I
nd
o
ne
s
ia
n J
o
urna
l o
f
E
lect
rica
l En
g
ineering
a
nd
Co
m
pu
t
er
Science
Vo
l.
23
,
No
.
2
,
A
u
g
u
s
t
2
0
2
1
,
p
p
.
1
0
5
9
~
1
0
6
7
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
23
.i
2
.
pp
1
0
5
9
-
1
0
6
7
1059
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Accelera
ti
ng
t
he
upda
te
o
f
a
DL
-
b
a
sed IDS for
Io
T usin
g
deep
trans
fer
l
ea
rning
I
dris
s
I
dr
is
s
i,
M
o
s
t
a
f
a
Aziz
i,
O
m
a
r
M
o
us
s
a
o
ui
M
ATS
I
Re
se
a
rc
h
Lab
.
,
ES
TO,
M
o
h
a
m
m
e
d
F
irst
Un
iv
e
rsit
y
,
Ou
jd
a
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
8
,
2
0
2
1
R
ev
is
ed
J
u
l 4
,
2
0
2
1
Acc
ep
ted
J
u
l 7
,
2
0
2
1
D
e
e
p
le
a
r
n
i
n
g
(
D
L
)
m
o
d
e
l
s
a
re
n
o
w
a
d
a
y
s
b
r
o
a
d
l
y
a
p
p
l
i
e
d
a
n
d
h
a
v
e
s
h
o
w
n
o
u
t
s
t
a
n
d
i
n
g
p
e
r
f
o
r
m
a
n
c
e
i
n
a
v
a
r
i
e
t
y
o
f
f
i
e
l
d
s
,
i
n
c
l
u
d
i
n
g
o
u
r
f
o
c
u
s
t
o
p
i
c
o
f
"
I
o
T
c
y
b
e
r
s
e
c
u
r
i
t
y
"
.
D
e
e
p
l
e
a
r
n
i
n
g
-
b
a
s
e
d
i
n
t
r
u
s
i
o
n
d
e
t
e
c
t
i
o
n
s
y
s
t
e
m
(
D
L
-
I
D
S
)
m
o
d
e
l
s
a
r
e
m
o
r
e
f
i
x
a
te
d
a
n
d
d
e
p
e
n
d
e
d
o
n
t
h
e
t
r
a
i
n
e
d
d
a
t
a
s
e
t
.
Th
i
s
p
o
s
e
s
a
p
r
o
b
l
e
m
f
o
r
t
h
e
s
e
D
L
-
ID
S
,
e
s
p
e
c
i
a
l
l
y
w
i
t
h
t
h
e
k
n
o
w
n
m
u
t
a
t
i
o
n
a
n
d
b
e
h
a
v
i
o
r
c
h
a
n
g
e
s
o
f
a
t
t
a
c
k
s
,
w
h
i
c
h
c
a
n
r
e
n
d
e
r
t
h
e
m
u
n
d
e
t
e
c
t
e
d
.
As
a
r
e
s
u
l
t
,
th
e
D
L
-
I
D
S
h
a
s
b
e
c
o
m
e
o
u
t
d
a
t
e
d
.
I
n
t
h
i
s
w
o
r
k
,
w
e
p
re
se
n
t
a
s
o
l
u
t
i
o
n
f
o
r
u
p
d
a
t
i
n
g
D
L
-
I
D
S
e
m
p
l
o
y
i
n
g
a
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
t
h
a
t
a
l
l
o
w
s
u
s
t
o
r
e
t
r
a
i
n
a
n
d
f
i
n
e
-
t
u
n
e
p
r
e
-
t
r
a
i
n
e
d
m
o
d
e
ls
o
n
s
m
a
l
l
d
a
t
a
s
e
ts
w
i
t
h
n
e
w
a
t
t
a
c
k
b
e
h
a
v
i
o
r
s
.
I
n
o
u
r
e
x
p
e
r
i
m
e
n
t
s
,
w
e
b
u
i
l
t
C
NN
-
b
a
se
d
I
D
S
o
n
t
h
e
B
o
t
-
I
o
T
d
a
t
a
se
t
,
a
n
d
u
p
d
a
t
e
d
i
t
o
n
s
m
a
l
l
d
a
ta
f
r
o
m
a
n
e
w
d
a
ta
se
t
n
a
m
e
d
TO
N
-
I
o
T
.
W
e
o
b
t
a
i
n
e
d
p
r
o
m
i
s
i
n
g
r
e
s
u
l
t
s
i
n
m
u
l
t
i
p
l
e
m
e
t
r
i
c
s
r
e
g
a
r
d
in
g
t
h
e
d
e
t
e
c
t
i
o
n
r
a
t
e
a
n
d
t
h
e
t
r
a
i
n
i
n
g
b
e
t
w
e
e
n
t
h
e
i
n
i
t
i
a
l
t
r
a
i
n
i
n
g
f
o
r
t
h
e
o
r
i
g
i
n
a
l
m
o
d
e
l
a
n
d
t
h
e
u
p
d
a
t
e
d
o
n
e
,
i
n
t
h
e
m
a
t
t
e
r
o
f
d
e
t
e
c
t
i
n
g
n
e
w
a
t
ta
c
k
s
b
e
h
a
v
i
o
r
s
a
n
d
i
m
p
r
o
v
i
n
g
t
h
e
d
e
t
e
c
t
i
o
n
r
a
t
e
f
o
r
s
o
m
e
c
l
a
ss
e
s
b
y
o
v
e
r
c
o
m
i
n
g
t
h
e
l
a
c
k
o
f
t
h
e
i
r
l
a
b
e
l
e
d
d
a
t
a
.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Dee
p
lear
n
in
g
I
n
ter
n
et
o
f
th
in
g
s
I
n
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
Sy
s
tem
u
p
d
ates
T
r
an
s
f
er
lear
n
i
n
g
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
:
I
d
r
is
s
I
d
r
is
s
i
MA
T
SI
L
ab
o
r
ato
r
y
,
E
STO
Mo
h
am
m
ed
First Un
iv
er
s
ity
B
P 4
7
3
C
am
p
u
s
u
n
iv
er
s
itair
e
Al
Qo
d
s
,
Ou
jd
a
6
0
0
0
0
,
Mo
r
o
c
co
E
m
ail: id
r
is
s
i
@
u
m
p
.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
D
e
e
p
l
ea
r
n
i
n
g
h
a
s
s
u
c
ce
e
d
e
d
in
a
v
a
r
i
e
t
y
o
f
f
i
el
d
s
i
n
r
e
c
e
n
t y
e
a
r
s
,
i
n
c
l
u
d
i
n
g
m
e
d
i
c
al
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
,
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
,
an
d
c
y
b
e
r
s
e
c
u
r
i
t
y
.
I
n
t
h
e
I
o
T
c
y
b
e
r
s
e
c
u
r
i
t
y
[
1
]
,
t
h
e
m
a
c
h
i
n
e
t
r
a
i
n
s
o
n
n
u
m
e
r
o
u
s
g
a
t
h
e
r
e
d
a
n
d
l
a
b
el
e
d
c
y
b
e
r
a
tt
ac
k
s
as
we
l
l
a
s
n
o
r
m
a
l
t
r
a
f
f
i
c
to
l
e
a
r
n
t
h
e
m
;
t
h
is
d
at
a
c
o
m
es
f
r
o
m
m
a
n
y
d
a
t
a
s
et
s
c
o
n
t
a
i
n
i
n
g
s
e
v
e
r
al
a
t
t
ac
k
s
,
f
r
o
m
w
h
i
c
h
t
h
i
s
m
ac
h
i
n
e
m
a
y
e
v
e
n
t
u
a
l
l
y
d
e
t
e
ct
n
e
w
s
i
m
i
la
r
a
t
ta
c
k
s
.
H
o
w
e
v
e
r
,
t
h
e
a
r
e
a
o
f
c
y
b
e
r
s
e
c
u
r
i
t
y
c
o
n
t
i
n
u
e
s
t
o
c
o
n
f
r
o
n
t
s
e
v
e
r
a
l
p
r
o
b
l
e
m
s
i
n
i
d
e
n
t
i
f
y
i
n
g
e
v
o
l
v
i
n
g
t
h
r
e
a
t
s
.
M
o
d
e
l
s
t
r
a
i
n
e
d
o
n
k
n
o
w
n
a
t
t
a
c
k
d
a
t
a
s
et
s
p
r
o
d
u
c
e
e
x
c
e
l
l
e
n
t
r
es
u
l
ts
[
2
]
.
T
h
e
s
e
f
i
n
d
i
n
g
s
,
h
o
w
e
v
e
r
,
a
r
e
d
i
r
e
c
t
l
y
d
e
p
e
n
d
e
n
t
o
n
t
h
e
d
a
t
a
o
n
t
h
e
t
r
a
i
n
i
n
g
d
at
a
s
e
t
.
T
h
e
is
s
u
e
i
s
t
h
a
t
t
h
e
s
e
a
tt
a
c
k
s
c
h
an
g
e
i
n
b
e
h
a
v
i
o
r
(
o
r
s
o
m
e
o
f
th
e
m
a
r
e
z
e
r
o
h
o
u
r
)
m
a
k
i
n
g
t
h
e
m
u
n
d
e
t
e
c
t
e
d
b
y
t
h
e
m
o
d
e
l
,
w
h
i
c
h
m
u
s
t
b
e
r
et
r
a
i
n
ed
o
n
t
h
e
s
e
n
e
w
b
e
h
a
v
i
o
r
s
[
3
]
.
T
h
is
p
r
o
b
lem
atic
o
f
d
ata
d
ep
en
d
en
cy
[
4
]
alo
n
g
with
th
e
lack
o
f
b
ig
d
atasets
th
at
co
n
tai
n
n
ew
an
d
u
p
d
ated
attac
k
s
b
e
h
av
io
r
s
,
m
a
k
es
th
e
c
r
ea
tio
n
o
f
an
u
p
d
ated
I
DS
a
b
ig
ch
allen
g
e.
An
d
h
er
e,
tr
an
s
f
er
lear
n
i
n
g
co
m
es
to
h
elp
s
o
lv
in
g
th
ese
p
r
o
b
lem
s
wh
er
e
tr
ain
i
n
g
an
d
te
s
tin
g
d
ata
is
n
o
t
m
an
d
ato
r
y
.
T
h
is
tech
n
iq
u
e
g
ai
n
s
th
e
ex
is
tin
g
k
n
o
wled
g
e
f
r
o
m
a
tr
ain
ed
m
o
d
el
o
n
a
ce
r
tain
d
at
aset
an
d
m
er
g
es
it
with
th
e
n
ew
o
n
e
tr
ain
ed
o
n
a
s
m
all
d
ataset
th
at
co
n
tain
s
n
e
w
attac
k
s
b
eh
av
io
r
s
[
5
]
.
T
h
u
s
,
th
e
n
ew
u
p
d
ate
d
m
o
d
el
will
n
o
t
b
e
b
u
ilt
f
r
o
m
s
cr
atch
o
r
f
r
o
m
a
p
o
o
r
d
ataset
[
4
]
.
T
r
an
s
f
er
lea
r
n
in
g
is
a
tech
n
iq
u
e
th
at
b
en
e
f
its
f
r
o
m
a
n
alr
ea
d
y
tr
ai
n
ed
weig
h
t
o
n
b
ig
d
ata
s
ets
f
o
r
a
lo
n
g
p
e
r
io
d
o
f
tim
e
an
d
tr
an
s
f
er
s
th
is
k
n
o
wled
g
e
[
6
]
.
As
an
ex
am
p
le,
th
e
Go
o
g
le’
s
lan
g
u
ag
e
r
ep
r
esen
tatio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2502
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t 2
0
2
1
:
1
0
5
9
-
1
0
6
7
1060
m
o
d
el
“BER
T
”
h
as
1
1
0
m
illi
o
n
p
a
r
am
eter
s
an
d
was
tr
ain
e
d
f
o
r
4
d
ay
s
o
n
1
6
C
l
o
u
d
T
P
Us
[
7
]
,
th
e
Op
en
AI
’
s
lan
g
u
a
g
e
g
en
er
ato
r
GPT3
w
h
ich
is
a
m
o
d
el
t
h
at
h
as
1
7
5
b
illi
o
n
p
a
r
am
eter
s
[
8
]
,
o
r
th
e
r
ea
l
-
tim
e
o
b
ject
d
etec
tio
n
“YO
L
O”
with
8
9
m
illi
o
n
p
ar
am
ete
r
s
in
t
h
e
f
i
f
th
v
er
s
io
n
[
9
]
.
T
h
e
i
d
ea
to
r
etr
ai
n
th
ese
m
o
d
els
f
r
o
m
s
cr
atch
with
n
ew
d
ata
is
v
er
y
ex
p
en
s
iv
e,
b
o
t
h
in
ter
m
s
o
f
tim
e
an
d
r
eso
u
r
ce
s
.
