T
E
L
K
O
M
NIKA
T
elec
o
mm
un
ica
t
io
n Co
m
pu
t
i
ng
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
2
3
,
No
.
2
,
A
p
r
il 2
0
2
5
,
p
p
.
393
~
401
I
SS
N:
1
6
9
3
-
6
9
3
0
,
DOI
: 1
0
.
1
2
9
2
8
/
T
E
L
KOM
NI
K
A
.
v
2
3
i2
.
2
6
6
1
4
393
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//telko
mn
ika
.
u
a
d
.
a
c.
i
d
Adv
ersa
ria
l
-
ro
bu
st stega
na
ly
sis
sy
ste
m
levera
g
ing
ad
v
ersa
ria
l
training
and
E
f
fic
ientNe
t
T
ha
k
w
a
n A
k
ra
m
J
a
w
a
d
1,
2
,
J
a
m
s
hid
B
a
g
herz
a
deh M
o
ha
s
ef
i
1
,
M
o
ha
mm
ed
Sa
la
h Re
da
Abdelg
ha
ny
3
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
F
a
c
u
l
t
y
o
f
El
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
U
r
mi
a
U
n
i
v
e
r
si
t
y
,
U
r
mi
a
,
I
r
a
n
2
D
e
p
a
r
t
me
n
t
o
f
S
y
st
e
m a
n
d
C
o
n
t
r
o
l
E
n
g
i
n
e
e
r
i
n
g
,
C
o
l
l
e
g
e
o
f
El
e
c
t
r
o
n
i
c
E
n
g
i
n
e
e
r
i
n
g
,
N
i
n
e
v
a
h
U
n
i
v
e
r
si
t
y
,
M
o
su
l
,
I
r
a
q
3
D
e
p
a
r
t
me
n
t
o
f
c
o
mp
u
t
e
r
sci
e
n
c
e
,
F
a
c
u
l
t
y
o
f
c
o
mp
u
t
e
r
s a
n
d
a
r
t
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
c
e
,
B
e
n
i
-
S
u
e
f
U
n
i
v
e
r
si
t
y
,
B
e
n
i
S
u
e
f
,
Eg
y
p
t
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
24
,
2024
R
ev
i
s
ed
Dec
27
,
2
0
2
4
A
cc
ep
ted
J
an
22
,
2
0
2
5
S
teg
a
n
a
l
y
sis
a
i
m
s
to
d
e
tec
t
h
i
d
d
e
n
m
e
ss
a
g
e
s
w
it
h
in
d
ig
it
a
l
m
e
d
ia,
p
re
se
n
ti
n
g
a
sig
n
if
ica
n
t
c
h
a
ll
e
n
g
e
in
th
e
field
o
f
in
f
o
rm
a
ti
o
n
se
c
u
rit
y
.
Th
is
p
a
p
e
r
in
tro
d
u
c
e
s
a
n
a
d
v
e
rsa
rial
-
ro
b
u
st
ste
g
a
n
a
l
y
sis
s
y
ste
m
le
v
e
ra
g
in
g
a
d
v
e
rsa
rial
train
in
g
a
n
d
th
e
p
o
w
e
rf
u
l
f
e
a
tu
re
e
x
trac
ti
o
n
c
a
p
a
b
il
it
ies
o
f
Ef
f
i
c
ie
n
tNe
t.
W
e
u
ti
li
z
e
Ef
f
icie
n
tNe
t
to
e
x
trac
t
ro
b
u
st
f
e
a
tu
re
s
f
ro
m
ima
g
e
s,
w
h
ich
a
re
su
b
se
q
u
e
n
tl
y
c
las
si
f
ied
b
y
a
d
e
n
se
n
e
u
ra
l
n
e
tw
o
rk
to
d
isti
n
g
u
ish
b
e
tw
e
e
n
ste
g
a
n
o
g
ra
p
h
ic
a
n
d
n
o
n
-
ste
g
a
n
o
g
ra
p
h
ic
c
o
n
ten
t.
T
o
e
n
h
a
n
c
e
th
e
s
y
ste
m
’
s
re
sili
e
n
c
e
a
g
a
in
st
a
d
v
e
rsa
rial
a
tt
a
c
k
s,
we
i
m
p
le
m
e
n
t
a
c
u
sto
m
a
d
v
e
rsa
rial
train
in
g
l
o
o
p
th
a
t
g
e
n
e
ra
tes
a
d
v
e
rsa
rial
e
x
a
m
p
le
s
u
sin
g
th
e
f
a
st
g
ra
d
ien
t
sig
n
m
e
th
o
d
(
F
G
S
M
)
a
n
d
i
n
teg
ra
tes
th
e
se
e
x
a
m
p
les
in
to
t
h
e
train
i
n
g
p
r
o
c
e
ss
.
Ou
r
re
su
lt
s
d
e
m
o
n
stra
te
th
a
t
th
e
p
r
o
p
o
se
d
s
y
ste
m
n
o
t
o
n
ly
a
c
h
iev
e
s
h
ig
h
a
c
c
u
ra
c
y
in
d
e
tec
ti
n
g
ste
g
a
n
o
g
ra
p
h
ic
c
o
n
t
e
n
t
b
u
t
a
lso
m
a
in
tain
s
ro
b
u
st
n
e
ss
a
g
a
in
st
a
d
v
e
rsa
rial
p
e
rtu
rb
a
ti
o
n
s.
T
h
is
d
u
a
l
a
p
p
r
o
a
c
h
o
f
lev
e
ra
g
in
g
sta
te
-
of
-
th
e
-
a
rt
d
e
e
p
lea
rn
i
n
g
a
rc
h
it
e
c
tu
re
s
a
n
d
a
d
v
e
rsa
rial
train
in
g
p
ro
v
id
e
s
a
s
ig
n
if
ica
n
t
a
d
v
a
n
c
e
m
e
n
t
in
t
h
e
f
ield
o
f
ste
g
a
n
a
ly
sis,
e
n
su
rin
g
m
o
re
re
li
a
b
le
d
e
tec
ti
o
n
o
f
h
id
d
e
n
m
e
ss
a
g
e
s in
d
ig
it
a
l
im
a
g
e
s.
K
ey
w
o
r
d
s
:
A
d
v
er
s
ar
ial
attac
k
s
Dee
p
lear
n
in
g
E
f
f
icien
tNet
Steg
a
n
al
y
s
i
s
Steg
a
n
o
g
r
ap
h
y
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
:
J
am
s
h
id
B
ag
h
er
za
d
eh
Mo
h
a
s
e
f
i
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
Fac
u
lt
y
o
f
E
lectr
ical
an
d
C
o
m
p
u
ter
E
n
g
in
ee
r
i
n
g
,
Ur
m
ia
Un
iv
er
s
it
y
11
Km
Ser
o
w
R
o
ad
,
Ur
m
ia,
W
est A
ze
r
b
aij
an
,
I
r
an
E
m
ail: j
.
b
ag
h
er
za
d
eh
@
u
r
m
ia.
ac
.
ir
1.
I
NT
RO
D
UCT
I
O
N
I
n
to
d
a
y
’
s
d
ig
i
tal
ag
e,
e
n
s
u
r
in
g
th
e
s
ec
u
r
it
y
o
f
i
n
f
o
r
m
atio
n
tr
an
s
m
i
s
s
io
n
h
a
s
b
ec
o
m
e
a
cr
itica
l
co
n
ce
r
n
.
Steg
a
n
o
g
r
ap
h
y
,
th
e
p
r
ac
tice
o
f
h
id
in
g
s
ec
r
et
m
es
s
ag
e
s
w
it
h
i
n
d
ig
ital
m
ed
ia,
p
o
s
es a
u
n
iq
u
e
ch
alle
n
g
e
a
s
it c
an
b
e
u
s
ed
f
o
r
b
o
th
leg
iti
m
a
te
an
d
m
alic
io
u
s
p
u
r
p
o
s
es.
Ste
g
a
n
a
l
y
s
i
s
,
th
e
ar
t
o
f
d
etec
tin
g
th
e
s
e
h
id
d
en
m
es
s
ag
e
s
,
is
th
er
ef
o
r
e
o
f
p
ar
am
o
u
n
t i
m
p
o
r
tan
c
e
[
1
]
-
[
3
]
.
T
r
a
d
itio
n
al
s
te
g
an
al
y
s
is
m
e
th
o
d
s
o
f
te
n
s
tr
u
g
g
le
w
ith
r
o
b
u
s
t
n
es
s
,
esp
ec
iall
y
w
h
e
n
f
ac
ed
w
it
h
ad
v
er
s
ar
ial
attac
k
s
d
esi
g
n
ed
to
d
ec
eiv
e
d
etec
tio
n
s
y
s
te
m
s
[
4
]
,
[
5]
.
T
h
e
n
ee
d
to
en
h
a
n
ce
m
o
d
el
r
o
b
u
s
t
n
ess
ag
ai
n
s
t a
d
v
er
s
ar
ial
a
t
tack
s
h
as
b
ec
o
m
e
a
cr
itica
l a
r
e
a
o
f
f
o
c
u
s
in
t
h
e
f
ield
o
f
m
ac
h
i
n
e
lear
n
i
n
g
.
A
d
v
er
s
ar
ial
a
ttack
s
p
o
s
e
s
i
g
n
i
f
ican
t
th
r
ea
t
s
b
y
s
u
b
tl
y
al
ter
in
g
in
p
u
t
d
ata,
w
h
ic
h
ca
n
d
ec
eiv
e
m
o
d
els
i
n
to
m
ak
i
n
g
er
r
o
n
eo
u
s
p
r
ed
ictio
n
s
,
p
o
ten
tiall
y
lead
in
g
to
s
e
v
er
e
co
n
s
e
q
u
en
ce
s
i
n
s
e
n
s
iti
v
e
d
o
m
ai
n
s
li
k
e
f
in
a
n
ce
,
h
ea
lt
h
c
ar
e,
an
d
au
to
n
o
m
o
u
s
s
y
s
te
m
s
[
6
]
.
T
h
ese
attac
k
s
ex
p
lo
it
th
e
v
u
l
n
er
ab
ilit
ie
s
in
a
m
o
d
el
’
s
d
ec
is
io
n
b
o
u
n
d
ar
y
,
r
ev
ea
li
n
g
th
a
t
m
o
d
els
m
i
g
h
t
p
er
f
o
r
m
w
el
l
in
co
n
tr
o
lled
en
v
ir
o
n
m
en
ts
b
u
t
ar
e
f
r
ag
ile
w
h
en
f
ac
ed
w
it
h
ad
v
er
s
ar
ial
in
p
u
t
s
.
Stre
n
g
th
e
n
i
n
g
t
h
e
r
o
b
u
s
tn
e
s
s
o
f
m
o
d
els
is
es
s
e
n
tial
to
en
s
u
r
e
th
eir
r
eliab
ilit
y
a
n
d
ac
cu
r
ac
y
i
n
r
e
al
-
w
o
r
ld
ap
p
licatio
n
s
,
w
h
er
e
ad
v
er
s
ar
ial
co
n
d
itio
n
s
m
ig
h
t
b
e
en
co
u
n
ter
ed
[
7
]
.
A
p
p
r
o
ac
h
es
s
u
c
h
as
ad
v
er
s
ar
i
al
tr
ain
i
n
g
,
w
h
ic
h
i
n
v
o
lv
es
tr
ain
i
n
g
m
o
d
els
w
i
th
ad
v
er
s
ar
ial
ex
a
m
p
le
s
,
an
d
t
h
e
u
s
e
o
f
r
o
b
u
s
t
o
p
ti
m
izat
io
n
tech
n
iq
u
e
s
,
h
av
e
s
h
o
w
n
p
r
o
m
i
s
e
in
m
iti
g
ati
n
g
t
h
ese
v
u
l
n
er
ab
ilit
ie
s
[
8
]
.
A
d
d
itio
n
all
y
,
i
m
p
le
m
en
t
in
g
d
ef
e
n
s
iv
e
s
tr
at
eg
ies
l
ik
e
g
r
ad
ien
t
m
a
s
k
in
g
a
n
d
in
p
u
t
tr
an
s
f
o
r
m
atio
n
ca
n
f
u
r
th
er
e
n
h
a
n
ce
t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
2
,
A
p
r
il
20
25
:
3
9
3
-
401
394
m
o
d
el
’
s
ab
ilit
y
to
w
it
h
s
t
an
d
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
[
9
]
.
B
y
f
o
cu
s
i
n
g
o
n
d
ev
elo
p
in
g
r
o
b
u
s
t
m
ac
h
in
e
lear
n
in
g
m
o
d
el
s
,
th
e
f
ield
ca
n
b
etter
p
r
o
tect
th
ese
s
y
s
te
m
s
f
r
o
m
ad
v
er
s
ar
ial
th
r
ea
t
s
,
en
s
u
r
i
n
g
t
h
eir
s
a
f
et
y
a
n
d
ef
f
ec
tiv
e
n
e
s
s
ac
r
o
s
s
v
ar
io
u
s
a
p
p
licatio
n
s
.
I
n
t
h
is
p
ap
er
,
w
e
p
r
o
p
o
s
e
an
ad
v
an
ce
d
s
teg
a
n
al
y
s
is
s
y
s
te
m
t
h
at
lev
er
a
g
es
t
h
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
ca
p
ab
ilit
ies
o
f
E
f
f
icie
n
tNet
[
1
0
]
an
d
in
co
r
p
o
r
ates
ad
v
er
s
ar
ial
tr
ain
in
g
to
en
h
a
n
ce
r
o
b
u
s
tn
e
s
s
ag
ain
s
t
ad
v
er
s
ar
ial
attac
k
s
.
Ou
r
a
p
p
r
o
a
ch
i
n
v
o
lv
es t
r
a
in
in
g
a
d
en
s
e
n
eu
r
a
l
n
e
tw
o
r
k
o
n
f
e
a
tu
r
e
s
ex
t
r
a
ct
e
d
b
y
E
f
f
ic
i
e
n
t
N
et
,
c
o
m
b
in
e
d
w
ith
a
c
u
s
t
o
m
a
d
v
e
r
s
a
r
i
a
l
t
r
ai
n
i
n
g
l
o
o
p
u
s
in
g
th
e
f
as
t
g
r
a
d
i
en
t s
i
g
n
m
e
th
o
d
(
F
GS
M
)
[
1
1
]
.
G
o
o
d
f
e
l
l
o
w
et
a
l
.
[
1
1
]
in
tr
o
d
u
ce
d
th
e
FGSM to
cr
ea
te
ad
v
er
s
ar
ial
ex
a
m
p
les.
T
h
is
tech
n
iq
u
e
lev
er
ag
e
s
th
e
d
er
iv
at
iv
e
o
f
th
e
n
et
w
o
r
k
’
s
lo
s
s
f
u
n
ctio
n
co
n
ce
r
n
in
g
t
h
e
i
n
p
u
t
f
ea
t
u
r
e
v
ec
to
r
.
Fo
r
a
g
iv
en
i
n
p
u
t
i
m
a
g
e,
FG
SM
p
er
tu
r
b
s
ea
ch
f
ea
t
u
r
e
b
y
ad
j
u
s
tin
g
it
in
t
h
e
d
ir
ec
tio
n
o
f
th
e
g
r
ad
ien
t.
