I
AE
S
I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
Ar
t
if
icial
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
Vol.
14
,
No.
4
,
Augus
t
2025
,
pp.
3324
~
3333
I
S
S
N:
2252
-
8938
,
DO
I
:
10
.
11591/i
jai
.
v
14
.i
4
.
pp
33
24
-
3333
3324
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
ai
.
iaes
c
or
e
.
c
om
C
on
t
r
ac
t
-
b
as
e
d
f
e
d
e
r
at
e
d
l
e
a
r
n
in
g
f
r
am
e
w
o
r
k
f
or
i
n
t
r
u
si
on
d
e
t
e
c
t
io
n
syst
e
m
i
n
i
n
t
e
r
n
e
t
o
f
t
h
i
n
gs n
e
t
w
or
k
s
Yuris
M
u
lya
S
ap
u
t
r
a
1
,
Divi
Gal
ih
P
r
as
e
t
yo
P
u
t
r
i
1
,
Jim
m
y
T
r
io
P
u
t
r
a
1
,
B
u
d
i
B
ayu
M
u
r
t
i
1
,
Wah
yon
o
2
1
D
e
pa
r
tm
e
nt
of
E
le
c
tr
ic
a
l
E
ngi
ne
e
r
in
g a
nd I
nf
or
ma
ti
c
s
, V
oc
a
ti
ona
l
C
ol
le
ge
, U
ni
ve
r
s
it
a
s
G
a
dj
a
h M
a
da
, Y
ogya
k
a
r
ta
, I
ndone
s
ia
2
D
e
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ie
nc
e
a
nd E
le
c
tr
oni
c
s
, F
a
c
ul
ty
of
M
a
th
e
ma
ti
c
s
a
nd N
a
tu
r
a
l
S
c
ie
nc
e
s
, U
ni
ve
r
s
it
a
s
G
a
dj
a
h M
a
da
,
Y
ogya
ka
r
ta
, I
ndone
s
ia
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
Nov
7,
2024
R
e
vis
e
d
J
un
12,
2025
Ac
c
e
pted
J
ul
10,
2025
A
p
l
et
h
o
r
a
o
f
n
at
i
o
n
al
v
i
t
al
i
n
fras
t
ru
ct
u
res
co
n
n
ec
t
ed
t
o
i
n
t
ern
et
o
f
t
h
i
n
g
s
(Io
T
)
n
et
w
o
r
k
s
ma
y
t
r
i
g
g
er
s
eri
o
u
s
d
a
t
a
s
ec
u
ri
t
y
v
u
l
n
er
ab
i
l
i
t
i
e
s
.
T
o
ad
d
res
s
t
h
e
i
s
s
u
e,
i
n
t
r
u
s
i
o
n
d
et
ec
t
i
o
n
s
y
s
t
ems
(ID
S)
w
ere
i
n
v
es
t
i
g
a
t
ed
w
h
ere
t
h
e
b
eh
a
v
i
o
r
an
d
t
raff
i
c
o
f
I
o
T
n
e
t
w
o
r
k
s
are
mo
n
i
t
o
re
d
t
o
d
et
ermi
n
e
w
h
et
h
er
mal
i
c
i
o
u
s
at
t
ack
s
o
r
n
o
t
o
ccu
r
t
h
r
o
u
g
h
cen
t
ral
i
zed
l
earn
i
n
g
o
n
a
cl
o
u
d
.
N
o
n
et
h
el
e
s
s
,
s
u
c
h
a
met
h
o
d
req
u
i
re
s
Io
T
d
ev
i
ces
t
o
t
ran
s
m
i
t
t
h
ei
r
l
o
cal
n
et
w
o
r
k
t
raff
i
c
d
a
t
a
t
o
t
h
e
cl
o
u
d
,
t
h
ereb
y
l
ea
d
i
n
g
t
o
d
at
a
b
reach
e
s
.
T
h
i
s
p
ap
er
p
r
o
p
o
s
e
s
a
fed
erat
e
d
l
ear
n
i
n
g
(FL
)
-
b
as
e
d
ID
S
o
n
Io
T
n
e
t
w
o
rk
s
ai
mi
n
g
at
i
m
p
ro
v
i
n
g
t
h
e
i
n
t
r
u
s
i
o
n
d
e
t
ec
t
i
o
n
acc
u
racy
w
i
t
h
o
u
t
p
ri
v
ac
y
l
ea
k
ag
e
fro
m
t
h
e
I
o
T
d
ev
i
ces
.
S
p
eci
f
i
cal
l
y
,
a
n
Io
T
s
er
v
i
ce
p
ro
v
i
d
er
c
an
fi
r
s
t
m
o
t
i
v
a
t
e
Io
T
d
ev
i
ces
t
o
p
a
rt
i
ci
p
at
e
i
n
t
h
e
FL
p
ro
ce
s
s
v
i
a
a
c
o
n
t
ract
-
b
a
s
ed
i
n
ce
n
t
i
v
e
mech
an
i
s
m
acc
o
rd
i
n
g
t
o
t
h
e
i
r
l
o
cal
d
at
a.
T
h
en
,
t
h
e
FL
p
ro
ces
s
i
s
e
x
ecu
t
ed
t
o
p
red
i
ct
Io
T
n
et
w
o
r
k
t
raff
i
c
t
y
p
es
w
i
t
h
o
u
t
s
en
d
i
n
g
Io
T
d
ev
i
ce
s
’
l
o
ca
l
d
at
a
t
o
t
h
e
cl
o
u
d
.
H
ere,
each
Io
T
d
ev
i
ce
p
erfo
rms
t
h
e
l
earn
i
n
g
p
ro
ces
s
l
o
cal
l
y
an
d
o
n
l
y
s
en
d
s
t
h
e
t
rai
n
ed
mo
d
el
t
o
t
h
e
cl
o
u
d
fo
r
t
h
e
mo
d
el
u
p
d
at
e.
T
h
e
p
ro
p
o
s
ed
FL
-
b
as
e
d
s
y
s
t
em
ach
i
ev
e
s
a
h
i
g
h
er
u
t
i
l
i
t
y
(u
p
t
o
4
4
%
)
t
h
a
n
t
h
at
o
f
a
n
o
n
-
c
o
n
t
ract
-
b
a
s
ed
i
n
cen
t
i
v
e
mech
a
n
i
s
m
an
d
a
h
i
g
h
er
p
red
i
c
t
i
o
n
acc
u
racy
(u
p
t
o
3
%
)
t
h
a
n
t
h
at
o
f
t
h
e
l
o
cal
l
ear
n
i
n
g
met
h
o
d
u
s
i
n
g
a
real
-
w
o
r
l
d
Io
T
n
et
w
o
r
k
t
raffi
c
d
at
a
s
et
.
K
e
y
w
o
r
d
s
:
C
ontr
a
c
t
theor
y
F
e
de
r
a
ted
lea
r
ning
I
nc
e
nti
ve
mec
ha
nis
m
I
nter
ne
t
of
thi
ngs
Ne
twor
k
s
e
c
ur
it
y
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
W
a
hyono
De
pa
r
tm
e
nt
of
C
omput
e
r
S
c
ienc
e
a
nd
E
lec
tr
onics
,
F
a
c
ult
y
of
M
a
thema
ti
c
s
a
nd
Na
tur
a
l
S
c
ienc
e
s
Unive
r
s
it
a
s
Ga
djah
M
a
da
Yogya
ka
r
ta,
I
ndone
s
ia
E
mail:
wa
hyo@ugm.
a
c
.
id
1.
