ug
19
,
202
4
A
c
c
e
pt
e
d
A
ug
26
,
202
4
Co
n
v
o
l
u
t
i
o
n
al
n
eu
ral
n
e
t
w
o
r
k
s
(CN
N
)
ar
e
w
i
d
el
y
u
s
e
d
b
y
r
e
s
e
ar
c
h
e
rs
d
u
e
t
o
t
h
e
i
r
e
x
t
e
n
s
i
v
e
ad
v
an
t
ag
e
s
o
v
e
r
v
ari
o
u
s
ap
p
l
i
c
at
i
o
n
s
.
H
o
w
ev
e
r,
i
m
ag
e
s
ar
e
h
i
g
h
l
y
s
u
s
ce
p
t
i
b
l
e
t
o
m
al
i
c
i
o
u
s
at
t
ac
k
s
u
s
i
n
g
p
e
r
t
u
rb
at
i
o
n
s
t
h
at
are
u
n
r
ec
o
g
n
i
z
e
d
e
v
en
u
n
d
e
r
h
u
m
an
i
n
t
e
r
v
e
n
t
i
o
n
.
T
h
i
s
c
au
s
e
s
s
i
g
n
i
f
i
c
an
t
s
e
cu
ri
t
y
p
e
ri
l
s
an
d
c
h
a
l
l
en
g
e
s
t
o
C
N
N
-
r
e
l
at
e
d
ap
p
l
i
c
at
i
o
n
s
.
I
n
t
h
i
s
art
i
c
l
e,
a
n
e
ffi
ci
e
n
t
ad
v
e
rs
ari
al
t
rai
n
i
n
g
mo
d
e
l
a
g
ai
n
s
t
m
al
e
v
o
l
en
t
at
t
ac
k
s
i
s
d
emo
n
s
t
rat
e
d
.
T
h
i
s
mo
d
el
i
s
h
i
g
h
l
y
ro
b
u
s
t
t
o
b
l
a
c
k
-
b
o
x
m
al
i
c
i
o
u
s
e
x
am
p
l
e
s
,
i
t
i
s
p
ro
ce
s
s
ed
w
i
t
h
d
i
ff
e
r
e
n
t
m
a
l
i
ci
o
u
s
s
am
p
l
e
s
.
I
n
i
t
i
al
l
y
,
m
a
l
i
ci
o
u
s
t
rai
n
i
n
g
mo
d
e
l
s
l
i
k
e
fas
t
g
rad
i
e
n
t
d
e
s
cen
t
(FG
S),
r
ec
u
rs
i
v
e
-
FG
SM
(I
-
FG
S),
D
ee
p
-
Fo
o
l
,
a
n
d
Carl
i
n
i
an
d
W
ag
n
e
r
(CW
)
t
ech
n
i
q
u
e
s
ar
e
u
t
i
l
i
z
e
d
t
h
at
g
en
e
rat
e
ad
v
e
rs
ari
al
i
n
p
u
t
b
y
me
an
s
o
f
t
h
e
C
N
N
a
ck
n
o
w
l
e
d
g
e
d
t
o
t
h
e
at
t
ac
k
e
r.
I
n
t
h
e
e
x
p
e
ri
men
t
at
i
o
n
p
ro
ce
s
s
,
t
h
e
MN
I
ST
d
at
as
e
t
c
o
m
p
ri
s
i
n
g
6
0
K
an
d
1
0
K
t
rai
n
i
n
g
an
d
t
e
s
t
i
n
g
g
r
ey
-
s
c
al
e
i
m
ag
e
s
ar
e
u
t
i
l
i
z
e
d
.
I
n
t
h
e
e
x
p
e
r
i
me
n
t
al
s
e
c
t
i
o
n
,
t
h
e
a
d
v
e
rs
ari
a
l
t
rai
n
i
n
g
mo
d
el
r
ed
u
ce
s
t
h
e
at
t
ac
k
a
cc
u
ra
cy
rat
e
(A
SR)
b
y
a
n
av
e
ra
g
e
o
f
2
9
.
2
%
fo
r
d
i
ff
e
r
e
n
t
m
al
i
c
i
o
u
s
i
n
p
u
t
s
,
w
h
en
p
re
s
e
rv
i
n
g
t
h
e
a
ccu
ra
cy
o
f
9
8
.
9
%
c
o
n
ce
rn
i
n
g
a
c
t
u
a
l
i
m
ag
e
s
i
n
t
h
e
MN
I
ST
d
at
ab
as
e
.
T
h
e
s
i
m
u
l
at
i
o
n
o
u
t
co
me
s
s
h
o
w
t
h
e
p
reemi
n
en
ce
o
f
t
h
e
mo
d
el
ag
ai
n
s
t
ad
v
e
rs
ari
al
at
t
ack
s
.
K
e
y
w
o
r
d
s
:
A
d
v
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
A
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
s
C
o
n
v
o
l
ut
i
o
n
a
l
n
e
ur
a
l
n
e
t
wo
r
k
De
e
p
l
e
a
r
ni
ng
De
f
e
ns
i
ve
m
e
c
ha
ni
s
m
I
m
a
ge
m
a
ni
pu
l
a
t
i
o
n
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
cen
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
S
r
i
ni
va
s
A
.
Va
dda
d
i
De
pa
r
t
m
e
n
t
o
f
I
nf
o
r
m
a
t
i
o
n
T
e
c
hn
o
l
o
g
y
,
Un
i
ve
r
s
i
t
y
o
f
t
h
e
C
u
m
be
r
l
a
n
d
s
6178
C
o
l
l
e
ge
S
t
a
t
i
o
n
Dr
i
ve
,
W
il
li
a
m
s
bur
g,
K
Y
40769,
Uni
t
e
d
S
t
a
t
e
s
E
m
a
i
l
:
Vs
a
d93
@
g
m
a
i
l
.
c
o
m
1.
I
NT
RODU
C
T
I
ON
No
wa
da
y
s
,
pa
tt
e
r
n
i
de
n
t
i
f
i
c
a
t
i
o
n
a
n
d
vi
s
u
a
l
i
z
a
t
i
o
n
a
s
s
o
c
i
a
t
e
d
wi
t
h
c
o
m
put
e
r
s
c
i
e
n
c
e
h
a
v
e
ga
i
ne
d
m
uc
h
a
t
t
e
n
t
i
o
n
f
r
o
m
r
e
s
e
a
r
c
he
r
s
.
S
pe
c
i
f
i
c
a
ll
y
,
t
h
e
c
o
n
v
o
l
ut
i
o
n
a
l
n
e
ur
a
l
n
e
t
wo
r
k
(
C
NN
)
o
b
t
a
i
n
s
b
e
t
t
e
r
pe
r
f
o
r
m
a
n
c
e
i
n
va
r
i
o
us
a
pp
li
c
a
t
i
o
n
s
li
ke
m
e
d
i
c
a
l
d
i
a
g
n
o
s
i
s
,
v
o
i
c
e
i
de
n
t
i
f
i
c
a
t
i
o
n
,
a
n
d
im
a
ge
pr
o
c
e
s
s
i
ng
[
1]
.
Ne
v
e
r
t
h
e
l
e
s
s
,
t
h
e
s
e
m
o
de
l
s
s
u
f
f
e
r
f
r
o
m
i
n
c
r
e
a
s
e
d
a
dv
e
r
s
a
r
i
a
l
v
u
l
ne
r
a
bil
i
t
i
e
s
.
T
h
e
s
e
c
ur
i
t
y
i
s
s
ue
s
a
r
e
di
vi
de
d
i
n
t
o
t
w
o
g
r
o
ups
n
a
m
e
ly
pr
o
bi
ng
a
tt
a
c
k
a
n
d
c
a
us
a
l
a
tt
a
c
k.
T
h
e
c
a
us
a
l
a
t
t
a
c
ks
di
mi
n
i
s
h
t
h
e
pe
r
f
o
r
m
a
nc
e
o
f
t
h
e
C
NN
m
o
d
e
l
by
p
u
r
p
os
e
f
u
l
l
y
i
n
teg
r
a
t
i
n
g
a
dv
e
r
s
a
r
i
a
l
s
a
m
p
l
e
s
a
t
th
e
tr
a
i
ni
n
g
s
ta
ge
d
u
r
i
n
g
t
h
e
l
e
a
r
ni
n
g
pr
oc
e
s
s
[
2
]
.
S
o
m
e
o
f
t
h
e
c
o
m
m
o
n
c
a
us
a
l
a
tt
a
c
ks
a
r
e
p
o
i
s
o
nin
g
a
n
d
b
a
c
kdo
o
r
a
tt
a
c
ks
t
h
a
t
a
r
e
m
a
ni
pu
l
a
t
e
d
to
c
r
e
a
t
e
a
pa
r
t
i
c
u
l
a
r
e
f
f
e
c
t
.
T
h
e
pr
o
bi
n
g
a
t
t
a
c
k
i
s
o
n
e
o
f
t
h
e
e
v
a
s
i
ve
a
t
t
a
c
ks
t
h
a
t
m
o
d
i
f
i
e
s
t
h
e
t
e
s
t
da
t
a
t
h
a
t
a
r
e
a
l
r
e
a
d
y
b
e
e
n
t
r
a
i
n
e
d.
T
hi
s
t
y
pe
o
f
a
tt
a
c
k
i
s
r
e
a
l
i
s
t
i
c
s
i
nc
e
t
h
e
r
e
i
s
n
o
de
f
i
n
i
t
i
o
n
o
f
t
h
e
l
e
a
r
ni
ng
da
t
a
t
h
a
t
a
r
e
b
e
i
n
g
a
c
c
e
s
s
e
d
[
3]
.
T
h
e
a
dv
e
r
s
a
r
i
a
l
a
tt
a
c
k
i
s
o
n
e
i
m
po
r
t
a
n
t
e
x
a
m
p
l
e
o
f
a
pr
o
bi
n
g
a
t
t
a
c
k
t
h
a
t
i
n
t
e
gr
a
t
e
s
s
o
m
e
n
o
i
s
e
s
i
n
t
o
t
h
e
a
c
t
ua
l
da
t
a
t
h
a
t
c
a
n
b
e
e
a
s
il
y
i
de
n
t
i
f
i
e
d
by
h
u
m
a
ns
b
ut
by
t
h
e
r
e
c
e
n
t
de
e
p
l
e
a
r
ni
ng
(
DL
)
m
o
de
l
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
36
,
N
o
.
3
,
De
c
e
m
b
e
r
20
24
:
1
7
69
-
1
7
77
1770
I
n
s
uc
h
c
a
s
e
s
,
i
t
c
o
ul
d
c
a
u
s
e
e
r
r
o
r
s
wh
e
n
t
h
e
y
a
r
e
us
e
d
i
n
r
e
a
l
-
t
i
m
e
a
pp
l
i
c
a
t
i
o
ns
li
ke
m
e
d
i
c
a
l
im
a
g
i
ng,
a
uto
n
o
m
o
us
v
e
hi
c
l
e
s
,
a
n
d
b
us
i
ne
s
s
a
ppli
c
a
t
i
o
n
s
due
to
un
pr
e
d
i
c
t
a
bl
e
o
u
t
c
o
m
e
s
.
S
e
v
e
r
a
l
e
xi
s
t
i
ng
s
t
udi
e
s
h
a
v
e
b
e
e
n
ut
i
li
z
e
d
a
s
a
de
f
e
n
s
i
ve
m
e
t
h
o
d
a
ga
i
ns
t
m
a
l
e
v
o
l
e
n
t
a
tt
a
c
ks
a
n
d
t
h
e
y
a
r
e
o
f
t
wo
t
y
pe
s
.
On
e
o
f
t
h
e
s
e
i
s
da
t
a
m
a
ni
pu
l
a
t
i
o
n
a
n
d
t
h
e
o
t
h
e
r
i
s
m
a
k
i
n
g
DL
m
o
de
l
s
m
o
r
e
r
o
b
us
t
o
v
e
r
ot
h
e
r
t
e
c
h
ni
que
s
[
4]
.
T
h
e
da
t
a
m
a
ni
pu
l
a
t
i
o
n
pr
o
c
e
s
s
m
i
n
im
i
z
e
s
t
h
e
a
t
t
a
c
k
l
e
v
e
l
w
i
t
h
t
h
e
us
e
o
f
n
o
i
s
e
f
r
o
m
m
a
li
c
i
o
us
i
ns
t
a
n
c
e
s
by
f
i
l
t
e
r
i
n
g
o
r
r
e
s
i
z
i
ng
t
h
e
da
t
a
.
I
n
c
o
n
t
r
a
di
c
t
i
o
n
to
thi
s
,
t
h
e
m
o
de
l
s
o
f
m
a
k
i
ng
C
NN
n
e
t
wo
r
ks
m
o
r
e
r
e
li
a
bl
e
b
y
ut
i
li
z
i
ng
t
h
e
d
i
s
t
i
ll
a
t
i
o
n
m
e
t
h
o
d,
a
n
d
m
a
li
c
i
o
u
s
t
r
a
i
ni
ng
pr
o
c
e
s
s
.
T
h
e
d
i
s
t
i
ll
a
t
i
o
n
a
ppr
o
a
c
h
ut
i
l
i
z
e
d
dua
l
n
e
t
wo
r
k
m
o
de
l
s
to
r
e
s
i
s
t
t
h
e
c
r
e
a
t
i
o
n
o
f
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
nc
e
s
.
Ho
we
v
e
r
,
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
s
c
he
m
e
s
a
r
e
m
o
r
e
f
l
e
xi
b
l
e
by
j
o
i
n
t
l
y
s
c
r
ut
i
ni
z
i
ng
t
h
e
C
NN
m
o
de
l
o
n
m
a
li
c
i
o
u
s
s
a
m
p
l
e
s
t
h
a
t
a
r
e
o
b
t
a
i
n
e
d
f
r
o
m
t
h
e
t
r
a
di
t
i
o
na
l
D
L
m
o
de
l
s
[
5]
.
F
r
o
m
t
h
e
s
e
,
t
h
e
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
s
c
h
e
m
e
s
a
r
e
s
i
m
p
l
e
r
a
n
d
m
o
r
e
r
o
b
us
t
c
o
m
pa
r
e
d
t
o
c
o
n
v
e
n
t
i
o
n
a
l
s
t
ud
i
e
s
.