C
o
n
s
eq
u
e
n
tl
y
,
t
h
is
m
o
d
if
icat
io
n
alte
r
s
t
h
e
cla
s
s
i
f
icatio
n
o
u
tco
m
e
o
f
th
e
in
p
u
t
i
m
ag
e.
Fo
r
a
n
eu
r
al
n
et
w
o
r
k
w
i
th
a
cr
o
s
s
-
e
n
tr
o
p
y
lo
s
s
f
u
n
ct
io
n
,
w
h
er
e
th
e
in
p
u
t
i
m
a
g
e
an
d
tar
g
et
class
ar
e
s
p
ec
if
ied
,
th
e
ad
v
er
s
ar
ial
ex
a
m
p
le
is
g
e
n
er
ated
b
y
ad
d
in
g
a
s
m
al
l
p
er
tu
r
b
atio
n
to
th
e
o
r
ig
in
al
i
m
a
g
e.
T
h
is
p
er
tu
r
b
atio
n
is
co
n
tr
o
lled
b
y
a
p
ar
a
m
eter
t
h
at
d
eter
m
in
es
i
ts
s
ize,
en
s
u
r
i
n
g
th
e
c
h
a
n
g
e
s
ar
e
s
u
b
tle
y
e
t
ef
f
ec
tiv
e
i
n
m
is
lead
in
g
th
e
n
et
w
o
r
k
[
1
1
]
.
Dee
p
Fo
o
l
is
an
u
n
tar
g
eted
attac
k
m
et
h
o
d
th
at
g
e
n
er
ates
ad
v
er
s
ar
ial
ex
a
m
p
le
s
th
r
o
u
g
h
iter
ativ
e
p
er
tu
r
b
atio
n
s
o
f
an
i
m
a
g
e.
T
h
is
ap
p
r
o
ac
h
aim
s
to
f
in
d
th
e
n
e
ar
est d
ec
is
io
n
b
o
u
n
d
ar
y
b
y
s
l
i
g
h
t
l
y
m
o
d
i
f
y
in
g
th
e
i
m
a
g
e
in
ea
c
h
iter
atio
n
u
n
til
th
e
alter
ed
i
m
a
g
e
c
h
an
g
es
t
h
e
n
et
w
o
r
k
’
s
cla
s
s
i
f
icatio
n
[
1
2
]
.
T
h
e
C
ar
lin
i
a
n
d
W
ag
n
er
(
C
&
W
)
m
et
h
o
d
,
n
a
m
ed
af
ter
its
cr
ea
to
r
s
,
is
a
p
o
w
er
f
u
l
attac
k
th
at
ca
n
b
e
eit
h
er
tar
g
eted
o
r
u
n
tar
g
eted
.
I
t
u
tili
ze
s
t
h
r
ee
m
etr
ic
s
to
m
e
asu
r
e
d
is
to
r
tio
n
:
L
0
,
L
2
,
an
d
L
∞
n
o
r
m
s
,
w
it
h
t
h
e
u
n
tar
g
ete
d
L
2
n
o
r
m
v
er
s
io
n
b
ein
g
n
o
ted
f
o
r
it
s
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
.
T
h
is
m
e
th
o
d
g
e
n
er
ates
ad
v
er
s
ar
ia
l
ex
a
m
p
les
b
y
s
o
l
v
i
n
g
an
o
p
tim
izatio
n
p
r
o
b
lem
to
f
i
n
d
th
e
s
m
a
lles
t
p
er
tu
r
b
atio
n
th
a
t
m
is
lead
s
th
e
n
et
w
o
r
k
.
T
h
e
o
p
ti
m
izat
io
n
b
alan
ce
s
m
i
n
i
m
izi
n
g
t
h
e
p
er
tu
r
b
atio
n
a
n
d
alter
in
g
t
h
e
n
e
t
w
o
r
k
’
s
cla
s
s
if
ica
tio
n
,
co
n
tr
o
lled
b
y
a
h
y
p
er
p
ar
am
eter
.
T
h
is
attac
k
s
ee
k
s
th
e
m
i
n
i
m
al
L
2
n
o
r
m
p
er
tu
r
b
atio
n
to
ch
a
n
g
e
th
e
i
m
a
g
e
’
s
clas
s
i
f
icatio
n
w
h
ile
e
n
s
u
r
i
n
g
t
h
e
p
er
tu
r
b
ed
im
ag
e
r
e
m
ai
n
s
w
it
h
in
v
alid
p
ix
el
v
al
u
es.
T
h
e
m
et
h
o
d
also
in
co
r
p
o
r
ates
a
c
o
n
f
i
d
en
ce
p
ar
am
eter
to
g
en
er
ate
ad
v
er
s
ar
ial
e
x
a
m
p
le
s
w
it
h
h
ig
h
clas
s
i
f
icatio
n
co
n
f
id
en
ce
,
w
h
ic
h
en
h
an
ce
s
th
eir
tr
a
n
s
f
er
ab
ilit
y
to
o
th
er
m
o
d
el
s
.
T
h
e
C
&
W
m
eth
o
d
is
r
ec
o
g
n
ized
f
o
r
its
r
o
b
u
s
t
n
es
s
a
n
d
d
if
f
icu
l
t
y
to
d
ef
e
n
d
ag
ai
n
s
t
[
1
3
]
.
Qin
et
a
l.
[
1
4
]
lev
er
ag
es
v
ec
to
r
-
w
is
e
e
m
b
ed
d
in
g
w
it
h
co
lo
r
p
ix
el
v
ec
to
r
s
(
C
P
V)
f
o
r
c
o
lo
r
im
a
g
e
s
teg
a
n
o
g
r
ap
h
y
,
i
m
p
r
o
v
in
g
u
p
o
n
tr
ad
itio
n
al
ele
m
e
n
t
-
w
i
s
e
e
m
b
ed
d
in
g
.
I
n
p
ar
tic
u
lar
,
th
i
s
m
et
h
o
d
u
s
e
s
th
e
iter
ativ
e
f
a
s
t
g
r
ad
ie
n
t
s
i
g
n
m
eth
o
d
(I
-
FGSM)
to
g
en
er
ate
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
t
h
at
ef
f
ec
t
iv
el
y
d
ec
eiv
e
s
teg
a
n
al
y
tic
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
et
w
o
r
k
s
(
C
NN
s
)
.
B
y
in
cr
e
m
en
tall
y
ad
j
u
s
ti
n
g
ad
v
er
s
ar
ia
l
co
s
ts
w
it
h
i
n
n
o
n
-
o
v
er
lap
p
in
g
s
u
b
-
i
m
a
g
es,
th
i
s
ap
p
r
o
ac
h
r
e
d
u
ce
s
d
etec
tab
ilit
y
,
m
ai
n
tain
in
g
a
h
ig
h
s
u
cc
e
s
s
r
ate
ag
ain
s
t
C
NN
-
b
ased
s
teg
a
n
al
y
ze
r
s
.
R
o
b
u
s
t
n
ess
-
b
ased
d
ef
en
s
e
f
o
c
u
s
e
s
o
n
co
r
r
ec
tly
clas
s
i
f
y
in
g
ad
v
er
s
a
r
ial
ex
a
m
p
le
s
,
an
d
v
ar
io
u
s
m
et
h
o
d
s
h
a
v
e
b
ee
n
d
ev
elo
p
ed
to
ac
h
iev
e
th
is
.
O
n
e
co
m
m
o
n
ap
p
r
o
ac
h
is
ad
v
er
s
a
r
ial
tr
ain
in
g
,
w
h
ich
in
v
o
l
v
es a
u
g
m
e
n
ti
n
g
th
e
tr
ai
n
i
n
g
s
et
w
it
h
a
m
ix
o
f
n
o
r
m
al
a
n
d
ad
v
er
s
ar
ial
ex
a
m
p
le
s
to
i
m
p
r
o
v
e
th
e
n
et
w
o
r
k
’
s
r
esil
ien
ce
[
1
1
]
.
An
o
th
er
s
tr
ateg
y
is
p
r
ep
r
o
ce
s
s
i
n
g
i
n
p
u
t
i
m
ag
e
s
to
r
e
m
o
v
e
ad
v
er
s
ar
ial
p
er
tu
r
b
ati
o
n
s
th
r
o
u
g
h
tech
n
iq
u
es
s
u
c
h
as
p
r
in
cip
al
c
o
m
p
o
n
en
t
a
n
al
y
s
i
s
(
P
C
A
)
.
B
h
ag
o
j
i
et
a
l.
[
1
5
]
in
v
esti
g
a
te
th
e
en
h
a
n
ce
m
en
t
o
f
m
ac
h
in
e
lear
n
in
g
s
y
s
te
m
r
o
b
u
s
tn
e
s
s
t
h
r
o
u
g
h
v
ar
io
u
s
d
ata
tr
an
s
f
o
r
m
atio
n
s
.
T
h
eir
s
tu
d
y
r
e
v
ea
ls
t
h
at
ce
r
tai
n
tr
an
s
f
o
r
m
a
tio
n
s
,
s
u
c
h
a
s
in
p
u
t
p
r
ep
r
o
ce
s
s
in
g
,
f
ea
tu
r
e
m
an
ip
u
latio
n
,
an
d
P
C
A
,
ca
n
s
i
g
n
i
f
ica
n
tl
y
m
iti
g
ate
th
e
i
m
p
ac
t
o
f
ad
v
er
s
ar
ial
at
tack
s
,
t
h
er
eb
y
i
m
p
r
o
v
i
n
g
t
h
e
r
esil
ie
n
ce
o
f
n
e
u
r
al
n
et
wo
r
k
s
to
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
[
1
5
]
,
[
1
6
]
.
An
o
th
er
m
eth
o
d
is
J
P
E
G
co
m
p
r
ess
io
n
,
ad
d
in
g
n
o
is
e,
cr
o
p
p
i
n
g
,
a
n
d
r
o
tatin
g
.
Das
et
a
l.
[
1
7
]
p
r
o
p
o
s
e
u
s
i
n
g
J
P
E
G
c
o
m
p
r
ess
io
n
as
a
p
r
ep
r
o
ce
s
s
in
g
s
tep
to
d
ef
en
d
ag
ain
s
t
ad
v
er
s
ar
ial
attac
k
s
.
B
y
r
e
m
o
v
i
n
g
h
ig
h
-
f
r
eq
u
en
c
y
s
i
g
n
al
co
m
p
o
n
e
n
ts
th
r
o
u
g
h
s
elec
t
iv
e
b
l
u
r
r
in
g
,
J
P
E
G
c
o
m
p
r
ess
io
n
ca
n
m
iti
g
ate
th
e
i
m
p
ac
t
o
f
p
er
tu
r
b
atio
n
s
ad
d
ed
t
o
im
ag
e
s
.
T
h
eir
ap
p
r
o
ac
h
s
h
o
w
s
th
a
t
th
is
tech
n
iq
u
e
ca
n
s
i
g
n
i
f
ica
n
tl
y
r
ed
u
ce
th
e
ef
f
icac
y
o
f
s
e
v
er
al
attac
k
m
et
h
o
d
s
,
in
c
l
u
d
in
g
t
h
e
F
SGM
a
n
d
Dee
p
Fo
o
l,
w
ith
o
u
t
r
eq
u
ir
i
n
g
p
r
io
r
k
n
o
w
led
g
e
o
f
th
e
m
o
d
el
u
n
d
er
attac
k
.
Pa
p
e
r
n
o
t
e
t
a
l
.
[
1
8
]
i
n
t
r
o
d
u
ce
d
e
f
en
s
iv
e
d
i
s
t
i
ll
a
t
i
o
n
as
a
m
et
h
o
d
t
o
e
n
h
an
c
e
th
e
r
o
b
u
s
tn
ess
o
f
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
(
D
NN
s
)
ag
ai
n
s
t
a
d
v
e
r
s
a
r
i
al
a
t
t
ac
k
s
.
B
y
l
ev
e
r
ag
in
g
k
n
o
w
l
e
d
g
e
d
is
t
il
l
a
ti
o
n
t
e
ch
n
iq
u
e
s
,
t
h
i
s
a
p
p
r
o
a
ch
r
e
d
u
c
es
t
h
e
e
f
f
e
ct
iv
en
es
s
o
f
ad
v
e
r
s
a
r
i
a
l
s
am
p
l
e
g
en
e
r
at
i
o
n
,
s
ig
n
if
i
c
an
t
ly
m
it
ig
at
in
g
t
h
e
s
u
c
c
e
s
s
r
a
t
e
o
f
s
u
c
h
a
t
t
a
ck
s
.
T
h
e
s
tu
d
y
d
em
o
n
s
t
r
a
te
s
a
d
r
am
a
t
i
c
r
e
d
u
c
t
i
o
n
in
a
d
v
e
r
s
a
r
i
al
ef
f
e
ct
iv
en
e
s
s
,
h
i
g
h
l
ig
h
t
i
n
g
th
e
p
o
t
en
t
i
a
l
o
f
d
i
s
t
il
l
at
i
o
n
t
o
im
p
r
o
v
e
D
N
N
s
e
c
u
r
ity
a
n
d
r
es
i
li
en
c
e
.
D
ef
e
n
s
iv
e
d
is
t
il
l
a
ti
o
n
en
h
an
c
es
r
o
b
u
s
tn
es
s
b
y
h
i
d
in
g
th
e
g
r
a
d
i
en
t
b
etw
e
en
t
h
e
p
r
e
-
s
o
f
tm
a
x
l
ay
e
r
a
n
d
s
o
f
tm
ax
o
u
t
p
u
t
s
u
s
in
q
in
g
d
i
s
t
i
ll
a
t
i
o
n
t
r
ain
in
g
t
e
ch
n
i
q
u
e
s
[
1
6
]
,
[
1
8
]
.
A
d
d
itio
n
al
l
y
,
o
b
f
u
s
ca
ted
g
r
a
d
ien
ts
h
i
n
d
er
attac
k
er
s
b
y
m
ak
in
g
it
d
if
f
ic
u
lt
to
co
m
p
u
te
f
ea
s
ib
le
g
r
ad
ien
ts
f
o
r
g
en
er
ati
n
g
ad
v
er
s
ar
ial
ex
a
m
p
le
s
.
A
t
h
al
y
e
et
a
l.
[
1
9
]
i
d
en
tify
a
n
d
an
al
y
ze
o
b
f
u
s
ca
ted
g
r
ad
ien
t
s
,
a
f
o
r
m
o
f
g
r
ad
ien
t
m
as
k
i
n
g
t
h
at
f
alsel
y
ap
p
ea
r
s
to
o
f
f
er
s
ec
u
r
it
y
a
g
ai
n
s
t
ad
v
er
s
ar
ial
attac
k
s
.