I
NT
RODU
C
T
I
ON
T
he
gr
owing
popula
r
it
y
of
int
e
r
ne
t
of
th
ings
(
I
o
T
)
ne
twor
ks
pos
e
ne
w
c
ha
ll
e
nge
s
to
da
ta
s
e
c
ur
it
y
a
nd
pr
ivac
y.
S
pe
c
if
ica
ll
y
,
na
ti
ona
l
vit
a
l
inf
r
a
s
tr
uc
tur
e
s
li
nke
d
to
the
I
o
T
ne
twor
ks
,
s
uc
h
a
s
tr
a
ns
por
tation,
e
ne
r
gy,
a
nd
he
a
lt
hc
a
r
e
s
ys
tems
,
c
a
n
be
vulner
a
ble
to
c
ybe
r
-
a
tt
a
c
ks
that
may
lea
d
to
da
maging
c
ons
e
que
nc
e
s
f
or
na
t
ional
s
e
c
ur
it
y
a
nd
the
we
ll
-
be
ing
of
s
oc
iety.
F
or
that,
the
I
oT
ne
twor
k
s
e
c
ur
it
y
e
nha
nc
e
ment
i
s
c
r
uc
ial
f
or
na
ti
ona
l
vit
a
l
inf
r
a
s
tr
uc
tu
r
e
s
thr
ough
p
r
otec
ti
ve
mea
s
ur
e
s
s
uc
h
a
s
the
a
dopti
on
o
f
int
r
us
ion
d
e
tec
ti
on
s
ys
tems
(
I
DS)
,
da
ta
e
nc
r
ypti
on,
a
nd
r
e
gular
s
e
c
ur
it
y
upda
tes
to
r
e
duc
e
r
is
ks
a
nd
boos
t
inf
r
a
s
tr
uc
tur
e
r
e
s
il
ienc
e
[
1]
,
[
2]
.
O
ne
s
ol
ut
io
n
t
o
i
mp
r
ov
e
t
he
I
o
T
n
e
t
wo
r
k
s
e
c
u
r
i
ty
f
o
r
vi
ta
l
i
n
f
r
a
s
t
r
uc
tu
r
e
p
r
o
tec
ti
on
is
t
he
de
v
e
l
op
me
nt
o
f
a
ut
om
a
t
ic
I
DS
a
t
the
c
e
nt
r
a
l
ize
d
c
l
ou
d
s
e
r
ve
r
de
p
l
oye
d
by
I
o
T
s
e
r
v
ice
p
r
o
v
ide
r
s
(
I
S
P
s
)
[
3
]
.
He
r
e
,
th
e
us
e
o
f
m
a
c
hi
ne
lea
r
ni
ng
(
M
L
)
us
in
g
d
e
e
p
le
a
r
n
in
g
(
DL
)
a
pp
r
oa
c
he
s
f
o
r
I
D
S
im
pl
e
m
e
n
ta
ti
on
in
th
e
c
l
ou
d
s
e
r
ve
r
h
a
s
b
e
e
n
w
id
e
l
y
i
nv
e
s
t
i
ga
t
e
d
.
F
or
e
x
a
m
pl
e
,
the
wo
r
ks
in
[
4]
–
[
1
0
]
p
r
op
os
e
ne
tw
or
k
a
t
tac
k
c
las
s
i
f
i
c
a
t
i
on
ba
s
e
d
on
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
C
ontr
ac
t
-
bas
e
d
fede
r
ated
lear
ning
fr
ame
w
or
k
for
i
ntr
us
ion
de
tec
ti
on
s
y
s
tem
in
…
(
Y
ur
is
M
ulya
Saput
r
a
)
3325
d
e
e
p
n
e
u
r
a
l
ne
tw
or
ks
(
D
NN
)
ut
i
li
z
i
ng
va
r
io
us
n
e
t
wo
r
k
t
r
a
f
f
ic
i
n
f
o
r
ma
ti
on
da
tas
e
ts
wi
t
h
d
i
f
f
e
r
e
nt
c
a
pt
ur
e
t
im
e
s
.
S
p
e
c
i
f
ica
l
ly
,
th
e
a
u
t
ho
r
s
in
[
4
]
,
[
5
]
d
is
c
us
s
a
D
NN
-
b
a
s
e
d
I
D
S
t
o
c
las
s
i
f
y
ma
lwa
r
e
t
o
n
ot
i
f
y
us
e
r
s
a
bo
ut
p
o
ten
t
ia
l
a
t
tac
ks
.
Us
in
g
Ap
a
c
he
S
pa
r
k
,
M
i
gh
a
n
a
nd
Ka
ha
n
i
[
6
]
in
ve
s
ti
ga
tes
th
e
c
o
mp
a
r
is
on
o
f
I
DS
be
twe
e
n
th
e
p
r
op
os
e
d
DL
me
th
od
a
nd
s
e
ve
r
a
l
li
ne
a
r
M
L
me
th
ods
.
T
h
e
n
,
a
n
a
d
a
p
t
ive
DN
N
-
ba
s
e
d
I
D
S
f
r
a
me
wo
r
k
is
pr
op
os
e
d
in
[
7
]
,
a
i
mi
ng
a
t
c
las
s
if
y
in
g
ne
tw
or
k
a
tt
a
c
ks
.
T
o
ta
c
k
le
t
he
uns
up
e
r
v
is
e
d
f
e
a
t
u
r
e
l
e
a
r
n
in
g
,
a
n
on
-
s
ym
me
t
r
i
c
d
e
e
p
a
ut
o
-
e
nc
o
de
r
f
o
r
ne
tw
o
r
k
I
D
S
is
s
tu
di
e
d
i
n
[
8
]
,
[
9
]
.
M
o
r
e
ov
e
r
,
Z
ho
ng
e
t
a
l
.
[
10
]
i
nt
r
od
uc
e
s
a
h
i
e
r
a
r
c
h
ic
a
l
DL
l
e
v
e
r
a
g
in
g
b
e
h
a
v
io
r
a
nd
c
o
nt
e
n
t
f
e
a
tu
r
e
s
t
o
r
e
c
og
niz
e
t
he
ne
t
wo
r
k
t
r
a
f
f
ic
a
t
tac
ks
.
T
he
us
e
of
DL
a
ppr
oa
c
h
f
o
r
I
D
S
s
c
e
na
r
io
is
then
e
xtende
d
to
I
oT
ne
twor
k
-
ba
s
e
d
a
ppli
c
a
ti
ons
.
F
or
e
xa
mpl
e
,
a
DL
-
ba
s
e
d
I
DS
us
ing
f
e
e
d
-
f
or
wa
r
d
ne
ur
a
l
ne
twor
ks
(
F
NN
)
f
o
r
b
inar
y
a
nd
mul
t
i
-
c
las
s
tr
a
f
f
ic
f
low
c
las
s
if
ica
ti
on
in
pa
c
ke
t
leve
l
of
I
oT
de
vice
s
is
inves
ti
ga
ted
in
[
11]
.
Otoum
e
t
al
.
[
12
]
pr
opos
e
s
a
DL
-
ba
s
e
d
I
DS
uti
l
izing
s
tac
ke
d
-
de
e
p
polynom
ial
ne
twor
k
to
obtain
opt
im
a
l
s
e
c
ur
it
y
r
is
k
de
tec
ti
on
.
T
he
n,
a
n
I
oT
-
ba
s
e
d
I
DS
to
r
e
ve
a
l
dis
tr
ibut
e
d
de
nial
-
of
-
s
e
r
vice
botnet
a
tt
a
c
ks
us
ing
DN
N
a
pp
r
oa
c
h
is
inves
ti
ga
ted
in
[
13]
.