B
a
s
e
d
o
n
t
r
a
di
t
i
o
n
a
l
t
e
c
hni
que
s
,
t
h
e
m
a
li
c
i
o
us
s
c
r
ut
i
ni
z
a
t
i
o
n
i
s
c
a
l
i
br
a
t
e
d
to
m
a
i
n
t
a
i
n
r
o
b
us
t
t
h
a
n
o
t
h
e
r
m
a
n
ua
l
t
e
c
hni
que
s
i
n
t
e
r
m
s
o
f
h
a
n
d
s
-
o
n
de
f
e
ns
i
ve
m
e
c
ha
ni
s
m
s
.
B
ut
t
h
e
c
o
n
v
e
n
t
i
o
n
a
l
m
a
li
c
i
o
us
s
c
r
ut
i
ni
z
i
ng
m
o
de
l
t
r
a
i
ns
t
h
e
f
i
na
l
m
o
de
l
w
i
t
h
a
d
v
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
o
b
t
a
i
n
e
d
us
i
ng
s
i
ng
l
e
a
tt
a
c
k
m
o
de
l
s
[
6]
.
W
h
e
n
t
he
f
i
na
l
m
o
de
l
i
s
s
c
r
ut
i
ni
z
e
d
w
i
t
h
a
d
v
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
t
h
a
t
a
r
e
o
b
t
a
i
ne
d
f
r
o
m
s
e
ve
r
a
l
a
t
t
a
c
ks
r
a
t
h
e
r
t
h
a
n
f
o
c
us
i
ng
o
n
a
s
i
n
g
l
e
a
tt
a
c
k,
wi
ll
b
e
m
o
r
e
e
f
f
e
c
t
i
ve
a
ga
i
n
s
t
v
u
l
ne
r
a
b
l
e
m
a
l
i
c
i
o
us
a
t
t
a
c
ks
.
De
v
e
l
o
p
i
n
g
r
o
b
us
t
m
a
li
c
i
o
us
pe
r
t
ur
b
a
t
i
o
ns
un
d
e
r
v
a
r
yi
ng
e
nvi
r
o
nm
e
n
t
s
w
i
t
h
d
i
f
f
e
r
e
n
t
l
e
a
r
ni
ng
l
e
v
e
l
s
o
n
t
h
e
C
NN
m
o
de
l
h
a
s
be
e
n
de
li
be
r
a
t
e
d
i
n
v
a
r
i
o
us
f
r
a
m
e
wo
r
ks
[
7
]
,
[
8]
.
I
n
t
h
e
s
a
m
e
wa
y
,
t
h
e
de
s
i
g
n
o
f
a
de
f
e
ns
i
ve
m
e
c
h
a
ni
s
m
f
o
r
i
n
c
r
e
a
s
i
ng
t
h
e
r
e
l
i
a
bil
i
t
y
o
f
C
NN
c
o
n
t
r
a
r
y
to
m
a
l
e
v
o
l
e
n
t
pe
r
t
u
r
b
a
t
i
o
n
ha
s
un
de
r
go
n
e
t
w
o
m
a
j
o
r
pr
o
c
e
s
s
e
s
:
a
tt
a
c
k
r
e
c
o
gni
t
i
o
n
,
a
n
d
a
tt
a
c
k
r
e
t
r
i
e
v
a
l
m
e
t
ho
ds
.
T
h
e
de
v
e
l
o
pe
d
s
t
ud
y
a
im
s
t
o
c
l
a
s
s
if
y
t
h
e
im
a
ge
s
a
s
n
o
r
m
a
l
o
r
m
a
l
i
c
i
o
us
by
t
r
a
i
ni
ng
pr
o
c
e
s
s
a
n
d
dr
o
p
o
u
t
u
ni
t
s
.
M
ot
i
v
a
t
i
o
n
:
t
h
e
pr
e
s
e
n
c
e
o
f
a
d
v
a
n
c
e
d
DL
m
o
de
l
s
h
a
s
c
o
n
t
e
m
p
l
a
t
e
d
t
h
e
pe
r
f
o
r
m
a
n
c
e
a
b
o
v
e
t
h
e
h
u
m
a
n
i
n
t
e
r
v
e
n
t
i
o
n
l
e
v
e
l
b
a
s
e
d
o
n
i
m
a
ge
pr
o
c
e
s
s
i
n
g
t
a
s
ks
.
De
s
p
i
t
e
t
h
e
s
e
ve
r
a
l
m
o
de
l
s
,
C
NN
s
a
r
e
hi
g
hly
s
ubj
e
c
t
e
d
to
a
dv
e
r
s
a
r
i
a
l
a
t
t
a
c
ks
by
i
nj
e
c
t
i
n
g
pe
r
t
u
r
b
a
t
i
o
n
i
n
t
o
t
h
e
a
c
t
ua
l
i
m
a
ge
.
T
hi
s
pe
r
t
u
r
b
a
t
i
o
n
i
s
un
k
n
o
wn
to
h
u
m
a
n
pe
r
c
e
pt
i
o
n
a
n
d
d
i
s
t
r
a
c
t
s
t
h
e
m
o
de
l
s
f
r
o
m
pr
o
vi
d
i
n
g
b
e
t
t
e
r
pe
r
f
o
r
m
a
n
c
e
.
No
wa
da
y
s
,
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
s
c
h
e
m
e
s
a
r
e
p
l
a
yi
ng
a
n
i
n
t
e
gr
a
l
r
o
l
e
in
l
e
a
r
ni
n
g
v
u
l
ne
r
a
bl
e
a
t
t
a
c
ks
us
i
n
g
pr
a
c
t
i
c
a
l
d
e
f
e
n
s
i
ve
m
e
c
h
a
ni
s
m
s
.
Ho
we
v
e
r
,
t
r
a
di
t
i
o
n
a
l
a
d
v
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
t
e
c
h
ni
que
s
a
r
e
m
uc
h
m
o
r
e
e
f
f
e
c
t
i
v
e
a
s
t
h
e
y
c
a
n
t
r
a
i
n
i
n
put
s
wi
t
h
s
i
ng
l
e
a
tt
a
c
ks
.
T
h
e
s
e
k
i
n
d
s
o
f
m
a
j
o
r
c
o
n
s
,
m
o
t
i
v
a
t
e
us
to
de
v
e
l
o
p
a
r
o
b
us
t
DL
-
b
a
s
e
d
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
o
de
l
to
de
f
i
ne
v
a
r
i
o
us
m
a
li
c
i
o
us
a
t
t
a
c
ks
a
c
c
ur
a
t
e
l
y
.
T
h
e
ke
y
c
o
n
t
r
i
b
ut
i
o
n
s
o
f
t
h
e
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
e
t
h
o
d
a
r
e
c
o
n
que
r
e
d
b
e
l
o
w:
T
o
i
n
t
r
o
duc
e
a
di
v
e
r
s
e
a
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
n
g
m
o
de
l
in
t
h
e
C
NN
f
o
r
ge
n
e
r
a
t
i
n
g
m
a
li
c
i
o
us
s
a
m
p
l
e
s
f
r
o
m
t
h
e
a
c
t
ua
l
i
nput
.
T
o
pr
e
s
e
n
t
s
e
v
e
r
a
l
a
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
s
l
i
k
e
f
a
s
t
gr
a
d
i
e
n
t
de
s
c
e
n
t
(
F
GS)
,
i
t
e
r
a
t
i
v
e
-
F
G
S
M
(I
-
F
GS)
,
De
e
p
-
F
oo
l
,
a
n
d
C
a
r
li
n
i
a
n
d
W
a
g
n
e
r
(
C
W
)
f
o
r
c
r
e
a
t
i
n
g
m
a
l
i
c
i
o
u
s
e
x
a
m
p
l
e
s
i
n
t
h
e
C
NN
s
.
T
o
v
a
l
i
da
t
e
d
i
v
e
r
s
e
a
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
s
by
a
s
s
e
s
s
i
ng
d
i
f
f
e
r
e
n
t
m
e
a
s
ur
e
s
l
i
ke
a
c
c
ur
a
c
y
,
a
n
d
a
tt
a
c
k
a
c
c
ur
a
c
y
r
a
t
e
by
c
o
m
pa
r
i
ng
t
h
e
m
w
i
t
h
c
o
n
ve
n
t
i
o
n
a
l
t
e
c
hni
que
s
.
T
h
e
upc
o
m
i
ng
s
e
c
t
i
o
ns
a
r
e
or
ga
ni
z
e
d
a
s
de
p
i
c
t
e
d
b
e
l
o
w:
s
e
c
t
i
o
n
2
o
u
t
l
a
y
s
t
h
e
s
e
c
t
i
o
n
a
b
o
ut
r
e
l
a
t
e
d
wor
k,
s
e
c
t
i
o
n
3
de
l
i
be
r
a
t
e
s
o
v
e
r
t
h
e
s
ugge
s
t
e
d
a
ppr
o
a
c
h
,
s
e
c
t
i
o
n
4
pr
e
s
e
n
t
s
t
h
e
r
e
s
u
l
t
s
a
n
d
d
i
s
c
u
s
s
i
o
n
,
a
n
d
s
e
c
t
i
o
n
5
r
e
pr
e
s
e
n
t
s
t
h
e
c
o
n
c
l
u
s
i
o
n
o
f
t
h
e
de
v
e
l
o
pe
d
f
r
a
m
e
w
o
r
k
.
2.
RE
L
AT
E
D
WORK
S
M
a
ni
e
t
al.
[
9]
,
pu
t
f
o
r
t
h
DL
-
b
a
s
e
d
m
o
de
l
s
f
o
r
de
f
e
n
d
i
ng
a
d
v
e
r
s
a
r
i
a
l
s
c
r
ut
i
ny
o
ve
r
m
a
li
c
i
o
us
a
t
t
a
c
ks
.
He
r
e
,
t
h
e
R
e
s
Ne
t
-
b
a
s
e
d
im
a
ge
c
l
a
s
s
if
i
c
a
t
i
o
n
t
e
c
hni
que
wa
s
de
m
o
n
s
t
r
a
t
e
d
a
n
d
i
nv
e
s
t
i
g
a
t
e
d
to
de
f
e
n
d
a
ga
i
ns
t
t
h
e
GSM
a
dv
e
r
s
a
r
i
a
l
m
o
de
l
.
F
o
r
t
h
e
s
i
m
u
l
a
t
i
o
n
pr
o
c
e
s
s
,
t
h
e
C
I
F
A
R
-
10
da
t
a
b
a
s
e
wa
s
ut
i
l
i
z
e
d
i
n
t
hi
s
s
t
udy
.
I
n
t
h
e
r
e
s
ul
t
a
n
t
pa
r
t
,
a
c
c
ur
a
c
y
wa
s
a
n
a
ly
z
e
d
a
n
d
d
i
s
t
i
n
gu
i
s
h
e
d
f
r
o
m
ot
h
e
r
s
t
udi
e
s
.
H
o
we
v
e
r
,
t
hi
s
m
e
t
h
o
d
wa
s
hi
g
hly
s
e
ns
i
t
i
ve
w
h
e
n
s
c
r
ut
i
ni
z
i
ng
w
i
t
h
n
o
i
s
y
p
e
r
t
u
r
b
e
d
da
t
a
a
n
d
i
nc
r
e
a
s
e
d
t
h
e
e
r
r
o
r
dur
i
n
g
t
h
e
t
e
s
t
i
ng
pr
o
c
e
s
s
.
L
a
l
e
t
al.
[
10]
,
de
f
i
ne
d
t
h
e
DL
-
b
a
s
e
d
de
f
e
ns
i
ve
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
n
g
m
o
de
l
f
o
r
de
t
e
c
t
i
n
g
d
i
a
be
t
i
c
r
e
t
i
n
o
pa
t
hy
.
He
r
e
,
t
h
e
de
f
e
n
s
i
ve
m
e
t
h
o
d
a
ga
i
n
s
t
s
pe
c
k
l
e
n
o
i
s
e
a
tt
a
c
ks
,
m
a
li
c
i
o
us
t
r
a
i
ni
ng,
a
n
d
f
e
a
t
ur
e
f
us
i
o
n
m
e
c
h
a
ni
s
m
we
r
e
a
n
a
ly
z
e
d
to
s
e
c
ur
e
t
h
e
r
e
t
i
n
o
pa
t
hy
i
m
a
g
e
s
e
f
f
e
c
t
i
v
e
ly
.
S
o
m
e
o
f
t
h
e
f
e
a
t
ur
e
e
x
t
r
a
c
t
i
o
n
t
e
c
h
ni
que
s
l
i
ke
l
o
c
a
l
bi
na
r
y
pa
t
t
e
r
n
(
L
B
P
)
a
n
d
hi
s
t
o
gr
a
m
o
f
o
r
i
e
n
t
e
d
gr
a
di
e
n
t
s
(
HOG
)
we
r
e
c
on
s
i
d
e
r
e
d.
M
o
r
e
o
v
e
r
,
s
e
gm
e
n
t
a
t
i
o
n
-
b
a
s
e
d
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
e
t
h
o
ds
we
r
e
a
l
s
o
i
nve
s
t
i
ga
t
e
d.
T
h
e
pe
r
t
ur
b
e
d
da
t
a
wa
s
ge
n
e
r
a
t
e
d
vi
a
a
d
ve
r
s
a
r
i
a
l
m
o
d
e
l
s
li
ke
F
GSM
,
a
n
d
de
e
p
f
o
o
l
s
we
r
e
t
r
a
i
ne
d
o
n
t
h
e
a
c
t
ua
l
im
a
ge
s
to
o
b
t
a
i
n
m
a
li
c
i
o
us
i
ns
t
a
nc
e
s
.
I
n
t
h
e
r
e
s
u
l
t
a
n
t
pa
r
t
,
a
c
c
ur
a
c
y
wa
s
a
na
l
y
z
e
d
a
n
d
d
i
s
t
i
n
gu
i
s
he
d
f
r
o
m
o
t
h
e
r
s
t
udi
e
s
.
Ho
we
v
e
r
,
t
hi
s
m
e
t
h
o
d
f
a
c
e
s
hi
g
h
bl
ack
-
b
o
x
i
s
s
ue
s
a
n
d
r
e
duc
e
s
t
h
e
i
n
t
e
r
pr
e
t
a
bi
li
t
y
o
f
t
h
e
o
u
t
c
o
m
e
i
m
a
ge
s
.
P
a
l
e
t
al.
[
11]
e
s
t
a
bl
i
s
h
e
d
t
h
e
m
a
li
c
i
o
us
de
f
e
n
s
i
ve
m
e
c
h
a
ni
s
m
us
i
ng
t
h
e
i
n
t
e
gr
a
t
e
d
m
o
de
l
o
n
v
o
i
c
e
s
a
m
p
l
e
s
.