T
h
e
s
t
u
d
y
d
em
o
n
s
tr
ate
s
t
h
at
d
e
f
en
s
es
r
el
y
i
n
g
o
n
o
b
f
u
s
ca
ted
g
r
ad
ie
n
ts
ca
n
b
e
e
f
f
ec
tiv
el
y
c
ir
cu
m
v
en
ted
,
s
h
o
w
ca
s
i
n
g
tech
n
iq
u
es
to
b
y
p
ass
s
u
c
h
d
ef
en
s
es
an
d
r
ev
ea
li
n
g
th
eir
v
u
l
n
er
ab
ilit
y
in
a
w
h
i
te
-
b
o
x
s
ettin
g
.
T
h
e
p
ap
e
r
e
m
p
h
a
s
izes
th
a
t
m
a
n
y
d
ef
e
n
s
es
ac
ce
p
ted
at
I
n
ter
n
atio
n
al
C
o
n
f
er
en
ce
o
n
L
ea
r
n
i
n
g
R
ep
r
esen
tatio
n
s
(
I
C
L
R
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
d
ve
r
s
a
r
ia
l
-
r
o
b
u
s
t steg
a
n
a
lysi
s
s
yste
m
leve
r
a
g
in
g
a
d
ve
r
s
a
r
ia
l tra
in
in
g
…
(
Th
a
kwa
n
A
kra
m
Ja
w
a
d
)
395
2
0
1
8
,
w
h
ic
h
clai
m
r
o
b
u
s
t
n
ess
,
ac
tu
all
y
d
ep
en
d
o
n
o
b
f
u
s
ca
te
d
g
r
ad
ien
ts
,
m
ak
in
g
th
e
m
s
u
s
ce
p
tib
le
to
tar
g
eted
attac
k
s
[
1
9
]
.
Me
n
g
a
n
d
C
h
en
[
2
0
]
p
r
o
p
o
s
ed
Ma
g
Net,
a
f
r
a
m
e
w
o
r
k
f
o
r
d
ef
en
d
in
g
n
e
u
r
al
n
et
w
o
r
k
s
ag
ai
n
s
t
ad
v
er
s
ar
ial
ex
a
m
p
les.
T
h
is
f
r
a
m
e
w
o
r
k
n
e
ith
er
m
o
d
i
f
ie
s
th
e
p
r
o
tecte
d
class
if
ier
n
o
r
r
eq
u
ir
es
k
n
o
w
led
g
e
o
f
th
e
ad
v
er
s
ar
ial
ex
a
m
p
le
g
e
n
er
atio
n
p
r
o
ce
s
s
.
Ma
g
Net
e
m
p
lo
y
s
d
etec
to
r
n
et
w
o
r
k
s
to
d
is
ti
n
g
u
i
s
h
b
et
w
ee
n
n
o
r
m
a
l
an
d
ad
v
er
s
ar
ial
ex
a
m
p
les
b
y
a
p
p
r
o
x
im
a
tin
g
t
h
e
m
a
n
i
f
o
ld
o
f
n
o
r
m
al
ex
a
m
p
le
s
.
A
d
d
itio
n
a
ll
y
,
a
r
ef
o
r
m
er
n
et
w
o
r
k
is
u
s
ed
to
m
o
v
e
ad
v
er
s
ar
ial
e
x
a
m
p
les
clo
s
er
to
th
e
m
an
if
o
ld
o
f
n
o
r
m
al
ex
a
m
p
les,
e
n
ab
lin
g
co
r
r
ec
t
class
if
icatio
n
.
Ma
g
Net
h
a
s
d
e
m
o
n
s
tr
ated
ef
f
ec
tiv
e
n
ess
a
g
ain
s
t
s
tate
-
of
-
t
h
e
-
ar
t
attac
k
s
i
n
b
lack
-
b
o
x
a
n
d
g
r
a
y
-
b
o
x
s
ce
n
ar
io
s
w
it
h
o
u
t sacr
i
f
ici
n
g
f
alse p
o
s
iti
v
e
r
ates.
T
ar
an
et
a
l.
[
2
1
]
p
r
o
p
o
s
ed
a
k
e
y
-
b
a
s
ed
d
iv
er
s
if
ied
ag
g
r
eg
a
ti
o
n
(
KDA
)
m
ec
h
a
n
is
m
to
d
ef
e
n
d
DNNs
ag
ain
s
t
ad
v
er
s
ar
ia
l
attac
k
s
i
n
g
r
a
y
-
b
o
x
an
d
b
lack
-
b
o
x
s
ce
n
a
r
io
s
.
KDA
i
n
v
o
lv
e
s
a
m
u
lti
-
c
h
an
n
el
ar
ch
itect
u
r
e
w
it
h
r
a
n
d
o
m
ize
d
tr
an
s
f
o
r
m
ati
o
n
s
ap
p
lied
to
ea
ch
ch
a
n
n
el.
T
h
e
r
an
d
o
m
iza
tio
n
,
b
ased
o
n
a
s
ec
r
et
k
e
y
,
p
r
ev
en
ts
g
r
ad
ien
t
b
ac
k
p
r
o
p
ag
atio
n
an
d
h
in
d
er
s
th
e
attac
k
er
’
s
ab
ilit
y
to
cr
ea
te
ef
f
ec
ti
v
e
ad
v
er
s
ar
ial
ex
a
m
p
les.
B
y
ag
g
r
e
g
ati
n
g
t
h
e
o
u
tp
u
ts
f
r
o
m
m
u
lt
ip
le
ch
a
n
n
el
s
,
KD
A
i
m
p
r
o
v
es
t
h
e
r
o
b
u
s
t
n
es
s
an
d
r
eliab
ilit
y
o
f
th
e
class
i
f
icatio
n
p
r
o
ce
s
s
.
T
h
e
au
th
o
r
s
d
e
m
o
n
s
tr
ated
t
h
e
e
f
f
ec
t
iv
en
e
s
s
o
f
KD
A
ag
ai
n
s
t
v
ar
io
u
s
s
ta
te
-
of
-
th
e
-
ar
t
attac
k
s
,
i
n
cl
u
d
in
g
g
r
ad
ie
n
t
-
b
a
s
ed
an
d
n
o
n
-
g
r
ad
ien
t
-
b
ased
m
e
th
o
d
s
[
2
1
]
.
E
f
f
icien
tNetB
0
is
p
ar
t
o
f
t
h
e
E
f
f
icien
tNet
f
a
m
il
y
,
a
s
er
ies
o
f
C
NN
s
d
esi
g
n
ed
f
o
r
h
i
g
h
p
er
f
o
r
m
an
c
e
an
d
ef
f
icie
n
c
y
.
T
h
e
k
e
y
in
n
o
v
atio
n
i
n
E
f
f
icie
n
tNe
t
is
t
h
e
u
s
e
o
f
a
co
m
p
o
u
n
d
s
ca
li
n
g
m
et
h
o
d
th
at
u
n
if
o
r
m
l
y
s
ca
les
all
d
i
m
en
s
io
n
s
o
f
d
ep
th
,
w
id
th
,
an
d
r
eso
l
u
tio
n
u
s
in
g
a
s
i
m
p
le
y
e
t
ef
f
ec
ti
v
e
s
ca
lin
g
co
ef
f
icie
n
t.
E
f
f
icien
tNetB
0
,
th
e
b
aseli
n
e
m
o
d
el,
ac
h
ie
v
es
r
e
m
ar
k
ab
le
a
cc
u
r
ac
y
a
n
d
ef
f
icie
n
c
y
b
y
o
p
ti
m
izi
n
g
th
i
s
b
alan
ce
.
I
t
u
tili
ze
s
f
e
w
er
p
ar
a
m
eter
s
a
n
d
co
m
p
u
ta
tio
n
al
r
eso
u
r
ce
s
wh
ile
m
ai
n
tai
n
in
g
o
r
ev
e
n
i
m
p
r
o
v
in
g
p
er
f
o
r
m
a
n
ce
co
m
p
ar
ed
to
p
r
ev
io
u
s
a
r
ch
it
ec
tu
r
es.
E
f
f
icien
tNetB
0
e
m
p
l
o
y
s
a
co
m
b
i
n
atio
n
o
f
d
ep
th
w
i
s
e
co
n
v
o
lu
tio
n
s
,
s
q
u
ee
ze
-
a
n
d
-
ex
ci
tatio
n
o
p
ti
m
i
za
tio
n
,
an
d
th
e
MB
C
o
n
v
b
lo
ck
(
a
v
ar
ian
t
o
f
in
v
er
ted
r
esid
u
al
b
lo
ck
s
)
.
T
h
ese
ar
ch
itect
u
r
al
ch
o
ices
allo
w
it
to
ac
h
iev
e
s
tate
-
of
-
t
h
e
-
ar
t
p
er
f
o
r
m
an
ce
o
n
s
e
v
er
al
b
en
ch
m
ar
k
s
w
h
ile
b
ein
g
s
ig
n
i
f
ica
n
tl
y
m
o
r
e
co
m
p
u
tatio
n
all
y
e
f
f
icie
n
t [
1
0
]
.
R
an
d
o
m
f
o
r
est
is
a
n
e
n
s
e
m
b
l
e
lear
n
i
n
g
m
et
h
o
d
th
at
co
m
b
i
n
es
m
u
ltip
le
d
ec
is
io
n
tr
ee
s
to
i
m
p
r
o
v
e
class
i
f
icatio
n
ac
cu
r
ac
y
a
n
d
r
o
b
u
s
t
n
ess
[
2
2
]
.
T
h
is
m
o
d
el
is
k
n
o
w
n
f
o
r
its
ab
ili
t
y
to
h
a
n
d
l
e
lar
g
e
d
atasets
an
d
p
er
f
o
r
m
w
ell
w
i
th
u
n
s
tr
u
ct
u
r
e
d
d
ata,
m
a
k
i
n
g
it a
p
o
p
u
lar
ch
o
ice
f
o
r
v
ar
io
u
s
m
ac
h
in
e
lear
n
in
g
ta
s
k
s
.
Ho
w
e
v
er
,
r
an
d
o
m
f
o
r
est
s
ar
e
s
u
s
ce
p
tib
le
to
ad
v
er
s
ar
ial
ex
a
m
p
le
s
,
w
h
ich
ar
e
in
te
n
tio
n
all
y
p
er
t
u
r
b
ed
in
p
u
t
s
d
esig
n
ed
to
d
ec
eiv
e
m
ac
h
i
n
e
lear
n
i
n
g
m
o
d
els.
Un
l
ik
e
DNN
s
,
w
h
ich
ca
n
lear
n
co
m
p
le
x
r
ep
r
esen
tatio
n
s
a
n
d
p
o
ten
t
iall
y
d
ev
elo
p
s
o
m
e
le
v
el
o
f
r
esil
ie
n
ce
to
ad
v
er
s
ar
ial
attac
k
s
th
r
o
u
g
h
s
p
ec
ialized
tr
ain
in
g
tech
n
iq
u
es,
r
an
d
o
m
f
o
r
est
s
lack
i
n
h
er
e
n
t
m
ec
h
a
n
is
m
s
to
d
ea
l
w
it
h
s
u
ch
p
er
t
u
r
b
atio
n
s
e
f
f
ec
tiv
e
l
y
.
A
d
v
e
r
s
a
r
i
al
e
x
am
p
le
s
ex
p
l
o
i
t
th
e
l
in
e
a
r
d
e
c
i
s
i
o
n
b
o
u
n
d
a
r
i
e
s
o
f
e
ac
h
in
d
i
v
i
d
u
al
t
r
e
e
w
i
th
in
th
e
e
n
s
em
b
le
,
l
e
a
d
in
g
t
o
m
is
c
l
as
s
if
i
c
at
i
o
n
s
.
R
e
c
en
t
s
tu
d
i
es
h
av
e
s
h
o
w
n
th
a
t
a
d
v
e
r
s
a
r
i
al
a
t
ta
ck
s
c
a
n
r
e
d
u
c
e
t
h
e
a
c
cu
r
a
cy
o
f
r
an
d
o
m
f
o
r
e
s
t
m
o
d
el
s
s
ig
n
if
ic
an
t
ly
m
o
r
e
th
an
th
at
o
f
w
el
l
-
t
r
ain
e
d
D
NN
s
,
w
h
i
ch
c
an
b
e
f
o
r
t
if
i
e
d
th
r
o
u
g
h
a
d
v
e
r
s
a
r
i
a
l
t
r
a
in
in
g
an
d
r
e
g
u
la
r
i
z
a
ti
o
n
t
e
ch
n
i
q
u
e
s
[
1
8
]
,
[
2
3
]
.
T
h
er
ef
o
r
e,
w
h
ile
r
an
d
o
m
f
o
r
ests
p
r
o
v
id
e
a
s
i
m
p
le
an
d
in
t
er
p
r
etab
le
m
o
d
el,
t
h
eir
v
u
l
n
e
r
ab
ilit
y
to
ad
v
er
s
ar
ial
attac
k
s
h
i
g
h
lig
h
t
s
th
e
n
ee
d
f
o
r
ad
d
itio
n
al
d
ef
en
s
es
w
h
e
n
d
ep
lo
y
in
g
t
h
ese
m
o
d
els
in
ad
v
er
s
ar
ia
l
s
etti
n
g
s
.
I
n
th
is
p
ap
er
,
w
e
co
n
t
r
ib
u
te
th
e
f
o
llo
w
in
g
n
o
v
eltie
s
:
−
W
e
i
m
p
le
m
e
n
ted
t
h
r
ee
d
if
f
er
e
n
t
m
ac
h
in
e
lear
n
i
n
g
an
d
d
ee
p
l
ea
r
n
in
g
m
o
d
els
o
n
o
u
r
d
ataset
an
d
ev
al
u
ated
th
e
m
o
d
els
u
n
d
er
attac
k
.
−
W
e
u
s
e
d
E
f
f
i
c
i
en
t
N
e
t
t
o
ex
t
r
a
ct
f
e
a
tu
r
es
f
r
o
m
im
ag
es
a
n
d
t
r
a
in
e
d
th
e
m
o
d
e
ls
a
b
o
v
e
o
n
t
h
e
s
e
e
x
t
r
ac
t
e
d
d
a
t
a
.
−
W
e
ac
h
iev
ed
h
i
g
h
ac
cu
r
ac
y
a
n
d
p
r
ec
is
io
n
in
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
co
m
p
ar
ed
to
o
th
er
b
ase
m
o
d
el
s
.
T
h
is
p
ap
er
is
o
r
g
an
ized
as
f
o
l
lo
w
s
.
I
n
s
ec
tio
n
2
,
w
e
ex
p
lai
n
th
e
d
ata
p
r
ep
ar
atio
n
an
d
p
r
ep
r
o
ce
s
s
in
g
o
p
er
atio
n
s
.
T
h
en
in
s
ec
tio
n
3
,
w
e
ex
p
lai
n
th
e
p
r
o
p
o
s
ed
m
et
h
o
d
s
in
clu
d
in
g
d
ee
p
lear
n
in
g
an
d
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es.
L
ater
,
w
e
s
h
o
w
th
e
r
esu
lts
o
f
t
h
e
i
m
p
le
m
e
n
ted
m
et
h
o
d
s
in
s
ec
tio
n
4
.
Fin
all
y
,
s
ec
tio
n
5
p
r
esen
ts
t
h
e
d
is
cu
s
s
io
n
a
n
d
co
n
clu
d
e
s
th
e
p
ap
er
.