F
ur
ther
mor
e
,
a
n
a
nomaly
-
ba
s
e
d
I
DS
to
c
las
s
if
y
a
tt
a
c
ks
us
ing
c
onvolut
ional
ne
ur
a
l
ne
twor
ks
(
C
N
N)
-
ba
s
e
d
a
ppr
oa
c
h
f
or
I
oT
ne
twor
ks
is
s
tudi
e
d
in
[
14]
.
E
lna
kib
e
t
al.
[
15
]
e
xtends
the
pr
e
vious
wor
k
to
c
ove
r
t
he
mul
ti
-
c
las
s
c
a
tegor
iza
ti
on
us
ing
a
nomaly
-
ba
s
e
d
a
tt
a
c
k
da
tas
e
ts
f
or
the
I
oT
ne
twor
ks
.
None
thele
s
s
,
a
ll
of
thes
e
s
tudi
e
s
a
r
e
us
ing
the
c
e
ntr
a
li
z
e
d
lea
r
ning
pr
oc
e
s
s
a
t
the
c
loud
s
e
r
ve
r
.
I
n
thi
s
c
a
s
e
,
s
e
nding
ne
twor
k
tr
a
f
f
ic
da
ta
f
r
om
I
oT
de
vice
s
to
the
c
loud
s
e
r
ve
r
f
or
the
c
e
ntr
a
li
z
e
d
lea
r
ning
may
tr
igger
to
o
ther
is
s
ue
s
s
uc
h
a
s
da
ta
tr
a
f
f
ic
br
e
a
c
he
s
a
nd
p
r
ivac
y
lea
ka
ge
o
f
the
I
oT
de
vice
s
.
As
the
a
lt
e
r
na
ti
ve
s
olut
ion
,
e
a
c
h
I
oT
de
vice
c
a
n
pr
oc
e
s
s
the
ne
twor
k
tr
a
f
f
ic
da
ta
loca
ll
y,
howe
ve
r
,
t
his
I
DS
a
ppr
oa
c
h
will
not
a
c
hieve
high
int
r
us
ion
d
e
tec
ti
on
a
c
c
ur
a
c
y
due
t
o
li
mi
ted
loca
l
da
ta
a
nd
c
omput
a
ti
on
a
l
c
ons
tr
a
int
s
on
I
o
T
de
vice
s
[
16]
.
T
h
e
d
e
ve
l
op
men
t
of
e
d
ge
c
o
mp
ut
in
g
a
nd
dis
t
r
i
bu
ted
M
L
[
17
]
,
[
18
]
c
a
n
be
us
e
d
t
o
a
u
t
oma
t
ica
l
ly
de
t
e
c
t
i
n
t
r
us
io
ns
f
r
o
m
I
o
T
d
e
v
ice
s
wi
t
ho
ut
c
o
mp
r
o
m
is
i
ng
p
r
i
va
c
y
.
F
o
r
th
a
t
,
a
f
e
de
r
a
t
e
d
lea
r
ni
ng
(
F
L
)
a
p
p
r
o
a
c
h
ha
s
e
me
r
ge
d
a
s
on
e
o
f
t
he
m
os
t
po
te
nt
ia
l
s
o
lu
ti
on
s
to
a
c
hi
e
v
e
th
a
t
goa
l
.
S
pe
c
i
f
ica
l
ly
,
e
a
c
h
I
o
T
de
v
ic
e
e
x
e
c
ut
e
s
t
he
l
e
a
r
ni
ng
p
r
o
c
e
s
s
lo
c
a
ll
y
a
nd
o
nl
y
s
e
nds
t
he
t
r
a
in
e
d
m
o
d
e
l
t
o
th
e
c
lo
ud
f
o
r
th
e
m
od
e
l
u
pda
te
w
it
ho
ut
s
ha
r
in
g
the
I
o
T
d
e
v
ice
’
loc
a
l
d
a
t
a
.
T
he
us
e
o
f
F
L
ha
s
be
e
n
i
nv
e
s
t
i
ga
t
e
d
in
[
19
]
–
[
2
4
]
.
P
a
r
ti
c
u
la
r
l
y
,
t
he
wo
r
k
in
[
1
9
]
,
[
20
]
p
r
op
os
e
a
n
I
D
S
us
i
ng
F
L
w
i
th
a
t
ten
t
io
n
g
a
te
d
r
e
c
u
r
r
e
nt
un
i
t
(
t
hr
ou
gh
e
l
i
mi
na
ti
ng
ins
ig
n
if
ic
a
n
t
t
r
a
i
ne
d
m
o
de
l
to
t
he
c
lo
ud
)
a
nd
c
o
nv
e
n
t
io
na
l
DL
me
th
od
,
r
e
s
p
e
c
t
iv
e
l
y
.
T
he
n
,
a
n
F
L
-
ba
s
e
d
I
D
S
to
ta
c
k
le
c
y
be
r
a
tt
a
c
k
s
us
in
g
D
NN
,
C
NN
,
a
nd
r
e
c
u
r
r
e
nt
ne
ur
a
l
ne
tw
or
ks
(
R
N
N)
f
o
r
a
g
r
i
c
u
lt
u
r
a
l
I
o
T
e
nv
i
r
o
nm
e
n
t
is
dis
c
us
s
e
d
in
[
2
1]
.
Us
in
g
non
-
in
de
pe
nde
nt
a
n
d
i
de
nt
ic
a
l
ly
dis
t
r
ib
ute
d
s
e
c
u
r
i
t
y
a
tt
a
c
ks
da
ta
,
Alc
a
z
a
r
e
t
al
.
[
22
]
inc
o
r
p
o
r
a
te
s
F
L
v
ia
F
e
dAv
g
a
n
d
F
e
d+
a
p
pr
oa
c
he
s
f
o
r
I
DS
i
n
a
n
i
nd
us
tr
i
a
l
I
o
T
s
e
t
t
in
g
.
An
ot
he
r
pe
r
f
o
r
ma
nc
e
c
o
mp
a
r
is
on
b
e
t
we
e
n
F
e
dP
r
ox
a
nd
F
e
d
Av
g
m
e
t
ho
ds
f
o
r
d
is
t
r
ib
ut
e
d
n
e
t
wo
r
k
I
D
S
is
i
nv
e
s
t
ig
a
t
e
d
in
[
2
3
]
.
Ac
c
o
r
d
in
g
t
o
O
li
ve
i
r
a
e
t
al
.
[
2
4
]
,
a
n
F
L
-
e
n
a
b
le
d
I
D
S
w
i
th
a
s
ync
h
r
o
no
us
le
a
r
n
in
g
us
in
g
b
ina
r
y
a
n
d
m
ul
t
i
-
c
l
a
s
s
c
l
a
s
s
i
f
ica
t
io
n
is
a
ls
o
i
n
t
r
o
duc
e
d
.
N
e
v
e
r
t
he
l
e
s
s
,
t
he
s
e
s
tu
d
ies
u
t
il
ize
ou
tda
te
d
ne
tw
o
r
k
t
r
a
f
f
i
c
da
tas
e
ts
.
T
h
e
y
a
l
s
o
do
no
t
c
ons
ide
r
th
e
e
c
o
no
mi
c
a
s
pe
c
t
o
f
t
he
s
ys
tem
p
a
r
t
ic
ip
a
t
in
g
in
t
he
le
a
r
n
in
g
pr
oc
e
s
s
(
d
ue
t
o
th
e
s
e
l
f
is
h
ne
s
s
c
h
a
r
a
c
te
r
i
s
t
ic
o
f
I
o
T
d
e
v
ic
e
s
)
.