T
h
er
ef
o
r
e,
im
p
lem
en
tin
g
tr
an
s
f
er
lear
n
in
g
is
m
o
r
e
p
r
ac
t
ical
an
d
b
en
e
f
icial
f
o
r
t
h
is
s
itu
atio
n
.
T
h
e
r
est
o
f
th
is
p
ap
er
is
s
tr
u
ctu
r
ed
as
f
o
llo
ws.
I
n
th
e
s
ec
tio
n
2
,
we
d
escr
ib
e
th
e
b
ac
k
g
r
o
u
n
d
,
th
e
s
ec
tio
n
3
p
r
esen
ts
th
e
r
elate
d
wo
r
k
s
,
th
e
s
ec
tio
n
4
p
r
esen
ts
o
u
r
m
eth
o
d
o
lo
g
y
,
th
e
s
ec
tio
n
5
we
p
r
esen
t
an
d
d
is
cu
s
s
th
e
o
b
tain
ed
r
esu
lts
,
a
n
d
f
in
ally
a
co
n
clu
s
io
n
as a
s
ec
tio
n
6
.
2.
B
ACK
G
RO
UND
2
.
1
.
Co
nv
o
lutio
na
l
neura
l net
wo
rk
s
C
NN
o
r
C
o
n
v
Nets
ar
e
a
class
o
f
d
ee
p
n
eu
r
al
n
etwo
r
k
s
th
a
t
ar
e
u
s
ed
in
m
an
y
f
ield
s
b
u
t
m
o
s
tly
in
p
atter
n
r
ec
o
g
n
itio
n
[
10
]
.
I
t
is
c
o
m
p
o
s
ed
o
f
an
i
n
p
u
t
lay
er
,
m
a
n
y
h
i
d
d
en
lay
er
s
in
b
etwe
en
,
a
n
d
an
o
u
tp
u
t
la
y
er
as
s
h
o
wn
in
Fig
u
r
e
1
lik
e
th
e
m
u
lti
-
lay
er
p
er
ce
p
tr
o
n
(
M
L
P
)
n
etwo
r
k
s
.
B
est
k
n
o
wn
a
n
d
u
s
ed
la
y
er
s
ar
e:
co
n
v
o
l
u
tio
n
[1
1
]
,
ac
tiv
atio
n
o
r
R
eL
U,
a
n
d
p
o
o
lin
g
[1
2
]
,
[1
3
]
.
I
n
co
n
t
r
ast
to
s
tan
d
a
r
d
f
ea
tu
r
e
s
elec
tio
n
alg
o
r
ith
m
s
,
i
t
is
ca
p
ab
le
o
f
d
y
n
am
ically
lear
n
in
g
b
etter
f
ea
tu
r
es
an
d
ca
teg
o
r
izin
g
t
r
a
f
f
ic.
Fu
r
th
er
m
o
r
e,
b
ec
au
s
e
it
s
h
ar
es
th
e
s
am
e
co
n
v
o
lu
tio
n
m
atr
ix
(
k
er
n
el)
,
it
ca
n
ac
co
m
p
lis
h
b
etter
class
if
icatio
n
an
d
lear
n
n
e
w
f
ea
tu
r
es with
m
o
r
e
tr
af
f
ic
d
ata
,
r
ed
u
cin
g
th
e
n
u
m
b
er
o
f
p
ar
a
m
eter
s
an
d
tr
ain
in
g
ca
lcu
latio
n
to
tal
s
u
b
s
tan
tially
.
I
n
co
n
t
r
ar
y
t
o
o
th
e
r
d
ee
p
-
lear
n
in
g
o
r
m
ac
h
i
n
e
lear
n
in
g
alg
o
r
ith
m
s
th
at
ca
n
b
e
o
v
e
r
-
f
itted
with
h
u
g
e
v
ast
d
ata,
C
NN
ca
n
r
ec
o
g
n
ize
t
h
e
t
y
p
e
o
f
a
n
ass
au
lt
q
u
ick
ly
.
Fu
r
th
e
r
m
o
r
e,
liter
atu
r
e
f
in
d
in
g
s
s
u
g
g
est
th
at
u
s
in
g
C
NN
s
in
th
e
in
tr
u
s
io
n
d
etec
tio
n
f
iel
d
p
r
o
d
u
ce
s
s
u
p
er
io
r
r
esu
lts
th
an
o
th
er
m
eth
o
d
s
[1
4
]
.
I
n
p
u
t
O
u
t
pu
t
O
u
t
pu
t
L
ay
e
r
C
o
n
vo
l
u
t
i
o
n
a
l
l
a
y
er
P
o
o
l
i
n
g
l
a
y
er
F
u
l
l
y
c
o
n
n
e
c
t
e
d
Fig
u
r
e
1
.
C
NN
ar
ch
itectu
r
e
2
.
2
.
T
ra
ns
f
er
lea
rning
T
L
i
s
a
m
a
c
h
i
n
e
a
n
d
d
e
e
p
l
e
ar
n
i
n
g
r
e
s
e
a
r
c
h
a
r
e
a
t
h
at
a
i
m
s
to
t
r
a
n
s
f
e
r
k
n
o
w
l
e
d
g
e
f
r
o
m
o
n
e
o
r
m
o
r
e
s
o
u
r
c
e
t
as
k
s
t
o
o
n
e
o
r
m
o
r
e
t
a
r
g
e
t
t
as
k
s
[1
5
]
.
S
u
p
p
o
s
i
n
g
a
s
o
u
r
c
e
d
o
m
a
i
n
,
a
l
ea
r
n
i
n
g
t
as
k
,
a
t
a
r
g
e
t
d
o
m
ain
,
a
n
d
a
l
e
a
r
n
i
n
g
t
as
k
;
T
L
s
e
r
v
e
s
i
n
i
m
p
r
o
v
i
n
g
t
h
e
l
e
a
r
n
i
n
g
o
f
t
h
e
t
a
r
g
e
t
p
r
e
d
i
ct
i
v
e
f
u
n
c
ti
o
n
(
.
)
i
n
u
s
i
n
g
t
h
e
k
n
o
w
l
e
d
g
e
i
n
a
n
d
,
w
h
e
r
e
≠
,
o
r
≠
[1
6
]
.
T
L
i
s
a
l
l
a
b
o
u
t
u
s
i
n
g
t
h
e
f
e
a
t
u
r
e
s
l
e
a
r
n
e
d
o
n
o
n
e
p
r
o
b
l
e
m
a
n
d
l
e
v
e
r
a
g
i
n
g
t
h
e
m
o
n
a
n
e
w
s
i
m
i
l
a
r
p
r
o
b
l
e
m
.
F
o
r
e
x
a
m
p
l
e
,
t
a
k
i
n
g
t
h
e
c
h
a
r
a
c
t
e
r
is
t
i
cs
o
f
a
m
o
d
e
l
w
h
o
h
a
s
l
e
a
r
n
e
d
t
o
i
d
en
t
i
f
y
c
a
ts
,
i
t
ca
n
b
e
u
s
e
f
u
l
i
n
c
r
e
a
t
i
n
g
a
m
o
d
e
l
f
o
r
i
d
e
n
t
i
f
y
in
g
t
i
g
e
r
s
[1
7
]
.
T
L
is
t
y
p
i
c
a
l
l
y
p
e
r
f
o
r
m
e
d
f
o
r
t
a
s
k
s
w
h
e
r
e
t
h
e
r
e
i
s
a
li
t
tl
e
d
a
ta
s
et
t
o
t
r
a
i
n
a
f
u
l
l
-
s
c
al
e
m
o
d
e
l
f
r
o
m
s
c
r
a
t
c
h
.
I
n
o
u
r
e
x
p
e
r
i
m
e
n
t
s
,
t
h
i
s
i
n
v
o
l
v
es
u
p
d
a
t
i
n
g
a
n
a
l
r
e
a
d
y
e
x
i
s
t
i
n
g
I
DS
m
o
d
e
l
w
i
t
h
n
ew
a
t
t
ac
k
b
e
h
a
v
i
o
r
s
f
r
o
m
a
s
m
a
ll
d
a
t
a
s
et
c
o
n
t
a
i
n
i
n
g
t
h
es
e
n
e
w
b
e
h
a
v
i
o
r
s
a
n
d
w
i
t
h
o
u
t
t
h
e
n
e
e
d
o
f
b
u
i
l
d
i
n
g
a
n
e
w
b
i
g
d
a
t
a
s
e
t
an
d
r
e
t
r
a
i
n
i
n
g
it.
T
h
er
e
ar
e
d
if
f
er
en
t
m
eth
o
d
s
t
o
p
u
ll
o
u
t
th
e
t
r
an
s
f
er
lear
n
i
n
g
in
th
e
d
ee
p
lea
r
n
in
g
c
o
n
tex
t
,
it
d
ep
en
d
s
o
n
h
o
w
m
u
ch
d
ata
we
h
av
e
g
o
t.
Fo
r
ex
a
m
p
le,
it
co
u
l
d
f
r
ee
ze
all
lay
er
s
a
n
d
tr
ain
o
n
ly
o
n
th
e
last
o
n
e,
o
r
f
r
ee
ze
m
o
s
t
lay
er
s
an
d
tr
ain
t
h
e
last
o
n
es,
o
r
tr
ain
in
g
all
la
y
er
s
b
y
in
itializin
g
th
e
weig
h
t
s
o
n
th
e
p
r
e
-
tr
ain
ed
o
n
es.
I
n
o
u
r
ex
p
e
r
im
en
ts
,
we
f
r
ee
ze
m
o
s
t
o
f
th
e
lay
er
s
an
d
t
r
ain
in
g
th
e
last
o
n
es
u
s
in
g
C
NN
o
n
d
atasets
B
o
t
-
I
o
T
(
s
o
u
r
ce
d
o
m
ain
)
an
d
T
O
N
-
I
o
T
(
tar
g
et
d
o
m
ai
n
).
T
r
an
s
f
er
lear
n
i
n
g
ca
n
b
e
ac
h
ie
v
ed
b
y
r
em
o
v
in
g
th
e
o
r
ig
in
al
m
o
d
el
class
if
ier
,
th
en
a
d
d
in
g
a
n
ew
o
n
e
th
at
f
its
o
u
r
p
u
r
p
o
s
es
as
s
h
o
wn
in
Fig
u
r
e
2
,
a
n
d
f
in
ally
f
in
e
-
tu
n
in
g
t
h
is
n
ew
m
o
d
el
r
en
d
er
in
g
o
n
o
n
e
o
f
t
h
r
ee
ap
p
r
o
ac
h
es
[1
8
]
:
a)
T
r
ain
in
g
th
e
wh
o
le
m
o
d
el.
I
n
th
is
ap
p
r
o
ac
h
,
we
u
s
e
th
e
o
r
ig
in
al
ar
ch
itectu
r
e
o
f
th
e
p
r
e
-
tr
ain
ed
m
o
d
e
l
an
d
tr
ain
it
ac
co
r
d
in
g
ly
to
t
h
e
n
ew
d
ataset.
T
h
is
m
ea
n
s
t
h
at
t
h
e
m
o
d
el
will
b
e
r
etr
ain
ed
f
r
o
m
s
cr
atch
.
I
n
th
is
ap
p
r
o
ac
h
we
n
ee
d
a
b
ig
d
ataset;
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
cc
elera
tin
g
th
e
u
p
d
a
te
o
f
a
DL
-
b
a
s
ed
I
DS
fo
r
I
o
T
u
s
in
g
d
ee
p
tr
a
n
s
fer lea
r
n
in
g
(
I
d
r
is
s
I
d
r
is
s
i
)
1061
b)
T
r
ain
in
g
o
n
ly
s
o
m
e
lay
e
r
s
.
I
n
th
is
ap
p
r
o
ac
h
,
we
f
r
ee
ze
s
o
m
e
lay
er
s
wh
e
r
e
we
r
etr
ain
th
e
r
em
ai
n
in
g
o
n
es,
in
th
e
ca
s
e
o
f
a
s
m
all
d
ataset
an
d
a
lar
g
e
n
u
m
b
er
o
f
p
ar
am
eter
s
,
we
f
r
o
ze
m
o
r
e
la
y
er
s
to
a
v
o
id
o
v
er
f
itti
n
g
.
B
u
t
in
th
e
ca
s
e
o
f
a
lar
g
e
d
ataset
an
d
a
s
m
all
n
u
m
b
er
o
f
p
ar
am
ete
r
s
,
wh
er
e
o
v
er
f
itti
n
g
is
n
o
t
a
p
r
o
b
lem
,
we
ca
n
im
p
r
o
v
e
o
u
r
m
o
d
el
b
y
r
etr
ai
n
in
g
m
o
r
e
lay
er
s
to
o
u
r
n
ew
task
;
c)
Fre
ez
in
g
th
e
c
o
n
v
o
lu
tio
n
al
b
a
s
e.