2.
M
E
T
H
O
D
Ou
r
p
r
o
p
o
s
ed
m
et
h
o
d
in
teg
r
at
es
E
f
f
ic
ien
t
Net
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
ad
v
er
s
ar
ial
tr
ain
in
g
to
en
h
an
ce
th
e
r
o
b
u
s
t
n
es
s
o
f
a
s
teg
a
n
al
y
s
is
s
y
s
te
m
.
E
f
f
ic
ien
t
Net
is
c
h
o
s
en
f
o
r
its
ab
ilit
y
to
ca
p
tu
r
e
i
n
tr
icate
p
atter
n
s
in
i
m
a
g
e
d
ata
w
it
h
f
e
w
er
p
ar
a
m
et
er
s
,
m
a
k
i
n
g
it
b
o
th
ef
f
ec
ti
v
e
a
n
d
co
m
p
u
tatio
n
al
l
y
e
f
f
icie
n
t.
T
h
e
m
o
d
el
is
tr
ai
n
ed
u
s
i
n
g
a
d
ataset
au
g
m
e
n
ted
w
i
th
ad
v
er
s
ar
ial
e
x
a
m
p
les
g
en
er
ated
th
r
o
u
g
h
t
h
e
F
G
SM,
w
h
ic
h
in
tr
o
d
u
ce
s
s
li
g
h
t
p
er
tu
r
b
atio
n
s
to
th
e
in
p
u
t
d
ata.
T
h
is
p
r
o
ce
s
s
h
elp
s
th
e
m
o
d
e
l
lear
n
to
m
ai
n
tai
n
h
ig
h
p
er
f
o
r
m
an
ce
e
v
en
w
h
e
n
f
ac
i
n
g
ad
v
er
s
ar
ial
attac
k
s
.
T
h
e
ar
ch
itect
u
r
e
o
f
o
u
r
m
o
d
el
in
clu
d
es
la
y
er
s
s
u
ch
as
d
r
o
p
o
u
t
an
d
b
atch
n
o
r
m
aliza
t
io
n
to
i
m
p
r
o
v
e
g
en
er
aliza
tio
n
a
n
d
s
tab
ilit
y
b
y
r
ed
u
cin
g
o
v
er
f
itti
n
g
a
n
d
n
o
r
m
a
l
izin
g
f
ea
tu
r
e
i
n
p
u
ts
,
r
esp
ec
tiv
el
y
.
T
h
ese
la
y
er
s
ar
e
cr
itical
in
en
s
u
r
i
n
g
th
at
t
h
e
m
o
d
el
r
e
m
ain
s
r
o
b
u
s
t
u
n
d
er
v
ar
io
u
s
co
n
d
it
io
n
s
,
p
ar
ticu
lar
l
y
i
n
ad
v
er
s
ar
ial
s
et
t
in
g
s
.
C
o
m
p
ar
ed
to
tr
ad
itio
n
al
m
ac
h
in
e
lear
n
in
g
ap
p
r
o
ac
h
es
lik
e
r
an
d
o
m
f
o
r
est
,
o
u
r
m
o
d
el
o
f
f
er
s
s
u
p
er
io
r
ac
cu
r
ac
y
an
d
r
esil
ie
n
ce
,
d
e
m
o
n
s
t
r
atin
g
it
s
ca
p
ab
ilit
y
to
n
o
t
o
n
l
y
ac
c
u
r
atel
y
c
lass
if
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
2
,
A
p
r
il
20
25
:
3
9
3
-
401
396
i
m
a
g
es
b
u
t
also
w
it
h
s
ta
n
d
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
,
w
h
i
ch
is
ess
e
n
tial
f
o
r
r
ea
l
-
w
o
r
ld
ap
p
li
ca
tio
n
s
w
h
er
e
s
ec
u
r
it
y
an
d
r
eliab
ilit
y
ar
e
p
ar
a
m
o
u
n
t.
E
f
f
icien
tNet
is
a
h
i
g
h
-
p
er
f
o
r
m
an
ce
m
o
d
el
ar
ch
itect
u
r
e
o
p
ti
m
ized
f
o
r
co
m
p
u
ta
tio
n
al
e
f
f
icien
c
y
.
I
t
s
ca
les
d
ep
th
,
w
id
th
,
a
n
d
r
eso
lu
tio
n
in
a
co
m
p
o
u
n
d
m
a
n
n
er
,
en
ab
lin
g
s
m
a
ller
m
o
d
els
to
ac
h
ie
v
e
h
i
g
h
er
ac
cu
r
ac
y
th
an
m
a
n
y
co
n
v
e
n
tio
n
al
ar
ch
itect
u
r
es
[
1
0
]
.
I
n
ad
v
er
s
ar
i
al
tr
ain
i
n
g
,
a
m
o
d
el
i
s
iter
ativ
el
y
e
x
p
o
s
ed
to
ad
v
er
s
ar
iall
y
p
er
tu
r
b
ed
in
p
u
t
s
d
u
r
in
g
tr
ain
i
n
g
to
i
m
p
r
o
v
e
its
r
o
b
u
s
tn
e
s
s
.
T
h
is
m
et
h
o
d
u
s
es
a
d
v
er
s
ar
ial
ex
a
m
p
le
s
(
g
en
er
ated
v
ia
tec
h
n
iq
u
e
s
lik
e
FGSM)
a
lo
n
g
s
id
e
r
eg
u
lar
in
p
u
t
s
,
allo
w
in
g
t
h
e
m
o
d
el
t
o
ad
ap
t
an
d
lear
n
d
is
cr
i
m
i
n
ati
v
e
p
atter
n
s
t
h
at
r
e
d
u
ce
s
u
s
ce
p
tib
ilit
y
to
attac
k
s
.
T
h
is
o
n
g
o
i
n
g
ex
p
o
s
u
r
e
h
e
lp
s
t
h
e
m
o
d
el
g
e
n
er
alize
b
etter
to
u
n
s
ee
n
ad
v
er
s
ar
ial
s
c
en
ar
io
s
,
p
ar
ticu
lar
l
y
i
n
co
m
p
le
x
v
i
s
u
a
l ta
s
k
s
[
8
]
,
[
9
]
,
[
1
1
]
.
2
.
1
.
Da
t
a
prepa
ra
t
i
o
n a
nd
prepro
ce
s
s
ing
Fo
r
o
u
r
d
ataset,
w
e
u
tili
ze
d
2
0
,
0
0
0
im
ag
e
s
f
r
o
m
th
e
C
eleb
A
d
ataset
[
2
4
]
to
tr
ain
o
u
r
p
r
o
p
o
s
ed
m
o
d
el.
Fo
r
p
r
e
p
r
o
ce
s
s
in
g
,
f
ir
s
t,
to
cr
e
ate
s
teg
a
n
o
g
r
ap
h
ic
i
m
ag
e
s
,
w
e
i
m
p
le
m
e
n
ted
a
f
u
n
ctio
n
t
h
at
g
en
er
ated
r
an
d
o
m
s
tr
in
g
s
a
n
d
e
m
b
ed
d
ed
th
ese
s
t
r
in
g
s
i
n
to
th
e
i
m
a
g
es
u
s
in
g
t
h
e
least
s
i
g
n
i
f
ican
t
b
it
(
L
SB
)
m
et
h
o
d
.
T
h
e
d
ataset
w
a
s
th
e
n
s
p
lit in
to
tr
ain
in
g
,
v
a
lid
atio
n
,
an
d
test
s
ets,
w
it
h
an
8
0
:2
0
r
atio
b
et
w
ee
n
t
h
e
tr
ain
i
n
g
an
d
test
d
atasets
.
T
h
en
,
ea
ch
i
m
a
g
e
w
a
s
r
es
iz
ed
to
6
4
×
6
4
p
ix
els
to
e
n
s
u
r
e
ef
f
icie
n
t
p
r
o
ce
s
s
i
n
g
.
Af
ter
w
ar
d
,
w
e
g
en
er
ated
s
teg
a
n
o
g
r
ap
h
ic
i
m
a
g
es.
T
h
en
,
th
e
i
m
a
g
es
w
er
e
p
r
ep
ar
ed
u
s
in
g
E
f
f
icien
tNet
’
s
“
p
r
ep
r
o
ce
s
s
in
p
u
t
”
f
u
n
ctio
n
to
en
s
u
r
e
co
m
p
atib
ilit
y
w
i
th
t
h
e
p
r
e
-
tr
ain
ed
m
o
d
el.
Af
ter
ex
p
e
r
i
m
en
tin
g
w
it
h
v
ar
io
u
s
E
f
f
ici
en
tNet
m
o
d
el
s
an
d
ev
alu
a
tin
g
t
h
eir
p
er
f
o
r
m
a
n
ce
s
,
th
e
E
f
f
icien
tNetB
0
m
o
d
el
w
as
s
elec
ted
f
o
r
f
ea
tu
r
e
e
x
tr
ac
tio
n
,
g
en
er
ati
n
g
f
ea
tu
r
e
v
ec
to
r
s
f
o
r
ea
ch
i
m
a
g
e.
Ou
r
s
teg
a
n
al
y
s
i
s
s
y
s
te
m
co
n
s
i
s
ts
o
f
t
w
o
m
ai
n
co
m
p
o
n
e
n
ts
:
E
f
f
icien
tNet
f
o
r
f
ea
tu
r
e
e
x
tr
a
c
tio
n
a
n
d
a
d
en
s
e
n
eu
r
al
n
et
w
o
r
k
f
o
r
class
i
f
icatio
n
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
d
esi
g
n
ed
u
s
in
g
t
h
e
s
e
q
u
en
tial
A
P
I
f
r
o
m
T
en
s
o
r
Flo
w
’
s
Ker
a
s
lib
r
ar
y
,
f
ea
tu
r
in
g
m
u
ltip
le
f
u
ll
y
co
n
n
ec
ted
(
d
en
s
e
)
la
y
er
s
w
it
h
e
n
h
an
c
e
m
en
ts
f
o
r
tr
ain
in
g
ef
f
icien
c
y
a
n
d
p
er
f
o
r
m
an
ce
.
I
t
b
eg
in
s
w
it
h
an
i
n
p
u
t
la
y
er
tail
o
r
ed
to
th
e
d
im
e
n
s
io
n
alit
y
o
f
t
h
e
tr
ain
i
n
g
f
ea
t
u
r
es.
T
h
e
f
ir
s
t
h
id
d
en
la
y
er
co
n
tai
n
s
1
0
2
4
n
eu
r
o
n
s
,
f
o
llo
w
ed
b
y
b
atch
n
o
r
m
al
izatio
n
f
o
r
in
p
u
t
s
tab
ilizatio
n
,
l
ea
k
y
r
ec
tif
ied
li
n
ea
r
u
n
it
(
R
e
L
U)
ac
tiv
atio
n
w
it
h
a
n
e
g
ati
v
e
s
lo
p
e
o
f
0
.
0
1
to
ad
d
r
ess
th
e
d
y
i
n
g
R
eL
U
p
r
o
b
le
m
,
an
d
a
d
r
o
p
o
u
t
la
y
er
w
it
h
a
0
.
5
r
ate
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
Su
b
s
e
q
u
en
t
la
y
er
s
i
n
cl
u
d
e
5
1
2
,
2
5
6
,
1
2
8
,
6
4
,
3
2
,
1
6
,
8
,
an
d
4
n
eu
r
o
n
s
,
ea
ch
e
m
p
lo
y
i
ng
b
atc
h
n
o
r
m
aliza
tio
n
an
d
l
ea
k
y
R
e
L
U
ac
ti
v
atio
n
,
w
i
th
t
h
e
f
ir
s
t
t
h
r
ee
also
in
co
r
p
o
r
atin
g
d
r
o
p
o
u
t
.
L
ea
k
y
R
eL
U
ac
ti
v
atio
n
f
u
n
ctio
n
an
d
b
atch
n
o
r
m
aliza
tio
n
[
2
5
]
is
also
u
tili
ze
d
f
o
r
i
m
p
r
o
v
ed
tr
ain
i
n
g
[
2
6
]
.
T
h
e
o
u
tp
u
t
la
y
er
f
ea
t
u
r
es
a
s
in
g
le
n
e
u
r
o
n
w
it
h
a
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
f
o
r
b
in
ar
y
class
i
f
icatio
n
.
T
h
e
m
o
d
el
is
co
m
p
iled
w
it
h
th
e
A
d
a
m
o
p
ti
m
izer
[
2
7
]
(
lear
n
in
g
r
ate
o
f
0
.
0
0
0
3
)
,
B
in
ar
y
C
r
o
s
s
en
tr
o
p
y
lo
s
s
f
u
n
ct
io
n
,
an
d
ac
cu
r
ac
y
m
etr
ic
f
o
r
p
e
r
f
o
r
m
an
ce
ev
a
lu
at
io
n
.
T
h
is
ar
c
h
itect
u
r
e
lev
er
ag
e
s
b
atch
n
o
r
m
aliza
t
io
n
f
o
r
b
etter
co
n
v
er
g
en
ce
,
l
ea
k
y
R
e
L
U
t
o
av
o
id
d
ea
d
n
eu
r
o
n
s
,
an
d
d
r
o
p
o
u
t
to
en
h
an
ce
g
en
er
aliza
tio
n
,
m
a
k
i
n
g
it
r
o
b
u
s
t
an
d
ef
f
ec
t
iv
e
f
o
r
th
e
tas
k
at
h
an
d
.
Fig
u
r
e
1
s
h
o
w
s
th
e
ar
ch
i
tectu
r
e
o
f
th
e
m
o
d
el.
T
h
e
f
ig
u
r
e
is
g
e
n
er
ated
u
s
in
g
v
is
u
al
k
er
as [
2
8
]
.
Fig
u
r
e
1
.
Mo
d
el
ar
ch
itectu
r
e
W
e
em
p
lo
y
ed
t
h
e
FGSM
w
i
th
a
p
er
tu
r
b
atio
n
m
a
g
n
it
u
d
e
(
ep
s
ilo
n
)
s
et
to
0
.
0
1
to
g
en
er
ate
ad
v
er
s
ar
ial
ex
a
m
p
le
s
.
T
h
ese
ex
a
m
p
les
wer
e
u
s
ed
to
au
g
m
en
t
th
e
tr
ain
in
g
d
ata,
en
h
a
n
cin
g
th
e
m
o
d
el
’
s
r
o
b
u
s
tn
e
s
s
to
ad
v
er
s
ar
ial
attac
k
s
.
T
h
e
ad
v
e
r
s
ar
ial
tr
ain
i
n
g
lo
o
p
in
v
o
l
v
ed
g
e
n
er
atin
g
ad
v
er
s
ar
ial
e
x
a
m
p
les
o
n
t
h
e
f
l
y
a
n
d
co
m
b
i
n
i
n
g
t
h
e
m
w
i
th
o
r
ig
i
n
a
l
ex
a
m
p
les
d
u
r
i
n
g
ea
ch
tr
ain
in
g
ep
o
ch
.