I
n
o
the
r
wo
r
ds
,
th
e
a
bo
ve
s
ys
t
e
m
w
il
l
no
t
wo
r
k
u
n
les
s
I
o
T
d
e
v
ice
s
a
r
e
m
ot
iv
a
t
e
d
t
o
jo
in
in
t
he
F
L
p
r
oc
e
s
s
e
s
.
T
h
e
r
e
f
o
r
e
,
th
e
us
e
o
f
inc
e
n
ti
ve
s
a
s
a
r
e
w
a
r
d
f
o
r
I
o
T
d
e
v
ice
pa
r
ti
c
i
pa
ti
on
is
r
e
qu
ir
e
d
.
T
o
a
ddr
e
s
s
the
a
f
or
e
mentioned
pr
oblem
,
in
thi
s
p
a
pe
r
,
a
n
int
e
gr
a
ted
F
L
-
ba
s
e
d
I
DS
f
r
a
mew
or
k
with
c
ontr
a
c
t
-
ba
s
e
d
ince
nti
ve
mec
ha
nis
m
f
or
a
n
I
oT
ne
twor
k
is
pr
opos
e
d.
T
his
a
im
s
to
pr
e
dict
ne
twor
k
tr
a
f
f
ic
types
(
i.
e
.
,
nor
mal
pa
tt
e
r
ns
or
a
tt
a
c
k
pa
tt
e
r
ns
)
with
high
a
c
c
ur
a
c
y
while
maximi
z
ing
uti
li
ty
f
o
r
the
w
hole
I
o
T
ne
twor
k
in
the
F
L
pr
oc
e
s
s
e
s
.
S
pe
c
if
ica
ll
y,
a
n
I
S
P
c
a
n
f
ir
s
t
mot
ivate
a
s
e
t
of
I
oT
de
vice
s
in
the
c
o
ns
ider
e
d
a
r
e
a
t
o
joi
n
the
F
L
pr
oc
e
s
s
e
s
.
He
r
e
,
the
I
S
P
c
a
n
p
r
ovide
ince
nti
ve
mec
ha
nis
m
f
o
r
the
I
o
T
de
vice
s
by
s
olvi
ng
a
c
ontr
a
c
t
opti
mi
z
a
ti
on
pr
oblem
that
maxim
ize
s
the
uti
li
ty
f
or
the
I
S
P
a
nd
the
I
oT
de
vice
s
.
T
his
opti
mi
z
a
ti
on
will
pr
oduc
e
a
s
e
t
of
opti
mal
c
ontr
a
c
ts
c
ont
a
ini
ng
pe
r
f
or
manc
e
a
nd
r
e
wa
r
d
f
o
r
the
I
oT
de
vice
s
.
T
he
I
S
P
then
of
f
e
r
s
the
op
ti
mal
c
ontr
a
c
ts
to
the
I
oT
de
vice
s
i
n
whic
h
they
c
a
n
r
e
c
e
ive
or
r
e
jec
t
the
of
f
e
r
e
d
c
ontr
a
c
ts
a
c
c
or
ding
to
their
de
c
is
ions
.
I
n
thi
s
wa
y,
the
I
oT
de
vice
s
that
r
e
c
e
ive
the
c
ontr
a
c
ts
c
a
n
pa
r
ti
c
ipate
i
n
the
F
L
pr
oc
e
s
s
e
s
.
F
or
the
F
L
pr
oc
e
s
s
,
e
a
c
h
pa
r
t
icipa
ti
ng
I
o
T
de
vice
c
a
n
f
i
r
s
t
e
xe
c
ute
the
tr
a
ini
ng
pr
oc
e
s
s
loca
ll
y
us
ing
it
s
loca
l
ne
twor
k
t
r
a
f
f
ic
da
ta.
T
he
n,
the
tr
a
i
ne
d
model
f
r
om
the
t
r
a
ini
ng
p
r
oc
e
s
s
c
a
n
be
s
ha
r
e
d
to
the
I
S
P
’
s
c
loud
f
o
r
the
global
ne
twor
k
tr
a
f
f
ic
model
u
pda
te
without
r
e
ve
a
li
ng
a
ny
pr
ivate
inf
or
mation
o
f
the
I
oT
de
vice
s
.
T
hr
ough
e
xpe
r
i
menta
l
r
e
s
ult
s
us
ing
a
r
e
a
l
-
wor
ld
I
o
T
ne
twor
k
tr
a
f
f
ic
da
tas
e
t,
the
pr
opos
e
d
c
ontr
a
c
t
-
ba
s
e
d
F
L
f
r
a
mew
or
k
c
a
n
obtain
a
hig
he
r
uti
l
it
y
(
up
to
44%
)
than
that
of
non
-
c
ontr
a
c
t
-
ba
s
e
d
ince
nti
ve
mec
ha
nis
m
a
nd
a
higher
pr
e
diction
a
c
c
u
r
a
c
y
(
up
to
3%
)
than
that
o
f
the
loca
l
lea
r
ning
me
thod.
I
n
the
f
oll
owing,
the
de
tails
of
c
ont
r
a
c
t
-
ba
s
e
d
ince
nti
ve
mec
ha
nis
m
a
nd
F
L
a
ppr
oa
c
h
be
twe
e
n
the
I
S
P
a
nd
I
oT
de
vi
c
e
s
a
r
e
dis
c
us
s
e
d.
T
he
n
,
e
xtens
ive
c
ompar
is
ons
in
ter
ms
of
uti
li
ty
,
va
li
da
ti
on
a
c
c
ur
a
c
y,
va
li
da
t
ion
los
s
,
a
nd
lea
r
ning
pe
r
f
o
r
manc
e
s
a
r
e
pr
e
s
e
nted.
2.
M
E
T
HO
D
L
e
t
=
{
1
,
…
,
,
…
,
}
is
the
s
e
t
of
I
o
T
de
vice
s
a
nd
a
s
s
ume
that
a
c
lo
ud
-
ba
s
e
d
I
S
P
is
c
onne
c
ted
to
N
I
oT
de
vice
s
via
W
i
-
F
i
or
c
e
ll
ular
ne
two
r
ks
in
t
he
c
ons
ider
e
d
I
oT
ne
twor
k
f
or
a
c
e
r
tain
pe
r
iod.
He
r
e
,
the
I
S
P
uti
li
z
e
s
a
huge
c
omput
ing
r
e
s
our
c
e
while
the
I
oT
de
vice
s
ha
ve
li
mi
ted
c
omput
ing
r
e
s
our
c
e
s
.
T
o
pr
e
dict
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
332
4
-
3333
3326
ne
twor
k
tr
a
f
f
ic
types
of
I
DS
with
high
a
c
c
ur
a
c
y
while
maximi
z
ing
uti
li
ty
f
o
r
the
whole
I
o
T
ne
two
r
k
in
the
F
L
pr
oc
e
s
s
e
s
,
two
a
ppr
oa
c
he
s
a
r
e
inves
ti
ga
ted.
P
a
r
ti
c
ular
ly,
a
n
ince
nti
ve
mec
ha
nis
m
ba
s
e
d
on
a
c
ontr
a
c
t
theor
y
a
ppr
oa
c
h
be
twe
e
n
the
I
S
P
a
nd
pa
r
ti
c
ipati
ng
I
oT
de
vice
s
is
f
ir
s
t
de
s
igned.