T
h
is
ap
p
r
o
ac
h
is
m
o
r
e
lik
e
th
e
last
o
n
e
b
u
t
its
m
ain
id
ea
is
to
k
ee
p
th
e
co
n
v
o
l
u
tio
n
al
b
ase
(
wh
ic
h
is
a
s
tack
o
f
co
n
v
o
lu
tio
n
al
an
d
p
o
o
lin
g
lay
er
s
)
in
its
o
r
ig
i
n
al
f
o
r
m
,
t
h
en
u
s
e
its
o
u
tp
u
ts
to
f
ee
d
th
e
class
if
ier
(
wh
ich
is
g
e
n
er
ally
th
e
f
u
ll
y
co
n
n
ec
ted
lay
e
r
s
)
as
s
h
o
wn
in
Fig
u
r
e
2
.
I
n
th
is
ca
s
e,
we
ca
n
u
s
e
ju
s
t
a
s
m
all
d
ataset
with
a
m
i
n
im
u
m
o
f
co
m
p
u
tatio
n
al
p
o
wer
.
B
asin
g
o
n
th
ese
ad
v
an
tag
es,
we
will u
s
e
th
is
last
ap
p
r
o
ac
h
i
n
o
u
r
p
r
o
p
o
s
ed
m
eth
o
d
.
I
npu
t
O
u
t
p
u
t
O
u
t
pu
t
L
ay
e
r
C
o
n
vo
l
u
t
i
o
n
a
l
l
a
y
er
P
o
o
l
i
n
g
l
a
y
er
F
u
l
l
y
c
o
n
n
ec
t
ed
F
ro
z
e
n
L
ay
e
rs
Fig
u
r
e
2
.
T
r
an
s
f
er
lear
n
in
g
(
f
r
ee
zin
g
th
e
co
n
v
o
l
u
tio
n
al
b
ase)
2
.
3
.
Da
t
a
s
et
s
B
o
t
-
I
o
T:
I
t
is
a
cy
b
er
s
ec
u
r
ity
d
ataset
th
at
wa
s
cr
ea
ted
s
p
e
cif
ically
f
o
r
I
o
T
s
y
s
tem
s
b
y
an
ac
tu
al
n
etwo
r
k
m
ilieu
.
T
h
e
en
v
ir
o
n
m
en
t
in
co
r
p
o
r
ates
a
co
m
b
in
atio
n
o
f
u
s
u
al
n
o
r
m
al
an
d
b
a
d
tr
af
f
ic,
wit
h
4
ty
p
es
o
f
attac
k
s
an
d
1
0
s
u
b
ca
teg
o
r
ies.
Nam
ely
,
r
ec
o
n
n
aiss
an
ce
(
s
er
v
i
ce
s
ca
n
n
in
g
a
n
d
OS
f
in
g
er
p
r
i
n
tin
g
)
,
DDo
S (
T
C
P,
UDP,
an
d
HT
T
P),
Do
S (
T
C
P,
UDP,
an
d
HT
T
P),
th
ef
t (
k
ey
lo
g
g
in
g
a
n
d
d
ata
ex
f
iltra
tio
n
)
[1
9
]
.
TON
-
I
o
T:
I
t
is
al
s
o
a
cy
b
er
s
ec
u
r
ity
d
ataset
th
at
was
cr
ea
ted
s
p
ec
if
ically
f
o
r
I
n
d
u
s
tr
y
4
.
0
/I
o
T
an
d
I
n
d
u
s
tr
ial
I
o
T
(
I
I
o
T
)
in
2
0
1
9
.
I
t
co
n
tain
s
h
eter
o
g
en
e
o
u
s
d
ata,
co
llected
f
r
o
m
I
o
T
an
d
I
I
o
T
s
en
s
o
r
s
,
o
p
er
atin
g
s
y
s
tem
s
(
W
in
d
o
ws an
d
L
in
u
x
)
,
an
d
n
etwo
r
k
tr
a
f
f
ic.
T
h
ese
d
atasets
wer
e
co
llected
at
th
e
cy
b
er
r
an
g
e
a
n
d
I
o
T
L
a
b
s
o
f
th
e
UNSW
C
an
b
er
r
a
C
y
b
er
f
r
o
m
a
r
ea
lis
tic
an
d
lar
g
e
-
s
ca
le
n
etwo
r
k
d
esig
n
e
d
.
T
h
ese
d
atasets
co
n
tain
n
in
e
attac
k
ca
teg
o
r
ies.
Nam
ely
,
s
ca
n
n
in
g
attac
k
,
Do
S,
DDo
S,
r
an
s
o
m
w
ar
e
attac
k
,
b
ac
k
d
o
o
r
attac
k
,
i
n
jectio
n
attac
k
,
cr
o
s
s
-
s
ite
s
cr
i
p
tin
g
(
XSS)
attac
k
,
p
ass
wo
r
d
attac
k
,
a
n
d
m
an
-
in
-
th
e
-
m
id
d
le
(
MI
T
M
)
attac
k
,
ag
ai
n
s
t
v
u
ln
e
r
ab
le
elem
en
ts
o
f
I
o
T
/I
I
o
T
ap
p
licatio
n
s
,
o
p
e
r
atin
g
s
y
s
tem
s
,
an
d
n
etwo
r
k
s
y
s
tem
s
[
20
]
.
3.
RE
L
AT
E
D
WO
RK
S
I
n
2
0
1
9
,
Sin
g
la
et
a
l
.
[2
1
]
s
h
o
wed
ex
p
er
im
e
n
tally
o
n
th
e
UNSW
-
N
B
1
5
d
ataset
(
9
s
u
b
-
d
a
tasets
)
th
at
T
r
an
s
f
er
L
ea
r
n
i
n
g
is
ef
f
ec
tiv
e
in
r
ed
u
cin
g
t
h
e
q
u
an
tity
o
f
l
ab
eled
d
ata
r
eq
u
i
r
ed
f
o
r
tr
ain
i
n
g
n
eu
r
al
n
etwo
r
k
m
o
d
els
f
o
r
n
etwo
r
k
I
DS,
b
y
o
b
s
er
v
in
g
th
e
class
ificatio
n
ac
c
u
r
ac
y
(
u
p
to
2
6
.
4
% b
etter
ac
c
u
r
ac
y
)
an
d
p
r
ec
is
io
n
m
etr
ics f
o
r
co
m
p
ar
in
g
an
d
e
v
a
lu
atin
g
th
e
r
esu
ltin
g
n
eu
r
al
n
et
w
o
r
k
s
m
o
d
els.
I
n
2
0
1
9
,
Z
h
a
n
g
e
t
a
l
.
[2
2
]
p
r
o
p
o
s
e
d
a
n
a
u
t
o
n
o
m
o
u
s
c
l
a
s
s
i
fi
e
r
u
p
d
a
t
e
s
c
h
e
m
e
,
c
h
a
l
l
e
n
g
i
n
g
t
h
e
m
a
t
t
e
r
o
f
u
p
d
a
t
i
n
g
a
n
e
x
i
s
t
i
n
g
n
e
t
w
o
r
k
t
r
a
f
fi
c
c
l
a
s
s
i
fi
e
r
.
T
h
e
y
e
v
a
l
u
a
t
e
d
t
h
e
i
r
w
o
r
k
b
y
u
s
i
n
g
M
L
P
a
n
d
C
N
N
m
o
d
e
l
s
o
n
t
h
e
I
S
C
X
V
P
N
-
n
o
n
V
P
N
d
a
t
a
s
e
t
a
n
d
t
h
e
m
e
t
r
i
c
s
r
e
c
a
l
l
,
p
r
e
c
i
s
i
o
n
,
a
n
d
F
1
s
c
o
r
e
.
T
h
e
y
a
l
s
o
d
e
m
o
n
s
t
r
a
t
e
d
t
h
a
t
t
h
e
i
r
p
r
o
p
o
s
e
d
c
l
a
s
s
i
fi
e
r
u
p
d
a
t
e
s
c
h
e
m
e
c
a
n
h
e
l
p
b
u
i
l
d
i
n
g
a
d
a
t
a
s
e
t
o
f
t
h
e
n
e
w
a
p
p
l
i
c
a
t
i
o
n
f
r
o
m
a
n
a
c
t
i
v
e
n
e
t
w
o
r
k
t
r
a
f
fi
c
.
I
n
2
0
1
9
.
S
a
m
e
e
r
a
e
t
a
l
.
[2
3
]
a
p
p
l
i
e
d
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
t
o
d
e
t
e
c
t
u
n
l
a
b
e
l
e
d
R
2
L
f
r
o
m
t
h
e
N
S
L
-
KD
D
d
a
t
a
s
et
b
y
m
a
k
i
n
g
u
s
e
o
f
l
a
b
el
e
d
D
o
S
at
t
a
c
k
s
o
f
NS
L
-
K
DD
d
a
t
a
s
et
,
a
n
d
g
o
t
a
n
a
cc
u
r
a
c
y
o
f
8
9
.
7
9
%
)
a
n
d
a
n
F
PR
o
f
0
.
1
5
%
.
T
h
e
n
,
i
n
[2
4
]
,
th
e
a
u
t
h
o
r
s
p
r
o
p
o
s
e
d
a
d
e
e
p
t
r
an
s
d
u
c
t
i
v
e
t
r
a
n
s
f
e
r
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
f
o
r
d
e
t
e
c
t
i
n
g
z
e
r
o
-
d
a
y
a
t
t
a
c
k
s
e
x
h
i
b
i
t
e
d
as
th
e
t
a
r
g
e
t
d
o
m
a
i
n
w
it
h
n
o
l
a
b
el
e
d
e
x
a
m
p
l
e
s
.
T
h
e
y
e
m
p
l
o
y
t
h
e
D
A
M
A
m
a
n
i
f
o
l
d
a
l
i
g
n
m
e
n
t
a
p
p
r
o
a
c
h
,
c
r
e
a
t
i
n
g
ta
r
g
e
t
s
o
f
t l
a
b
el
s
t
o
c
o
m
p
e
n
s
a
t
e
f
o
r
t
h
e
a
b
s
e
n
c
e
o
f
l
a
b
el
e
d
t
a
r
g
e
t
i
n
s
t
a
n
c
es
t
h
r
o
u
g
h
t
h
e
a
p
p
l
i
c
a
ti
o
n
o
f
c
l
u
s
t
e
r
c
o
r
r
e
s
p
o
n
d
e
n
c
e
p
r
o
c
e
s
s
e
s
.
T
h
e
y
in
v
e
s
t
i
g
a
te
d
t
h
e
i
r
p
r
o
p
o
s
e
d
m
o
d
e
l
s
u
s
i
n
g
m
u
l
t
i
p
le
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
s
s
u
c
h
k
N
N
,
S
V
M
,
R
F
,
D
T
,
a
n
d
D
N
N
w
i
t
h
t
w
o
t
es
t
s
c
e
n
a
r
i
o
s
.
I
n
t
h
e
f
i
r
s
t
o
n
e
,
t
h
e
y
i
n
v
e
s
t
i
g
at
e
d
a
n
d
t
es
t
e
d
u
s
i
n
g
t
h
e
N
SL
-
K
DD
d
a
t
as
et
,
a
n
d
i
n
t
h
e
s
ec
o
n
d
o
n
e
,
t
h
e
y
d
et
e
c
t
e
d
z
e
r
o
-
d
a
y
a
t
t
a
c
k
s
o
f
t
h
e
C
I
D
D
c
l
o
u
d
d
a
t
a
s
e
t
b
y
c
o
n
s
i
d
er
i
n
g
t
h
e
D
o
S
p
a
r
t
o
f
t
h
e
N
S
L
-
K
D
D
as
a
s
o
u
r
c
e
d
a
t
as
e
t
.
I
n
2
0
2
0
,
Q
u
r
esh
i
et
a
l
.
[2
5
]
p
r
o
p
o
s
ed
a
d
ee
p
n
eu
r
al
n
etwo
r
k
an
d
a
d
ap
tiv
e
s
elf
-
tau
g
h
t
b
as
ed
tr
an
s
f
er
lear
n
in
g
(
n
am
e
d
DST
-
T
L
)
tec
h
n
iq
u
e
th
at
u
s
es
s
elf
-
tau
g
h
t
le
ar
n
in
g
to
d
ev
elo
p
th
ei
r
I
DS.