He
r
e
,
t
h
e
c
r
o
s
s
-
e
n
t
r
o
py
,
f
e
a
t
ur
e
s
c
a
tt
e
r
i
n
g,
a
n
d
m
a
r
g
i
n
a
l
l
o
s
s
e
s
we
r
e
j
o
i
n
t
l
y
us
e
d
to
i
nv
e
s
t
i
g
a
t
e
pe
r
t
u
r
b
e
d
s
a
m
p
l
e
s
.
M
o
r
e
o
v
e
r
,
t
h
e
de
e
p
n
e
ur
a
l
n
e
t
wo
r
k
(
DNN
)
m
o
de
l
wa
s
ut
i
li
z
e
d
t
o
i
d
e
n
t
i
f
y
t
h
e
a
tt
a
c
k
t
h
a
t
o
c
c
ur
r
e
d
o
n
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
s
.
T
h
e
s
i
m
u
lat
i
o
n
pr
o
c
e
s
s
wa
s
de
m
o
n
s
t
r
a
t
e
d
o
n
t
h
e
L
i
b
r
i
s
p
e
e
c
h
da
t
a
s
e
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
n
e
ff
icie
nt
c
onv
olut
ional
ne
ur
al
ne
t
w
or
k
f
or
ad
v
e
r
s
ar
ial
tr
aini
ng
agains
t
…
(
Sr
ini
v
a
s
A
.
V
addadi
)
1771
a
n
d
a
c
c
ur
a
c
y
-
b
a
s
e
d
c
o
m
put
a
t
i
o
n
a
l
pe
r
f
o
r
m
a
n
c
e
wa
s
a
na
ly
z
e
d.
M
o
r
e
o
v
e
r
,
C
W
a
tt
a
c
ks
a
n
d
pr
o
j
e
c
t
e
d
GSM
a
tt
a
c
ks
we
r
e
a
l
s
o
c
o
n
s
i
de
r
e
d.
Ho
we
v
e
r
,
t
hi
s
m
e
t
ho
d
wa
s
hi
g
hly
c
o
s
t
-
e
f
f
e
c
t
i
ve
a
n
d
r
e
qu
i
r
e
d
a
m
o
r
e
r
e
l
i
a
ble
t
r
a
i
ni
ng
pr
o
c
e
s
s
f
o
r
i
de
n
t
i
f
yi
ng
ot
h
e
r
h
a
r
m
f
u
l
a
t
t
a
c
ks
.
Ha
s
h
e
mi
,
a
n
d
M
o
z
a
f
f
a
r
i
[
12]
i
n
t
r
o
duc
e
d
a
C
NN
-
b
a
s
e
d
DL
m
o
de
l
f
o
r
t
r
a
i
ni
ng
t
h
e
pe
r
t
ur
b
e
d
da
t
a
us
i
n
g
a
c
t
ua
l
i
nput
s
.
He
r
e
,
t
h
e
C
NN
w
i
t
h
v
a
r
i
a
t
i
o
na
l
a
ut
o
e
n
c
o
de
r
(
A
E
)
m
o
de
l
wa
s
us
e
d
a
s
t
h
e
t
a
r
ge
t
t
h
a
t
wa
s
kn
o
wn
to
t
h
e
a
tt
a
c
ke
r
.
Va
r
i
o
us
a
tt
a
c
ks
o
n
gr
a
di
e
n
t
-
b
a
s
e
d
t
e
c
h
ni
que
s
,
di
s
t
a
n
c
e
f
u
n
c
t
i
o
n
o
p
t
i
mi
z
e
r
(
DF
O)
,
t
r
a
n
s
f
o
r
m
a
t
i
ve
m
o
de
l
s
(
T
M
)
,
a
n
d
pr
e
pr
o
c
e
s
s
i
n
g
t
e
c
h
ni
que
s
we
r
e
i
nv
e
s
t
i
ga
t
e
d.
M
or
e
o
v
e
r
,
uni
f
o
r
m
n
o
i
s
e
s
,
s
a
l
t
-
pe
ppe
r
n
o
i
s
e
s
,
ga
us
s
i
a
n
n
o
i
s
e
s
,
a
n
d
P
o
i
s
s
o
n
n
o
i
s
e
s
we
r
e
a
l
s
o
c
o
n
s
i
de
r
e
d
f
o
r
t
h
e
a
tt
a
c
k
c
l
a
s
s
i
f
i
c
a
t
i
o
n
pr
o
c
e
s
s
.
I
n
t
h
e
r
e
s
u
l
t
a
n
t
pa
r
t
,
a
c
c
ur
a
c
y
a
n
d
r
e
c
e
i
v
e
r
-
o
pe
r
a
t
i
n
g
c
ha
r
a
c
t
e
r
i
s
t
i
c
c
ur
v
e
(
R
O
C
)
we
r
e
a
na
l
y
z
e
d
a
nd
d
i
s
t
i
n
gu
i
s
he
d
f
r
o
m
o
t
h
e
r
s
t
udi
e
s
.
Ho
we
v
e
r
,
ha
r
m
f
u
l
C
W
a
tt
a
c
ks
f
a
il
e
d
t
o
c
o
n
s
i
d
e
r
t
h
a
t
c
a
u
s
e
s
hi
g
h
pe
r
t
u
r
b
a
t
i
o
n
t
h
a
t
wa
s
un
k
n
o
wn
to
t
h
e
h
u
m
a
n
i
n
t
e
r
ve
n
t
i
o
n
.
S
h
e
n
,
a
n
d
R
o
b
e
r
t
s
o
n
[
13]
pr
e
s
e
n
t
e
d
a
DN
N
-
b
a
s
e
d
n
e
t
wor
k
m
o
de
l
f
o
r
t
r
a
i
ni
ng
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
o
v
e
r
o
r
i
g
i
na
l
i
nput
s
.
He
r
e
,
v
a
r
i
o
us
k
i
n
ds
o
f
m
a
l
i
c
i
o
us
t
r
a
i
ni
ng
a
t
t
a
c
ks
l
i
ke
F
GSM
,
I
_F
GSM
,
R
_F
G
S
M
,
a
n
d
P
GD
we
r
e
c
o
n
s
i
de
r
e
d
f
o
r
t
h
e
tr
a
i
ni
ng
o
n
t
h
e
DN
N
m
o
de
l
.
F
o
r
t
h
e
s
i
m
u
l
a
t
i
o
n
pr
o
c
e
s
s
,
t
h
e
publ
i
c
ly
a
v
a
i
l
a
ble
M
NI
S
T
,
a
n
d
GT
S
R
B
da
t
a
b
a
s
e
s
we
r
e
ut
i
l
i
z
e
d
i
n
t
h
i
s
s
t
udy
.
I
n
t
h
e
r
e
s
u
l
t
a
n
t
pa
r
t,
s
uc
c
e
s
s
r
a
t
e
,
a
n
d
tr
a
ns
f
e
r
r
a
t
e
we
r
e
i
nv
e
s
t
i
ga
t
e
d
a
n
d
di
s
t
i
n
gu
i
s
he
d
f
r
o
m
ot
h
e
r
s
tud
i
e
s
.
B
u
t
,
t
hi
s
m
o
de
l
wa
s
o
pe
r
a
t
i
v
e
o
nl
y
w
he
n
pr
o
c
e
s
s
i
ng
s
i
n
g
l
e
a
t
t
a
c
ks
l
i
ke
F
GSM
dur
i
n
g
t
h
e
t
r
a
i
ni
ng
pr
o
c
e
s
s
.
T
h
e
de
gr
e
e
o
f
a
c
c
e
s
s
t
h
a
t
t
h
e
“
a
t
t
a
c
ke
r
”
h
a
s
to
i
nf
o
r
m
a
t
i
o
n
a
l
l
o
w
s
us
to
c
l
a
s
s
i
f
y
a
d
v
e
r
s
a
r
i
a
l
a
s
s
a
u
l
t
s
.
W
hi
l
e
a
n
a
tt
a
c
k
e
r
i
n
a
whi
t
e
-
b
o
x
s
c
e
n
a
r
i
o
c
a
n
s
e
e
a
n
d
c
h
a
n
g
e
t
h
e
tar
ge
t
m
o
d
e
l
'
s
pa
r
a
m
e
t
e
r
s
di
r
e
c
t
l
y
[
14
]
-
[
1
6]
,
a
bl
a
c
k
-
b
o
x
s
c
e
n
a
r
i
o
pr
e
v
e
n
t
s
t
h
e
m
f
r
o
m
do
i
n
g
s
o
.
T
h
us
,
i
n
m
a
ny
pr
a
c
t
i
c
a
l
c
a
s
e
s
,
bl
a
c
k
-
b
o
x
a
s
s
a
u
l
t
s
a
r
e
pr
e
f
e
r
a
bl
e
.
A
d
v
e
r
s
a
r
i
a
l
a
s
s
a
u
l
t
s
wo
u
l
d
r
e
n
d
e
r
C
NN
s
us
e
l
e
s
s
a
n
d
m
a
y
be
m
i
s
l
e
a
d
h
u
m
a
n
do
c
to
r
s
.
S
i
g
ni
f
i
c
a
n
t
l
y
,
t
hi
s
f
l
a
w
n
o
t
o
nl
y
c
r
e
a
t
e
s
s
e
r
i
o
us
s
e
c
ur
i
t
y
c
o
n
c
e
r
n
s
,
b
ut
i
t
a
l
s
o
hi
n
d
e
r
s
t
h
e
pr
a
c
t
i
c
a
l
us
e
o
f
a
uto
m
a
t
e
d
C
NN
-
b
a
s
e
d
s
y
s
t
e
m
s
,
whi
c
h
i
s
pa
r
t
i
c
u
lar
l
y
pr
o
bl
e
m
a
t
i
c
i
n
t
he
h
e
a
l
t
hc
a
r
e
i
n
du
s
t
r
y
w
h
e
r
e
pr
e
c
i
s
e
d
i
a
g
n
o
s
t
i
c
f
i
nd
i
ngs
a
r
e
c
r
i
t
i
c
a
l
f
o
r
pa
t
i
e
n
t
t
r
e
a
t
m
e
n
t
[
17
]
,
[
18]
.
Al
t
h
o
ugh
t
h
e
r
e
i
s
s
o
m
e
wr
i
t
t
e
n
m
a
t
e
r
i
a
l
o
n
a
dv
e
r
s
a
r
i
a
l
m
a
c
hi
ne
l
e
a
r
ni
ng,
t
h
e
v
a
s
t
m
a
j
o
r
i
t
y
o
f
i
t
de
a
l
s
w
i
t
h
r
e
a
l
-
wo
r
l
d
p
h
o
to
gr
a
ph
s
[
19]
.
Onl
y
a
s
m
a
l
l
n
u
m
be
r
o
f
s
t
ud
i
e
s
h
a
v
e
e
x
a
mi
ne
d
m
e
d
i
c
a
l
p
i
c
t
ur
e
s
i
n
de
pt
h
;
n
e
v
e
r
t
h
e
l
e
s
s
,
t
h
o
s
e
t
h
a
t
h
a
v
e
do
n
e
s
o
h
a
v
e
s
h
o
w
n
t
h
a
t
m
e
d
i
c
a
l
D
L
s
y
s
t
e
m
s
a
r
e
v
u
l
ne
r
a
bl
e
to
a
dv
e
r
s
a
r
i
a
l
a
s
s
a
u
l
t
s
us
i
ng
e
i
t
h
e
r
ge
n
e
r
i
c
o
r
i
m
a
g
e
-
s
pe
c
i
f
i
c
m
e
t
h
o
ds
[
20
]
,
[
21
]
.
A
l
a
r
ge
b
o
d
y
o
f
pr
i
o
r
r
e
s
e
a
r
c
h
li
nks
t
h
e
hi
g
h
d
im
e
n
s
i
o
n
a
li
t
y
o
f
t
r
a
i
ni
ng
da
t
a
o
f
t
e
n
c
o
m
pr
i
s
i
ng
i
m
a
ge
s
w
i
t
h
h
u
n
dr
e
ds
o
r
e
v
e
n
t
e
n
s
o
f
t
h
o
us
a
n
ds
o
f
p
i
x
e
l
s
t
o
t
h
e
v
u
l
ne
r
a
bil
i
t
y
t
o
a
d
v
e
r
s
a
r
i
a
l
a
tt
a
c
ks
[
22
]
,
[
23]
.
T
a
bl
e
1
pr
o
vi
de
s
t
h
e
s
e
v
e
r
a
l
po
ten
t
i
a
l
r
e
s
e
a
r
c
h
ga
ps
a
n
d
a
r
e
a
s
f
o
r
f
ur
t
h
e
r
e
x
p
l
o
r
a
t
i
o
n
.
T
a
bl
e
1.
P
ot
e
n
t
i
a
l
r
e
s
e
a
r
c
h
ga
ps
a
n
d
a
r
e
a
s
f
o
r
f
ur
t
he
r
e
x
p
l
o
r
a
t
i
o
n
A
r
e
a
R
e
s
e
a
r
c
h
ga
p
O
ppo
r
tu
ni
t
y
R
o
bus
tn
e
s
s
a
c
r
o
s
s
di
ve
r
s
e
da
ta
s
e
ts
C
ur
r
e
nt
r
e
s
e
a
r
c
h
ma
y
f
oc
us
o
n
s
p
e
c
i
f
i
c
da
ta
s
e
ts
,
la
c
ki
ng
a
c
ompr
e
he
ns
i
ve
e
v
a
lu
a
ti
o
n
a
c
r
o
s
s
di
ve
r
s
e
a
nd r
e
a
l
-
w
or
ld
da
ta
s
e
ts
.
E
x
p
l
o
r
e
th
e
r
o
bus
tn
e
s
s
of
th
e
pr
o
p
o
s
e
d
C
N
N
a
c
r
o
s
s
v
a
r
i
o
us
da
ta
s
e
ts
,
in
c
lu
di
ng
th
o
s
e
f
r
o
m
di
f
f
e
r
e
n
t
do
ma
in
s
s
uc
h
a
s
he
a
lt
hc
a
r
e
,
f
in
a
nc
e
,
a
nd
a
ut
o
n
o
m
o
us
dr
i
v
in
g,
t
o
e
ns
ur
e
it
s
e
f
f
e
c
ti
ve
n
e
s
s
i
n di
ve
r
s
e
s
c
e
na
r
i
o
s
.