T
his
is
c
a
r
r
ied
out
by
f
or
mul
a
ti
ng
a
c
ontr
a
c
t
opti
m
iza
ti
on
pr
oblem
to
f
in
d
the
opti
mal
c
ontr
a
c
ts
c
ontaining
pe
r
f
or
manc
e
in
ter
ms
of
qua
nti
ty
a
nd
qua
li
ty
of
I
o
T
ne
twor
k
tr
a
f
f
ic
da
ta
f
r
om
e
a
c
h
I
oT
de
vice
.
T
his
opti
mal
I
o
T
ne
twor
k
t
r
a
f
f
ic
da
ta
is
then
us
e
d
a
s
the
input
da
ta
f
or
the
lea
r
ning
p
r
o
c
e
s
s
e
s
thr
ough
us
ing
the
F
L
a
ppr
oa
c
h
without
s
ha
r
ing
a
ny
s
e
ns
it
ive
da
ta
of
the
I
o
T
de
vice
s
.
T
he
whole
model
a
r
c
hit
e
c
tur
e
is
s
hown
in
F
igur
e
1.
F
igur
e
1.
T
he
model
a
r
c
hit
e
c
tu
r
e
of
the
pr
opos
e
d
I
DS
in
the
I
oT
ne
twor
k
2.
1.
Cont
r
ac
t
-
b
as
e
d
in
c
e
n
t
ive
ap
p
r
oac
h
F
igur
e
2
s
hows
the
pr
oc
e
dur
e
s
f
or
c
ontr
a
c
t
-
ba
s
e
d
ince
nti
ve
a
nd
F
L
a
ppr
oa
c
h
f
or
the
I
DS
in
the
I
o
T
ne
twor
k.
I
n
th
is
c
ontext,
c
ontr
a
c
t
-
ba
s
e
d
ince
nti
ve
mec
ha
nis
m
is
im
pleme
nted
to
mo
ti
va
te
I
o
T
de
vice
s
with
high
qua
nti
ty
a
nd
q
ua
li
ty
loca
l
ne
twor
k
tr
a
f
f
ic
da
ta
in
joi
ning
the
F
L
pr
oc
e
s
s
,
a
im
ing
a
t
p
r
oduc
ing
high
-
a
c
c
ur
a
c
y
I
DS.
T
his
ince
nti
ve
mec
ha
nis
m
is
ba
s
e
d
on
the
c
ontr
a
c
t
theor
y,
a
n
e
c
onomi
c
a
ppr
o
a
c
h
that
ba
lanc
e
s
the
uti
li
ti
e
s
of
the
I
S
P
a
nd
I
o
T
de
vice
s
in
the
F
L
p
r
oc
e
s
s
und
e
r
inf
o
r
mation
a
s
ymm
e
tr
y
[
25
]
.
T
o
thi
s
e
nd,
the
I
S
P
wor
ks
a
s
a
p
r
incipa
l
whic
h
of
f
e
r
s
th
e
c
ontr
a
c
ts
to
the
I
oT
de
vice
s
a
s
obs
e
r
ve
d
in
F
ig
ur
e
2(
a
)
.
M
e
a
nwhile,
the
pa
r
ti
c
ipating
I
o
T
de
vice
s
a
c
t
a
s
a
ge
nts
that
ha
ve
r
igh
ts
t
o
r
e
c
e
ive
o
r
r
e
jec
t
the
of
f
e
r
e
d
c
ontr
a
c
ts
.
As
the
p
r
incipa
l,
the
I
S
P
will
pr
ovide
in
c
e
nti
ve
s
to
the
I
o
T
de
vice
s
a
s
pa
r
t
o
f
the
c
ontr
a
c
ts
in
r
e
tu
r
n
f
or
their
pa
r
ti
c
ipation
in
the
F
L
pr
oc
e
s
s
.
An
I
o
T
d
e
vice
that
pa
r
ti
c
ipate
s
mor
e
in
the
F
L
pr
oc
e
s
s
will
r
e
c
e
iv
e
mor
e
ince
nti
ve
s
f
r
om
the
I
S
P
.
None
thele
s
s
,
due
to
the
inf
or
mation
a
s
ymm
e
tr
y
be
twe
e
n
the
I
S
P
a
nd
I
oT
de
vice
s
(
i.
e
.
,
the
I
S
P
doe
s
not
know
the
pr
e
f
e
r
e
nc
e
s
a
s
we
ll
a
s
ne
twor
k
tr
a
f
f
ic
da
ta
qua
li
ty
a
nd
qua
nti
ty
of
the
I
oT
de
vice
s
due
to
their
pr
ivac
y)
,
the
I
S
P
wil
l
onl
y
obtain
the
ge
ne
r
a
l
in
f
or
mation
f
r
om
the
I
o
T
de
vi
c
e
s
,
e
.
g.
,
I
oT
de
vice
s
pe
c
if
ica
ti
on
a
nd
r
e
s
our
c
e
inf
o
r
mation
[
26]
.
Af
ter
the
ge
ne
r
a
l
inf
o
r
mation
is
c
oll
e
c
ted
f
r
om
the
I
oT
de
vice
s
,
the
I
S
P
c
a
n
pe
r
f
or
m
the
F
L
c
ontr
a
c
t
opti
mi
z
a
ti
on
that
maximi
z
e
s
uti
li
ti
e
s
o
f
the
I
S
P
a
nd
I
oT
de
vice
s
.
S
pe
c
i
f
ica
ll
y,
the
I
S
P
f
ir
s
t
divi
de
s
the
I
oT
de
vice
s
int
o
N
ty
pe
s
.
T
his
type
r
e
pr
e
s
e
nts
the
will
ingnes
s
of
a
n
I
oT
de
vice
to
pa
r
t
icipa
te
in
the
F
L
pr
oc
e
s
s
c
ons
ider
ing
it
s
ne
twor
k
tr
a
f
f
ic
da
ta
qua
li
ty
a
nd
qu
a
nti
ty.
L
e
t
β
de
note
a
n
I
oT
de
vice
with
type
-
n
,
in
whic
h
β
1
<
⋯
<
β
<
⋯
<
β
,
n
∈
{
1
,
…
,
N
}
.
T
he
lar
ge
r
β
r
e
f
lec
ts
the
higher
will
ingnes
s
to
pa
r
ti
c
ipate
in
the
F
L
pr
oc
e
s
s
due
to
the
higher
ince
nti
ve
(
a
t
the
e
xpe
ns
e
of
higher
da
ta
qua
nti
ty
a
nd
qua
li
ty)
[
26]
,
[
27]
.
I
n
t
his
c
a
s
e
,
the
I
S
P
doe
s
not
ha
ve
a
ny
knowle
dge
of
the
tr
ue
type
o
f
e
a
c
h
pa
r
ti
c
ipating
I
o
T
de
vice
in
the
F
L
pr
oc
e
s
s
.
How
e
ve
r
,
the
I
S
P
knows
the
li
ke
l
ihood
that
a
n
I
o
T
de
vice
be
longs
to
a
type
-
n
f
r
om
pr
ior
a
c
ti
vit
ies
of
the
I
oT
de
vice
s
[
26]
s
uc
h
that
∑
ρ
=
1
=
1
,
whe
r
e
ρ
is
the
pr
oba
bil
it
y
of
I
oT
de
vice
with
type
-
n.
Ne
xt,
the
F
L
c
ontr
a
c
t
opti
mi
z
a
ti
on
p
r
oblem
c
a
n
b
e
f
or
mul
a
ted
with
the
a
im
to
maximi
z
e
the
uti
li
ty
of
the
I
S
P
in
the
F
L
p
r
oc
e
s
s
,
in
a
ddit
ion
to
the
uti
l
it
y
of
I
oT
de
vice
s
.