T
h
ey
c
o
n
d
u
cted
t
h
eir
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2502
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t 2
0
2
1
:
1
0
5
9
-
1
0
6
7
1062
ex
p
er
im
en
t
b
y
co
m
p
a
r
in
g
t
h
eir
tech
n
iq
u
e
to
t
h
e
p
er
f
o
r
m
an
ce
o
f
co
n
v
en
tio
n
al
class
if
ier
s
(
lik
e
ML
P),
n
o
n
lin
ea
r
p
r
in
ci
p
al
co
m
p
o
n
e
n
t
an
aly
s
is
(
NL
PC
A)
,
an
d
d
ee
p
b
elief
n
etwo
r
k
(
DB
N)
)
.
First,
b
y
ex
tr
ac
tin
g
f
ea
tu
r
es
an
d
p
ass
in
g
th
e
o
r
ig
i
n
al
f
ea
tu
r
e
s
et
o
f
NSL
-
KDD
d
ataset
th
r
o
u
g
h
t
h
e
p
r
e
-
tr
ain
e
d
n
etwo
r
k
.
T
h
e
n
,
a
m
ix
tu
r
e
o
f
o
r
ig
in
al
an
d
ex
tr
a
cted
f
ea
tu
r
es
is
u
s
ed
in
th
e
tr
ain
in
g
o
f
th
e
s
p
a
r
s
e
au
to
-
en
c
o
d
er
.
T
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
(
DST
-
T
L
)
i
n
[2
5
]
s
h
o
wed
a
clea
r
im
p
r
o
v
e
d
p
r
e
d
ictio
n
ac
cu
r
ac
y
.
I
n
2
0
1
9
,
Z
h
a
n
g
et
al
.
p
r
o
p
o
s
ed
in
[
2
1
]
a
tr
a
n
s
f
er
lear
n
in
g
f
r
am
ewo
r
k
b
ased
o
n
t
h
e
d
o
m
ain
-
ad
v
er
s
ar
ial
tr
ain
i
n
g
tech
n
iq
u
e
f
o
r
i
n
tr
u
s
io
n
d
etec
tio
n
in
a
S
m
ar
t
Gr
id
en
v
i
r
o
n
m
e
n
t.
T
h
ey
u
s
e
a
r
ea
lis
tic
s
m
ar
t
g
r
id
s
ec
u
r
ity
d
ataset,
wh
ich
is
co
llected
th
r
o
u
g
h
h
ar
d
war
e
-
in
-
th
e
-
lo
o
p
s
im
u
lato
r
s
.
T
h
e
ir
f
r
am
ewo
r
k
was
ca
p
ab
le
o
f
m
ix
i
n
g
d
if
f
er
en
t
b
aselin
e
m
ac
h
in
e
lear
n
in
g
clas
s
ifier
s
in
o
r
d
er
to
im
p
r
o
v
e
p
er
f
o
r
m
a
n
ce
,
wh
er
e
th
ey
g
o
t
im
p
r
o
v
e
m
en
ts
f
r
o
m
7
%
u
p
to
3
6
.
8
%
u
s
in
g
Ad
B
,
k
NN,
SVM,
an
d
R
F
class
i
fie
r
s
,
an
d
an
a
v
er
ag
e
im
p
r
o
v
em
e
n
t o
v
e
r
2
.
5
% f
o
r
b
o
th
C
AR
T
an
d
ANN.
I
n
th
ese
r
elate
d
wo
r
k
s
,
s
o
m
e
r
esear
ch
er
s
u
s
ed
o
n
ly
o
n
e
d
at
aset
to
p
er
f
o
r
m
b
o
t
h
th
e
p
r
etr
ain
in
g
a
n
d
th
e
T
r
an
s
f
er
L
ea
r
n
in
g
,
wh
ile
o
th
er
s
ex
p
l
o
ited
d
if
f
e
r
en
t
d
a
tasets
f
o
r
th
e
m
o
d
el’
s
p
r
etr
ai
n
in
g
r
e
g
ar
d
in
g
th
e
tr
an
s
f
er
lear
n
i
n
g
m
o
d
el.
F
u
r
t
h
e
r
m
o
r
e
,
s
o
m
e
o
f
t
h
e
u
s
e
d
d
a
t
a
s
e
t
s
a
r
e
m
o
r
e
g
e
n
e
r
a
l
a
n
d
n
o
t
r
e
a
l
l
y
c
o
l
l
e
ct
e
d
f
r
o
m
a
t
t
a
c
k
s
o
n
I
o
T
s
y
s
t
e
m
s
,
o
r
h
av
e
w
o
r
k
e
d
w
i
t
h
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
.
I
n
o
u
r
e
x
p
e
r
i
m
e
n
t
s
,
w
e
p
r
e
s
e
n
t
a
p
r
o
o
f
o
f
c
o
n
c
e
p
t
o
f
t
h
e
D
L
-
I
DS
u
p
d
a
t
e
u
s
i
n
g
t
w
o
r
e
ce
n
t
I
o
T
d
at
as
e
ts
.
T
h
e
f
i
r
s
t
o
n
e
(
B
o
t
-
I
o
T
)
is
g
e
n
e
r
a
t
e
d
i
n
2
0
1
8
a
n
d
t
h
e
s
e
c
o
n
d
o
n
e
(
T
O
N
-
I
o
T
Ne
tw
o
r
k
)
i
s
g
e
n
e
r
a
t
e
d
i
n
2
0
1
9
u
s
in
g
t
h
e
b
e
s
t
-
k
n
o
w
n
D
e
e
p
L
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
C
N
N
.
4.
P
RO
P
O
SE
D
M
E
T
H
O
D
T
h
e
p
u
r
p
o
s
e
o
f
o
u
r
m
e
t
h
o
d
i
s
t
o
c
r
e
a
t
e
a
C
N
N
-
b
a
s
e
d
I
D
S
m
o
d
e
l
u
s
i
n
g
t
h
e
B
o
t
-
I
o
T
d
a
t
as
e
t
f
i
r
s
t
,
a
n
d
t
h
e
n
u
p
d
a
t
e
i
t
wi
t
h
t
h
e
T
ON
-
I
o
T
d
a
t
a
s
et
a
s
s
h
o
w
n
i
n
F
i
g
u
r
e
3
b
y
f
o
l
l
o
w
i
n
g
t
h
e
s
te
p
s
1
st
p
h
as
e
-
6
th
p
h
a
s
e
b
e
l
o
w
:
a)
1
s
t
Ph
ase:
B
o
t
-
I
o
T
Data
s
et
p
r
ep
r
o
ce
s
s
in
g
;
I
n
th
is
s
tep
,
we
p
r
ep
r
o
ce
s
s
ed
o
u
r
d
ataset
(
we
w
o
r
k
ed
with
th
e
en
tire
d
ataset;
m
o
r
e
th
an
7
2
m
illi
o
n
r
ec
o
r
d
s
o
f
d
ata,
with
all
f
ea
tu
r
es)
b
y
alter
in
g
th
e
r
aw
d
a
ta
an
d
n
o
r
m
alizin
g
its
v
alu
es,
a
n
d
th
e
n
co
n
v
er
t it
in
to
im
ag
e
s
h
ap
e;
b)
2
n
d
Ph
ase:
T
ON
-
I
o
T
Data
s
et
p
r
ep
r
o
ce
s
s
in
g
;
Firstl
y
,
we
n
ee
d
to
ad
ap
t
th
e
n
ew
d
ata
with
t
h
e
o
ld
d
ataset,
an
d
th
at
b
y
g
en
e
r
atin
g
th
e
m
i
s
s
in
g
f
ea
tu
r
es
lik
e
(
Std
d
ev
,
St
ate,
Me
an
,
Min
,
Ma
x
,
Seq
,
Sra
te,
Dr
ate,
…)
f
r
o
m
th
e
p
ca
p
f
iles
u
s
in
g
th
e
Ar
g
u
s
to
o
l
[2
6
]
.
T
h
e
n
s
ec
o
n
d
ly
,
we
p
r
ep
r
o
ce
s
s
ed
it
u
s
in
g
th
e
s
am
e
p
r
ep
r
o
ce
s
s
o
f
t
h
e
B
o
t
-
I
o
T
d
ataset
p
r
ep
r
o
ce
s
s
in
g
(
we
t
o
o
k
o
n
ly
5
0
0
0
0
r
ec
o
r
d
s
o
f
d
ata
f
r
o
m
ea
ch
class
;
3
0
0
0
0
f
o
r
th
e
tr
ai
n
in
g
d
ataset
,
1
0
0
0
0
f
o
r
th
e
v
alid
atio
n
d
at
aset,
an
d
th
e
r
em
ain
in
g
o
t
h
er
1
0
0
0
0
f
o
r
th
e
test
d
ataset)
,
to
en
s
u
r
e
th
e
s
am
e
ad
ap
tiv
ity
b
etwe
en
th
e
two
d
atasets
;
c)
3
r
d
Ph
ase:
B
u
ild
in
g
t
h
e
o
r
i
g
in
al
m
o
d
el;
a
f
ter
war
d
,
we
b
u
il
d
an
d
tr
ain
o
u
r
m
o
d
el
o
n
a
t
r
ain
in
g
d
ataset
(
6
0
%
o
f
th
e
B
o
t
-
I
o
T
d
ataset)
,
an
d
we
u
s
e
th
e
v
alid
atio
n
d
ataset
(
2
0
%
o
f
th
e
d
ataset)
t
o
v
alid
ate
th
e
ac
cu
r
ac
y
o
f
o
u
r
m
o
d
el;
d)
4
th
Ph
ase:
E
v
alu
atin
g
th
e
m
o
d
el
o
n
n
ew
d
ata;
af
ter
b
u
ild
in
g
th
e
o
r
i
g
in
al
m
o
d
el
,
f
ir
s
tly
e
v
alu
ated
with
th
e
test
d
ataset
o
f
th
e
B
o
t
-
I
o
T
d
ataset
(
th
e
r
em
ain
in
g
2
0
%
o
f
th
e
d
ataset)
,
th
en
we
ev
alu
at
e
o
n
th
e
n
ew
attac
k
’
s
b
eh
av
io
r
s
o
n
th
e
test
d
ataset
o
f
th
e
T
ON
-
I
o
T
d
ataset
(
2
0
% o
f
t
h
e
d
ataset)
b
y
p
r
e
d
ictin
g
attac
k
s
;
e)
5
th
Ph
ase:
Up
d
atin
g
t
h
e
m
o
d
el;
f
r
o
m
th
e
o
r
ig
in
al
an
d
alr
ea
d
y
tr
ain
ed
m
o
d
el,
we
f
r
ee
ze
th
e
C
o
n
v
o
lu
tio
n
al
b
ase
an
d
th
en
u
s
e
its
o
u
tp
u
ts
to
f
ee
d
th
e
C
lass
if
ier
.
W
e
r
etr
ain
th
e
class
if
ier
lay
er
s
o
n
to
p
o
f
th
e
f
r
o
ze
n
o
n
es
o
n
a
tr
ain
i
n
g
d
ataset
(
5
0
%
o
f
T
ON
-
I
o
T
d
ataset
an
d
1
0
%
o
f
B
o
T
-
I
o
T
d
ataset)
,
an
d
u
s
in
g
th
e
v
alid
atio
n
d
ataset
(
1
5
%
o
f
t
h
e
d
ataset
T
ON
-
I
o
T
d
ataset
an
d
5
%
o
f
B
o
T
-
I
o
T
d
ataset)
we
v
alid
ate
th
e
ac
cu
r
ac
y
o
f
o
u
r
m
o
d
el.
T
h
e
id
ea
b
eh
in
d
r
etr
ain
in
g
o
n
n
ew
an
d
o
ld
d
ata
is
t
h
at
we
d
o
n
o
t
wan
t o
u
r
m
o
d
el
to
b
e
in
f
l
u
en
c
ed
o
n
ly
o
n
t
h
e
n
ew
attac
k
’
s
b
e
h
av
io
r
s
;
f)
6
t
h
P
h
a
s
e
:
E
v
a
l
u
a
t
i
n
g
t
h
e
u
p
d
a
t
e
d
m
o
d
e
l
;
a
f
t
e
r
u
p
d
a
t
i
n
g
t
h
e
m
o
d
e
l
,
w
e
e
v
a
l
u
a
t
e
i
t
w
i
t
h
t
h
e
t
e
s
t
d
a
t
a
s
e
t
o
f
t
h
e
T
O
N
-
I
o
T
d
a
t
a
s
e
t
(
1
5
%
o
f
t
h
e
d
a
t
a
s
e
t
T
O
N
-
I
o
T
d
a
t
a
s
e
t
a
n
d
5
%
o
f
B
o
T
-
I
o
T
d
a
t
a
s
e
t
)
b
y
p
r
e
d
i
c
t
i
n
g
a
t
t
a
c
k
s
.
B
o
t
-
I
o
T
D
a
ta
se
t
T
o
n
-
I
o
T
D
a
ta
set
T
r
an
sfe
r
t
h
e
K
n
o
w
l
e
d
g
e
C
N
N
T
ra
i
n
i
n
g
I
D
S
M
o
d
e
l
U
p
d
a
ted
I
D
S
M
o
d
e
l
C
N
N
T
r
a
i
n
i
n
g
Fr
o
z
e
n
W
e
ig
h
t
s
Fig
u
r
e
3
.