E
f
f
i
c
i
e
n
c
y
a
nd
s
c
a
la
bi
li
t
y
W
hi
le
th
e
m
o
d
e
l
is
te
r
m
e
d
“
e
f
f
ic
i
e
nt
,
”
d
e
ta
il
e
d
a
na
l
y
s
is
of
it
s
c
o
mput
a
ti
o
na
l
e
f
f
i
c
i
e
nc
y
,
s
c
a
la
bi
li
t
y
,
a
nd
pe
r
f
o
r
ma
nc
e
o
n
la
r
ge
-
s
c
a
le
da
ta
s
e
ts
ma
y
b
e
l
i
mi
t
e
d.
C
o
nduc
t
th
o
r
o
ugh
e
v
a
lu
a
ti
o
ns
of
th
e
m
o
d
e
l’
s
c
o
mput
a
ti
o
na
l
r
e
qui
r
e
m
e
nt
s
a
nd
s
c
a
la
bi
li
t
y
.
E
x
pl
o
r
e
o
pt
im
i
z
a
ti
o
ns
li
k
e
m
o
de
l
c
ompr
e
s
s
io
n,
p
r
uni
ng,
a
nd
qua
nt
iz
a
ti
o
n
t
o
im
p
r
ov
e
e
f
f
i
c
i
e
nc
y
w
it
h
o
ut
s
a
c
r
i
f
i
c
in
g
r
o
bus
tn
e
s
s
.
G
e
ne
r
a
li
z
a
ti
o
n
t
o
ne
w
a
tt
a
c
k t
y
p
e
s
A
dve
r
s
a
r
ia
l
tr
a
in
in
g
me
th
o
ds
mi
ght
b
e
ta
il
or
e
d
to
s
pe
c
i
f
i
c
t
y
p
e
s
of
a
d
ve
r
s
a
r
ia
l
a
tt
a
c
ks
,
po
t
e
nt
ia
ll
y
li
mi
ti
ng
th
e
i
r
e
f
f
e
c
ti
ve
n
e
s
s
a
ga
in
s
t
ne
w
o
r
uns
e
e
n a
tt
a
c
k s
tr
a
te
gi
e
s
.
D
e
v
e
l
o
p
a
da
pt
iv
e
a
d
v
e
r
s
a
r
ia
l
tr
a
in
in
g
te
c
hni
qu
e
s
th
a
t
c
a
n
ge
n
e
r
a
li
z
e
t
o
a
w
id
e
r
a
nge
of
a
tt
a
c
k
t
y
pe
s
,
in
c
lu
di
ng
nove
l
a
nd
s
o
phi
s
ti
c
a
t
e
d
a
tt
a
c
ks
.
I
n
v
e
s
ti
ga
t
e
th
e
us
e
of
me
ta
-
l
e
a
r
ni
ng
a
nd
tr
a
ns
f
e
r
l
e
a
r
ni
ng
t
o
e
nha
n
c
e
ge
n
e
r
a
li
z
a
t
i
o
n.
E
x
p
la
in
a
bi
li
t
y
a
nd
in
te
r
p
r
e
ta
bi
l
it
y
T
h
e
in
t
e
r
p
r
e
ta
bi
l
it
y
of
C
N
N
s
,
e
s
pe
c
ia
ll
y
w
he
n
a
dve
r
s
a
r
ia
l
tr
a
in
in
g
is
in
v
o
l
ve
d,
r
e
ma
in
s
a
s
ig
ni
f
i
c
a
nt
c
ha
ll
e
ng
e
.
U
nd
e
r
s
ta
ndi
ng
h
o
w
a
nd
w
h
y
th
e
m
o
d
e
l
r
e
s
is
ts
a
dv
e
r
s
a
r
ia
l
a
tt
a
c
ks
is
c
r
u
c
ia
l
f
o
r
tr
us
t
a
nd de
pl
oy
m
e
nt
.
I
mpl
e
m
e
nt
t
e
c
hni
que
s
t
o
im
pr
ove
th
e
in
t
e
r
pr
e
ta
bi
li
t
y
a
nd
e
x
pl
a
in
a
bi
li
t
y
of
th
e
a
d
ve
r
s
a
r
ia
ll
y
t
r
a
in
e
d
C
N
N
.
U
s
e
v
is
ua
li
z
a
ti
o
n
me
th
o
ds
a
nd
mo
d
e
l
-
a
gn
o
s
ti
c
e
x
pl
a
in
a
bi
li
t
y
te
c
hni
que
s
t
o
pr
ov
id
e
in
s
ig
ht
s
in
t
o
th
e
m
o
d
e
l’
s
d
e
c
is
io
n
-
ma
ki
ng pr
oc
e
s
s
.
R
e
a
l
-
ti
m
e
a
ppl
ic
a
ti
o
n a
nd
de
pl
oy
m
e
nt
R
e
s
e
a
r
c
h
ma
y
la
c
k
a
f
o
c
us
o
n
th
e
pr
a
c
ti
c
a
l
de
pl
oy
m
e
nt
of
a
d
ve
r
s
a
r
ia
ll
y
t
r
a
in
e
d
C
N
N
s
in
r
e
a
l
-
ti
me
s
y
s
t
e
ms
w
he
r
e
la
t
e
n
c
y
a
nd
r
e
s
o
ur
c
e
c
o
ns
tr
a
in
ts
a
r
e
c
r
it
ic
a
l.
I
nve
s
ti
ga
te
th
e
d
e
pl
oy
m
e
nt
c
ha
ll
e
ng
e
s
of
a
d
ve
r
s
a
r
ia
ll
y
tr
a
in
e
d
m
o
de
ls
in
r
e
a
l
-
ti
m
e
a
ppl
i
c
a
ti
o
ns
.
E
x
pl
o
r
e
ha
r
dw
a
r
e
a
c
c
e
l
e
r
a
ti
o
ns
,
e
dg
e
c
o
mpu
ti
ng,
a
nd
r
e
a
l
-
ti
m
e
in
f
e
r
e
n
c
e
o
pt
im
i
z
a
ti
o
ns
to
f
a
c
il
i
ta
te
th
e
pr
a
c
ti
c
a
l
us
e
of
th
e
s
e
m
o
d
e
ls
i
n r
e
a
l
-
w
o
r
ld
e
n
v
i
r
o
nme
n
ts
.
2.
1.
P
r
ob
lem
s
t
at
e
m
e
n
t
I
t
i
s
e
nc
o
m
p
a
s
s
e
d
t
h
a
t
v
a
r
i
o
us
i
s
s
ue
s
h
a
v
e
be
e
n
ob
s
e
r
v
e
d
l
i
k
e
hi
g
h
t
i
m
e
c
o
m
p
l
e
xi
t
y
,
poo
r
t
r
a
i
ni
ng,
a
n
d
f
o
c
us
m
a
i
n
ly
o
n
a
s
i
n
g
l
e
v
u
l
ne
r
a
bl
e
a
tt
a
c
k.
No
wa
da
y
s
,
DL
m
o
de
l
s
a
r
e
hi
g
hly
e
f
f
e
c
t
i
v
e
i
n
c
l
a
s
s
if
yi
ng
a
dv
e
r
s
a
r
i
a
l
a
t
t
a
c
ks
c
a
us
e
d
by
d
if
f
e
r
e
n
t
a
tt
a
c
k
s
tr
a
t
e
gi
e
s
.
T
h
e
DL
m
o
de
l
s
li
ke
C
N
N,
DN
N,
a
n
d
r
e
c
ur
r
e
n
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
36
,
N
o
.
3
,
De
c
e
m
b
e
r
20
24
:
1
7
69
-
1
7
77
1772
n
e
ur
a
l
n
e
t
wo
r
k
(
R
NN
)
m
o
de
l
s
a
r
e
c
o
n
s
i
de
r
e
d
a
s
t
h
e
a
l
t
e
r
n
a
t
i
v
e
s
o
l
ut
i
o
n
t
h
a
t
pr
o
vi
d
e
s
f
a
s
c
i
n
a
t
i
n
g
pe
r
f
o
r
m
a
n
c
e
o
v
e
r
a
n
t
a
go
ni
s
t
i
c
a
tt
a
c
ks
.
He
n
c
e
,
thi
s
s
t
udy
i
n
t
r
o
duc
e
d
t
h
e
kn
o
wn
C
NN
m
o
de
l
t
o
c
l
a
s
s
i
f
y
m
u
l
t
i
p
l
e
a
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
a
t
t
a
c
ks
us
i
n
g
m
a
li
c
i
o
us
i
n
put
s
.
T
o
o
ur
kn
o
wl
e
dge
,
t
h
e
de
v
e
l
o
pe
d
m
e
t
h
o
d
o
u
t
pe
r
f
o
r
m
s
a
b
e
t
t
e
r
pe
r
f
o
r
m
a
n
c
e
by
s
o
l
vi
ng
v
a
r
i
o
us
i
s
s
ue
s
c
a
u
s
e
d
by
c
o
nv
e
n
t
i
o
n
a
l
s
c
he
m
e
s
.
3.
DE
VE
L
O
P
E
D
M
E
T
HO
D
I
n
t
hi
s
a
r
t
i
c
l
e
,
a
n
e
f
f
i
c
i
e
n
t
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
a
ga
i
ns
t
m
a
l
e
v
o
l
e
n
t
a
tt
a
c
ks
i
s
de
m
o
n
s
t
r
a
t
e
d.
T
hi
s
m
o
de
l
i
s
hi
g
hly
r
o
b
us
t
to
bl
a
c
k
-
b
o
x
m
a
li
c
i
o
us
e
x
a
m
p
l
e
s
,
i
t
i
s
pr
o
c
e
s
s
e
d
w
i
t
h
d
i
f
f
e
r
e
n
t
m
a
li
c
i
o
us
s
a
m
p
l
e
s
.
I
ni
t
i
a
ll
y
,
m
a
li
c
i
o
us
t
r
a
i
ni
ng
m
o
de
l
s
l
i
ke
F
GS,
I
-
F
GS,
De
e
p
-
F
oo
l
,
a
n
d
C
W
t
e
c
hni
qu
e
s
a
r
e
u
t
i
li
z
e
d
t
h
a
t
ge
n
e
r
a
t
e
a
dv
e
r
s
a
r
i
a
l
i
nput
us
i
n
g
t
h
e
C
NN
m
o
de
l
k
n
o
wn
to
t
h
e
a
tt
a
c
ke
r
.
3.
1.
Adver
s
a
r
ia
l
t
r
ain
in
g
m
e
t
h
od
T
h
e
c
o
n
s
i
d
e
r
e
d
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
e
d
s
c
h
e
m
e
h
a
s
t
wo
ph
a
s
e
s
:
t
h
e
c
r
e
a
t
i
o
n
o
f
m
a
li
c
i
o
u
s
c
a
s
e
s
,
a
n
d
h
o
w
t
h
e
s
c
h
e
m
e
a
c
qu
i
r
e
s
t
h
e
s
e
i
nput
s
.
I
ni
t
i
a
ll
y
,
i
t
pr
o
duc
e
s
s
e
v
e
r
a
l
m
a
l
e
v
o
l
e
n
t
i
n
put
s
a
n
d
i
nc
l
ude
s
t
h
e
m
a
s
s
upp
l
e
m
e
n
t
a
r
y
da
t
a
to
t
h
e
C
NN
a
ppr
o
a
c
h
f
o
r
e
nh
a
nc
i
n
g
t
h
e
r
o
b
us
t
n
e
s
s
a
ga
i
ns
t
m
a
li
c
i
o
us
a
tt
a
c
ks
.
T
h
e
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
s
ge
n
e
r
a
t
e
s
e
v
e
r
a
l
a
d
v
e
r
s
a
r
i
a
l
i
ns
t
a
nc
e
s
vi
a
F
GSM
,
R
-
F
GSM
,
De
e
p
-
F
o
o
l
,
a
n
d
C
W
t
e
c
h
ni
que
s
e
m
p
l
o
y
i
ng
t
h
e
C
N
N
t
h
a
t
i
s
a
c
k
n
o
w
l
e
d
ge
d
to
t
h
e
a
s
s
a
u
l
t
e
r
.
B
a
s
e
d
o
n
t
hi
s
pr
o
c
e
dur
e
,
t
h
e
r
e
l
i
a
bil
i
t
y
o
f
t
h
e
f
i
na
l
a
ppr
o
a
c
h
o
v
e
r
m
a
li
c
i
o
us
o
c
c
ur
r
e
n
c
e
s
c
a
n
b
e
e
nh
a
n
c
e
d.
T
h
e
m
a
n
e
u
ve
r
f
u
nc
t
i
o
n
o
f
t
h
e
C
NN
i
s
r
e
pr
e
s
e
n
ted
a
s
,
(
)
a
n
d
i
t
i
s
pr
o
c
e
s
s
e
d
w
i
t
h
t
h
e
a
c
t
ua
l
t
r
a
i
ni
ng
d
a
t
a
b
a
s
e
.
As
s
u
m
e
t
h
e
C
NN
m
o
de
l
,
t
h
e
a
c
t
ua
l
t
r
a
i
ni
ng
da
t
a
∈
,
wi
t
h
i
t
s
e
qu
i
v
a
l
e
n
t
c
l
a
s
s
e
s
∈
,
a
n
d
f
i
na
l
c
l
a
s
s
e
s
′
∈
,
t
h
e
n
t
h
e
o
p
t
i
mi
z
a
t
i
o
n
i
s
s
ue
s
a
r
e
t
a
c
k
l
e
d
by
ge
n
e
r
a
t
i
n
g
a
dve
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
′
c
a
n
b
e
m
a
t
h
e
m
a
t
i
c
a
ll
y
a
s
(
1)
.
′
:
′
(
,
′
)
.
.
(
′
)
=
′
(
1)
He
r
e
,
(
,
′
)
de
f
i
ne
s
t
h
e
d
i
s
t
a
n
c
e
b
e
t
we
e
n
t
h
e
a
c
t
ua
l
i
ns
t
a
n
c
e
s
a
n
d
pe
r
t
ur
b
a
t
e
d
i
n
s
t
a
n
c
e
s
′
,
(
)
r
e
pr
e
s
e
n
t
s
t
h
a
t
(
)
c
o
n
v
e
r
t
s
to
m
i
n
im
a
l
b
a
s
e
d
o
n
th
e
v
a
l
u
e
o
f
.
(
′
)
i
nd
i
c
a
t
e
s
t
h
e
C
NN
m
o
de
l
t
h
a
t
i
de
n
t
i
f
i
e
s
t
h
e
i
n
put
v
a
l
ue
.