F
ir
s
t,
the
uti
li
ty
o
f
the
I
S
P
that
e
mpl
oys
a
n
I
oT
de
vice
with
type
-
n
c
a
n
be
e
xpr
e
s
s
e
d
a
s
t
he
c
ombi
na
ti
on
be
twe
e
n
the
be
ne
f
it
a
nd
c
os
t
f
unc
ti
ons
in
e
xe
c
uti
ng
the
F
L
pr
oc
e
s
s
a
s
(
1)
:
μ
=
α
−
σ
(
1)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
C
ontr
ac
t
-
bas
e
d
fede
r
ated
lear
ning
fr
ame
w
or
k
for
i
ntr
us
ion
de
tec
ti
on
s
y
s
tem
in
…
(
Y
ur
is
M
ulya
Saput
r
a
)
3327
W
he
r
e
α
indi
c
a
tes
the
be
ne
f
it
f
unc
ti
on
f
or
the
I
S
P
with
α
>
0
is
a
c
onve
r
s
ion
va
r
iable
that
im
pli
e
s
the
moneta
r
y
unit
of
the
I
o
T
ne
twor
k
t
r
a
f
f
ic
da
ta
qua
n
ti
ty
a
nd
qua
li
ty
[
28]
.
M
e
a
nwhile,
is
the
ince
nti
ve
f
or
the
pa
r
ti
c
ipating
I
oT
de
vice
s
a
nd
σ
r
e
pr
e
s
e
nts
unit
c
os
t
of
the
ince
nti
ve
.
S
ince
ther
e
e
xis
ts
N
types
of
the
pa
r
ti
c
ipating
I
oT
de
vice
s
with
pr
oba
bil
it
y
ρ
,
∀
∈
,
th
e
n
the
e
xpe
c
ted
uti
li
ty
of
the
I
S
P
c
a
n
be
f
or
mul
a
ted
in
(
2)
.
μ
=
∑
μ
ρ
=
1
(
2)
S
e
c
ond,
the
uti
l
it
y
of
a
n
I
oT
de
vice
wi
th
type
-
n
that
a
ls
o
c
ontains
the
be
ne
f
it
a
nd
c
os
ts
f
unc
ti
on
s
c
a
n
be
de
f
ined
a
s
(
3)
.
μ
=
β
γ
(
)
−
η
(
3)
W
h
e
r
e
γ
(
Y
n
)
=
√
i
s
a
s
t
r
i
c
t
l
y
i
n
c
r
e
a
s
i
n
g
c
o
n
c
a
v
e
b
e
n
e
f
i
t
f
u
n
c
t
i
o
n
w
i
t
h
γ
(
0
)
=
0
,
γ
′
(
)
<
0
,
γ
′′
(
)
<
0
,
∀
[
2
7
]
.
Additi
ona
ll
y,
η
c
or
r
e
s
ponds
to
the
c
omput
a
ti
on
a
nd
memo
r
y
c
os
ts
f
o
r
the
I
oT
de
vice
with
ty
pe
-
n
in
tr
a
ini
ng
it
s
loca
l
ne
twor
k
t
r
a
f
f
ic
da
ta
in
the
F
L
pr
o
c
e
s
s
.
T
o
obtain
the
c
ontr
a
c
t
f
e
a
s
ibi
li
ty
,
e
a
c
h
of
f
e
r
e
d
c
o
ntr
a
c
t
pa
c
ka
ge
,
i.
e
.
,
(
,
)
,
∀
∈
,
mus
t
mee
t
indi
vidual
r
a
ti
ona
li
ty
(
I
R
)
a
nd
ince
nti
ve
c
ompatibi
li
ty
(
I
C
)
c
ons
tr
a
int
s
[
25]
,
[
27
]
.
T
he
I
R
c
ons
tr
a
int
s
gua
r
a
ntee
that
a
n
I
o
T
de
vice
with
type
-
n
will
o
btain
the
uti
li
ty
that
is
g
r
e
a
ter
than
or
e
qua
l
to
z
e
r
o
a
s
de
s
c
r
ibed
a
s
(
4)
.
μ
=
β
γ
(
)
−
η
≥
0
,
∀
∈
(
4)
M
e
a
nwhile,
the
I
C
c
ons
tr
a
int
s
e
ns
ur
e
that
a
ll
I
oT
de
vice
s
only
a
c
c
e
pt
c
ontr
a
c
t
pa
c
ka
ge
s
de
s
igned
f
or
their
r
e
s
pe
c
ti
ve
types
unde
r
the
pr
e
s
e
nc
e
of
in
f
or
mation
a
s
ymm
e
tr
y,
a
s
given
in
(
5)
.
β
n
γ
(
Y
n
)
−
η
X
n
≥
β
n
γ
(
Y
m
)
−
η
X
m
,
m
≠
n
,
∀
m
,
n
∈
(
5)
T
o
thi
s
e
nd,
the
F
L
c
ontr
a
c
t
opti
mi
z
a
ti
on
pr
oblem
that
maximi
z
e
s
the
e
xpe
c
ted
u
ti
li
ty
of
the
I
S
P
und
e
r
the
I
R
a
nd
I
C
c
ons
tr
a
int
s
of
the
I
o
T
de
vice
s
c
a
n
be
f
or
mu
late
d
by
(
6
)
.
m
a
x
(
X
,
Y
)
∑
μ
n
I
S
P
ρ
n
N
n
=
1
(
6)
S
ubjec
t
to
the
I
R
,
I
C
,
a
nd
monot
onicity
c
ons
tr
a
int
s
a
s
s
hown
in
(
7
)
to
(
9)
.
β
γ
(
)
−
η
≥
0
,
∀
∈
(
7)
β
n
γ
(
Y
n
)
−
η
X
n
≥
β
n
γ
(
Y
m
)
−
η
X
m
,
m
≠
n
,
∀
m
,
n
∈
(
8)
β
1
<
⋯
<
β
<
⋯
<
β
,
n
∈
{
1
,
…
,
N
}
(
9)
W
he
r
e
=
[
1
,
…
,
,
…
,
]
a
nd
=
[
1
,
…
,
,
…
,
]
.
Us
ing
the
s
a
me
method
a
s
in
[
26]
–
[
28]
,
the
opti
mal
c
ontr
a
c
ts
(
∗
,
∗
)
c
a
n
be
f
ound
thr
ough
s
im
pli
f
yi
ng
the
I
R
a
nd
I
C
c
ons
tr
a
int
s
s
uc
h
that
the
pr
oblem
be
c
omes
(
10)
.
m
a
x
(
X
,
Y
)
∑
μ
n
I
S
P
ρ
n
N
n
=
1
(
10)
S
ubjec
t
to
the
monot
onicity
c
ondit
ion
in
(
9)
,
a
nd
t
he
s
im
pli
f
ied
I
R
a
nd
I
C
c
ons
tr
a
int
s
a
s
given
in
(
11
)
to
(
12)
.
β
1
γ
(
1
)
−
η
1
=
0
(
11)
(
)
−
=
(
−
1
)
−
−
1
,
∀
∈
(
12)
2.
2.