Ou
r
s
ch
em
e
o
f
u
p
d
at
in
g
th
e
DL
-
b
ased
I
DS m
o
d
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
cc
elera
tin
g
th
e
u
p
d
a
te
o
f
a
DL
-
b
a
s
ed
I
DS
fo
r
I
o
T
u
s
in
g
d
ee
p
tr
a
n
s
fer lea
r
n
in
g
(
I
d
r
is
s
I
d
r
is
s
i
)
1063
4
.
1
.
B
uil
din
g
t
he
o
rig
ina
l
m
o
del
W
e
b
u
ild
o
u
r
I
DS
m
o
d
el
r
eso
r
tin
g
o
n
C
NN
as
in
o
u
r
p
r
e
v
io
u
s
wo
r
k
[
1
]
,
th
is
m
o
d
el
was
d
efin
ed
b
y
an
in
p
u
t
lay
e
r
with
1
6
in
p
u
t
n
eu
r
o
n
s
,
f
iv
e
h
id
d
e
n
lay
er
s
C
o
n
v
o
lu
tio
n
1
D
lay
er
,
Ma
x
Po
o
li
n
g
1
D
lay
er
,
Flatten
lay
er
,
R
eL
U
lay
er
,
Den
s
e
lay
er
,
an
d
an
o
u
tp
u
t
lay
er
.
W
e
in
itially
tr
ain
ed
o
u
r
m
o
d
el
in
1
0
ep
o
ch
s
(
th
e
wh
o
le
d
ataset
is
p
ass
ed
th
r
o
u
g
h
t
h
e
n
eu
r
al
n
etwo
r
k
1
0
tim
es)
w
ith
a
b
atch
s
ize
o
f
3
2
(
th
e
a
m
o
u
n
t
o
f
tr
ain
in
g
s
am
p
les
in
a
s
in
g
le
b
atch
is
3
2
)
.
T
h
e
n
e
u
r
al
n
etwo
r
k
in
cl
u
d
es
1
6
in
p
u
t
n
e
u
r
o
n
s
(
th
e
s
am
e
n
u
m
b
er
as
th
e
f
ea
tu
r
es),
with
4
in
ter
m
ed
iate
(
h
id
d
en
)
lay
er
s
,
1
6
(
C
o
n
v
o
l
u
tio
n
1
D)
,
8
(
Ma
x
Po
o
lin
g
1
D)
,
2
5
6
(
Flatten
)
,
2
5
6
(
R
eL
U)
n
eu
r
o
n
s
,
4
4
(
Den
s
e)
n
eu
r
o
n
s
,
an
d
4
o
u
t
p
u
t
n
eu
r
o
n
s
f
o
r
th
e
m
u
lticlas
s
class
i
ficatio
n
as
s
h
o
wn
in
Fig
u
r
e
4
.
W
e
tr
ain
ed
a
n
d
test
ed
o
u
r
m
o
d
el
o
n
th
e
B
o
t
-
I
o
T
d
ataset
th
at
co
n
tain
s
ar
o
u
n
d
7
2
m
illi
o
n
r
ec
o
r
d
s
o
f
d
ata
tr
af
f
ic
s
im
u
lated
I
o
T
en
v
ir
o
n
m
e
n
t.
T
h
e
tr
ain
in
g
an
d
test
d
ataset
co
n
tain
s
1
1
class
es
wh
ich
r
ef
lect
1
0
ty
p
es
o
f
attac
k
s
with
in
4
attac
k
s
ca
teg
o
r
ies
an
d
t
h
e
n
o
r
m
al
tr
af
f
ic;
m
ea
n
in
g
5
class
es
if
we
wo
r
k
o
n
ly
with
ca
teg
o
r
ies,
wh
ich
is
in
o
u
r
ca
s
e
in
th
e
g
o
al
to
h
a
v
e
th
e
s
am
e
class
with
th
e
tar
g
et
d
ataset.
?×
1
6
×
1
C
o
n
v
1
D
ke
r
n
e
l
〈
4
×
1
×
3
2
〉
b
i
a
s
〈
32
〉
M
a
xP
o
o
l
i
n
g
1
D
F
l
a
t
t
e
n
R
e
L
U
D
e
n
s
e
ke
r
n
e
l
〈
2
5
6
×
4
4
〉
b
i
a
s
〈
44
〉
D
e
n
s
e
ke
r
n
e
l
〈
4
4
×
4
〉
b
i
a
s
〈
4
〉
i
n
p
u
t
d
e
n
s
e
_
1
Fig
u
r
e
4
.
Sp
ec
if
icatio
n
o
f
o
u
r
I
DS m
o
d
el
lay
er
s
4
.
2
.
B
uil
din
g
t
he
up
da
t
ed
m
o
del
.
As
s
h
o
wn
in
Fig
u
r
e
3
,
f
r
o
m
th
e
p
r
ev
io
u
s
an
d
alr
ea
d
y
tr
ai
n
ed
m
o
d
el
o
n
th
e
o
r
ig
in
al
d
a
taset
(
B
o
t
-
I
o
T
)
,
we
f
r
ee
ze
th
e
co
n
v
o
lu
tio
n
al
b
ase
(
k
ee
p
it
in
its
o
r
ig
in
al
f
o
r
m
to
av
o
id
d
estro
y
in
g
an
y
o
f
th
e
in
f
o
r
m
atio
n
th
ey
co
n
tain
d
u
r
in
g
f
u
tu
r
e
tr
ain
in
g
r
o
u
n
d
s
;
m
ea
n
i
n
g
t
r
an
s
f
er
r
in
g
t
h
e
p
ar
am
eter
s
)
;
wh
i
ch
is
th
e
s
tack
o
f
C
o
n
v
o
lu
tio
n
1
D
lay
er
,
Ma
x
Po
o
lin
g
1
D
la
y
er
,
Flatten
lay
e
r
,
an
d
R
eL
U
lay
er
.
T
h
e
im
p
o
r
tan
t
g
o
al
o
f
th
e
co
n
v
o
l
u
tio
n
al
b
ase
is
to
g
e
n
er
ate
f
ea
tu
r
es
f
r
o
m
th
e
b
ase
d
ata
s
et
(
B
o
t
-
I
o
T
)
,
an
d
th
en
u
s
e
its
o
u
tp
u
ts
to
f
ee
d
th
e
class
if
ier
;
wh
ich
is
th
e
s
tack
o
f
th
e
d
en
s
e
la
y
er
s
,
an
d
th
e
o
u
tp
u
t
lay
e
r
(
th
ese
a
r
e
f
u
lly
th
e
co
n
n
ec
te
d
lay
e
r
s
)
.
W
e
r
etr
ain
th
e
class
if
ier
lay
e
r
s
o
n
to
p
o
f
th
e
f
r
o
ze
n
o
n
es.
T
h
ese
r
etr
ain
ed
lay
er
s
will
le
ar
n
to
t
u
r
n
th
e
o
l
d
f
ea
tu
r
es in
to
p
r
ed
ictio
n
s
o
n
th
e
n
ew
d
ataset
(
T
ON
-
I
o
T
)
.
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
5
.
1
.
H
a
rdwa
re
cha
ra
ct
er
is
t
ics
T
h
e
r
esu
lts
we
o
b
tain
e
d
wer
e
p
er
f
o
r
m
ed
o
n
a
h
ig
h
-
p
e
r
f
o
r
m
an
ce
co
m
p
u
tin
g
(
HPC
)
in
f
r
astru
ctu
r
e
with
th
e
f
o
llo
win
g
h
ar
d
war
e
c
h
ar
ac
ter
is
tics
:
a)
C
PU: two
I
n
tel
Xeo
n
Go
ld
6
1
4
8
(
2
.
4
GHz
/2
0
c
o
r
es)
b)
R
AM
: 1
9
2
Gb
c)
GPU: two
NVI
DI
A
T
esla P1
0
0
(
1
2
Gb
)
with
cu
d
a
v
1
0
.
1
I
n
o
u
r
ex
p
e
r
im
en
ts
,
we
wo
r
k
ed
with
Ker
as
(
2
.
4
.
0
)
[2
7
]
;
an
o
p
e
n
-
s
o
u
r
ce
p
y
th
o
n
Dee
p
L
ea
r
n
in
g
lib
r
ar
y
wh
ich
is
r
u
n
n
in
g
o
n
t
o
p
o
f
G
o
o
g
le’
s
o
p
en
-
s
o
u
r
ce
d
ata
f
lo
w
s
o
f
twar
e,
an
d
u
s
es
T
en
s
o
r
Flo
w
-
GPU
(
2
.
3
.
0
)
[2
8
]
as a
b
ac
k
en
d
en
g
i
n
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2502
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t 2
0
2
1
:
1
0
5
9
-
1
0
6
7
1064
5
.
2
.
E
v
a
lua
t
io
n
m
et
rics
T
o
ev
al
u
ate
o
u
r
m
o
d
els,
we
u
s
ed
th
e
s
p
ec
if
ied
m
etr
ics:
Acc
u
r
ac
y
(
)
,
L
o
s
s
(
)
,
Pre
cisi
o
n
(
)
,
R
ec
all
(
)
,
F1
Sco
r
e
(
1
)
,
a
n
d
C
o
n
f
u
s
io
n
Ma
tr
ix
(
)
.
T
h
ese
m
etr
i
cs
ar
e
ca
lcu
lated
u
s
in
g
th
e
f
o
ll
o
win
g
f
o
u
r
d
if
f
er
en
t m
ea
s
u
r
es
[2
9
]
:
1)
T
r
u
e
Po
s
itiv
e
(
)
: is th
e
n
u
m
b
er
o
f
p
o
s
itiv
e
class
r
ec
o
r
d
s
co
r
r
e
ctly
class
if
ied
.
2)
T
r
u
e
Neg
ativ
e
(
)
: is th
e
n
u
m
b
e
r
o
f
n
eg
ativ
e
class
r
ec
o
r
d
s
co
r
r
e
ctly
class
if
ied
.
3)
Fals
e
Po
s
i
tiv
e
(
)
: is th
e
n
u
m
b
er
o
f
n
eg
ativ
e
class
r
ec
o
r
d
s
wr
o
n
g
ly
class
if
ied
.
4)
Fals
e
Neg
ativ
e
(
)
: is th
e
n
u
m
b
er
o
f
p
o
s
itiv
e
class
r
ec
o
r
d
s
wr
o
n
g
ly
class
if
ied
.
−
Acc
u
r
a
cy
(
)
:
is
th
e
p
er
ce
n
tag
e
o
f
tr
u
e
d
etec
tio
n
o
v
e
r
th
e
to
tal
tr
af
f
ic
r
ec
o
r
d
s
an
d
ca
lcu
lat
ed
u
s
in
g
th
e
f
o
llo
win
g
f
o
r
m
u
la
(
1
)
:
=
+
+
+
+
(
1
)
−
L
o
s
s
(
)
:
is
th
e
d
if
f
er
en
ce
b
etw
ee
n
th
e
p
r
ed
icted
v
alu
e
an
d
th
e
tr
u
e
v
alu
e.
T
h
e
m
o
s
t
u
s
ed
lo
s
s
f
u
n
ctio
n
in
d
ee
p
n
e
u
r
al
n
etwo
r
k
s
is
cr
o
s
s
-
en
tr
o
p
y
[
30
]
,
an
d
ca
lcu
lated
u
s
in
g
th
e
f
o
llo
win
g
f
o
r
m
u
la
(
2
)
:
=
−
∑
,
(
,
)
=
1
(
2
)
W
h
ile:
: is th
e
n
u
m
b
e
r
o
f
class
es (
5
i
n
o
u
r
ca
s
e)
: is th
e
b
in
ar
y
i
n
d
icato
r
(
tr
u
e
1
o
r
f
alse 0
)
if
class
lab
el
is
t
h
e
co
r
r
ec
t c
lass
if
icatio
n
f
o
r
o
b
s
er
v
atio
n
: is th
e
p
r
ed
icted
p
r
o
b
a
b
ilit
y
o
b
s
er
v
atio
n
o
f
class
−
Pre
cisi
o
n
(
)
:
is
th
e
p
er
ce
n
tag
e
o
f
p
r
e
d
icted
attac
k
s
tr
a
f
f
ic
th
at
ar
e
tr
u
ly
attac
k
s
,
an
d
ca
lcu
l
ated
u
s
in
g
th
e
f
o
llo
win
g
f
o
r
m
u
la
(
3
)
:
=
+
(
3
)
−
R
ec
all
(
)
: is th
e
p
er
ce
n
ta
g
e
o
f
attac
k
s
tr
af
f
ic
v
er
s
u
s
all
th
e
attac
k
s
tr
af
f
ic
o
b
tain
a
b
le,
an
d
ca
lcu
lated
u
s
in
g
th
e
f
o
llo
win
g
f
o
r
m
u
la
(
4
)
:
=
+
(
4
)
−
F1
Sco
r
e
(
1
)
:
is
a
m
ea
s
u
r
e
o
f
th
e
test
'
s
ac
cu
r
ac
y
.