F
o
r
c
r
e
a
t
i
n
g
t
h
e
s
e
′
,
e
v
e
r
y
a
d
v
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
i
s
de
f
i
ne
d
us
i
ng
F
GSM
,
I
-
F
GSM
,
De
e
p
-
F
o
o
l
,
a
n
d
C
W
s
c
h
e
m
e
s
.
3.
1.
1.
F
GS
M
s
c
h
e
m
e
T
hi
s
m
e
t
h
o
d
c
a
n
ge
n
e
r
a
t
e
′
by
ut
i
li
z
i
ng
∞
a
n
d
i
t
c
a
n
b
e
m
a
t
h
e
m
a
t
i
c
a
ll
y
f
o
r
m
u
l
a
t
e
d
a
s
(
2)
.
′
=
+
×
(
,
(
)
)
(
2)
H
e
r
e
,
a
n
d
r
e
p
r
e
s
e
n
t
s
t
h
e
o
b
tai
n
e
d
c
l
a
s
s
a
n
d
m
e
t
h
o
d
p
r
oc
e
s
s
f
un
c
t
i
o
n
r
e
s
p
e
c
t
i
v
e
l
y
.
Us
i
n
g
th
e
n
or
m
a
l
i
n
s
tan
c
e
,
th
e
g
r
a
di
e
n
t
d
e
s
c
e
n
t
i
s
i
m
pr
o
v
e
d
by
p
a
r
a
m
e
t
e
r
.
W
hi
l
e
′
i
s
g
e
n
e
r
a
t
e
d
by
op
t
i
m
i
z
e
r
s
.
T
hi
s
p
r
oc
e
s
s
i
s
s
i
m
p
l
e
a
n
d
p
r
o
vi
de
s
b
e
tt
e
r
pe
r
f
or
m
a
n
c
e
.
F
i
gur
e
1
i
l
l
us
tr
a
tes
th
e
w
or
kf
l
o
w
o
f
t
h
e
a
dv
e
r
s
a
r
i
a
l
s
c
r
u
t
i
ni
z
i
n
g
m
ode
l
.
F
i
gur
e
1.
W
o
r
kf
l
o
w
o
f
t
h
e
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
o
de
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
n
e
ff
icie
nt
c
onv
olut
ional
ne
ur
al
ne
t
w
or
k
f
or
ad
v
e
r
s
ar
ial
tr
aini
ng
agains
t
…
(
Sr
ini
v
a
s
A
.
V
addadi
)
1773
3.
1.
2.
I
_F
GS
M
s
c
h
e
m
e
T
hi
s
m
o
de
l
i
s
t
h
e
e
x
t
e
n
s
i
o
n
o
f
t
h
e
F
GS
M
t
e
c
hni
qu
e
.
He
r
e
,
a
s
m
a
l
l
e
r
a
m
o
u
n
t
i
s
e
v
e
n
t
ua
l
ly
c
h
a
n
g
e
d
a
n
d
c
li
ppe
d
i
ns
t
e
a
d
o
f
c
h
a
n
g
i
ng
i
n
e
a
c
h
s
t
e
p.
I
t
c
a
n
b
e
m
a
t
h
e
m
a
t
i
c
a
ll
y
i
n
t
e
r
pr
e
t
e
d
a
s
(
3)
.
′
=
−
1
′
−
(
×
(
,
(
)
)
)
(
3)
Dur
i
n
g
e
a
c
h
i
t
e
r
a
t
i
o
n
,
t
h
e
I
_F
GSM
ge
n
e
r
a
t
e
s
a
n
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
o
n
a
C
NN
m
o
de
l
.
W
h
e
n
c
o
m
p
a
r
e
d
w
i
t
h
F
GSM
,
t
h
e
I
_F
GSM
t
e
c
h
ni
que
h
a
s
b
e
t
t
e
r
pr
e
v
e
n
t
i
o
n
pe
r
f
o
r
m
a
n
c
e
b
a
s
e
d
o
n
w
hi
t
e
-
b
o
x
a
t
t
a
c
ks
.
3.
1.
3.
De
e
p
-
f
ool
s
c
h
e
m
e
T
hi
s
a
ppr
o
a
c
h
ge
n
e
r
a
t
e
s
a
n
a
n
t
a
go
ni
s
t
i
c
i
ns
t
a
nc
e
w
i
t
h
m
i
n
im
a
l
f
a
l
s
if
i
c
a
t
i
o
n
f
r
o
m
t
h
e
a
c
t
ua
l
il
l
us
t
r
a
t
i
o
n
.
I
t
c
r
e
a
t
e
s
′
vi
a
t
h
e
li
ne
a
r
a
ppr
o
xi
m
a
t
i
o
n
a
n
d
due
t
o
t
h
e
n
o
nl
i
ne
a
r
DL
m
o
de
l
s
,
t
hi
s
t
e
c
hni
que
i
s
m
o
r
e
c
o
m
p
l
e
x
t
h
a
n
F
GS
M
.
3.
1.
4.
De
e
p
-
f
ool
s
c
h
e
m
e
T
hi
s
m
o
de
l
i
s
t
h
e
C
a
r
l
i
n
i
a
t
t
a
c
k
t
h
a
t
c
r
e
a
t
e
s
100%
a
dv
e
r
s
a
r
i
a
l
e
x
a
m
p
l
e
s
by
us
i
ng
v
a
r
i
o
us
o
bj
e
c
t
i
ve
f
u
n
c
t
i
o
n
s
.
I
t
c
a
n
b
e
e
x
pr
e
s
s
e
d
a
s
g
i
ve
n
(
4)
.
(
,
′
)
+
×
(
′
)
(
4)
T
hi
s
m
o
de
l
e
v
a
l
ua
t
e
s
t
h
e
c
o
r
r
e
s
po
n
d
i
n
g
bi
na
r
y
va
l
ue
to
de
f
i
ne
a
gr
e
a
t
e
r
a
c
c
ur
a
c
y
r
a
t
e
o
n
a
tt
a
c
k
i
ns
t
a
nc
e
.
M
o
r
e
o
v
e
r
,
i
t
r
e
gu
l
a
t
e
s
t
h
e
b
o
ut
a
c
c
ur
a
c
y
r
a
t
e
a
t
t
h
e
c
o
s
t
o
f
i
nc
r
e
a
s
e
d
f
a
l
s
if
i
c
a
t
i
o
n
by
f
i
n
e
-
t
uni
n
g
s
e
l
f
-
a
s
s
ur
a
n
c
e
r
a
t
e
a
s
de
pi
c
t
e
d
(
5
)
.
(
′
)
=
[
{
(
′
)
:
≠
]
−
{
(
′
)
(
5)
He
r
e
,
i
n
d
i
c
a
t
e
s
t
h
e
a
c
t
ua
l
c
l
a
s
s
,
a
n
d
{
(
.
)
m
a
ni
pu
l
a
t
e
s
t
h
e
pr
e
l
i
mi
na
r
y
S
o
f
t
M
a
x
c
l
a
s
s
i
f
i
c
a
t
i
o
n
v
e
c
t
or
o
u
t
c
o
m
e
.
T
h
e
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
nc
e
s
o
b
t
a
i
n
e
d
by
e
v
e
r
y
m
o
de
l
a
r
e
c
o
m
bi
ne
d
a
t
t
h
e
t
r
a
i
ni
ng
s
e
t
to
f
ur
t
h
e
r
tr
a
i
n
t
h
e
C
NN
m
o
de
l
.
T
h
e
pr
o
c
e
s
s
t
h
a
t
t
a
ke
s
pl
a
c
e
i
n
t
h
e
C
NN
m
o
de
l
i
s
r
e
pr
e
s
e
n
t
e
d
a
s
,
(
)
.
T
h
e
C
NN
m
o
de
l
i
s
i
ni
t
i
a
ll
y
t
r
a
i
ne
d
w
i
t
h
t
h
e
a
c
t
ua
l
tr
a
i
ni
ng
da
t
a
b
a
s
e
.
As
s
u
m
e
t
h
e
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
′
∈
,
t
h
e
a
c
t
ua
l
c
l
a
s
s
∈
a
n
d
t
h
e
C
NN
c
l
a
s
s
e
s
′
∈
wi
t
h
t
h
e
i
r
pa
r
t
i
c
u
l
a
r
l
a
b
e
l
a
s
t
h
e
a
c
t
ua
l
c
l
a
s
s
a
r
e
f
o
r
m
u
l
a
t
e
d
a
s
g
i
v
e
n
(
6)
.
(
′
)
=
(
6)
Us
i
n
g
t
hi
s
pr
o
c
e
dur
e
,
t
h
e
C
NN
i
s
a
c
c
o
m
p
li
s
h
e
d
w
i
t
h
d
i
f
f
e
r
e
n
t
a
n
t
a
g
o
ni
s
t
i
c
c
a
s
e
s
a
n
d
f
i
na
l
ly
,
t
h
e
r
e
l
i
a
bil
i
t
y
c
o
n
t
r
a
r
y
to
un
wa
n
t
e
d
a
n
t
a
g
o
ni
s
t
i
c
i
n
s
t
a
nc
e
s
i
s
e
nh
a
n
c
e
d.
Al
go
r
i
t
hm
1
i
n
d
i
c
a
t
e
s
t
h
e
a
n
t
a
go
ni
s
t
i
c
t
r
a
i
ni
ng
pr
o
g
r
e
s
s
i
o
n
.
Al
go
r
i
t
hm
1
.
P
s
e
udo
c
o
de
f
o
r
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
e
t
h
o
d
Start:
Initialize
various
parameters
like
Actual
training
dataset
′
∈
,
Actual
class
∈
,
validation
data
,
CNN
model
,
malicious
tra
ining
models
like
fast
gradient
descent
(
FGS),
recursive
-
FGSM (I
-
FGS), Deep
-
Fool, and Carlini
-
Wagner (CW) technique.
Adversarial training models: (
,
, FGS, I_FGS, Deep
-
fool, CW)
′
←
Creation of adversarial cases (
,
, FGS)
′
←
Creation of adversarial cases (
,
, FGS)
′
←
Creation of adversarial cases (
,
, FGS)
′
←
Creation of adversarial cases (
,
, FGS)
Scrutinize the final model
←
(
,
)
+
(
′
,
)
Store the accuracy of the final CNN model
(
)
Return
Stop
4.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
h
e
de
v
e
l
o
pe
d
m
e
t
h
o
d
i
s
pr
o
c
e
s
s
e
d
a
n
d
s
i
mu
l
a
t
e
d
by
t
h
e
M
NI
S
T
[
2
4
]
b
a
s
e
d
h
a
n
dwr
i
t
t
e
n
r
e
pr
e
s
e
n
t
a
t
i
v
e
da
t
a
ba
s
e
i
s
ut
i
li
z
e
d
i
n
t
hi
s
s
t
ud
y
.
I
t
c
o
n
s
i
s
t
s
o
f
n
u
m
e
r
i
c
a
l
d
i
g
i
t
s
t
h
a
t
v
a
r
y
f
r
o
m
0
t
o
9
i
n
t
h
e
f
o
r
m
o
f
gr
e
y
s
c
a
l
e
i
m
a
ge
s
.
He
r
e
,
a
tot
a
l
o
f
60
K
t
r
a
i
ni
ng
a
n
d
10
K
t
e
s
t
i
n
g
i
m
a
ge
s
a
r
e
pr
e
s
e
n
t
.
T
h
e
i
m
a
ge
s
pr
e
s
e
n
t
i
n
t
h
e
da
t
a
b
a
s
e
a
r
e
i
n
t
h
e
s
i
z
e
28
×
28
.
S
om
e
e
xi
s
t
i
n
g
m
o
de
l
s
:
w
i
t
h
o
u
t
a
b
a
s
e
li
n
e
,
a
n
d
w
i
t
h
b
a
s
e
li
ne
m
e
t
h
o
ds
[
2
5
]
a
r
e
c
o
m
p
a
r
e
d
w
i
t
h
t
h
e
de
v
e
l
o
pe
d
a
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
t
o
p
r
o
v
e
i
t
s
e
f
f
i
c
i
e
n
c
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
36
,
N
o
.
3
,
De
c
e
m
b
e
r
20
24
:
1
7
69
-
1
7
77
1774
4.
1.
As
s
e
s
s
m
e
n
t
m
e
t
r
ics
A
t
t
a
c
k
s
uc
c
e
s
s
r
a
t
e
m
e
a
s
ur
e
s
h
o
w
e
f
f
e
c
t
i
v
e
a
n
a
d
v
e
r
s
a
r
i
a
l
a
t
t
a
c
k
i
s
i
n
f
o
o
l
i
ng
a
m
o
de
l
by
qua
n
t
i
f
yi
ng
t
h
e
pr
o
p
o
r
t
i
o
n
o
f
s
uc
c
e
s
s
f
u
l
a
t
t
a
c
k
s
.
Nu
m
b
e
r
o
f
s
u
c
c
e
s
s
f
u
l
a
tt
a
c
ks
r
e
pr
e
s
e
n
t
s
h
o
w
m
a
ny
a
dv
e
r
s
a
r
i
a
l
a
t
t
a
c
ks
we
r
e
a
bl
e
t
o
f
o
o
l
t
h
e
m
a
c
hi
n
e
l
e
a
r
ni
ng
m
o
de
l
,
c
a
us
i
ng
i
t
to
m
i
s
c
l
a
s
s
if
y
o
r
b
e
h
a
v
e
un
e
x
pe
c
t
e
d
l
y
.
Ac
c
ur
a
c
y
i
s
us
e
d
to
e
v
a
l
ua
t
e
t
h
e
o
v
e
r
a
l
l
pe
r
f
o
r
m
a
n
c
e
o
f
a
c
l
a
s
s
if
i
c
a
t
i
o
n
m
o
de
l
by
de
t
e
r
m
i
n
i
ng
t
h
e
pr
o
p
or
t
i
o
n
o
f
c
o
r
r
e
c
t
l
y
c
l
a
s
s
if
i
e
d
i
n
s
t
a
n
c
e
s
(
b
o
t
h
p
o
s
i
t
i
v
e
a
n
d
ne
g
a
t
i
v
e
)
o
u
t
o
f
t
h
e
tot
a
l
n
u
m
be
r
o
f
i
ns
t
a
nc
e
s
.
(
%
)
=
(
)
×
100
(
7)
=
+
+
+
+
(
8)
He
r
e
,
,
,
,
a
n
d
i
n
d
i
c
a
t
e
s
t
h
e
t
r
ue
n
e
ga
t
i
v
e
(
T
N
)
,
t
r
ue
p
o
s
i
t
i
v
e
(
T
P
)
,
f
a
l
s
e
n
e
ga
t
i
v
e
(
F
N)
,
a
n
d
f
a
l
s
e
po
s
i
t
i
v
e
(
F
P
)
r
e
s
pe
c
t
i
v
e
ly
.