T
h
is
lo
o
p
w
a
s
d
esig
n
ed
to
g
en
er
ate
ad
v
er
s
ar
ial
ex
a
m
p
le
s
d
y
n
a
m
i
ca
ll
y
d
u
r
in
g
ea
ch
tr
ain
in
g
ep
o
ch
.
T
h
at
in
v
o
lv
ed
co
n
tin
u
o
u
s
l
y
p
r
o
d
u
cin
g
n
e
w
ad
v
er
s
ar
ial
i
n
s
ta
n
ce
s
th
a
t
w
er
e
th
e
n
m
i
x
ed
w
it
h
t
h
e
o
r
i
g
in
al
t
r
ain
in
g
d
ata.
T
h
is
ap
p
r
o
ac
h
n
o
t
o
n
l
y
i
n
cr
ea
s
es
th
e
d
iv
er
s
it
y
o
f
tr
ain
i
n
g
ex
a
m
p
l
es
b
u
t
also
h
elp
s
th
e
m
o
d
el
d
ev
elo
p
a
m
o
r
e
n
u
a
n
ce
d
u
n
d
er
s
tan
d
i
n
g
o
f
t
h
e
u
n
d
er
l
y
in
g
d
ata
d
is
tr
ib
u
t
io
n
,
t
h
u
s
i
m
p
r
o
v
in
g
its
o
v
er
all
r
esil
i
en
ce
to
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
d
ve
r
s
a
r
ia
l
-
r
o
b
u
s
t steg
a
n
a
lysi
s
s
yste
m
leve
r
a
g
in
g
a
d
ve
r
s
a
r
ia
l tra
in
in
g
…
(
Th
a
kwa
n
A
kra
m
Ja
w
a
d
)
397
T
h
e
d
ec
is
io
n
to
ap
p
ly
t
h
e
FG
S
M
attac
k
o
n
t
h
e
f
ea
t
u
r
es e
x
tr
ac
ted
b
y
E
f
f
icien
tNet,
r
ath
er
t
h
an
d
ir
ec
tl
y
o
n
th
e
o
r
ig
i
n
al
i
m
a
g
es,
is
d
r
iv
en
b
y
s
e
v
er
al
k
e
y
co
n
s
id
er
atio
n
s
:
−
Hig
h
-
lev
el
f
ea
tu
r
e
m
a
n
ip
u
lati
o
n
:
E
f
f
icien
t
Net
is
d
esig
n
ed
to
ex
tr
ac
t
h
ig
h
-
le
v
el,
ab
s
tr
ac
t
f
ea
t
u
r
es
f
r
o
m
i
m
a
g
es
t
h
at
ar
e
m
o
r
e
r
ep
r
esen
tativ
e
o
f
th
e
u
n
d
er
l
y
i
n
g
p
atter
n
s
a
n
d
s
tr
u
ct
u
r
es
r
ele
v
a
n
t
f
o
r
class
i
f
icatio
n
task
s
.
B
y
ap
p
l
y
i
n
g
FGS
M
o
n
th
ese
h
i
g
h
-
lev
e
l
f
ea
tu
r
e
s
,
w
e
en
s
u
r
e
th
at
th
e
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
ar
e
cr
af
ted
to
tar
g
et
th
e
m
o
s
t
cr
iti
ca
l
asp
ec
ts
o
f
th
e
d
ata
u
s
ed
b
y
th
e
clas
s
i
f
icatio
n
m
o
d
el.
T
h
is
ap
p
r
o
ac
h
is
lik
el
y
to
g
e
n
er
ate
m
o
r
e
e
f
f
ec
t
iv
e
ad
v
er
s
ar
ial
e
x
a
m
p
les
t
h
at
ca
n
ch
al
len
g
e
t
h
e
m
o
d
el
’
s
r
o
b
u
s
t
n
ess
m
o
r
e
d
ir
ec
tl
y
.
−
Di
m
e
n
s
io
n
alit
y
r
ed
u
ct
io
n
:
t
h
e
f
ea
t
u
r
e
s
p
ac
e
p
r
o
d
u
ce
d
b
y
E
f
f
i
cien
tNet
is
t
y
p
icall
y
o
f
lo
w
er
d
i
m
e
n
s
io
n
alit
y
co
m
p
ar
ed
to
th
e
o
r
ig
i
n
al
i
m
a
g
e
s
p
ac
e.
Op
er
atin
g
i
n
th
i
s
r
ed
u
ce
d
f
ea
t
u
r
e
s
p
ac
e
allo
w
s
f
o
r
m
o
r
e
e
f
f
icie
n
t
co
m
p
u
ta
tio
n
o
f
ad
v
er
s
ar
ial
ex
a
m
p
les.
T
h
e
FGS
M
attac
k
,
wh
ich
in
v
o
l
v
es
ca
lc
u
lati
n
g
g
r
a
d
ien
ts
,
ca
n
b
e
co
m
p
u
tatio
n
all
y
in
te
n
s
iv
e;
t
h
u
s
,
ap
p
ly
i
n
g
it
in
t
h
e
f
ea
tu
r
e
s
p
a
ce
ca
n
s
ig
n
i
f
ica
n
tl
y
r
ed
u
ce
th
e
co
m
p
u
tatio
n
al
b
u
r
d
en
an
d
ex
p
ed
ite
th
e
g
e
n
er
atio
n
o
f
ad
v
er
s
ar
ial
e
x
a
m
p
le
s
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
o
ev
alu
ate
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
s
y
s
te
m
,
w
e
d
esig
n
ed
t
w
o
s
c
en
ar
io
s
.
I
n
th
e
f
ir
s
t
s
ce
n
ar
io
,
th
e
tr
ain
s
e
t
an
d
test
s
et
h
a
v
e
n
o
ad
v
er
s
ar
ial
in
s
ta
n
ce
s
.
T
h
e
p
u
r
p
o
s
e
in
s
ce
n
ar
io
1
is
to
ev
alu
ate
th
e
g
en
e
r
al
p
er
f
o
r
m
a
n
ce
o
f
o
u
r
s
teg
a
n
a
l
y
s
is
m
o
d
el
ag
a
in
s
t
n
o
r
m
al
s
te
g
a
n
o
g
r
ap
h
y
i
m
a
g
es.
C
o
n
v
er
s
el
y
,
in
t
h
e
s
ec
o
n
d
s
ce
n
ar
io
,
w
e
ad
d
ad
v
er
s
ar
ial
in
s
tan
ce
s
to
o
u
r
tr
ain
s
et
o
n
-
th
e
-
f
l
y
.
T
h
e
ev
al
u
atio
n
al
s
o
tak
es
p
lace
o
n
th
e
test
th
at
co
n
tai
n
s
ad
v
er
s
ar
ial
in
s
tan
ce
s
.
W
e
tr
ain
ed
th
r
ee
m
o
d
els:
r
an
d
o
m
f
o
r
es
t
m
o
d
el,
b
ase
m
o
d
el,
an
d
ad
v
er
s
ar
ial
m
o
d
el.
3
.
1
.
Scena
rio
1
I
n
th
e
f
ir
s
t
s
ce
n
ar
io
,
th
e
d
ata
s
et
d
id
n
o
t
co
n
tai
n
a
n
y
ad
v
er
s
a
r
ial
in
s
tan
ce
s
.
A
d
d
itio
n
all
y
,
we
tr
ain
ed
a
r
an
d
o
m
f
o
r
est
m
o
d
el
f
o
r
co
m
p
ar
is
o
n
w
it
h
o
u
r
p
r
o
p
o
s
ed
a
r
c
h
itect
u
r
e.
T
h
e
r
esu
lts
o
f
th
i
s
test
ar
e
p
r
esen
ted
in
T
ab
le
1
.
T
ab
le
1
.
P
er
f
o
r
m
a
n
ce
co
m
p
ar
i
s
o
n
o
f
m
o
d
el
s
in
s
ce
n
ar
io
1
M
e
t
r
i
c
B
a
se
mo
d
e
l
R
a
n
d
o
m
f
o
r
e
st
A
c
c
u
r
a
c
y
0
.
8
1
4
5
0
0
.
6
5
0
7
5
P
r
e
c
i
si
o
n
0
.
8
4
2
5
5
0
.
6
4
9
5
2
R
e
c
a
l
l
0
.
7
7
1
9
5
0
.
6
4
9
5
2
F1
-
s
c
o
r
e
0
.
8
0
5
7
0
0
.
6
4
9
5
2
R
O
C
-
A
U
C
0
.
8
1
4
3
5
0
.
6
5
0
7
4
3
.
2
.
Scena
rio
2
Fo
r
th
e
s
ec
o
n
d
p
ar
t
o
f
th
e
tes
t,
w
e
in
cl
u
d
ed
ad
v
er
s
ar
ial
i
n
s
ta
n
ce
s
i
n
t
h
e
d
ataset.
T
h
e
ad
v
er
s
a
r
ial
m
o
d
el,
w
h
ic
h
co
n
ta
in
s
d
r
o
p
o
u
t a
n
d
b
atch
n
o
r
m
aliza
t
io
n
la
y
er
s
,
w
a
s
tr
ain
ed
w
it
h
t
h
ese
ad
v
er
s
ar
ial
ex
a
m
p
le
s
.
Fi
g
u
r
e
2
d
ep
icts
th
e
ac
cu
r
ac
y
a
n
d
lo
s
s
p
lo
ts
f
o
r
th
e
ad
v
er
s
ar
ial
m
o
d
el.
I
n
co
n
tr
ast,
t
h
e
b
ase
m
o
d
el
an
d
r
an
d
o
m
f
o
r
est
d
id
n
o
t e
n
co
u
n
ter
ad
v
er
s
ar
ial
e
x
a
m
p
les d
u
r
i
n
g
t
h
eir
tr
ai
n
in
g
.
T
h
e
r
esu
l
ts
o
f
t
h
i
s
test
ar
e
p
r
esen
ted
in
T
ab
le
2
.
Fig
u
r
e
2
.
T
r
ain
in
g
p
lo
t o
f
th
e
ad
v
er
s
ar
ial
m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
2
,
A
p
r
il
20
25
:
3
9
3
-
401
398
T
h
e
ev
alu
atio
n
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
s
y
s
te
m
w
as
co
n
d
u
ct
ed
u
s
in
g
t
w
o
d
is
ti
n
ct
d
atasets
.
T
h
e
f
ir
s
t
d
ataset
d
id
n
o
t
co
n
tain
an
y
ad
v
er
s
ar
ial
in
s
ta
n
ce
s
,
an
d
w
e
u
s
ed
it
to
b
en
ch
m
ar
k
t
h
e
p
er
f
o
r
m
an
ce
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
a
g
ain
s
t
a
r
an
d
o
m
f
o
r
es
t
m
o
d
el.
A
s
s
h
o
w
n
i
n
T
ab
le
1
,
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
s
i
g
n
i
f
ic
an
tl
y
o
u
tp
er
f
o
r
m
ed
th
e
r
an
d
o
m
f
o
r
est
ac
r
o
s
s
all
m
etr
ics.
Sp
ec
i
f
icall
y
,
o
u
r
m
o
d
el
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
0
.
8
1
4
5
0
,
p
r
ec
is
io
n
o
f
0.
8
4
2
5
5
,
r
ec
all
o
f
0
.
7
7
1
9
5
,
F1
-
s
co
r
e
o
f
0
.
8
0
5
7
0
,
an
d
R
OC
-
A
U
C
o
f
0
.
8
1
4
3
5
,
w
h
i
le
th
e
r
an
d
o
m
f
o
r
est
la
g
g
ed
b
eh
in
d
w
it
h
an
ac
c
u
r
ac
y
o
f
0
.
6
5
0
7
5
,
p
r
ec
is
io
n
o
f
0
.
6
4
9
5
2
,
r
ec
all
o
f
0
.
6
4
9
5
2
,
F1
-
s
co
r
e
o
f
0
.
6
4
9
5
2
,
an
d
R
OC
-
A
U
C
o
f
0
.
6
5
0
7
4
.
I
n
th
e
s
ec
o
n
d
p
ar
t
o
f
th
e
test
,
ad
v
er
s
ar
ial
in
s
tan
ce
s
w
er
e
i
n
clu
d
ed
in
th
e
d
ataset
to
f
u
r
th
er
ev
alu
ate
th
e
r
o
b
u
s
tn
es
s
o
f
o
u
r
p
r
o
p
o
s
ed
m
o
d
el.
Un
lik
e
th
e
b
ase
m
o
d
el
an
d
r
an
d
o
m
f
o
r
est,
o
u
r
ad
v
er
s
ar
ial
m
o
d
el
w
a
s
tr
ain
ed
w
it
h
t
h
ese
ad
v
er
s
ar
ial
ex
a
m
p
l
es
an
d
f
ea
tu
r
ed
d
r
o
p
o
u
t
an
d
b
a
tch
n
o
r
m
al
iza
tio
n
la
y
er
s
.
T
h
e
r
esu
lt
s
,
p
r
esen
ted
in
T
ab
le
2
,
illu
s
tr
ate
a
s
i
g
n
i
f
ica
n
t
p
er
f
o
r
m
a
n
ce
d
r
o
p
f
o
r
th
e
b
ase
m
o
d
el
an
d
r
an
d
o
m
f
o
r
est
w
h
e
n
f
ac
ed
w
it
h
ad
v
er
s
ar
ial
in
s
ta
n
ce
s
,
w
it
h
ac
c
u
r
ac
ies
o
f
0
.
6
5
4
0
0
an
d
0
.
5
4
8
6
2
,
r
esp
ec
tiv
el
y
.
T
h
eir
o
th
er
m
etr
ics
also
d
ec
lin
ed
su
b
s
ta
n
tial
l
y
.
T
ab
le
2
.
P
er
f
o
r
m
a
n
ce
co
m
p
ar
i
s
o
n
o
f
d
if
f
er
e
n
t
m
o
d
els i
n
s
ce
n
ar
io
2
M
e
t
r
i
c
B
a
se
mo
d
e
l
R
a
n
d
o
m fo
r
e
st
A
d
v
e
r
sari
a
l
mo
d
e
l
A
c
c
u
r
a
c
y
0
.
6
5
4
0
0
0
.
5
4
8
6
2
0
.
7
2
0
8
7
P
r
e
c
i
si
o
n
0
.
6
2
0
1
6
0
.
6
1
8
2
9
0
.
7
0
9
8
3
R
e
c
a
l
l
0
.
7
8
8
5
0
0
.
2
4
5
8
6
0
.
7
4
3
8
5
F1
-
s
c
o
r
e
0
.
6
9
4
2
7
0
.
3
5
1
8
2
0
.
7
2
6
4
4
R
O
C
-
A
U
C
0
.
6
5
4
4
6
0
.
5
4
7
5
6
0
.
7
2
0
9
5
C
o
n
v
er
s
el
y
,
o
u
r
ad
v
er
s
ar
ial
m
o
d
el
d
em
o
n
s
tr
ated
s
u
p
er
io
r
r
esil
ie
n
ce
to
ad
v
er
s
ar
ial
attac
k
s
,
ac
h
iev
i
n
g
an
ac
cu
r
ac
y
o
f
0
.
7
2
0
8
7
,
p
r
ec
is
io
n
o
f
0
.