F
e
d
e
r
at
e
d
lear
n
in
g
ap
p
r
oac
h
Upon
ob
tai
ning
the
o
pti
mal
c
ont
r
a
c
t
pa
c
ka
ge
s
(
∗
,
∗
)
f
o
r
a
ll
pa
r
t
icipa
ti
n
g
I
o
T
de
vice
s
,
t
he
lea
r
ni
ng
pr
oc
e
s
s
us
i
ng
F
L
be
twe
e
n
the
I
S
P
a
nd
t
he
pa
r
t
ic
ipati
ng
I
o
T
de
v
ice
s
in
that
a
c
c
e
p
t
t
he
o
f
f
e
r
e
d
op
ti
mal
c
ont
r
a
c
ts
c
a
n
be
e
xe
c
u
ted
a
nd
i
ll
us
t
r
a
te
d
in
F
i
gu
r
e
2(
b
)
.
S
pe
c
i
f
ica
ll
y
,
f
or
e
a
c
h
le
a
r
n
ing
r
oun
d,
I
o
T
d
e
v
ice
s
f
i
r
s
t
tr
a
i
n
t
he
i
r
in
div
idua
l
ne
two
r
k
tr
a
f
f
ic
da
ta
loca
l
ly
a
n
d
then
on
ly
s
e
n
d
th
e
t
r
a
in
e
d
I
DS
mod
e
ls
t
o
t
he
I
S
P
’
s
c
lo
ud
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
332
4
-
3333
3328
withi
n
a
p
r
e
-
de
f
i
ne
d
l
im
i
ted
pe
r
i
od
,
e
ns
u
r
ing
the
da
ta
p
r
ivac
y
of
the
I
o
T
de
vic
e
s
.
T
o
ob
tain
the
g
l
oba
l
I
DS
model
,
th
e
I
S
P
’
s
c
loud
c
a
n
a
ggr
e
ga
te
a
ll
the
r
e
c
e
i
ve
d
tr
a
ined
I
DS
models
a
n
d
us
e
thi
s
a
g
gr
e
ga
ted
I
DS
mode
l
to
upda
te
the
c
u
r
r
e
nt
globa
l
I
D
S
mo
de
l
.
He
r
e
,
the
c
ur
r
e
nt
g
loba
l
I
D
S
mode
l
is
us
e
d
f
or
the
ne
xt
lea
r
ning
F
L
it
e
r
a
ti
o
n
p
r
oc
e
s
s
by
the
c
lou
d
a
n
d
t
he
I
o
T
d
e
vice
s
.
T
h
is
p
r
oc
e
s
s
r
e
pe
a
ts
un
ti
l
t
he
g
loba
l
I
DS
mo
de
l
c
onve
r
ge
s
or
the
lea
r
ni
ng
dur
a
ti
on
r
e
a
c
he
s
the
s
pe
c
if
i
e
d
de
a
dl
ine
ti
me
.
He
nc
e
,
us
ing
a
s
uc
h
F
L
a
ppr
oa
c
h
,
the
a
c
c
ur
a
c
y
o
f
I
DS
i
n
the
I
o
T
ne
tw
or
k
c
a
n
be
i
mpr
ove
d
w
hil
e
pr
e
s
e
r
vin
g
pr
ivate
in
f
or
mat
ion
a
n
d
r
e
d
uc
ing
c
omm
u
nica
ti
on
ove
r
he
a
d
(
s
i
nc
e
I
o
T
ne
t
wor
k
tr
a
f
f
ic
da
ta
is
t
ypica
ll
y
m
uc
h
la
r
ge
r
than
t
he
tr
a
ini
ng
mo
de
l
)
in
the
F
L
p
r
oc
e
s
s
.
T
o
im
pleme
nt
the
DL
pr
oc
e
s
s
in
the
F
L
pr
oc
e
s
s
,
a
DN
N
a
ppr
oa
c
h
[
29
]
is
e
mpl
oye
d
.
P
a
r
ti
c
ula
r
ly,
input
da
ta
c
ontaining
tabula
r
da
ta
with
many
s
a
mpl
e
s
a
nd
tr
a
ini
ng
f
e
a
tur
e
s
(
s
uc
h
a
s
pa
c
ke
t
type,
s
e
r
vice
,
pr
otocol,
a
nd
other
r
e
leva
nt
ne
twor
k
tr
a
f
f
ic
f
e
a
tur
e
s
)
a
long
with
tr
a
ini
ng
labe
ls
,
i
.
e
.
,
ne
twor
k
tr
a
f
f
ic
p
a
tt
e
r
n,
is
f
ir
s
t
c
oll
e
c
ted
f
r
om
the
r
e
a
l
ne
twor
k
tr
a
f
f
ic
a
c
ti
v
it
y
on
e
a
c
h
I
o
T
de
vice
.
T
o
r
e
duc
e
the
c
ompl
e
xit
y
of
the
lea
r
ning
pr
oc
e
s
s
,
the
f
e
a
tur
e
s
e
lec
ti
on
pr
oc
e
s
s
us
ing
the
c
or
r
e
lation
be
twe
e
n
f
e
a
tur
e
s
a
nd
labe
l
is
then
e
xe
c
uted.
I
n
thi
s
c
a
s
e
,
the
f
e
a
tur
e
s
with
the
c
o
r
r
e
l
a
ti
on
va
lue
les
s
than
0.
1
c
a
n
be
dr
oppe
d
f
r
om
t
he
tr
a
ini
ng
pr
oc
e
s
s
.
Upon
s
e
lec
ti
ng
the
r
e
leva
nt
f
e
a
tur
e
s
,
the
input
da
ta
is
f
e
d
int
o
the
DN
N
on
e
a
c
h
I
oT
de
vice
.
He
r
e
,
the
DN
N
model
include
s
a
n
input
laye
r
,
s
e
ve
r
a
l
hi
dde
n
laye
r
s
with
a
c
ti
va
ti
on
f
unc
ti
ons
,
s
ome
dr
opo
ut
laye
r
s
,
a
nd
the
output
laye
r
with
a
n
output
a
c
ti
va
ti
on
f
unc
ti
on
f
or
the
ne
twor
k
tr
a
f
f
ic
pa
tt
e
r
n
c
las
s
if
ica
ti
on.
Onc
e
the
DN
N
model
is
c
r
e
a
ted
on
e
a
c
h
I
oT
de
vice
,
the
I
oT
de
vice
c
a
n
pe
r
f
or
m
the
lea
r
ning
p
r
oc
e
s
s
loca
ll
y
to
ge
ne
r
a
te
a
tr
a
ined
model
γ
,
whe
r
e
n
is
the
index
of
I
oT
de
vice
a
nd
t
is
the
it
e
r
a
ti
on
o
f
F
L
pr
oc
e
s
s
.
T
he
a
ggr
e
ga
ti
on
of
t
r
a
ined
models
then
lea
ds
to
the
glo
ba
l
I
DS
model
that
c
a
n
be
e
xpr
e
s
s
e
d
a
s
(
13)
.
=
1
∑
γ
=
1
(
13)
Us
ing
the
global
model
,
e
a
c
h
I
o
T
de
vice
c
a
n
pe
r
f
or
m
the
ne
xt
it
e
r
a
ti
on
’
s
tr
a
ini
ng
pr
oc
e
s
s
to
obtain
+
1
,
+
2
,
…
,
∗
.
T
he
f
inal
∗
,
whic
h
is
the
f
inal
global
I
DS
model
,
i
s
then
us
e
d
to
va
li
da
te
the
a
c
c
ur
a
c
y
of
I
DS
us
ing
ne
w
ne
twor
k
tr
a
f
f
ic
da
ta
ge
ne
r
a
ted
by
t
he
I
oT
de
vice
s
f
or
othe
r
pe
r
iods
.
(
a
)
(
b)
F
igur
e
2.