4.
2.
S
im
u
l
at
ion
an
al
ys
is
of
d
e
ve
l
op
e
d
m
e
t
h
od
ove
r
c
on
ve
n
t
ion
al
s
c
h
e
m
e
s
I
n
t
hi
s
s
e
c
t
i
o
n
,
t
h
e
pe
r
f
o
r
m
a
n
c
e
a
c
hi
e
v
e
d
by
t
h
e
de
v
e
l
o
pe
d
m
e
t
h
o
d
i
s
a
n
a
ly
z
e
d
vi
a
gr
a
phi
c
a
l
i
n
t
e
r
p
r
e
t
a
t
i
o
n
.
He
r
e
,
t
h
e
a
tt
a
c
k
s
uc
c
e
s
s
r
a
t
e
a
nd
a
c
c
ur
a
c
y
a
r
e
a
n
a
ly
z
e
d
a
n
d
c
o
m
p
a
r
e
d
w
i
t
h
t
r
a
d
i
t
i
o
n
a
l
m
a
li
c
i
o
us
s
c
r
ut
i
ni
z
i
ng
m
o
de
l
s
.
T
h
e
de
t
a
i
l
e
d
a
n
a
ly
s
i
s
o
f
t
h
e
o
b
t
a
i
n
e
d
pe
r
f
o
r
m
a
nc
e
i
s
de
p
i
c
t
e
d
b
e
l
o
w.
F
i
gur
e
2
il
l
us
t
r
a
t
e
s
t
h
e
a
tt
a
c
k
s
uc
c
e
s
s
r
a
t
e
a
n
a
ly
s
i
s
o
v
e
r
g
e
ne
r
a
t
e
d
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
w
i
t
h
d
i
f
f
e
r
e
n
t
t
e
c
h
ni
qu
e
s
.
F
i
gur
e
2.
A
tt
a
c
k
s
uc
c
e
s
s
r
a
t
e
(
A
S
R
)
a
n
a
ly
s
i
s
o
ve
r
ge
n
e
r
a
t
e
d
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
w
i
t
h
d
i
f
f
e
r
e
n
t
t
e
c
hni
que
s
W
hil
e
c
o
ns
i
de
r
i
n
g
t
h
e
w
i
t
h
o
ut
m
o
de
l
a
s
t
h
e
f
i
na
l
m
o
de
l
do
e
s
n
ot
pe
r
f
o
r
m
a
ny
de
f
e
n
s
i
ve
m
e
c
h
a
ni
s
m
o
v
e
r
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
o
de
l
.
W
hil
e
c
o
n
s
i
d
e
r
i
n
g
t
h
e
b
a
s
e
l
i
ne
m
o
de
l
a
s
a
f
i
na
l
m
o
de
l
f
o
r
t
r
a
i
ni
ng
t
h
e
a
dv
e
r
s
a
r
i
a
l
s
a
m
p
l
e
s
,
i
t
c
a
n
us
e
o
n
l
y
s
i
ng
l
e
t
y
pe
o
f
a
tt
a
c
k
m
o
de
l
n
a
m
e
ly
F
GSM
.
W
hil
e
c
o
n
s
i
de
r
i
n
g
t
h
e
de
v
e
l
o
pe
d
t
r
a
i
ni
ng
m
o
de
l
a
s
t
h
e
f
i
na
l
m
o
d
e
l
,
i
t
c
a
n
c
r
e
a
t
e
v
a
r
i
o
us
a
dve
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
us
i
ng
F
GS,
I
-
F
GS,
De
e
p
-
F
o
o
l
,
a
n
d
C
W
t
e
c
h
ni
que
s
.
T
h
e
a
tt
a
c
k
s
uc
c
e
s
s
r
a
t
e
f
o
r
t
h
e
wi
t
h
o
ut
m
e
t
h
o
d
m
i
s
c
l
a
s
s
if
i
e
d
l
a
r
ge
r
t
h
a
n
89.
88%
o
f
t
h
e
m
a
li
c
i
o
u
s
i
ns
t
a
n
c
e
s
.
T
h
o
ugh
,
t
h
e
de
v
e
l
o
pe
d
m
o
de
l
mi
n
im
i
z
e
s
t
h
e
A
S
R
t
o
32
.
88%
a
n
d
t
h
e
m
e
a
n
c
o
m
put
a
t
i
o
n
i
s
i
nv
e
s
t
i
ga
t
e
d
to
i
m
pr
o
v
e
m
o
r
e
t
h
a
n
44.
79%
di
s
t
i
n
gu
i
s
h
e
d
f
r
o
m
t
h
e
b
a
s
e
li
n
e
m
o
de
l
.
He
n
c
e
,
i
t
i
s
c
l
e
a
r
t
h
a
t
t
h
e
C
NN
m
o
de
l
(
f
i
na
l
m
o
d
e
l
)
i
s
hi
g
hly
vi
go
r
o
us
a
ga
i
ns
t
m
a
l
i
c
i
o
us
a
t
t
a
c
ks
.
F
i
gur
e
3
de
p
i
c
t
s
t
h
e
a
c
c
ur
a
c
y
a
n
a
ly
s
i
s
o
n
t
e
s
t
i
n
g
w
i
t
h
M
NI
S
T
da
t
a
b
a
s
e
.
E
v
e
n
t
h
o
ugh
,
t
h
e
a
dd
i
t
i
o
n
a
l
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
nc
e
s
a
r
e
tr
a
i
ne
d,
t
h
e
C
NN
m
o
de
l
s
t
i
ll
r
e
m
a
i
ns
t
h
e
s
i
mi
l
a
r
a
c
c
ur
a
c
y
c
o
n
c
e
r
ni
ng
t
h
e
a
c
t
ua
l
da
t
a
.
I
t
i
s
n
o
t
e
d
t
h
a
t
t
h
e
c
o
n
s
i
de
r
e
d
a
d
v
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
s
c
h
e
m
e
a
c
hi
e
v
e
d
a
lm
o
s
t
s
i
mi
l
a
r
pe
r
f
o
r
m
a
n
c
e
w
h
e
n
c
o
m
pa
r
e
d
wi
t
h
t
h
e
w
i
t
h
o
u
t
m
o
de
l
a
n
d
b
a
s
e
li
ne
m
o
de
l
dur
i
n
g
t
h
e
tes
t
i
n
g
pr
o
c
e
s
s
.
4.
3.
Ou
t
c
om
e
an
al
ys
is
I
n
t
hi
s
s
e
c
t
i
o
n
,
t
h
e
o
u
t
c
o
m
e
s
a
c
hi
e
v
e
d
by
d
if
f
e
r
e
n
t
t
e
c
h
ni
que
s
f
o
r
t
h
e
C
NN
c
o
n
s
t
r
uc
t
e
d
o
n
t
h
e
M
NI
S
T
da
t
a
b
a
s
e
a
r
e
a
n
a
ly
z
e
d
.
F
i
gur
e
4
s
h
o
ws
t
h
e
o
u
t
c
o
m
e
s
o
b
t
a
i
n
e
d
f
r
o
m
e
a
c
h
i
ns
t
a
n
c
e
.
F
i
gur
e
4
(
a
)
to
4
(
e
)
i
nd
i
c
a
t
e
s
t
h
e
o
r
i
g
i
na
l
im
a
ge
,
F
GSM
,
I
-
F
GSM
,
D
e
e
p
-
f
o
o
l
,
a
n
d
C
W
m
e
t
h
o
ds
r
e
s
pe
c
t
i
v
e
ly
.
I
t
i
s
n
o
t
e
d
t
h
a
t
F
GSM
,
I
-
F
GSM
,
a
n
d
De
e
p
-
f
o
o
l
s
h
o
ws
gr
e
a
t
e
r
di
s
t
o
r
t
i
o
n
s
c
o
m
p
a
r
e
d
to
t
h
e
C
W
m
o
de
l
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
n
e
ff
icie
nt
c
onv
olut
ional
ne
ur
al
ne
t
w
or
k
f
or
ad
v
e
r
s
ar
ial
tr
aini
ng
agains
t
…
(
Sr
ini
v
a
s
A
.
V
addadi
)
1775
F
i
gur
e
3.
A
c
c
ur
a
c
y
a
n
a
ly
s
i
s
o
n
t
e
s
t
i
n
g
w
i
t
h
M
NI
S
T
da
t
a
b
a
s
e
(
a
)
(
b
)
(
c
)
(
d)
(
e
)
F
i
gur
e
4.
Ou
t
c
o
m
e
s
o
b
t
a
i
n
e
d
f
r
o
m
e
a
c
h
i
ns
t
a
n
c
e
:
(
a
)
o
r
i
g
i
na
l
im
a
g
e
,
(
b
)
F
GSM
,
(
c
)
I
-
F
GSM
,
(
d
)
de
e
p
-
f
o
o
l
,
a
n
d
(
e
)
C
W
5.
CONC
L
USI
ON
T
h
e
de
v
e
l
o
pe
d
m
e
t
h
o
d
de
m
o
n
s
t
r
a
t
e
d
a
n
d
i
nv
e
s
t
i
ga
t
e
d
a
r
o
b
us
t
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
s
c
h
e
m
e
f
o
r
m
a
li
c
i
o
us
c
l
a
s
s
if
i
c
a
t
i
o
n
a
t
t
a
c
ks
.
I
n
t
hi
s
s
t
ud
y
,
t
h
e
m
a
li
c
i
o
us
i
ns
t
a
n
c
e
s
a
r
e
i
ni
t
i
a
ll
y
c
r
e
a
t
e
d
us
i
n
g
v
a
r
i
o
u
s
a
ppr
o
a
c
h
e
s
s
uc
h
a
s
I
-
F
GS
M
,
De
e
p
-
F
o
o
l
,
F
GSM
,
a
n
d
C
W
a
n
d
t
h
e
n
t
r
a
i
n
t
h
e
C
NN
m
o
d
e
l
m
a
k
i
ng
i
t
m
o
r
e
r
e
l
i
a
bl
e
a
ga
i
ns
t
un
kn
o
wn
m
a
l
e
v
o
l
e
n
t
a
tt
a
c
ks
.
Ou
tco
m
e
s
o
b
t
a
i
n
e
d
by
t
h
e
s
e
m
o
de
l
s
a
r
e
di
s
t
or
t
e
d
to
v
a
r
i
o
us
de
gr
e
e
s
.
I
t
i
s
c
o
n
c
l
ud
e
d
t
h
a
t
a
m
o
n
g
I
-
F
GSM
,
De
e
p
-
F
oo
l
,
F
GSM
,
a
n
d
C
W
m
o
de
l
s
,
t
h
e
C
W
t
e
c
hni
qu
e
c
r
e
a
t
e
s
a
dv
e
r
s
a
r
i
a
l
i
ns
t
a
n
c
e
s
w
i
t
h
l
e
s
s
e
r
d
i
s
t
o
r
t
i
o
n
.
M
o
r
e
ov
e
r
,
t
h
e
A
da
m
o
pt
i
m
i
z
e
r
(
A
O)
i
s
ut
i
l
i
z
e
d
to
m
i
nim
i
z
e
t
h
e
n
e
t
wo
r
k
c
o
m
p
l
e
xi
t
i
e
s
by
c
o
n
s
i
de
r
i
ng
c
r
o
s
s
e
n
t
r
o
p
y
a
s
t
h
e
o
bj
e
c
t
i
v
e
f
u
nc
t
i
o
n
.
T
h
e
e
x
pe
r
i
m
e
n
t
s
a
r
e
c
a
r
r
i
e
d
o
u
t
i
n
t
h
e
P
y
t
h
o
n
p
l
a
t
f
o
r
m
us
i
n
g
t
h
e
publ
i
c
ly
a
va
i
labl
e
M
NI
S
T
da
t
a
b
a
s
e
.
I
n
t
h
e
e
x
pe
r
i
m
e
n
t
a
l
s
e
c
t
i
o
n
,
t
h
e
a
dv
e
r
s
a
r
i
a
l
t
r
a
i
ni
ng
m
o
de
l
r
e
duc
e
s
t
h
e
A
S
R
by
a
n
a
v
e
r
a
ge
o
f
27.
19%
f
o
r
di
f
f
e
r
e
n
t
m
a
l
i
c
i
o
u
s
i
npu
t
s
,
whi
l
e
pr
e
s
e
r
vi
ng
t
h
e
a
c
c
ur
a
c
y
o
f
98.
89%
c
o
n
c
e
r
ni
ng
a
c
t
ua
l
im
a
ge
s
i
n
t
h
e
M
NI
S
T
da
t
a
b
a
s
e
.
T
h
e
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
o
de
l
i
s
hi
g
hly
us
e
f
u
l
i
n
a
ut
o
n
om
o
us
v
e
hi
c
u
l
a
r
a
pp
l
i
c
a
t
i
o
ns
,
a
n
d
m
e
d
i
c
a
l
a
pp
li
c
a
t
i
o
ns
.
Ho
we
v
e
r
,
t
h
e
r
e
a
r
e
m
a
ny
o
t
h
e
r
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
e
c
h
a
ni
s
m
s
a
n
d
i
t
i
s
f
a
il
e
d
t
o
c
o
n
s
i
de
r
i
n
t
hi
s
s
t
udy
.
I
n
f
ut
ur
e
s
t
udi
e
s
,
m
o
r
e
a
dv
e
r
s
a
r
i
a
l
s
c
r
ut
i
ni
z
i
ng
m
e
c
h
a
ni
s
m
s
w
i
ll
b
e
i
n
t
r
o
duc
e
d
by
c
o
n
s
i
de
r
i
n
g
l
a
r
ge
r
da
t
a
s
e
t
s
,
a
n
d
t
h
e
i
r
pe
r
f
o
r
m
a
n
c
e
i
s
a
n
a
ly
z
e
d.
AC
K
NOWL
E
DGE
M
E
NT
S
T
h
e
A
ut
h
o
r
wi
t
h
a
de
e
p
s
e
ns
e
o
f
gr
a
t
i
t
ude
wo
ul
d
t
h
a
n
k
t
h
e
s
upe
r
vi
s
o
r
f
o
r
hi
s
gu
i
da
n
c
e
a
n
d
c
o
n
s
t
a
n
t
s
uppo
r
t
r
e
n
de
r
e
d
dur
i
n
g
t
hi
s
r
e
s
e
a
r
c
h
.