7
0
9
8
3
,
r
ec
all
o
f
0
.
7
4
3
8
5
,
F1
-
s
co
r
e
o
f
0
.
7
2
6
4
4
,
an
d
R
O
C
-
AUC
o
f
0
.
7
2
0
9
5
.
T
h
ese
r
esu
lts
i
n
d
icat
e
th
at
o
u
r
ad
v
er
s
ar
ial
t
r
ai
n
in
g
ap
p
r
o
ac
h
an
d
th
e
in
co
r
p
o
r
ati
o
n
o
f
d
r
o
p
o
u
t
an
d
b
atch
n
o
r
m
aliza
tio
n
la
y
er
s
e
f
f
e
ctiv
el
y
e
n
h
a
n
ce
d
t
h
e
m
o
d
el
’
s
r
o
b
u
s
tn
e
s
s
,
m
ai
n
tai
n
i
n
g
h
i
g
h
er
p
er
f
o
r
m
a
n
ce
lev
el
s
in
t
h
e
p
r
esen
ce
o
f
ad
v
er
s
ar
ial
ex
a
m
p
le
s
co
m
p
ar
ed
to
m
o
d
els
n
o
t
tr
ain
ed
w
it
h
s
u
c
h
ex
a
m
p
le
s
.
T
h
is
u
n
d
er
s
co
r
es
th
e
i
m
p
o
r
tan
ce
o
f
ad
v
er
s
ar
ial
tr
ain
i
n
g
i
n
d
ev
elo
p
in
g
r
o
b
u
s
t
m
ac
h
i
n
e
lear
n
in
g
m
o
d
els
c
ap
ab
le
o
f
r
esis
ti
n
g
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
.
T
h
e
k
e
y
d
i
f
f
er
e
n
ce
b
et
w
ee
n
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
an
d
t
h
e
r
an
d
o
m
f
o
r
est
m
o
d
el
lies
in
t
h
eir
ar
ch
itect
u
r
e
an
d
ad
ap
tab
ilit
y
to
co
m
p
le
x
d
ata
p
atter
n
s
,
p
ar
ticu
lar
l
y
w
h
e
n
f
ac
ed
w
it
h
ad
v
er
s
ar
ial
i
n
s
ta
n
ce
s
.
O
u
r
p
r
o
p
o
s
ed
m
o
d
el
is
a
n
e
u
r
al
n
et
w
o
r
k
th
at
in
co
r
p
o
r
ates
ad
v
an
ce
d
tech
n
iq
u
e
s
s
u
c
h
as
b
atch
n
o
r
m
ali
za
tio
n
an
d
d
r
o
p
o
u
t
la
y
er
s
,
allo
w
i
n
g
it
to
lear
n
m
o
r
e
in
tr
ic
ate
f
ea
t
u
r
es f
r
o
m
t
h
e
d
ata
an
d
m
ain
tain
r
o
b
u
s
t
p
er
f
o
r
m
an
ce
u
n
d
er
v
ar
ied
co
n
d
itio
n
s
.
As
s
h
o
w
n
i
n
t
h
e
r
esu
lt
s
f
r
o
m
T
ab
le
1
,
th
i
s
d
esi
g
n
led
to
a
clea
r
p
er
f
o
r
m
an
ce
ad
v
an
ta
g
e
o
v
er
t
h
e
r
an
d
o
m
f
o
r
est
m
o
d
el,
w
h
ich
r
elies o
n
en
s
e
m
b
le
lear
n
in
g
b
u
t
lack
s
t
h
e
d
ee
p
le
ar
n
in
g
m
o
d
el
’
s
ab
ilit
y
to
ca
p
tu
r
e
n
o
n
-
li
n
ea
r
r
elatio
n
s
h
ip
s
e
f
f
ec
ti
v
el
y
.
T
h
e
r
an
d
o
m
f
o
r
est
m
o
d
el
ex
h
ib
ited
s
i
g
n
i
f
ica
n
tl
y
lo
w
er
a
cc
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
R
OC
-
AUC
co
m
p
ar
ed
to
o
u
r
p
r
o
p
o
s
e
d
m
o
d
el,
h
ig
h
li
g
h
ti
n
g
t
h
e
li
m
i
tatio
n
s
o
f
tr
ad
itio
n
a
l
m
ac
h
in
e
lear
n
i
n
g
m
et
h
o
d
s
in
h
an
d
li
n
g
co
m
p
lex
d
ata
s
ets
w
it
h
o
u
t a
d
v
er
s
ar
ial
ex
a
m
p
les.
T
h
e
in
clu
s
io
n
o
f
ad
v
er
s
ar
ial
e
x
a
m
p
le
s
d
u
r
i
n
g
tr
ain
i
n
g
f
u
r
t
h
e
r
u
n
d
er
s
co
r
es
t
h
e
i
m
p
o
r
ta
n
ce
o
f
ex
p
o
s
i
n
g
m
o
d
el
s
to
d
iv
er
s
e
d
ata
v
ar
iat
i
o
n
s
to
e
n
h
a
n
ce
r
o
b
u
s
t
n
e
s
s
.
O
u
r
ad
v
er
s
ar
ial
m
o
d
el,
d
esi
g
n
e
d
w
it
h
t
h
e
ca
p
ab
ilit
y
to
tr
ain
o
n
ad
v
er
s
ar
ial
ex
a
m
p
l
es,
s
h
o
w
ed
r
e
m
ar
k
ab
le
r
esil
ie
n
ce
ag
ai
n
s
t a
d
v
er
s
ar
ial
attac
k
s
,
as e
v
id
en
ce
d
b
y
it
s
s
u
p
er
io
r
p
er
f
o
r
m
a
n
ce
m
etr
ics
i
n
T
ab
le
2
.
I
n
co
n
tr
ast,
b
o
th
t
h
e
b
ase
m
o
d
el
an
d
th
e
r
a
n
d
o
m
f
o
r
est
ex
p
er
ien
ce
d
a
n
o
tab
le
d
ec
lin
e
in
ac
cu
r
ac
y
an
d
o
th
er
p
e
r
f
o
r
m
a
n
ce
m
et
r
ics
w
h
en
test
ed
ag
ai
n
s
t
ad
v
er
s
ar
ial
ex
a
m
p
les,
d
em
o
n
s
tr
ati
n
g
t
h
eir
v
u
l
n
er
a
b
ilit
y
.
T
h
e
ad
v
er
s
ar
ial
tr
ai
n
in
g
p
r
o
ce
s
s
,
co
u
p
led
w
it
h
th
e
ar
ch
i
tectu
r
al
en
h
a
n
ce
m
en
ts
o
f
o
u
r
m
o
d
el,
p
r
o
v
ed
ef
f
ec
ti
v
e
in
i
m
p
r
o
v
i
n
g
it
s
ad
ap
tab
ilit
y
an
d
r
o
b
u
s
tn
e
s
s
,
m
ai
n
tai
n
in
g
h
ig
h
er
p
er
f
o
r
m
a
n
ce
lev
el
s
co
m
p
ar
ed
to
m
o
d
els
th
at
w
er
e
n
o
t
tr
ain
ed
w
it
h
ad
v
er
s
ar
ial
ex
a
m
p
les.
T
h
is
h
ig
h
er
p
er
f
o
r
m
a
n
ce
is
n
o
ticea
b
le
in
F
ig
u
r
e
3
w
h
er
e
d
i
f
f
er
e
n
t
m
etr
ic
s
ar
e
co
m
p
ar
ed
ac
r
o
s
s
m
o
d
els.
E
f
f
icien
tNet
h
a
s
d
e
m
o
n
s
tr
ated
ex
ce
p
tio
n
al
ca
p
ab
ilit
y
as
a
f
ea
tu
r
e
ex
tr
ac
to
r
,
ca
p
tu
r
in
g
in
tr
icate
d
etails
th
at
co
n
tr
ib
u
te
to
p
r
ec
is
e
clas
s
if
icatio
n
i
n
o
u
r
s
te
g
a
n
al
y
s
is
s
y
s
te
m
.
I
t
s
ar
ch
itect
u
r
al
e
f
f
icie
n
c
y
allo
w
s
i
t
to
f
o
c
u
s
o
n
cr
u
cial
p
atter
n
s
w
it
h
i
n
th
e
d
ata,
en
ab
lin
g
a
r
o
b
u
s
t
id
en
tific
atio
n
o
f
f
ea
tu
r
e
s
n
ec
es
s
ar
y
f
o
r
d
etec
tin
g
h
id
d
e
n
m
es
s
ag
e
s
.
Ou
r
r
esu
l
ts
s
h
o
w
t
h
at
tr
ain
i
n
g
w
ith
ad
v
er
s
ar
ial
e
x
a
m
p
le
s
s
i
g
n
i
f
ica
n
tl
y
e
n
h
an
ce
s
th
e
r
o
b
u
s
tn
e
s
s
o
f
th
e
s
y
s
te
m
,
a
s
r
ef
lecte
d
in
th
e
i
m
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
m
et
r
ics
u
n
d
er
ad
v
er
s
ar
ial
co
n
d
iti
o
n
s
.
T
h
is
ap
p
r
o
ac
h
ef
f
ec
tiv
e
l
y
b
o
ls
ter
s
th
e
m
o
d
el
’
s
r
esis
ta
n
ce
to
ad
v
er
s
ar
ia
l
atta
ck
s
,
w
h
ic
h
ar
e
d
esig
n
ed
to
m
i
s
lead
th
e
m
o
d
el
b
y
in
tr
o
d
u
ci
n
g
s
u
b
tle
p
er
tu
r
b
atio
n
s
.
T
h
e
b
ase
m
o
d
el,
d
esp
ite
it
s
in
itial
ef
f
ec
t
iv
e
n
es
s
,
r
ev
ea
le
d
a
s
u
s
ce
p
tib
ilit
y
to
th
ese
a
d
v
er
s
ar
ial
m
an
ip
u
latio
n
s
,
u
n
d
er
s
co
r
in
g
th
e
n
ec
e
s
s
it
y
f
o
r
ad
v
er
s
ar
ial
tr
ain
i
n
g
i
n
r
ein
f
o
r
ci
n
g
m
o
d
e
l
r
esil
ien
ce
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
d
ve
r
s
a
r
ia
l
-
r
o
b
u
s
t steg
a
n
a
lysi
s
s
yste
m
leve
r
a
g
in
g
a
d
ve
r
s
a
r
ia
l tra
in
in
g
…
(
Th
a
kwa
n
A
kra
m
Ja
w
a
d
)
399
Fig
u
r
e
3
.
Me
tr
ics co
m
p
ar
is
o
n
b
et
w
ee
n
tr
ain
ed
m
o
d
els
T
h
e
r
an
d
o
m
f
o
r
est
m
o
d
el,
w
h
ile
s
i
m
p
ler
an
d
p
r
o
v
id
in
g
a
co
m
p
e
titi
v
e
b
aselin
e,
lack
s
th
e
r
o
b
u
s
tn
e
s
s
an
d
ad
ap
tab
ilit
y
o
f
f
er
ed
b
y
d
ee
p
lear
n
in
g
m
o
d
els
l.
A
l
th
o
u
g
h
r
an
d
o
m
f
o
r
ests
ar
e
ad
v
a
n
t
ag
eo
u
s
i
n
s
ce
n
ar
io
s
w
it
h
li
m
ited
co
m
p
u
tatio
n
al
r
eso
u
r
ce
s
a
n
d
o
f
f
er
q
u
ic
k
,
s
tr
aig
h
t
f
o
r
w
ar
d
i
m
p
le
m
e
n
tat
io
n
,
t
h
e
y
d
o
n
o
t
m
atc
h
t
h
e
ca
p
ab
ilit
y
o
f
d
ee
p
lear
n
i
n
g
m
o
d
els
in
h
a
n
d
li
n
g
co
m
p
lex
d
ata
tr
an
s
f
o
r
m
atio
n
s
a
n
d
ad
v
er
s
a
r
ial
ch
alle
n
g
e
s
.
T
h
e
co
m
b
i
n
atio
n
o
f
E
f
f
icie
n
tNet
a
n
d
ad
v
er
s
ar
ial
tr
ai
n
i
n
g
e
m
er
g
e
s
as
a
p
o
w
er
f
u
l stra
te
g
y
f
o
r
s
te
g
an
al
y
s
is
,
e
n
ab
lin
g
th
e
d
etec
tio
n
o
f
h
id
d
en
m
e
s
s
ag
e
s
w
h
ile
m
ai
n
tai
n
i
n
g
r
o
b
u
s
t
n
es
s
a
g
ai
n
s
t
ad
v
er
s
ar
ial
p
er
tu
r
b
atio
n
s
.
T
h
is
h
ig
h
li
g
h
ts
t
h
e
cr
itical
i
m
p
o
r
tan
ce
o
f
ad
v
er
s
ar
ial
tr
ai
n
i
n
g
i
n
d
ev
elo
p
in
g
m
ac
h
in
e
lear
n
i
n
g
m
o
d
el
s
th
a
t
ar
e
n
o
t
o
n
l
y
ac
c
u
r
ate
b
u
t
also
r
esil
ie
n
t
to
m
a
n
ip
u
l
ati
v
e
p
er
tu
r
b
atio
n
s
.
Su
c
h
r
o
b
u
s
t
n
ess
e
n
s
u
r
es
r
el
iab
ilit
y
i
n
p
r
ac
tical
ap
p
licatio
n
s
w
h
er
e
ad
v
er
s
ar
ia
l
attac
k
s
p
o
s
e
a
s
ig
n
i
f
ica
n
t
t
h
r
ea
t,
r
ein
f
o
r
cin
g
t
h
e
m
o
d
el
’
s
ef
f
ec
ti
v
e
n
es
s
a
n
d
d
ep
en
d
ab
ilit
y
in
r
ea
l
-
w
o
r
ld
s
ce
n
ar
io
s
.
B
y
i
n
co
r
p
o
r
atin
g
th
ese
ele
m
e
n
ts
,
o
u
r
ap
p
r
o
ac
h
n
o
t
o
n
l
y
i
m
p
r
o
v
e
s
d
etec
tio
n
ca
p
ab
ilit
ies b
u
t a
l
s
o
s
ets a
f
o
u
n
d
atio
n
f
o
r
f
u
t
u
r
e
en
h
an
ce
m
e
n
ts
i
n
m
ac
h
in
e
lear
n
i
n
g
s
ec
u
r
it
y
.
3
.
3
.
Co
m
pa
ri
s
o
n
Giv
e
n
th
e
n
o
v
elt
y
o
f
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
,
d
ir
ec
t
c
o
m
p
ar
i
s
o
n
w
it
h
e
x
is
t
in
g
m
eth
o
d
s
i
s
c
h
alle
n
g
i
n
g
d
u
e
to
th
e
lack
o
f
d
ir
ec
tl
y
co
m
p
ar
ab
le
s
tu
d
ies.