I
t
is
ca
lcu
lat
ed
f
r
o
m
th
e
p
r
ec
is
io
n
an
d
r
e
ca
ll
o
f
th
e
test
,
an
d
ca
lcu
lated
u
s
in
g
th
e
f
o
llo
win
g
f
o
r
m
u
la
(
5
)
:
1
=
2
−
1
+
−
1
(
5
)
−
C
o
n
f
u
s
io
n
Ma
tr
ix
(
)
:
is
a
tab
le
th
at
g
iv
es
a
v
is
u
aliza
tio
n
o
f
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el
b
y
r
ep
r
esen
tin
g
t
h
e
in
s
tan
ce
s
in
t
h
e
p
r
e
d
icted
class
in
a
r
o
w
w
h
ile
r
ep
r
esen
ts
th
e
i
n
s
tan
ce
s
in
an
ac
t
u
al
class
in
th
e
co
l
u
m
n
s
.
5
.
3
.
E
v
a
lua
t
ing
t
he
m
o
del
I
n
t
h
e
f
o
llo
win
g
Fig
u
r
e
s
5
-
7
,
we
p
r
esen
t
th
e
ac
cu
r
ac
y
tr
ai
n
in
g
,
lo
s
s
tr
ain
in
g
,
ac
cu
r
a
cy
v
alid
atio
n
,
lo
s
s
v
alid
atio
n
,
an
d
f
o
r
ea
c
h
c
lass
,
we
p
r
esen
t
th
e
R
ec
all
an
d
Pre
cisi
o
n
,
a
n
d
with
th
e
F1
s
co
r
e
f
o
r
th
e
o
r
i
g
in
al
m
o
d
el
wh
ich
was
tr
ain
ed
o
v
e
r
1
0
ep
o
ch
s
.
As
s
h
o
wn
in
Fig
u
r
e
5
f
o
r
th
e
o
r
ig
in
al
m
o
d
el
in
1
0
ep
o
c
h
s
(
8
.
7
8
h
o
u
r
s
o
f
t
r
ain
in
g
)
th
e
ac
cu
r
ac
y
r
ea
ch
e
d
9
9
,
9
9
%
an
d
th
e
lo
s
s
attain
ed
0
,
1
5
%
in
tr
ain
in
g
a
n
d
f
o
r
th
e
v
alid
atio
n
as sh
o
wn
in
Fig
u
r
e
6
it r
ea
ch
ed
9
9
,
9
9
% in
ac
cu
r
ac
y
an
d
0
,
1
2
% in
lo
s
s
,
an
d
a
1
0
0
% f
o
r
th
e
test
in
g
s
et.
B
u
t th
e
ac
cu
r
ac
y
f
ails
to
co
n
tr
o
l
f
o
r
s
ize
im
b
alan
ce
s
in
th
e
clas
s
es
th
at’
s
wh
y
it
d
o
esn
’
t
allo
w
u
s
to
h
av
e
a
clea
r
er
v
iew
o
n
h
o
w
th
e
m
o
d
el
is
d
o
i
n
g
f
o
r
ea
ch
class
;
it
d
o
es
n
o
t
g
iv
e
a
p
er
-
class
m
etr
ic
f
o
r
m
u
lti
-
class
p
r
o
b
lem
s
.
R
eg
ar
d
in
g
t
h
e
R
ec
all
an
d
th
e
Pre
cisi
o
n
,
we
g
et
a
clea
r
er
r
esu
lt
f
o
r
ea
ch
class
.
W
e
s
ee
h
er
e
th
at
in
s
o
m
e
class
es
s
u
ch
as
5
:
“T
h
ef
t”
we
o
b
tain
ed
v
er
y
lo
w
m
ea
s
u
r
es
i
n
all
th
e
m
etr
ics
(
0
%),
an
d
f
o
r
th
e
f
o
r
t
h
e
class
3
:
“No
r
m
al
tr
af
f
ic”,
we
o
b
tain
ed
v
er
y
lo
w
m
ea
s
u
r
es
in
b
o
th
r
ec
all
an
d
F1
s
co
r
e
as
s
h
o
wn
in
Fig
u
r
e
7
,
m
ea
n
wh
ile
th
e
p
r
ec
is
io
n
is
f
a
ls
ely
h
ig
h
(
1
0
0
%)
d
u
e
to
wea
k
n
ess
o
f
th
is
m
etr
ic,
a
n
d
it
r
ar
ely
p
r
ed
icts
th
is
class
.
Fo
r
th
e
C
las
s
2
:
“Do
S”,
th
e
r
ec
all
m
etr
ic
also
h
as
a
wea
k
n
ess
wh
er
e
it
ca
n
ac
h
iev
e
v
er
y
h
ig
h
m
ea
s
u
r
em
en
t
in
th
is
class
b
y
alwa
y
s
p
r
ed
ictin
g
it
an
d
th
at
co
u
ld
m
ea
n
l
o
ts
o
f
i
n
co
r
r
ec
t
g
u
ess
es
(
in
th
e
D
o
S
class
w
ith
th
e
n
ew
d
ata
in
F
ig
u
r
e
8
)
.
E
ac
h
m
etr
ic
ca
n
b
e
b
iased
f
o
r
d
if
f
er
en
t
class
es
an
d
th
at
d
u
e
to
th
e
u
n
b
alan
ce
d
d
ata
in
th
ese
class
es,
s
o
u
s
in
g
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
Pre
cisi
o
n
an
d
R
ec
all,
th
e
F1
-
s
co
r
e
th
at
g
iv
es
u
s
a
b
etter
r
esu
lt o
f
t
h
e
in
co
r
r
e
ctly
class
if
ied
ca
s
es th
an
th
e
ac
cu
r
ac
y
m
etr
ic.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
cc
elera
tin
g
th
e
u
p
d
a
te
o
f
a
DL
-
b
a
s
ed
I
DS
fo
r
I
o
T
u
s
in
g
d
ee
p
tr
a
n
s
fer lea
r
n
in
g
(
I
d
r
is
s
I
d
r
is
s
i
)
1065
W
e
b
en
ch
m
ar
k
ed
th
e
o
r
ig
in
a
l
m
o
d
el
o
n
n
ew
d
at
a
f
r
o
m
t
h
e
T
ON
-
I
o
T
n
etwo
r
k
d
ataset
an
d
g
o
t
a
d
ec
r
ea
s
e
d
o
wn
to
5
6
%
o
n
th
e
F1
s
co
r
e
m
etr
ic;
an
d
esp
ec
ially
f
o
r
th
e
n
o
r
m
al
an
d
th
ef
t
cl
ass
es,
we
o
b
tain
ed
r
esp
ec
tiv
ely
o
n
ly
3
%
an
d
2
7
%
f
o
r
th
e
s
am
e
m
etr
ic
as
s
h
o
wn
in
Fig
u
r
e
8
.
T
h
is
d
ec
r
ea
s
e
was
ac
tu
ally
d
u
e
to
two
ca
u
s
es;
th
e
f
ir
s
t
o
n
e
is
th
a
t
we
test
ed
o
u
r
m
o
d
el
o
n
b
ala
n
ce
d
d
ata
o
v
er
d
if
f
e
r
en
t
class
es
wh
ich
allo
wed
to
m
ak
e
b
alan
ce
d
m
etr
ic
r
esu
lts
;
wh
ile
th
e
s
ec
o
n
d
ca
u
s
e
is
r
elate
d
to
n
ew
b
e
h
av
io
r
s
o
r
m
u
t
atio
n
s
in
th
e
attac
k
s
s
witch
in
g
f
r
o
m
th
e
o
ld
to
th
e
n
ew
d
a
tasets
.
I
n
Fig
u
r
e
s
9
-
1
1
,
we
p
r
esen
t
th
e
ac
cu
r
ac
y
tr
ai
n
in
g
,
lo
s
s
tr
ain
in
g
,
ac
cu
r
ac
y
v
alid
atio
n
,
lo
s
s
v
alid
atio
n
,
an
d
f
o
r
ea
ch
class
,
we
p
r
esen
t
th
e
R
ec
all
an
d
Pre
cisi
o
n
,
an
d
alo
n
g
with
F1
s
co
r
e
o
f
th
e
u
p
d
ated
m
o
d
el
th
at
was
r
etr
ain
ed
o
v
er
1
0
ep
o
ch
s
in
ju
s
t
1
7
0
s
ec
o
n
d
s
o
f
tr
ain
in
g
(
1
7
s
in
ev
er
y
ep
o
ch
)
.
Af
ter
u
p
d
atin
g
th
e
m
o
d
el
u
s
in
g
tr
an
s
f
er
lear
n
in
g
,
we
r
ea
ch
ed
an
ac
cu
r
ac
y
o
f
9
9
.
4
3
%
an
d
a
lo
s
s
o
f
0
.
3
6
%
in
tr
ain
in
g
an
d
f
o
r
th
e
v
alid
atio
n
,
it
also
r
ea
ch
es
9
9
.
4
7
%
in
ac
cu
r
ac
y
an
d
0
.
2
3
%
in
lo
s
s
,
f
o
r
th
e
F1
s
co
r
e
it
r
ea
ch
e
s
9
9
%
as
s
h
o
wn
in
Fig
u
r
e
1
1
W
e
ca
n
r
em
ar
q
u
e
a
n
im
p
r
o
v
em
en
t
f
o
r
t
h
e
Do
S,
DDo
S,
an
d
Scan
attac
k
s
,
an
d
a
b
ig
g
er
o
n
e
f
o
r
th
e
r
em
ain
in
g
class
es;
T
h
ef
t
an
d
No
r
m
al
tr
af
f
ic,
m
ea
n
in
g
th
a
t
o
u
r
I
DS
h
as
n
o
t
o
n
ly
b
ee
n
u
p
d
ated
b
u
t
also
o
v
er
co
m
e
th
e
lack
o
f
lab
e
led
d
ata
in
th
ese
class
e
s
,
esp
ec
iall
y
th
e
n
o
r
m
al
o
n
e
b
ec
au
s
e
its
im
p
o
r
tan
ce
is
wh
en
d
if
f
er
en
cin
g
b
etwe
en
g
o
o
d
an
d
b
ad
tr
a
f
f
ic.
W
e
al
s
o
co
m
p
ar
ed
th
e
tr
ain
in
g
tim
e
f
o
r
th
e
in
itial
tr
ain
in
g
f
o
r
th
e
o
r
ig
in
al
m
o
d
el
an
d
th
e
u
p
d
ated
o
n
e
as
s
h
o
wn
in
Fig
u
r
e
1
2
wh
er
e
we
s
ee
a
b
ig
d
if
f
er
en
ce
b
e
twee
n
th
e
two
,
f
r
o
m
3
1
5
9
0
s
e
co
n
d
s
in
th
e
in
itial
tr
ain
in
g
o
f
th
e
o
r
ig
in
al
I
DS
(
tr
ain
ed
o
v
e
r
1
0
ep
o
c
h
s
;
ar
o
u
n
d
3
1
5
9
0
s
ec
o
n
d
s
in
e
v
er
y
e
p
o
ch
)
t
o
o
n
l
y
1
7
0
s
ec
o
n
d
s
wer
e
u
p
d
atin
g
it
(
th
e
u
p
d
ate
o
f
th
e
DDo
S,
Do
s
,
an
d
R
ec
o
n
n
aiss
an
ce
wa
s
d
o
n
e
o
n
ly
in
3
ep
o
ch
s
,
an
d
we
p
r
o
lo
n
g
e
d
th
e
tr
ain
in
g
7
m
o
r
e
ep
o
c
h
s
to
tr
ain
th
e
o
th
er
class
e
s
;
ar
o
u
n
d
1
7
s
ec
o
n
d
s
in
ev
er
y
ep
o
ch
)
.
T
h
is
p
r
o
v
es
th
at
t
r
an
s
f
er
lear
n
in
g
is
a
s
o
lu
tio
n
f
o
r
tr
an
s
f
er
lear
n
in
g
with
m
i
n
im
al
co
m
p
u
tin
g
p
o
wer
an
d
m
u
c
h
f
ewe
r
d
ata
co
m
p
ar
ed
t
o
th
e
o
r
i
g
in
al
tr
ain
in
g
.
Fig
u
r
e
5
.
T
h
e
ac
cu
r
ac
y
in
th
e
tr
ain
in
g
an
d
th
e
v
alid
atio
n
(
o
r
ig
in
al
m
o
d
el
)
Fig
u
r
e
6
.