RE
F
E
R
E
NC
E
S
[
1]
B
.
C
he
n,
J
.
Y
in
,
S
.
C
he
n,
B
.
C
he
n,
a
nd
X
.
L
iu
,
“
A
n
a
da
pt
iv
e
m
o
d
e
l
e
ns
e
mbl
e
a
d
v
e
r
s
a
r
ia
l
a
tt
a
c
k
f
o
r
b
oo
s
ti
ng
a
d
v
e
r
s
a
r
ia
l
tr
a
ns
f
e
r
a
bi
li
t
y
,
”
I
n
P
r
oc
e
e
di
ngs
of
th
e
I
E
E
E
/C
V
F
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
is
io
n
,
pp.
4489
-
4498.
2023,
do
i:
10.1109/i
c
c
v
51070
.2023.00414.
[
2]
A
.
C
ha
kr
a
bo
r
t
y
,
M
.
A
la
m,
V
.
D
e
y
,
A
.
C
ha
tt
o
pa
dh
y
a
y
,
a
nd
D
.
M
ukho
pa
dh
y
a
y
,
“
A
s
ur
ve
y
o
n
a
d
ve
r
s
a
r
ia
l
a
tt
a
c
ks
a
nd
d
e
f
e
nc
e
s
,
”
C
A
A
I
T
r
ans
ac
ti
ons
on I
nt
e
ll
ig
e
nc
e
T
e
c
hnol
ogy
, v
o
l.
6, n
o
. 1, p
p. 25
-
45, 2021, do
i
:
10.1049/c
it
2.12028.
[
3]
X
.
J
ia
,
Y
.
Z
ha
ng,
B
.
W
u,
K
.
M
a
,
J
.
W
a
ng,
a
nd
X
.
C
a
o
,
“
L
A
S
-
A
T
:
a
dve
r
s
a
r
ia
l
t
r
a
in
in
g
w
it
h
le
a
r
na
bl
e
a
tt
a
c
k
s
tr
a
t
e
g
y
,
”
I
n
P
r
oc
e
e
di
ngs
of
th
e
I
E
E
E
/C
V
F
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
is
io
n
and
P
at
te
r
n
R
e
c
ogni
ti
on
,
pp.
13398
-
13408.
2022,
do
i:
10.1109/c
v
pr
52688.2022.01304.
[
4]
Y
.
X
ia
o
,
a
nd
C
.
M.
P
un,
“
I
mpr
ov
in
g
a
d
v
e
r
s
a
r
ia
l
a
tt
a
c
ks
o
n
de
e
p
ne
ur
a
l
ne
twor
ks
v
ia
c
o
ns
tr
i
c
t
e
d
gr
a
di
e
n
t
-
ba
s
e
d
pe
r
tu
r
ba
ti
ons
,
”
I
nf
or
m
at
io
n Sc
ie
n
c
e
s
, v
o
l.
571, pp. 104
-
132, 2021, d
o
i:
10.101
6/
j.
in
s
.2021.04.033.
[
5]
C
.
Z
ha
ng,
X
.
C
o
s
ta
-
P
e
r
e
z
,
a
nd
P
.
P
a
t
r
a
s
,
“
A
dve
r
s
a
r
ia
l
a
tt
a
c
ks
a
ga
in
s
t
de
e
p
l
e
a
r
ni
ng
-
ba
s
e
d
ne
twor
k
in
t
r
us
io
n
d
e
te
c
ti
o
n
s
y
s
t
e
ms
a
nd
de
f
e
ns
e
me
c
ha
ni
s
ms
,
”
I
E
E
E
/
A
C
M
T
r
ans
ac
ti
ons
on
N
e
tw
or
k
in
g
,
vo
l.
30,
n
o
.
3,
pp.
1294
-
1311,
2022,
do
i:
10.1109/
T
N
E
T
.2021.3137084.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
36
,
N
o
.
3
,
De
c
e
m
b
e
r
20
24
:
1
7
69
-
1
7
77
1776
[
6]
J
.
G
u,
H
.
Z
ha
o
,
V
.
T
r
e
s
p,
a
nd
P
.
H
.
T
or
r
,
“
S
E
G
P
G
D
:
a
n
e
f
f
e
c
ti
v
e
a
nd
e
f
f
i
c
i
e
nt
a
dve
r
s
a
r
ia
l
a
tt
a
c
k
f
o
r
e
v
a
lu
a
ti
ng
a
nd
boo
s
ti
ng
s
e
gme
nt
a
ti
o
n
r
o
bus
tn
e
s
s
,
”
I
n
E
ur
ope
an
C
on
f
e
r
e
nc
e
on
C
om
put
e
r
V
is
io
n,
C
ham
:
Sp
r
in
ge
r
N
at
ur
e
Sw
it
z
e
r
la
nd
,
pp.
308
-
325,
2
022,
do
i:
10.1007/978
-
3
-
031
-
19818
-
2_18.
[
7]
M
.
R
.
V
e
mpa
r
a
la
e
t
al
.,
“
Br
e
a
ki
ngbe
d:
B
r
e
a
ki
ng
bi
na
r
y
a
nd
e
f
f
i
c
i
e
nt
d
e
e
p
n
e
ur
a
l
n
e
tw
o
r
ks
b
y
a
d
v
e
r
s
a
r
ia
l
a
tt
a
c
ks
,
”
I
n
I
nt
e
ll
i
ge
nt
Sy
s
te
m
s
and A
ppl
ic
at
io
ns
:
P
r
oc
e
e
di
ngs
of
t
he
2021 I
nt
e
ll
ig
e
nt
Sy
s
te
m
s
C
on
fe
r
e
n
c
e
(
I
nt
e
ll
iS
y
s
)
, Spr
in
ge
r
I
nt
e
r
nat
io
nal
P
ubl
is
h
in
g
,
pp. 148
-
167, 2022,
do
i
:
10.1007/978
-
3
-
030
-
82193
-
7
10.
[
8]
S
. A
dde
pa
ll
i,
a
nd S
. J
a
in
,
“
E
f
f
i
c
i
e
nt
a
nd
e
f
f
e
c
ti
ve
a
ugme
nt
a
ti
o
n
s
tr
a
te
g
y
f
or
a
d
ve
r
s
a
r
ia
l
t
r
a
in
in
g,
”
A
dv
anc
e
s
i
n N
e
ur
al
I
nf
or
m
at
io
n
P
r
oc
e
s
s
in
g Sy
s
te
m
s
, v
ol
. 35, pp. 1488
-
1501, 2022, d
o
i:
10.4018
/i
js
w
is
.297038.
[
9]
N
.
M
a
ni
,
M
.
M
o
h,
a
nd
T
.
S
.
M
o
h,
“
D
e
f
e
ndi
ng
d
e
e
p
l
e
a
r
ni
ng
m
o
de
ls
a
ga
in
s
t
a
d
ve
r
s
a
r
ia
l
a
tt
a
c
ks
,
”
I
nt
e
r
nat
io
nal
J
our
nal
of
Sof
tw
ar
e
Sc
ie
nc
e
and C
om
put
at
io
nal
I
nt
e
ll
ig
e
nc
e
(
I
J
SSC
I
)
, vo
l.
13, no
. 1, pp. 72
-
89, 2021, d
o
i
:
0.4018/I
J
S
S
C
I
.2021010105.
[
10]
S
.
L
a
l
e
t
al
.,
“
A
dv
e
r
s
a
r
ia
l
a
tt
a
c
k
a
nd
de
f
e
n
c
e
th
r
o
ugh
a
dve
r
s
a
r
i
a
l
tr
a
in
in
g
a
nd
f
e
a
tu
r
e
f
us
io
n
f
or
di
a
be
ti
c
r
e
ti
n
o
pa
th
y
r
e
c
o
gni
t
i
o
n,
”
Se
ns
or
s
, v
ol
. 21, n
o
. 11, pp. 3922, 2021, d
o
i:
10.3390/s
21113922.
[
11]
M
.
P
a
l,
A
.
J
a
ti
,
R
.
P
e
r
i,
C
.
C
.
H
s
u,
W
.
A
bdA
lm
a
ge
e
d,
a
nd
S
.
N
a
r
a
y
a
na
n,
“
A
dve
r
s
a
r
ia
l
de
f
e
ns
e
f
or
d
e
e
p
s
pe
a
k
e
r
r
e
c
o
gni
ti
o
n
us
in
g
h
y
br
id
a
dve
r
s
a
r
ia
l
t
r
a
in
in
g,
”
I
n
I
C
A
SS
P
2021
-
2021
I
E
E
E
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
A
c
ous
ti
c
s
,
Spe
e
c
h
and
Si
gnal
P
r
oc
e
s
s
in
g
(
I
C
A
SS
P
)
, I
E
E
E
, pp. 6164
-
6168, 2021, d
oi
:
10.1109/I
C
A
S
S
P
3
9728.2021.9414843.
[
12]
A
.
S
.
H
a
s
he
mi
,
a
nd
S
.
M
oz
a
f
f
a
r
i,
“
C
N
N
a
dve
r
s
a
r
ia
l
a
tt
a
c
k
mi
ti
ga
ti
o
n
us
in
g
p
e
r
tu
r
b
e
d
s
a
mpl
e
s
tr
a
in
in
g,
”
M
ul
ti
m
e
di
a
T
ool
s
and
A
ppl
ic
at
io
ns
, vo
l.
80, pp.
22077
-
22095, 2021, d
o
i:
10.1007/s
11
042
-
020
-
10379
-
6.
[
13]
J
.
S
he
n,
a
nd
N
.
R
o
be
r
ts
o
n,
“
B
B
A
S
:
t
o
w
a
r
ds
la
r
ge
s
c
a
l
e
e
f
f
e
c
ti
v
e
e
ns
e
mbl
e
a
d
ve
r
s
a
r
ia
l
a
tt
a
c
ks
a
ga
in
s
t
de
e
p
n
e
ur
a
l
n
e
t
w
o
r
k
le
a
r
ni
ng,
”
I
nf
or
m
at
io
n Sc
ie
nc
e
s
, v
o
l.
569, pp. 469
-
478, 2021, d
o
i:
10.1016/j
.
in
s
.2020.11.026.
[
14]
K
.
X
u,
G
.
Z
ha
ng,
S
.
L
iu
,
Q
.
F
a
n,
M
.
S
un,
H
.
C
h
e
n,
P
.
-
Y
.
C
he
n,
Y
.
W
a
ng,
a
nd
X
.
L
in
,
“
A
d
v
e
r
s
a
r
ia
l
t
-
s
hi
r
t
!
e
v
a
di
ng
p
e
r
s
o
n
de
t
e
c
t
or
s
i
n a
ph
y
s
i
c
a
l
w
or
ld
,
”
in
P
r
oc
e
e
di
ngs
of
t
he
E
ur
op
e
an
C
onf
e
r
e
nc
e
on C
om
put
e
r
V
is
io
n
,
2020, pp. 665
–
681.
[
15]
X
. M
a
, Y
. N
iu
, L
.
G
u, Y
. W
a
ng, Y. Z
ha
o
, J
. B
a
il
e
y
, a
nd F
.
L
u,
“
U
nde
r
s
ta
ndi
ng a
dve
r
s
a
r
ia
l
a
tt
a
c
ks
o
n d
e
e
p l
e
a
r
ni
ng
ba
s
e
d m
e
d
ic
a
l
im
a
ge
a
na
l
y
s
is
s
y
s
t
e
ms
,
”
P
a
tt
e
r
n R
e
c
ogni
ti
on
,
vo
l.
107, 2020,
A
r
t.
no
. 107332.
[
16]
X
.
L
i
a
nd D
.
Z
hu,
“
R
o
bus
t
d
e
t
e
c
ti
o
n
of
a
d
v
e
r
s
a
r
ia
l
a
tt
a
c
ks
o
n
me
di
c
a
l
im
a
g
e
s
,
”
in
P
r
oc
e
e
di
ngs
of
t
he
I
nt
e
r
nat
io
nal
Sy
m
pos
iu
m
on
B
io
m
e
di
c
al
I
m
agi
ng
(
I
S
B
I
)
, 2020, pp
. 1154
–
1158.
[
17]
M
.
P
a
s
c
ha
li
,
S
.
C
o
nj
e
ti
,
F
.
N
a
v
a
r
r
o
,
a
nd
N
.
N
a
v
a
b,
“
G
e
n
e
r
a
li
z
a
bi
li
t
y
v
s
.
r
o
bus
tn
e
s
s
:
in
ve
s
ti
ga
ti
ng
m
e
di
c
a
l
im
a
gi
ng
n
e
tw
o
r
ks
us
in
g
a
dv
e
r
s
a
r
ia
l
e
x
a
mpl
e
s
,
”
in
P
r
oc
e
e
di
ngs
o
f
th
e
I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on
M
e
di
c
al
I
m
age
C
om
put
in
g
and
C
om
put
e
r
-
A
s
s
is
te
d I
nt
e
r
v
e
nt
io
n (
M
I
C
C
A
I
)
,
2018, pp. 493
–
501.
[
18]
S
.
G
.
F
in
la
y
s
o
n,
J
.
D
.
B
o
w
e
r
s
,
J
.
I
to
,
J
.
L
.
Z
it
t
r
a
in
,
A
.
L
.
B
e
a
m,
a
nd
I
.
S
.
K
o
ha
n
e
,
“
A
dve
r
s
a
r
ia
l
a
tt
a
c
ks
o
n
me
di
c
a
l
ma
c
hi
ne
le
a
r
ni
ng,
”
Sc
ie
nc
e
, vo
l.
363, n
o
. 6433, pp. 1287
–
1289, 2019.
[
19]
A
.
M
a
dr
y
,
A
.
M
a
ke
l
ov
,
L
.
S
c
hmi
dt
,
D
.
T
s
ip
r
a
s
,
a
nd
A
.
V
la
du,
“
T
o
w
a
r
ds
de
e
p
l
e
a
r
ni
ng
m
o
d
e
ls
r
e
s
is
ta
nt
to
a
d
ve
r
s
a
r
ia
l
a
tt
a
c
ks
,
”
in
P
r
oc
e
e
di
ngs
of
t
he
I
nt
e
r
nat
io
nal
C
on
f
e
r
e
nc
e
on L
e
ar
ni
ng R
e
pr
e
s
e
nt
at
io
ns
(
I
C
L
R
)
,
2018.
[
20]
X
.
H
a
n,
Y
.
H
u,
L
.
F
o
s
c
hi
ni
,
L
.
C
hi
ni
t
z
,
L
.