Ho
w
e
v
er
,
to
p
r
o
v
id
e
co
n
tex
t
f
o
r
t
h
e
p
er
f
o
r
m
an
ce
o
f
o
u
r
m
e
th
o
d
,
w
e
h
a
v
e
s
elec
ted
a
b
ase
lin
e
ap
p
r
o
ac
h
an
d
also
o
f
f
er
a
t
h
eo
r
etica
l
co
m
p
ar
is
o
n
to
h
i
g
h
lig
h
t
th
e
u
n
iq
u
e
asp
ec
t
s
o
f
o
u
r
w
o
r
k
.
T
ab
le
3
s
u
m
m
ar
izes
k
e
y
d
i
f
f
er
e
n
ce
s
a
n
d
th
e
p
o
ten
tial a
d
v
an
ta
g
e
s
o
f
o
u
r
m
e
th
o
d
.
T
ab
le
3
.
C
o
m
p
ar
is
o
n
o
f
m
et
h
o
d
s
M
e
t
h
o
d
o
l
o
g
y
T
e
c
h
n
i
q
u
e
I
mag
e
si
z
e
N
o
t
a
b
l
e
c
h
a
r
a
c
t
e
r
i
st
i
c
s
P
r
o
p
o
se
d
me
t
h
o
d
D
N
N
w
i
t
h
a
d
v
e
r
sari
a
l
t
r
a
i
n
i
n
g
6
4
×
6
4
C
o
mp
l
e
x
i
n
t
r
a
i
n
i
n
g
,
h
i
g
h
p
r
e
c
i
si
o
n
a
n
d
a
c
c
u
r
a
c
y
,
r
o
b
u
st
a
g
a
i
n
st
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
R
a
n
d
o
m
f
o
r
e
st
En
se
mb
l
e
6
4
×
6
4
S
i
mp
l
e
,
v
u
l
n
e
r
a
b
l
e
a
g
a
i
n
st
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s,
l
o
w
a
c
c
u
r
a
c
y
a
n
d
p
r
e
c
i
si
o
n
i
n
c
o
mp
a
r
i
so
n
w
i
t
h
o
t
h
e
r
mo
d
e
l
s.
B
a
se
mo
d
e
l
DNN
6
4
×
6
4
S
i
mp
l
e
t
r
a
i
n
i
n
g
,
h
i
g
h
a
c
c
u
r
a
c
y
i
n
d
e
t
e
c
t
i
n
g
n
o
r
mal
e
x
a
mp
l
e
s,
v
u
l
n
e
r
a
b
l
e
a
g
a
i
n
st
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
3
.
4
.
L
i
m
it
a
t
io
ns
I
n
th
is
s
t
u
d
y
,
o
n
e
o
f
th
e
m
ain
li
m
itatio
n
s
is
th
e
lac
k
o
f
w
e
ll
-
estab
lis
h
ed
b
aselin
e
m
o
d
el
s
f
o
r
co
m
p
ar
is
o
n
.
Gi
v
en
th
at
t
h
i
s
f
ield
is
r
elativ
e
l
y
n
e
w
a
n
d
r
ap
id
l
y
ev
o
l
v
i
n
g
,
t
h
er
e
ar
e
f
e
w
,
if
a
n
y
,
b
en
c
h
m
ar
k
m
et
h
o
d
s
w
it
h
s
ta
n
d
ar
d
ized
d
a
tasets
o
r
p
er
f
o
r
m
a
n
ce
m
etr
ics
th
at
w
o
u
ld
p
r
o
v
id
e
a
d
ir
ec
t
c
o
m
p
ar
is
o
n
f
o
r
o
u
r
m
o
d
el.
C
o
n
s
eq
u
en
tl
y
,
t
h
is
l
i
m
i
ts
o
u
r
ab
ilit
y
to
q
u
a
n
tita
tiv
e
l
y
ass
es
s
o
u
r
m
o
d
el
’
s
p
er
f
o
r
m
a
n
ce
ag
ain
s
t
a
d
iv
er
s
e
s
et
o
f
ap
p
r
o
ac
h
es.
Fu
tu
r
e
w
o
r
k
in
th
i
s
ar
ea
w
ill
lik
el
y
b
en
ef
it
f
r
o
m
th
e
d
ev
elo
p
m
en
t
o
f
m
o
r
e
estab
lis
h
ed
b
en
ch
m
ar
k
s
,
w
h
ich
w
o
u
ld
p
r
o
v
id
e
clea
r
er
s
tan
d
ar
d
s
f
o
r
e
v
alu
a
tin
g
a
n
d
co
m
p
ar
in
g
ad
v
an
ce
m
e
n
ts
ac
r
o
s
s
d
if
f
er
e
n
t
m
o
d
els.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
23
,
No
.
2
,
A
p
r
il
20
25
:
3
9
3
-
401
400
4.
CO
N
CL
U
SI
O
N
T
h
is
p
ap
er
p
r
esen
ts
a
n
o
v
el
s
teg
a
n
al
y
s
i
s
s
y
s
te
m
t
h
at
lev
e
r
ag
es
E
f
f
icie
n
tNet
f
o
r
ef
f
icie
n
t
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
ad
v
er
s
ar
ial
tr
ai
n
in
g
to
en
h
a
n
ce
r
o
b
u
s
t
n
es
s
.
Ou
r
r
es
u
lts
d
e
m
o
n
s
tr
ate
th
at
t
h
e
p
r
o
p
o
s
ed
s
y
s
te
m
ac
h
iev
e
s
h
ig
h
ac
c
u
r
ac
y
in
d
ete
ctin
g
s
te
g
a
n
o
g
r
ap
h
ic
co
n
ten
t
a
n
d
m
ai
n
tai
n
s
r
esil
ie
n
ce
a
g
ain
s
t
ad
v
er
s
ar
ial
attac
k
s
.
T
h
is
d
u
al
ap
p
r
o
ac
h
r
ep
r
esen
ts
a
s
ig
n
i
f
ica
n
t
ad
v
a
n
ce
m
en
t
in
s
teg
a
n
al
y
s
i
s
,
o
f
f
er
i
n
g
a
r
o
b
u
s
t
s
o
lu
tio
n
f
o
r
s
ec
u
r
e
in
f
o
r
m
atio
n
tr
a
n
s
m
is
s
io
n
i
n
th
e
r
ea
l
s
ce
n
ar
io
s
w
h
er
e
th
e
h
ig
h
p
er
f
o
r
m
an
ce
o
f
d
etec
tio
n
s
y
s
te
m
is
v
ital.
Fu
tu
r
e
r
esear
ch
co
u
ld
e
x
p
lo
r
e
th
e
in
t
eg
r
atio
n
o
f
ad
v
a
n
ce
d
g
en
er
ati
v
e
m
o
d
el
s
a
n
d
ad
v
er
s
ar
ial
tec
h
n
iq
u
es
to
f
u
r
t
h
er
i
m
p
r
o
v
e
th
e
s
y
s
te
m
’
s
p
er
f
o
r
m
an
ce
an
d
r
o
b
u
s
t
n
es
s
.
RE
F
E
R
E
NC
E
S
[
1
]
N
.
F
.
Jo
h
n
so
n
a
n
d
S
.
Ja
j
o
d
i
a
,
“
Ex
p
l
o
r
i
n
g
st
e
g
a
n
o
g
r
a
p
h
y
:
se
e
i
n
g
t
h
e
u
n
se
e
n
,
”
C
o
m
p
u
t
e
r
,
v
o
l
.
3
1
,
n
o
.
2
,
p
p
.
2
6
-
3
4
,
F
e
b
.
1
9
9
8
,
d
o
i
:
1
0
.
1
1
0
9
/
M
C
.
1
9
9
8
.
4
6
5
5
2
8
1
.
[
2
]
N
.
M
e
g
h
a
n
a
t
h
a
n
a
n
d
L
.
N
a
y
a
k
,
“
S
t
e
g
a
n
a
l
y
si
s
a
l
g
o
r
i
t
h
ms
f
o
r
d
e
t
e
c
t
i
n
g
t
h
e
h
i
d
d
e
n
i
n
f
o
r
mat
i
o
n
i
n
i
mag
e
,
a
u
d
i
o
a
n
d
v
i
d
e
o
c
o
v
e
r
me
d
i
a
,
”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
N
e
t
w
o
r
k
S
e
c
u
r
i
t
y
&
I
t
s A
p
p
l
i
c
a
t
i
o
n
(
I
J
N
S
A)
,
v
o
l
.
2
,
n
o
.
1
,
p
p
.
43
-
55
,
2
0
1
0
.
[
3
]
V.
V
e
r
ma,
S.
K
.
M
u
t
t
o
o
,
a
n
d
V
.
B.
S
i
n
g
h
,
“
D
e
t
e
c
t
i
n
g
st
e
g
o
mal
w
a
r
e
:
mal
i
c
i
o
u
s
i
mag
e
st
e
g
a
n
o
g
r
a
p
h
y
a
n
d
i
t
s
i
n
t
r
u
si
o
n
i
n
w
i
n
d
o
w
s
,
”
i
n
S
e
c
u
r
i
t
y
,
Pr
i
v
a
c
y
a
n
d
D
a
t
a
An
a
l
y
t
i
c
s:
S
e
l
e
c
t
Pro
c
e
e
d
i
n
g
s
o
f
I
S
PD
A
2
0
2
1
,
S
i
n
g
a
p
o
r
e
:
S
p
r
i
n
g
e
r
,
2
0
2
2
,
p
p
.
1
0
3
-
116
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
981
-
16
-
9
0
8
9
-
1
_
9
.
[
4
]
Y.
S
h
a
n
g
,
S.
Ji
a
n
g
,
D
.
Y
e
,
a
n
d
J
.
H
u
a
n
g
,
“
En
h
a
n
c
i
n
g
t
h
e
se
c
u
r
i
t
y
o
f
d
e
e
p
l
e
a
r
n
i
n
g
s
t
e
g
a
n
o
g
r
a
p
h
y
v
i
a
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
,
”
Ma
t
h
e
m
a
t
i
c
s
,
v
o
l
.
8
,
n
o
.
9
,
p
p
.
1
-
1
0
,
2
0
2
0
,
d
o
i
:
1
0
.
3
3
9
0
/
m
a
t
h
8
0
9
1
4
4
6
.
[
5
]
K.
K
a
r
a
mp
i
d
i
s,
E
.
K
a
v
a
l
l
i
e
r
a
t
o
u
,
a
n
d
G
.
P
a
p
a
d
o
u
r
a
k
i
s,
“
A
r
e
v
i
e
w
o
f
i
mag
e
st
e
g
a
n
a
l
y
si
s
t
e
c
h
n
i
q
u
e
s
f
o
r
d
i
g
i
t
a
l
f
o
r
e
n
si
c
s
,
”
J
o
u
r
n
a
l
o
f
i
n
f
o
rm
a
t
i
o
n
se
c
u
ri
t
y
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
4
0
,
p
p
.
217
-
235
,
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
i
s
a
.
2
0
1
8
.
0
4
.
0
0
5
.
[
6
]
N
.
A
k
h
t
a
r
a
n
d
A
.
M
i
a
n
,
“
T
h
r
e
a
t
o
f
a
d
v
e
r
sari
a
l
a
t
t
a
c
k
s
o
n
d
e
e
p
l
e
a
r
n
i
n
g
i
n
c
o
mp
u
t
e
r
v
i
s
i
o
n
:
a
s
u
r
v
e
y
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
6
,
p
p
.
1
4
4
1
0
-
1
4
4
3
0
,
2
0
1
8
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
8
0
1
.
0
0
5
5
3
.
[
7
]
X
.
Y
u
a
n
,
P
.
H
e
,
Q
.
Z
h
u
a
n
d
X
.
L
i
,
“
A
d
v
e
r
sari
a
l
Ex
a
mp
l
e
s:
A
t
t
a
c
k
s
a
n
d
D
e
f
e
n
se
s
f
o
r
D
e
e
p
L
e
a
r
n
i
n
g
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
rks
a
n
d
L
e
a
rn
i
n
g
S
y
st
e
m
s
,
v
o
l
.
3
0
,
n
o
.
9
,
p
p
.
2
8
0
5
-
2
8
2
4
,
S
e
p
t
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
T
N
N
L
S
.
2
0
1
8
.
2
8
8
6
0
1
7
.
[
8
]
A.
K
u
r
a
k
i
n
,
I
.
G
o
o
d
f
e
l
l
o
w
,
a
n
d
S
.
B
e
n
g
i
o
,
“
A
d
v
e
r
sari
a
l
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
t
sca
l
e
,
”
I
C
L
R
2
0
1
7
C
o
n
f
e
r
e
n
c
e
T
r
a
c
k
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
e
a
rn
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s
,
2
0
1
7
.
[
9
]
F.
T
r
a
mè
r
,
A.
K
u
r
a
k
i
n
,
N.
P
a
p
e
r
n
o
t
,
I.
G
o
o
d
f
e
l
l
o
w
,
D
.
B
o
n
e
h
,
a
n
d
P
.
M
c
D
a
n
i
e
l
,
“
En
se
mb
l
e
a
d
v
e
r
sari
a
l
t
r
a
i
n
i
n
g
:
a
t
t
a
c
k
s
a
n
d
d
e
f
e
n
se
s
,
”
a
r
Xi
v
,
2
0
1
7
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
7
0
5
.
0
7
2
0
4
.
[
1
0
]
M.
T
a
n
a
n
d
Q.
Le
,
“
Ef
f
i
c
i
e
n
t
N
e
t
:
r
e
t
h
i
n
k
i
n
g
mo
d
e
l
sc
a
l
i
n
g
f
o
r
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
,
2
0
1
9
,
v
o
l
9
7
,
p
p
.
6
1
0
5
-
6
1
1
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
9
0
5
.
1
1
9
4
6
.
[
1
1
]
I
.
J
.
G
o
o
d
f
e
l
l
o
w
,
J
.
S
h
l
e
n
s,
a
n
d
C
.
S
z
e
g
e
d
y
,
“
Ex
p
l
a
i
n
i
n
g
a
n
d
h
a
r
n
e
ssi
n
g
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
,
”
a
rXi
v
,
2
0
1
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
4
1
2
.
6
5
7
2
.
[
1
2
]
S
.
M
.
M
o
o
sav
i
d
e
z
f
o
o
l
i
,
A
.
F
a
w
z
i
,
a
n
d
P
.
F
r
o
ssard
,
“
D
e
e
p
F
o
o
l
:
a
si
mp
l
e
a
n
d
a
c
c
u
r
a
t
e
me
t
h
o
d
t
o
f
o
o
l
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
p
p.
2
5
7
4
–
2
5
8
2
,
2
0
1
5
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
5
1
1
.
0
4
5
9
9
.
[
1
3
]
N
.
C
a
r
l
i
n
i
a
n
d
D
.
W
a
g
n
e
r
,
“
T
o
w
a
r
d
s
e
v
a
l
u
a
t
i
n
g
t
h
e
r
o
b
u
s
t
n
e
ss
o
f
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
S
e
c
u
r
i
t
y
a
n
d
Pri
v
a
c
y
,
2
0
1
7
,
d
o
i
:
1
0
.
1
1
0
9
/
S
P
.
2
0
1
7
.
4
9
.
[
1
4
]
X
.
Q
i
n
,
B
.
L
i
,
S
.
T
a
n
,
W
.
T
a
n
g
,
a
n
d
J.