T
h
e
lo
s
s
in
th
e
tr
ain
i
n
g
an
d
th
e
v
alid
atio
n
(
o
r
ig
in
al
m
o
d
el
)
Fig
u
r
e
7
.
C
o
n
f
u
s
io
n
m
atr
i
x
(
f
o
r
th
e
o
r
ig
in
al
m
o
d
el
o
n
th
e
o
r
ig
in
al
d
ataset
Fig
u
r
e
8
.
C
o
n
f
u
s
io
n
m
atr
i
x
(
f
o
r
th
e
o
r
ig
in
al
m
o
d
el
o
n
th
e
n
ew
d
ataset)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2502
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci,
Vo
l.
23
,
No
.
2
,
Au
g
u
s
t 2
0
2
1
:
1
0
5
9
-
1
0
6
7
1066
Fig
u
r
e
9
.
T
h
e
ac
cu
r
ac
y
in
th
e
tr
ain
in
g
an
d
th
e
v
alid
atio
n
(
u
p
d
ated
m
o
d
el
)
Fig
u
r
e
1
0
.
T
h
e
lo
s
s
in
th
e
tr
ai
n
in
g
an
d
th
e
v
alid
atio
n
(
u
p
d
ated
m
o
d
el
)
Fig
u
r
e
1
1
.
C
o
n
f
u
s
io
n
m
atr
ix
(
t
h
e
u
p
d
ated
m
o
d
el)
Fig
u
r
e
1
2
.
T
r
ain
in
g
tim
e
6.
CO
NCLU
SI
O
N
T
h
e
en
o
r
m
o
u
s
n
etwo
r
k
tr
af
f
ic
d
ata
b
etwe
en
I
o
T
o
b
jects
d
is
p
atch
ed
ar
o
u
n
d
th
e
wo
r
ld
h
av
e
tak
en
a
b
ig
ch
allen
g
e
to
t
r
ad
itio
n
al
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
(
I
DS
)
.
R
esear
ch
er
s
ten
d
to
b
u
ild
I
DS
b
ased
o
n
Dee
p
lear
n
in
g
d
u
e
to
its
o
u
ts
tan
d
in
g
p
er
f
o
r
m
an
ce
in
v
ar
i
o
u
s
f
ield
s
,
wh
ich
its
elf
g
o
t
s
o
m
e
p
r
o
b
lem
s
lik
e
d
ata
-
d
ep
en
d
e
n
t
o
r
lack
o
f
lab
eled
d
ata.
Ou
r
p
r
o
p
o
s
ed
“
Up
d
ate
d
d
ee
p
tr
an
s
f
er
lear
n
in
g
-
b
ased
in
tr
u
s
io
n
d
etec
tio
n
s
y
s
tem
f
o
r
I
o
T
”
was
b
u
ilt
in
itially
with
co
n
v
o
lu
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NN)
o
n
B
o
t
-
I
o
T
d
ataset
an
d
b
ee
n
u
p
d
ated
o
n
a
s
m
all
am
o
u
n
t
o
f
d
ata
f
r
o
m
th
e
T
ON
-
I
o
T
d
ataset,
af
ter
s
ev
er
al
ex
p
e
r
im
e
n
ts
,
we
o
b
tain
ed
a
r
ea
s
o
n
ab
le
d
etec
tio
n
r
ate
o
n
th
is
n
ew
u
p
d
ated
I
DS.
B
y
an
aly
zin
g
th
e
o
b
tain
ed
r
esu
lts
,
we
co
n
clu
d
ed
th
at
T
r
an
s
f
er
L
ea
r
n
i
n
g
ca
n
b
e
an
i
d
ea
l
s
o
lu
tio
n
n
o
t
o
n
ly
to
c
o
m
p
en
s
ate
th
e
lack
o
f
d
ata
in
s
o
m
e
attac
k
class
es
b
u
t
also
to
u
p
d
ate
th
e
I
DS
s
y
s
tem
s
with
ju
s
t
a
m
in
im
al
co
m
p
u
tin
g
p
o
wer
a
n
d
ef
f
o
r
t.
As
f
u
tu
r
e
wo
r
k
s
,
we
will
d
ep
lo
y
o
u
r
I
DS
in
a
r
ea
l
I
o
T
en
v
ir
o
n
m
e
n
t,
a
n
d
o
n
a
lig
h
tweig
h
t
I
o
T
d
ev
ice
w
h
ile
o
p
tim
izin
g
it
with
n
o
ac
cu
r
ac
y
lo
s
s
wh
ile
s
tu
d
y
in
g
i
ts
p
er
f
o
r
m
a
n
ce
o
n
r
ea
l I
o
T
n
et
wo
r
k
tr
af
f
ic
d
ata.
ACK
NO
WL
E
DG
E
M
E
NT
S
T
h
is
r
esear
ch
was
s
u
p
p
o
r
ted
t
h
r
o
u
g
h
co
m
p
u
tatio
n
al
r
eso
u
r
c
es
o
f
HPC
-
MA
R
W
AN
p
r
o
v
id
ed
b
y
th
e
Natio
n
al
C
en
ter
f
o
r
Scien
tific
an
d
T
ec
h
n
ical
R
esear
ch
(
C
NR
ST)
R
ab
at,
Mo
r
o
cc
o
.
RE
F
E
R
E
NC
E
S
[1
]
I.
Id
rissi,
M
.
B
o
u
k
a
b
o
u
s,
M
.
A
z
izi,
O.
M
o
u
ss
a
o
u
i,
a
n
d
H.
El
F
a
d
il
i,
“
To
wa
r
d
a
d
e
e
p
lea
rn
in
g
-
b
a
se
d
in
tru
si
o
n
d
e
tec
ti
o
n
s
y
ste
m
fo
r
i
o
t
a
g
a
in
st
b
o
tn
e
t
a
tt
a
c
k
s,”
IA
ES
I
n
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Arti
f
icia
l
I
n
telli
g
e
n
c
e
(IJ
-
AI)
,
v
o
l.
1
0
,
n
o
.
1
,
p
p
.
1
1
0
–
1
2
0
,
M
a
r.
2
0
2
1
,
d
o
i:
1
0
.
1
1
5
9
1
/i
jai.
v
1
0
.
i
1
.
p
p
1
1
0
-
1
2
0
.
[2
]
I.
Id
rissi,
M
.
M
o
sta
fa
Az
izi,
a
n
d
O.
M
o
u
ss
a
o
u
i
,
“
A
Li
g
h
twe
ig
h
t
Op
ti
m
ize
d
De
e
p
Lea
rn
in
g
-
b
a
se
d
Ho
st
-
In
tr
u
sio
n
De
tec
ti
o
n
S
y
ste
m
De
p
l
o
y
e
d
o
n
t
h
e
Ed
g
e
f
o
r
Io
T,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
mp
u
ti
n
g
a
n
d
Di
g
it
a
l
S
y
ste
ms
,
2
0
2
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
I
SS
N:
2502
-
4
7
5
2
A
cc
elera
tin
g
th
e
u
p
d
a
te
o
f
a
DL
-
b
a
s
ed
I
DS
fo
r
I
o
T
u
s
in
g
d
ee
p
tr
a
n
s
fer lea
r
n
in
g
(
I
d
r
is
s
I
d
r
is
s
i
)
1067
[3
]
K.
Ba
r
to
s,
M
.
S
o
fk
a
,
a
n
d
V.
F
ra
n
c
,
Op
ti
mize
d
In
v
a
ria
n
t
Rep
re
se
n
ta
ti
o
n
o
f
Ne
two
rk
T
r
a
ff
ic
f
o
r
D
e
tec
ti
n
g
Un
se
e
n
M
a
lw
a
re
Va
ri
a
n
ts
.
2
0
1
6
.
[4
]
C.
Tan
,
F
.
S
u
n
,
T.
Ko
n
g
,
W.
Z
h
a
n
g
,
C
.
Ya
n
g
,
a
n
d
C
.
Li
u
,
“
A
su
rv
e
y
o
n
d
e
e
p
tran
sfe
r
lea
rn
i
n
g
,
”
in
L
e
c
tu
re
No
tes
in
Co
mp
u
ter
S
c
ien
c
e
(i
n
c
lu
d
in
g
su
b
se
rie
s
L
e
c
tu
re
No
tes
i
n
A
rtif
icia
l
I
n
telli
g
e
n
c
e
a
n
d
L
e
c
t
u
re
No
tes
i
n
Bi
o
in
fo
rm
a
t
ics
)
,
v
o
l.
1
1
1
4
1
LNC
S
,
2
0
1
8
,
p
p
.
2
7
0
–
2
7
9
.
[5
]
S
.
T
.
Kris
h
n
a
a
n
d
H.
K
.
Ka
ll
u
ri,
“
De
e
p
lea
rn
i
n
g
a
n
d
tran
sfe
r
lea
rn
in
g
a
p
p
r
o
a
c
h
e
s
fo
r
ima
g
e
c
las
sifica
ti
o
n
,
”
In
ter
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Rec
e
n
t
T
e
c
h
n
o
l
o
g
y
a
n
d
En
g
i
n
e
e
rin
g
,
v
o
l.
7
,
n
o
.
5
S
4
,
p
p
.
4
2
7
–
4
3
2
,
2
0
1
9
.
[6
]
L.
M
o
u
a
n
d
Z.
Ji
n
,
T
re
e
-
Ba
se
d
C
o
n
v
o
l
u
ti
o
n
a
l
Ne
u
r
a
l
Ne
two
rk
s
,
S
p
rin
g
e
r
S
i
n
g
a
p
o
re
,
2
0
1
8
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
9
8
1
-
13
-
1
8
7
0
-
2.
[7
]
Q.
Ya
n
g
,
Y.
Z
h
a
n
g
,
W.
Da
i,
a
n
d
S
.
J.
P
a
n
,
T
ra
n
sfe
r
lea
rn
in
g
,
Ca
m
b
rig
e
,
Un
it
e
d
Ki
n
g
d
o
m
:
Ca
m
b
rid
g
e
U
n
iv
e
rsit
y
P
re
ss
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
1
7
/9
7
8
1
1
3
9
0
6
1
7
7
3
.
[8
]
T.
B.
Br
o
wn
,
e
t
a
l.
,
“
Lan
g
u
a
g
e
M
o
d
e
ls are
F
e
w
-
S
h
o
t
Le
a
rn
e
rs,”
a
rXiv:2
0
0
5
.
1
4
1
6
5
,
2
0
2
0
.
[9
]
“
G
it
Hu
b
-
u
lt
ra
ly
ti
c
s/y
o
l
o
v
5
:
YO
LOv
5
i
n
P
y
T
o
rc
h
>
ON
NX
>
Co
re
M
L
>
iOS
.
”
[On
li
n
e
].
Av
a
il
a
b
le:
h
tt
p
s:/
/g
it
h
u
b
.
c
o
m
/
u
lt
ra
ly
ti
c
s/y
o
lo
v
5
(a
c
c
e
ss
e
d
Au
g
.
1
2
,
2
0
2
0
).
[1
0
]
M
.
Be
rra
h
a
l
a
n
d
M
.
Az
izi,
“
Re
v
iew
o
f
DL
-
Ba
se
d
G
e
n
e
ra
ti
o
n
Tec
h
n
iq
u
e
s
o
f
Au
g
m
e
n
te
d
Im
a
g
e
s
u
sin
g
P
o
rtraits
S
p
e
c
ifi
c
a
ti
o
n
,
”
in
4
th
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
I
n
telli
g
e
n
t
Co
m
p
u
ti
n
g
in
Da
t
a
S
c
ien
c
e
s,
ICDS
2
0
2
0
,
No
v
.
2
0
2
0
,
p
p
.
1
–
8
,
d
o
i:
1
0
.
1
1
0
9
/ICD
S
5
0
5
6
8
.
2
0
2
0
.
9
2
6
8
7
1
0
.
[1
1
]
K.
Uc
h
id
a
,
M
.
Ta
n
a
k
a
,
a
n
d
M
.
Ok
u
to
m
i
,
“
Co
u
p
led
c
o
n
v
o
l
u
ti
o
n
lay
e
r
fo
r
c
o
n
v
o
lu
t
io
n
a
l
n
e
u
ra
l
n
e
two
rk
,
”
Ne
u
ra
l
Ne
two
rk
s
,
v
o
l
.
1
0
5
,
p
p
.
1
9
7
–
2
0
5
,
S
e
p
.
2
0
1
8
,
d
o
i:
1
0
.
1
0
1
6
/
j.
n
e
u
n
e
t.
2
0
1
8
.
0
5
.
0
0
2
.
[1
2
]
M
.
S
u
n
,
Z.
S
o
n
g
,
X.
Jia
n
g
,
J.