J
a
nke
ls
o
n,
a
nd
R
.
R
a
n
ga
na
th
,
“
D
e
e
p
le
a
r
ni
ng
m
o
de
ls
f
or
e
l
e
c
tr
oc
a
r
di
o
g
r
a
ms
a
r
e
s
us
c
e
pt
ib
l
e
t
o
a
d
v
e
r
s
a
r
ia
l
a
tt
a
c
k,
”
N
at
ur
e
M
e
di
c
in
e
,
v
ol
. 26, pp.
1
–
4, 2020.
[
21]
T
.
K
.
Y
oo
a
nd J
. Y
. C
ho
i,
“
O
ut
c
o
m
e
s
of
a
d
ve
r
s
a
r
ia
l
a
tt
a
c
ks
o
n
de
e
p l
e
a
r
ni
ng m
o
d
e
ls
f
o
r
o
ph
th
a
lm
o
l
o
g
y
i
ma
gi
ng d
o
ma
in
s
,
”
J
A
M
A
O
pht
hal
m
ol
ogy
,
vo
l.
138, n
o
. 11, pp. 1213
–
1215, 2020.
[
22]
A
.
C
ha
kr
a
bo
r
t
y
,
M
.
A
la
m,
V
.
D
e
y
,
A
.
C
ha
tt
o
pa
dh
y
a
y
,
a
nd
D
.
M
ukho
pa
dh
y
a
y
,
“
A
d
v
e
r
s
a
r
ia
l
a
tt
a
c
ks
a
nd
de
f
e
n
c
e
s
:
a
s
ur
ve
y
,
”
ar
X
iv
pr
e
pr
in
t
ar
X
iv
:
1810.00069
, 2018.
[
23]
C
.
X
ie
,
Y
.
W
u,
L
.
v
a
n
d
e
r
M
a
a
te
n
,
A
.
L
.
Y
ui
ll
e
,
a
nd
K
.
H
e
,
“
F
e
a
tu
r
e
d
e
n
o
is
in
g
f
or
im
pr
ov
in
g
a
d
ve
r
s
a
r
ia
l
r
o
bus
tn
e
s
s
,
”
in
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
C
onf
e
r
e
nc
e
on C
o
m
put
e
r
V
is
io
n and
P
a
tt
e
r
n R
e
c
ogni
ti
on (
C
V
P
R
)
,
2019, pp. 501
–
509.
[
24]
H
.
K
h
o
da
ba
khs
h
“
M
N
I
S
T
da
ta
s
e
t,
”
K
aggl
e
,
2018,
ht
tp
s
:/
/ww
w
.ka
ggl
e
.c
o
m
/d
a
ta
s
e
ts
/h
o
j
ja
tk
/m
ni
s
t
-
da
ta
s
e
t
[
25]
H
.
K
w
o
n, a
nd J
.
L
e
e
,
“
D
i
ve
r
s
it
y
a
d
v
e
r
s
a
r
ia
l
tr
a
in
in
g a
ga
in
s
t
a
dve
r
s
a
r
ia
l
a
tt
a
c
k
o
n d
e
e
p n
e
ur
a
l
n
e
tw
o
r
ks
,
”
Sy
m
m
e
tr
y
, v
o
l.
13, n
o
. 3,
pp. 428, 2021, do
i:
10.3390/s
y
m13030428.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Sri
n
i
v
a
s
A
.
V
a
dda
di
i
s
a
d
y
n
a
m
i
c
a
n
d
fo
r
w
ard
-
t
h
i
n
k
i
n
g
p
ro
f
e
s
s
i
o
n
a
l
i
n
t
h
e
fi
el
d
o
f
C
l
o
u
d
a
n
d
D
ev
S
ec
O
p
s
.
W
i
t
h
a
s
o
l
i
d
ed
u
c
at
i
o
n
al
fo
u
n
d
at
i
o
n
i
n
c
o
m
p
u
t
e
r
s
c
i
e
n
ce,
he
em
b
ark
ed
o
n
a
j
o
u
r
n
ey
o
f
co
n
t
i
n
u
o
u
s
l
e
ar
n
i
n
g
an
d
p
ro
fe
s
s
i
o
n
al
g
ro
w
t
h
.
T
h
e
i
r
r
e
l
en
t
l
e
ss
p
u
rs
u
i
t
o
f
k
n
o
w
l
ed
g
e
a
n
d
co
mm
i
t
me
n
t
t
o
s
t
ay
i
n
g
at
t
h
e
fo
re
fr
o
n
t
o
f
i
n
d
u
s
t
r
y
ad
v
an
cemen
t
s
h
as
e
ar
n
e
d
t
h
em
r
ec
o
g
n
i
t
i
o
n
as
a
t
h
o
u
g
h
t
l
e
a
d
e
r
i
n
t
h
e
Cl
o
u
d
an
d
D
ev
S
ec
O
p
s
s
p
ac
e
.
H
e
c
an
b
e
co
n
t
a
c
t
ed
at
em
ai
l
:
V
s
ad
9
3
@
g
m
ai
l
.
co
m
.
Sa
n
ja
i
k
a
n
t
h
E
.
V
a
da
k
k
et
hi
l
So
m
a
n
a
th
a
n
P
i
l
l
a
i
h
o
l
d
s
an
MS
i
n
S
o
ft
w
ar
e
E
n
g
i
n
ee
r
i
n
g
fr
o
m
T
h
e
U
n
i
v
e
rs
i
t
y
o
f
T
e
x
as
at
A
u
s
t
i
n
,
T
e
x
as
,
U
SA
,
a
n
d
a
BE
fr
o
m
t
h
e
U
n
i
v
e
rs
i
t
y
o
f
Cal
i
c
u
t
,
K
eral
a
,
In
d
i
a.
Cu
rr
e
n
t
l
y
p
u
rs
u
i
n
g
a
Ph
.
D
.
i
n
C
o
m
p
u
t
e
r
S
c
i
en
ce
at
t
h
e
U
n
i
v
e
rs
i
t
y
o
f
N
o
rt
h
D
a
k
o
t
a,
G
ran
d
Fo
r
k
s
,
N
o
rt
h
D
a
k
o
t
a,
U
SA
,
h
i
s
re
s
e
ar
c
h
s
p
an
s
d
i
v
e
rs
e
are
as
s
u
c
h
as
m
o
b
i
l
e
n
e
t
w
o
r
k
s
,
n
e
t
w
o
rk
s
ecu
ri
t
y
,
p
ri
v
acy
,
l
o
c
at
i
o
n
-
b
as
e
d
s
e
rv
i
ce
s
,
a
n
d
mi
s
i
n
f
o
rm
at
i
o
n
d
e
t
ec
t
i
o
n
.
H
e
i
s
a
p
ro
u
d
mem
b
e
r
o
f
S
i
g
m
a
X
i
,
T
h
e
S
ci
e
n
t
i
f
i
c
R
e
s
e
ar
c
h
H
o
n
o
r
S
o
c
i
e
t
y
,
u
n
d
e
rl
i
n
i
n
g
h
i
s
c
o
mmi
t
men
t
t
o
ad
v
an
ci
n
g
s
c
i
en
t
i
fi
c
k
n
o
w
l
e
d
g
e
an
d
re
s
e
ar
ch
e
x
ce
l
l
e
n
ce
.
H
e
c
a
n
b
e
c
o
n
t
ac
t
e
d
at
em
ai
l
:
s
.
ev
a
d
ak
k
e
t
h
i
l
@
u
n
d
.
e
d
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
n
e
ff
icie
nt
c
onv
olut
ional
ne
ur
al
ne
t
w
or
k
f
or
ad
v
e
r
s
ar
ial
tr
aini
ng
agains
t
…
(
Sr
ini
v
a
s
A
.
V
addadi
)
1777
Sa
n
to
s
h
R
eddy
A
dd
u
l
a
a
S
e
n
i
o
r
M
em
b
e
r
o
f
t
h
e
I
E
E
E
,
h
o
l
d
s
a
Mas
t
e
r
o
f
Sci
en
ce
i
n
I
n
f
o
r
m
at
i
o
n
T
ech
n
o
l
o
g
y
fr
o
m
t
h
e
U
n
i
v
e
rs
i
t
y
o
f
t
h
e
Cu
m
b
e
rl
a
n
d
s
i
n
K
en
t
u
ck
y
,
U
SA
.
W
i
t
h
e
x
t
e
n
s
i
v
e
e
x
p
e
ri
en
ce
i
n
t
h
e
I
T
i
n
d
u
s
t
ry
,
h
e
h
as
d
em
o
n
s
t
rat
e
d
e
x
p
e
rt
i
s
e
a
c
r
o
s
s
mu
l
t
i
p
l
e
d
o
m
a
i
n
s
.
Sa
n
t
o
s
h
i
s
an
i
n
n
o
v
at
o
r
w
i
t
h
a
s
t
ro
n
g
p
o
rt
fo
l
i
o
o
f
p
at
e
n
t
s
an
d
h
as
s
i
g
n
i
fi
c
an
t
l
y
co
n
t
ri
b
u
t
ed
t
o
a
c
ad
emi
c
r
e
s
e
ar
c
h
t
h
ro
u
g
h
h
i
s
art
i
c
l
e
s
as
a
n
au
t
h
o
r
a
n
d
co
-
a
u
t
h
o
r
.
A
d
d
i
t
i
o
n
al
l
y
,
h
e
s
e
r
v
e
s
as
a
rev
i
ew
er
fo
r
e
s
t
eemed
j
o
u
r
n
al
s
,
r
e
f
l
ec
t
i
n
g
h
i
s
d
ed
i
c
at
i
o
n
t
o
a
d
v
a
n
c
i
n
g
k
n
o
w
l
e
d
g
e
an
d
e
n
s
u
r
i
n
g
t
h
e
q
u
al
i
t
y
o
f
s
ch
o
l
arl
y
p
u
b
l
i
c
at
i
o
n
s
i
n
h
i
s
fi
el
d
.
H
e
c
a
n
b
e
c
o
n
t
ac
t
e
d
at
em
a
i
l
:
s
an
t
o
s
h
ad
d
u
l
ai
t
@
g
m
ai
l
.
co
m
.
R
o
hi
t
h
V
a
l
l
a
bh
a
n
e
n
i
i
s
a
d
e
d
i
c
at
ed
w
o
rk
e
r
w
i
t
h
a
s
t
ro
n
g
w
o
rk
e
t
h
i
c
i
n
l
e
a
d
i
n
g
t
e
a
m
s
t
o
s
o
l
v
e
o
rg
a
n
i
zat
i
o
n
al
i
s
s
u
e
s
.
H
e
i
s
cap
ab
l
e
o
f
l
e
arn
i
n
g
al
l
as
p
ec
t
s
o
f
i
n
fo
r
m
at
i
o
n
w
i
t
h
i
n
a
c
o
m
p
an
y
a
n
d
u
s
i
n
g
t
h
e
t
ec
h
n
i
c
a
l
k
n
o
w
l
e
d
g
e
an
d
b
u
s
i
n
e
s
s
b
ac
k
g
r
o
u
n
d
t
o
e
ff
ec
t
i
v
e
l
y
an
a
l
y
z
e
s
ec
u
r
i
t
y
me
as
u
r
e
s
t
o
d
e
t
e
r
m
i
n
e
t
h
ei
r
e
ff
ec
t
i
v
en
e
s
s
i
n
o
rd
e
r
t
o
s
t
re
n
g
t
h
e
n
t
h
e
o
v
e
ra
l
l
s
ec
u
ri
t
y
p
o
s
t
u
r
e
.
H
e
h
as
g
r
e
at
w
o
rk
e
t
h
i
c
an
d
o
u
t
s
t
an
d
i
n
g
t
e
a
m
l
e
a
d
e
rs
h
i
p
s
k
i
l
l
s
a
n
d
s
eek
t
o
a
cc
o
m
p
l
i
s
h
o
rg
a
n
i
zat
i
o
n
a
l
g
o
al
s
,
w
h
i
l
e
g
ro
w
i
n
g
i
n
k
n
o
w
l
e
d
g
e
an
d
e
x
p
e
r
i
e
n
ce
.
H
e
c
a
n
b
e
c
o
n
t
ac
t
e
d
at
em
a
i
l
:
ro
h
i
t
.
v
al
l
ab
h
an
en
i
.
2
2
2
2
@
g
m
ai
l
.
c
o
m
.
Bhu
v
a
n
es
h
A
n
a
n
t
h
a
n
rece
i
v
e
d
t
h
e
B.
E
.
d
eg
r
ee
i
n
E
l
ec
t
ri
c
al
a
n
d
E
l
ec
t
ro
n
i
cs
E
n
g
i
n
ee
r
i
n
g
fro
m
A
n
n
a
U
n
i
v
e
rs
i
t
y
i
n
2
0
1
2
,
M.
T
e
c
h
.
i
n
Po
w
e
r
S
y
s
t
em
E
n
g
i
n
ee
r
i
n
g
fro
m
K
a
l
as
al
i
n
g
a
m
U
n
i
v
e
rs
i
t
y
i
n
2
0
1
4
an
d
P
h
.
D
.
d
e
g
r
ee
fr
o
m
Fa
cu
l
t
y
o
f
E
l
e
c
t
ri
c
a
l
E
n
g
i
n
ee
ri
n
g
o
f
A
n
n
a
U
n
i
v
e
rs
i
t
y
i
n
2
0
1
9
.
H
e
h
as
p
u
b
l
i
s
h
ed
mo
r
e
t
h
a
n
6
5
p
ap
e
rs
i
n
r
e
p
u
t
ed
i
n
t
e
r
n
at
i
o
n
al
j
o
u
rn
al
s
,
2
5
p
ap
e
rs
i
n
i
n
t
e
rn
at
i
o
n
a
l
c
o
n
f
e
r
e
n
ce
s
an
d
1
0
b
o
o
k
s
.
H
e
i
s
a
l
i
f
e
t
i
me
mem
b
e
r
o
f
I
n
t
e
rn
at
i
o
n
a
l
S
o
c
i
e
t
y
fo
r
R
e
s
e
ar
ch
a
n
d
D
ev
e
l
o
p
men
t
,
I
n
t
e
rn
at
i
o
n
a
l
A
s
s
o
ci
at
i
o
n
o
f
E
n
g
i
n
ee
rs
.
H
e
c
a
n
b
e
c
o
n
t
ac
t
ed
at
em
ai
l
:
b
h
u
v
an
e
s
h
.
a
n
a
n
t
h
an
@
g
m
ai
l
.
co
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.