H
u
a
n
g
,
“
G
r
a
d
u
a
l
l
y
e
n
h
a
n
c
e
d
a
d
v
e
r
sari
a
l
p
e
r
t
u
r
b
a
t
i
o
n
s
o
n
c
o
l
o
r
p
i
x
e
l
v
e
c
t
o
r
s
f
o
r
i
mag
e
st
e
g
a
n
o
g
r
a
p
h
y
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
C
i
r
c
u
i
t
s a
n
d
S
y
s
t
e
m
s
f
o
r
Vi
d
e
o
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
3
2
,
n
o
.
8
,
p
p
.
5
1
1
0
-
5
1
2
3
,
A
u
g
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
S
V
T
.
2
0
2
2
.
3
1
4
8
4
0
6
.
[
1
5
]
A.
N.
B
h
a
g
o
j
i
,
D.
C
u
l
l
i
n
a
,
C
.
S
i
t
a
w
a
r
i
n
,
a
n
d
P
.
M
i
t
t
a
l
,
“
En
h
a
n
c
i
n
g
r
o
b
u
s
t
n
e
ss
o
f
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
s
y
st
e
ms
v
i
a
d
a
t
a
t
r
a
n
sf
o
r
ma
t
i
o
n
s
,
”
i
n
2
0
1
8
5
2
n
d
A
n
n
u
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
a
n
d
S
y
s
t
e
m
s
(
C
I
S
S
)
,
2
0
1
8
,
p
p
.
1
-
5
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
7
0
4
.
0
2
6
5
4
.
[
1
6
]
J.
L
i
u
,
W.
Z
h
a
n
g
,
Y.
Z
h
a
n
g
,
D
.
H
o
u
,
Y.
L
i
u
,
H
.
Z
h
a
,
a
n
d
N
.
Y
u
,
“
D
e
t
e
c
t
i
o
n
b
a
se
d
d
e
f
e
n
se
a
g
a
i
n
st
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
f
r
o
m
t
h
e
st
e
g
a
n
a
l
y
si
s
p
o
i
n
t
o
f
v
i
e
w
,
”
in
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
I
EEE
/
C
VF
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r V
i
s
i
o
n
a
n
d
P
a
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
,
2
0
1
9
,
p
p
.
4
8
2
5
-
4
8
3
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
8
0
6
.
0
9
1
8
6
.
[
1
7
]
N
.
D
a
s,
M
.
S
h
a
n
b
h
o
g
u
e
,
S
.
-
T
.
C
h
e
n
,
F
.
H
o
h
man
,
L
.
C
h
e
n
,
M
.
E
.
K
o
u
n
a
v
i
s,
a
n
d
D
.
H
.
Ch
au
,
“
K
e
e
p
i
n
g
t
h
e
b
a
d
g
u
y
s o
u
t
:
p
r
o
t
e
c
t
i
n
g
a
n
d
v
a
c
c
i
n
a
t
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
w
i
t
h
j
p
e
g
c
o
mp
r
e
ssi
o
n
,
”
a
rX
i
v
,
2
0
1
7
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
7
0
5
.
0
2
9
0
0
.
[
1
8
]
N
.
P
a
p
e
r
n
o
t
,
P
.
M
c
D
a
n
i
e
l
,
X
.
W
u
,
S
.
Jh
a
a
n
d
A
.
S
w
a
mi
,
“
D
i
st
i
l
l
a
t
i
o
n
a
s
a
d
e
f
e
n
se
t
o
a
d
v
e
r
sari
a
l
p
e
r
t
u
r
b
a
t
i
o
n
s
a
g
a
i
n
st
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
1
6
I
EE
E
S
y
m
p
o
si
u
m
o
n
S
e
c
u
ri
t
y
a
n
d
Pr
i
v
a
c
y
(
S
P)
,
S
a
n
Jo
se
,
C
A
,
U
S
A
,
2
0
1
6
,
p
p
.
5
8
2
-
5
9
7
,
d
o
i
:
1
0
.
1
1
0
9
/
S
P
.
2
0
1
6
.
4
1
.
[
1
9
]
A
.
A
t
h
a
l
y
e
,
N
.
C
a
r
l
i
n
i
,
a
n
d
D
.
W
a
g
n
e
r
,
“
O
b
f
u
sc
a
t
e
d
g
r
a
d
i
e
n
t
s
g
i
v
e
a
f
a
l
se
se
n
se
o
f
se
c
u
r
i
t
y
:
c
i
r
c
u
mv
e
n
t
i
n
g
d
e
f
e
n
se
s t
o
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
,
”
in
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
3
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
,
S
t
o
c
k
h
o
l
m,
S
w
e
d
e
n
,
P
M
L
R
8
0
,
2
0
1
8
.
[
2
0
]
D
.
M
e
n
g
a
n
d
H
.
C
h
e
n
,
“
M
a
g
n
e
t
:
a
t
w
o
-
p
r
o
n
g
e
d
d
e
f
e
n
se
a
g
a
i
n
s
t
a
d
v
e
r
sari
a
l
e
x
a
mp
l
e
s
,
”
C
C
S
‘
1
7
:
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
7
AC
M
S
I
G
S
A
C
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
e
c
u
ri
t
y
,
2
0
1
7
,
p
p
.
1
3
5
-
1
4
7
,
d
o
i
:
1
0
.
1
1
4
5
/
3
1
3
3
9
5
6
.
3
1
3
4
0
5
7
.
[
2
1
]
O.
T
a
r
a
n
,
S.
R
e
z
a
e
i
f
a
r
,
T
.
H
o
l
o
t
y
a
k
,
a
n
d
S
.
V
o
l
o
sh
y
n
o
v
sk
i
y
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
t
h
r
o
u
g
h
c
r
y
p
t
o
g
r
a
p
h
i
c
g
l
a
sse
s:
c
o
mb
a
t
i
n
g
a
d
v
e
r
sari
a
l
a
t
t
a
c
k
s
b
y
k
e
y
-
b
a
se
d
d
i
v
e
r
si
f
i
e
d
a
g
g
r
e
g
a
t
i
o
n
,
”
EU
RA
S
I
P
j
o
u
rn
a
l
o
n
i
n
f
o
rm
a
t
i
o
n
se
c
u
ri
t
y
,
2
0
2
0
,
p
p
.
1
-
1
8
,
d
o
i
:
1
0
.
1
1
8
6
/
s
1
3
6
3
5
-
0
2
0
-
0
0
1
0
6
-
x
.
[
2
2
]
L.
B
r
e
i
man
,
“
R
a
n
d
o
m fo
r
e
st
s
,
”
Ma
c
h
i
n
e
l
e
a
r
n
i
n
g
,
v
o
l
.
4
5
,
p
p
.
5
-
32
,
2
0
0
1
,
d
o
i
:
1
0
.
1
0
2
3
/
A
:
1
0
1
0
9
3
3
4
0
4
3
2
4
.
[
2
3
]
H.
Z
h
a
n
g
,
Y.
Y
u
,
J.
J
i
a
o
,
E
.
X
i
n
g
,
L
.
E
l
G
h
a
o
u
i
,
a
n
d
M
.
Jo
r
d
a
n
,
“
T
h
e
o
r
e
t
i
c
a
l
l
y
p
r
i
n
c
i
p
l
e
d
t
r
a
d
e
-
o
f
f
b
e
t
w
e
e
n
r
o
b
u
st
n
e
ss
a
n
d
a
c
c
u
r
a
c
y
,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
f
e
re
n
c
e
o
n
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
201
9
,
p
p
.
7
4
7
2
-
7
4
8
2
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
9
0
1
.
0
8
5
7
3
.
[
2
4
]
Z
.
L
i
u
,
P
.
L
u
o
,
X
.
W
a
n
g
a
n
d
X
.
T
a
n
g
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
a
c
e
a
t
t
r
i
b
u
t
e
s
i
n
t
h
e
w
i
l
d
,
”
i
n
2
0
1
5
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
(
I
C
C
V)
,
S
a
n
t
i
a
g
o
,
C
h
i
l
e
,
2
0
1
5
,
p
p
.
3
7
3
0
-
3
7
3
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
.
2
0
1
5
.
4
2
5
.
[
2
5
]
S
.
I
o
f
f
e
a
n
d
C
.
S
z
e
g
e
d
y
,
“
B
a
t
c
h
n
o
r
mal
i
z
a
t
i
o
n
:
a
c
c
e
l
e
r
a
t
i
n
g
d
e
e
p
n
e
t
w
o
r
k
t
r
a
i
n
i
n
g
b
y
r
e
d
u
c
i
n
g
i
n
t
e
r
n
a
l
c
o
v
a
r
i
a
t
e
s
h
i
f
t
,
”
i
n
I
n
t
e
r
n
a
t
i
o
n
a
l
c
o
n
f
e
r
e
n
c
e
o
n
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
2
0
1
5
,
p
p
.
4
4
8
-
4
5
6
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
5
0
2
.
0
3
1
6
7
.
[
2
6
]
A.
K.
D
u
b
e
y
,
a
n
d
V
.
Ja
i
n
,
“
C
o
mp
a
r
a
t
i
v
e
st
u
d
y
o
f
c
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
’
s
R
e
L
U
a
n
d
L
e
a
k
y
-
R
e
LU
a
c
t
i
v
a
t
i
o
n
f
u
n
c
t
i
o
n
s
,
”
i
n
Ap
p
l
i
c
a
t
i
o
n
s
o
f
C
o
m
p
u
t
i
n
g
,
Au
t
o
m
a
t
i
o
n
a
n
d
Wi
r
e
l
e
ss
S
y
st
e
m
s
i
n
E
l
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
:
Pr
o
c
e
e
d
i
n
g
s
o
f
M
ARC
2
0
1
8
,
2
0
1
9
,
p
p
.
8
7
3
-
8
8
0
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
981
-
13
-
6
7
7
2
-
4
_
7
6
.
[
2
7
]
K
i
n
g
ma,
P
.
D
i
e
d
e
r
i
k
,
a
n
d
J.
Ba
,
“
A
d
a
m:
a
me
t
h
o
d
f
o
r
st
o
c
h
a
s
t
i
c
o
p
t
i
mi
z
a
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
e
a
r
n
i
n
g
Re
p
r
e
se
n
t
a
t
i
o
n
s
(
I
C
L
R)
,
2
0
1
4
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
4
1
2
.
6
9
8
0
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
d
ve
r
s
a
r
ia
l
-
r
o
b
u
s
t steg
a
n
a
lysi
s
s
yste
m
leve
r
a
g
in
g
a
d
ve
r
s
a
r
ia
l tra
in
in
g
…
(
Th
a
kwa
n
A
kra
m
Ja
w
a
d
)
401
[
2
8
]
P
.
G
a
v
r
i
k
o
v
,
“
v
i
su
a
l
k
e
r
a
s
[
C
o
m
p
u
t
e
r
so
f
t
w
a
r
e
]
,
”
G
i
t
H
u
b
r
e
p
o
si
t
o
r
y
,
2
0
2
0
,
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
g
i
t
h
u
b
.
c
o
m/
p
a
u
l
g
a
v
r
i
k
o
v
/
v
i
s
u
a
l
k
e
r
a
s (A
c
c
e
sse
d
:
N
o
v
20
,
2
0
2
4
)
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Th
a
k
w
a
n
A
k
r
a
m
J
a
w
a
d
M.S
.
m
a
ste
r
o
f
sc
ien
c
e
in
C
o
m
p
u
ter
E
n
g
in
e
e
rin
g
f
ro
m
No
rth
e
rn
Cy
p
ru
s
(EM
U)
in
T
ü
rk
iy
e
,
A
ss
i
sta
n
t
L
e
c
tu
re
r
a
t
Nin
e
v
e
h
Un
iv
e
rsity
,
De
p
a
rt
m
e
n
t
o
f
S
y
ste
m
a
n
d
Co
n
tr
o
l
E
n
g
in
e
e
rin
g
Co
ll
e
g
e
o
f
El
e
c
tro
n
ic
En
g
in
e
e
rin
g
Nin
e
v
a
h
Un
iv
e
rsit
y
M
o
su
l
4
1
0
0
2
,
Ira
q
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
th
a
k
w
a
n
.
ja
w
a
d
@
u
o
n
in
e
v
a
h
.
e
d
u
.
iq
.
J
a
m
sh
i
d
B
a
g
h
e
r
z
a
d
e
h
M
o
h
a
s
e
fi
h
o
ld
s
a
P
h
.
D
.
i
n
C
o
m
p
u
te
r
S
c
ien
c
e
f
ro
m
In
d
ian
I
n
stit
u
te
o
f
T
e
c
h
n
o
lo
g
y
,
In
d
ia.
He
is
c
u
rre
n
tl
y
P
ro
f
e
ss
o
r
o
f
th
e
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
t
th
e
Urm
i
a
U
n
iv
e
rsity
.
His
r
e
se
a
r
c
h
to
p
ics
in
c
l
u
d
e
a
rti
f
icia
l
in
telli
g
e
n
c
e
,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
n
e
tw
o
rk
se
c
u
rit
y
.
His
re
s
e
a
rc
h
h
a
s
b
e
e
n
f
u
n
d
e
d
b
y
th
e
Urm
i
a
Un
iv
e
rsit
y
,
Ira
n
ian
T
e
lec
o
m
M
i
n
istry
,
a
n
d
Ira
n
ian
M
i
n
istry
o
f
S
c
ien
c
e
,
Re
se
a
rc
h
,
a
n
d
T
e
c
h
n
o
lo
g
y
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
j.
b
a
g
h
e
rz
a
d
e
h
@u
rm
ia.a
c
.
ir.
M
o
h
a
m
m
e
d
S
a
la
h
Re
d
a
Abd
e
lg
h
a
n
y
h
o
l
d
s
a
P
h
.
D
.
i
n
C
o
m
p
u
te
r
S
c
ien
c
e
f
ro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
t
Da
m
iett
a
Un
iv
e
rsit
y
,
Eg
y
p
t.
H
e
a
lso
e
a
rn
e
d
a
M
a
ste
r
’
s
d
e
g
r
e
e
a
n
d
a
P
re
-
M
a
ste
r
D
ip
lo
m
a
in
C
o
m
p
u
t
e
r
S
c
ien
c
e
f
ro
m
th
e
F
a
c
u
lt
y
o
f
S
c
i
e
n
c
e
a
t
M
in
ia
Un
iv
e
rsit
y
.
Cu
rre
n
tl
y
,
h
e
is
a
ff
il
iate
d
w
it
h
th
e
F
a
c
u
lt
y
o
f
Co
m
p
u
ters
a
n
d
A
rti
fici
a
l
In
telli
g
e
n
c
e
a
t
Be
n
i
-
S
u
e
f
u
n
iv
e
rsity
.
His
re
se
a
r
c
h
in
te
re
sts
sp
a
n
a
c
ro
ss
v
a
rio
u
s
a
sp
e
c
ts
o
f
c
o
m
p
u
ter
sc
ien
c
e
a
n
d
a
rti
f
icia
l
in
telli
g
e
n
c
e
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
m
sr_
c
s_
8
7
@d
u
.
e
d
u
.
e
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.