P
a
n
,
a
n
d
Y.
P
a
n
g
,
“
Lea
rn
i
n
g
P
o
o
li
n
g
f
o
r
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
two
rk
,
”
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
2
2
4
,
p
p
.
9
6
–
1
0
4
,
F
e
b
.
2
0
1
7
,
d
o
i:
1
0
.
1
0
1
6
/J.
NEUCO
M
.
2
0
1
6
.
1
0
.
0
4
9
.
[1
3
]
M
.
Bo
u
k
a
b
o
u
s
a
n
d
M
.
Az
izi,
“
Re
v
iew
o
f
Lea
rn
in
g
-
Ba
se
d
Tec
h
n
iq
u
e
s
o
f
S
e
n
ti
m
e
n
t
An
a
l
y
si
s
fo
r
S
e
c
u
rit
y
P
u
rp
o
se
s,”
in
In
n
o
v
a
t
io
n
s
in
S
m
a
rt
Cit
ies
Ap
p
li
c
a
ti
o
n
s
Vo
l
u
me
4
,
2
0
2
1
,
p
p
.
1
–
1
4
,
d
o
i:
d
o
i.
o
rg
/
1
0
.
1
0
0
7
/9
7
8
-
3
-
0
3
0
-
6
6
8
4
0
-
2
_
8
.
[1
4
]
I.
Id
rissi,
M
.
Az
izi,
a
n
d
O.
M
o
u
ss
a
o
u
i,
“
Io
T
se
c
u
rit
y
with
De
e
p
Lea
rn
in
g
-
b
a
se
d
In
tru
si
o
n
De
tec
ti
o
n
S
y
ste
m
s:
A
sy
ste
m
a
ti
c
li
tera
tu
re
re
v
iew
,
”
in
4
th
I
n
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
In
telli
g
e
n
t
Co
m
p
u
ti
n
g
i
n
Da
t
a
S
c
ien
c
e
s,
ICDS
2
0
2
0
,
2
0
2
0
,
p
p
.
1
–
1
0
,
d
o
i:
1
0
.
1
1
0
9
/
ICDS5
0
5
6
8
.
2
0
2
0
.
9
2
6
8
7
1
3
.
[1
5
]
M
.
Bo
u
k
a
b
o
u
s
a
n
d
M
.
Az
izi,
“
A
Co
m
p
a
ra
ti
v
e
S
t
u
d
y
o
f
DL
-
Ba
se
d
Lan
g
u
a
g
e
Re
p
re
se
n
tati
o
n
Lea
rn
in
g
M
o
d
e
ls,”
In
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
E
lec
trica
l
E
n
g
i
n
e
e
rin
g
a
n
d
C
o
mp
u
ter
S
c
ien
c
e
(IJ
EE
C
S
)
,
v
o
l.
2
2
,
n
o
.
2
,
2
0
2
1
,
d
o
i:
1
0
.
1
1
5
9
1
/
ij
e
e
c
s.v
2
2
.
i2
.
p
p
1
0
3
2
-
1
0
4
0
.
[1
6
]
S
.
J.
P
a
n
a
n
d
Q.
Ya
n
g
,
"
A
S
u
rv
e
y
o
n
Tra
n
sfe
r
Lea
rn
i
n
g
,
"
in
IEE
E
T
r
a
n
sa
c
ti
o
n
s
o
n
Kn
o
wle
d
g
e
a
n
d
Da
t
a
En
g
i
n
e
e
rin
g
,
v
o
l.
2
2
,
n
o
.
1
0
,
p
p
.
1
3
4
5
-
1
3
5
9
,
Oc
t.
2
0
1
0
,
d
o
i:
1
0
.
1
1
0
9
/T
KD
E.
2
0
0
9
.
1
9
1
.
[1
7
]
Y.
G
u
o
,
H.
S
h
i
,
A.
Ku
m
a
r,
K.
G
ra
u
m
a
n
,
T.
Ro
si
n
g
,
a
n
d
R
.
F
e
ris,
“
S
p
o
t
Tu
n
e
:
Tran
sfe
r
Lea
rn
in
g
T
h
r
o
u
g
h
Ad
a
p
ti
v
e
F
in
e
-
Tu
n
i
n
g
.
”
a
rXiv:1
8
1
1
.
0
8
7
3
7
,
p
p
.
4
8
0
5
–
4
8
1
4
,
2
0
1
9
.
[1
8
]
“
Tran
sfe
r
lea
rn
in
g
fro
m
p
re
-
train
e
d
m
o
d
e
ls
|
b
y
P
e
d
ro
M
a
rc
e
li
n
o
|
To
wa
rd
s
Da
ta
S
c
ien
c
e
.
”
2
0
1
8
.
[On
li
n
e
].
Av
a
il
a
b
le:
h
t
tp
s://
to
wa
rd
sd
a
tas
c
i
e
n
c
e
.
c
o
m
/t
ra
n
sfe
r
-
lea
rn
in
g
-
fr
o
m
-
p
re
-
train
e
d
-
m
o
d
e
ls
-
f2
3
9
3
f
1
2
4
7
5
1
(a
c
c
e
ss
e
d
S
e
p
.
0
7
,
2
0
2
0
).
[1
9
]
“
Th
e
Bo
T
-
Io
T
Da
tas
e
t.
”
[On
li
n
e
].
Av
a
il
a
b
le:
h
t
tp
s://
ww
w.
u
n
sw
.
a
d
fa
.
e
d
u
.
a
u
/u
n
sw
-
c
a
n
b
e
rra
-
c
y
b
e
r/cy
b
e
rse
c
u
rit
y
/ADFA
-
NB1
5
-
Da
tas
e
ts/b
o
t_
io
t.
p
h
p
(a
c
c
e
ss
e
d
F
e
b
.
2
2
,
2
0
2
0
).
[2
0
]
N
.
M
o
u
sta
fa
,
"
T
o
N_
Io
T
d
a
tas
e
ts
,
"
IEE
E
D
a
ta
p
o
rt
,
2
0
1
9
,
d
o
i:
1
0
.
2
1
2
2
7
/fes
z
-
d
m
9
7
.
[2
1
]
A.
S
in
g
la,
E.
Be
rti
n
o
,
a
n
d
D.
Ve
r
m
a
,
"
Ov
e
rc
o
m
in
g
th
e
Lac
k
o
f
La
b
e
led
Da
ta:
Train
in
g
I
n
tru
si
o
n
De
tec
ti
o
n
M
o
d
e
ls
Us
in
g
Tran
sfe
r
Lea
rn
i
n
g
,
"
2
0
1
9
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
S
ma
rt
Co
m
p
u
t
in
g
(S
M
AR
T
COM
P)
,
2
0
1
9
,
p
p
.
6
9
-
7
4
,
d
o
i:
1
0
.
1
1
0
9
/S
M
ART
COMP
.
2
0
1
9
.
0
0
0
3
1
.
[2
2
]
J.
Zh
a
n
g
,
F
.
Li
,
H.
W
u
,
a
n
d
F
.
Y
e
,
"
Au
t
o
n
o
m
o
u
s
M
o
d
e
l
Up
d
a
te
S
c
h
e
m
e
fo
r
De
e
p
Lea
rn
in
g
Ba
se
d
Ne
two
rk
Traffic
Clas
sifiers
,
"
2
0
1
9
IE
EE
Glo
b
a
l
C
o
mm
u
n
ica
ti
o
n
s
Co
n
fer
e
n
c
e
(GLOBE
COM
)
,
2
0
1
9
,
p
p
.
1
-
6
,
d
o
i
:
1
0
.
1
1
0
9
/G
LOBECOM
3
8
4
3
7
.
2
0
1
9
.
9
0
1
4
0
3
6
.
[2
3
]
N.
S
a
m
e
e
ra
a
n
d
M
.
S
h
a
sh
i
,
“
Tran
sfe
r
lea
rn
in
g
b
a
se
d
p
ro
t
o
t
y
p
e
fo
r
z
e
ro
-
d
a
y
a
tt
a
c
k
d
e
tec
ti
o
n
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
E
n
g
i
n
e
e
rin
g
a
n
d
A
d
v
a
n
c
e
d
T
e
c
h
n
o
lo
g
y
(IJ
EA
T
)
,
v
o
l.
8
,
n
o
.
4
,
p
p
.
1
3
2
6
–
1
3
2
9
,
A
p
r.
2
0
1
9
.
[2
4
]
N.
S
a
m
e
e
ra
a
n
d
M
.
S
h
a
sh
i,
“
De
e
p
tran
sd
u
c
ti
v
e
tran
sfe
r
lea
rn
in
g
fra
m
e
wo
rk
fo
r
z
e
ro
-
d
a
y
a
tt
a
c
k
d
e
tec
t
io
n
,
”
ICT
Exp
re
ss
,
M
a
r.
2
0
2
0
,
d
o
i:
1
0
.
1
0
1
6
/j
.
icte
.
2
0
2
0
.
0
3
.
0
0
3
.
[2
5
]
A.
S
.
Qu
re
sh
i
,
A.
Kh
a
n
,
N.
S
h
a
m
im,
a
n
d
M
.
H
.
Du
ra
d
,
“
In
tr
u
si
o
n
d
e
tec
ti
o
n
u
sin
g
d
e
e
p
sp
a
rse
a
u
to
-
e
n
c
o
d
e
r
a
n
d
se
lf
-
tau
g
h
t
lea
rn
in
g
,
”
Ne
u
ra
l
Co
mp
u
t.
Ap
p
l.
,
v
o
l.
3
2
,
n
o
.
8
,
p
p
.
3
1
3
5
-
3
1
4
7
,
Ap
r.
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/s0
0
5
2
1
-
0
1
9
-
0
4
1
5
2
-
6.
[2
6
]
“
o
p
e
n
a
rg
u
s
-
Us
i
n
g
Arg
u
s
.
”
[On
li
n
e
].
A
v
a
iala
b
le:
h
tt
p
s:/
/o
p
e
n
a
rg
u
s.o
r
g
/i
n
d
e
x
.
p
h
p
/u
si
n
g
-
a
r
g
u
s
(a
c
c
e
ss
e
d
S
e
p
.
1
1
,
2
0
2
0
).
[2
7
]
N.
Ke
tk
a
r,
“
In
tro
d
u
c
ti
o
n
t
o
Ke
ra
s,”
in
De
e
p
L
e
a
rn
i
n
g
wit
h
Pyt
h
o
n
,
Be
rk
e
ley
,
CA,
U
n
it
e
d
S
tate
s:
Ap
re
ss
,
2
0
1
7
,
p
p
.
9
7
-
1
1
1
,
d
o
i:
1
0
.
1
0
0
7
/
9
7
8
-
1
-
4
8
4
2
-
2
7
6
6
-
4
_
7
.
[2
8
]
M
.
A
b
a
d
i
,
e
t
a
l
.
,
"
Ten
so
rF
lo
w:
A
S
y
ste
m
f
o
r
Larg
e
-
S
c
a
le
M
a
c
h
in
e
Lea
rn
i
n
g
,"
Pro
c
e
e
d
in
g
s
o
f
t
h
e
1
2
th
US
ENIX
S
y
mp
o
si
u
m o
n
Op
e
ra
ti
n
g
S
y
ste
ms
De
sig
n
a
n
d
Imp
lem
e
n
t
a
ti
o
n
(O
S
DI ’1
6
)
,
2
0
1
6
,
p
p
.
2
6
5
–
2
8
3
.
[2
9
]
N.
Ja
p
k
o
wic
z
,
“
Wh
y
Qu
e
stio
n
M
a
c
h
in
e
Lea
rn
in
g
Ev
a
l
u
a
ti
o
n
M
e
th
o
d
s?
(An
il
l
u
stra
ti
v
e
re
v
iew
o
f
th
e
sh
o
rtco
m
in
g
s
o
f
c
u
rre
n
t
m
e
th
o
d
s),”
AA
AI
wo
rk
sh
o
p
o
n
e
v
a
l
u
a
ti
o
n
me
th
o
d
s f
o
r ma
c
h
in
e
le
a
rn
i
n
g
,
p
p
.
6
-
1
1
,
2
0
0
6
.
[3
0
]
Q.
Wan
g
,
Y.
M
a
,
K.
Zh
a
o
,
a
n
d
Y.
Ti
a
n
,
“
A
Co
m
p
re
h
e
n
si
v
e
S
u
rv
e
y
o
f
Lo
ss
F
u
n
c
ti
o
n
s
i
n
M
a
c
h
in
e
Lea
rn
in
g
,
”
An
n
a
ls o
f
Da
t
a
S
c
ien
c
e
,
p
p
.
1
-
2
6
,
2
0
2
0
,
d
o
i:
1
0
.
1
0
0
7
/S
4
0
7
4
5
-
0
2
0
-
0
0
2
5
3
-
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.