T
he
pr
oc
e
dur
e
s
f
o
r
(
a
)
c
ontr
a
c
t
-
ba
s
e
d
ince
nti
ve
a
nd
(
b)
F
L
a
ppr
oa
c
h
f
or
the
I
DS
in
the
I
o
T
ne
twor
k
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
o
e
va
luate
the
s
upe
r
ior
i
ty
o
f
the
p
r
opos
e
d
c
ontr
a
c
t
-
ba
s
e
d
F
L
f
r
a
mew
or
k
,
a
r
e
a
l
-
ti
me
I
oT
ne
twor
k
tr
a
f
f
ic
da
tas
e
t
f
r
om
UC
I
M
a
c
hine
L
e
a
r
ning
R
e
pos
it
or
y
[
30]
that
c
ontains
83
f
e
a
tur
e
s
a
nd
100K
s
a
mp
les
with
nor
mal
a
nd
a
tt
a
c
k
ne
twor
k
a
c
ti
vi
ti
e
s
.
T
he
s
e
s
a
mpl
e
s
a
r
e
divi
de
d
int
o
s
ubs
a
mpl
e
s
a
c
c
or
ding
to
the
nu
mber
of
pa
r
ti
c
ipating
I
oT
de
vice
s
.
F
or
the
ince
nti
ve
mec
ha
nis
m,
the
pr
opos
e
d
c
ontr
a
c
t
-
ba
s
e
d
F
L
s
y
s
tem
is
c
ompar
e
d
with
the
inf
o
r
mation
s
ymm
e
tr
y
(
i.
e
.
,
the
I
S
P
c
omp
lete
ly
knows
the
tr
ue
type
o
f
I
o
T
de
vice
s
)
a
nd
the
ba
s
e
li
ne
method
(
i.
e
.
,
the
I
S
P
pr
ov
ides
the
pr
opor
ti
ona
l
i
nc
e
nti
ve
f
or
the
pa
r
ti
c
ipating
I
oT
de
vice
s
)
.
I
n
t
his
c
a
s
e
,
10
pa
r
ti
c
ipating
I
oT
de
vice
s
a
r
e
c
ons
ider
e
d
to
r
e
c
e
ive
the
opti
mal
c
ont
r
a
c
ts
that
c
or
r
e
s
ponds
to
10
types
of
I
oT
de
vice
s
.
Ne
xt,
the
F
L
pr
oc
e
s
s
is
then
im
pleme
nted
us
ing
the
DN
N
model
with
T
e
ns
or
F
low
NV
I
DI
A
T
4
T
e
ns
or
C
or
e
GPU.
P
a
r
t
icula
r
ly,
thr
e
e
hidden
laye
r
s
with
r
e
c
ti
f
ied
li
ne
a
r
unit
(
R
e
L
U
)
a
c
ti
va
ti
on
f
unc
ti
on,
two
dr
opout
laye
r
s
,
a
nd
a
n
output
laye
r
with
S
of
tM
a
x
a
c
ti
va
ti
on
f
unc
ti
on
a
r
e
e
mpl
oye
d.
T
o
f
ur
the
r
s
how
the
F
L
pe
r
f
or
manc
e
,
the
p
r
opos
e
d
f
r
a
mew
or
k
is
c
om
pa
r
e
d
with
the
c
e
ntr
a
li
z
e
d
lea
r
ning
(
i.
e
.
,
DN
N
gl
oba
l)
a
nd
the
loca
l
lea
r
ning
(
i
.
e
.
,
DN
N
loca
l)
.
Additi
ona
ll
y,
2
-
labe
l
a
nd
12
-
labe
l
s
c
e
na
r
ios
a
r
e
us
e
d.
S
pe
c
if
ica
ll
y,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
C
ontr
ac
t
-
bas
e
d
fede
r
ated
lear
ning
fr
ame
w
or
k
for
i
ntr
us
ion
de
tec
ti
on
s
y
s
tem
in
…
(
Y
ur
is
M
ulya
Saput
r
a
)
3329
2
-
labe
l
s
c
e
n
a
r
io
include
s
the
2
types
of
nor
mal
a
nd
a
tt
a
c
k
pa
tt
e
r
ns
.
M
e
a
nwhile,
12
-
labe
l
s
c
e
na
r
io
c
ont
a
ins
the
r
e
a
l
ne
twor
k
tr
a
f
f
ic
pa
tt
e
r
ns
s
uc
h
a
s
DO
S
_S
YN
_Hping,
AR
P
_pois
ioni
ng,
NM
AP_UDP
_S
C
AN
,
NM
AP_XM
AS_T
R
E
E
_S
C
AN
,
NM
AP_OS
_DE
T
E
C
T
I
ON
,
NM
AP_T
C
P
_s
c
a
n,
DD
OS_S
l
owlor
is
,
M
e
tas
ploi
t_B
r
ute_For
c
e
_S
S
H,
NM
AP_F
I
N_SC
AN
,
M
QT
T
,
T
h
ing_s
pe
a
k,
W
ipr
o_bulb_Da
tas
e
t,
Ama
z
on
-
Ale
xa
.
Additi
ona
ll
y,
a
di
f
f
e
r
e
nt
number
o
f
pa
r
t
icipa
ti
ng
I
oT
de
vice
s
is
a
ls
o
c
ons
ider
e
d.
3.
1.
Ut
il
it
y
p
e
r
f
or
m
an
c
e
P
r
ior
to
e
va
luating
the
F
L
pe
r
f
or
manc
e
,
the
uti
li
t
y
pe
r
f
or
manc
e
s
of
the
I
S
P
a
nd
pa
r
ti
c
ipating
I
oT
de
vice
s
ba
s
e
d
on
the
c
ontr
a
c
t
theor
y
a
r
e
f
ir
s
t
de
mons
tr
a
ted
a
s
s
hown
in
F
igur
e
3
.
P
a
r
ti
c
ular
ly
,
a
s
s
hown
in
F
igur
e
3(
a
)
,
the
I
S
P
a
lwa
ys
obtains
pos
it
ive
uti
li
t
y
f
or
I
o
T
de
vice
with
type
1
to
10.
T
his
pr
ove
s
th
a
t
the
I
R
c
ons
tr
a
int
s
a
r
e
s
a
ti
s
f
ie
d
f
or
a
ll
types
of
the
I
oT
de
vice
s
.
Additi
ona
ll
y,
the
uti
li
ty
of
the
I
S
P
f
ol
lows
a
n
incr
e
a
s
ing
f
unc
ti
on
r
e
ga
r
ding
the
types
of
I
oT
de
v
ice
s
.
T
his
is
be
c
a
us
e
the
I
oT
de
vice
with
a
higher
type
ha
s
mor
e
will
ingnes
s
to
joi
n
the
F
L
pr
oc
e
s
s
,
ther
e
by
lea
ding
to
hi
ghe
r
uti
li
ty
of
the
I
S
P
in
te
r
ms
of
the
gl
oba
l
I
DS
model
a
c
c
ur
a
c
y.
M
or
e
ove
r
,
the
I
S
P
’
s
no
r
malize
d
uti
li
ty
of
pr
opos
e
d
c
ontr
a
c
t
-
ba
s
e
d
s
ys
tem
is
be
twe
e
n
the
inf
or
mation
-
s
ymm
e
tr
y
a
nd
ba
s
e
li
ne
mec
ha
nis
ms
,
i.
e
.
,
a
t
0.
28
whe
n
I
oT
de
vice
ha
s
type
10.
I
n
thi
s
c
a
s
e
,
the
inf
or
mation
-
s
ymm
e
tr
y
mec
ha
nis
m
a
c
ts
a
s
the
uppe
r
-
bound
s
olut
ion
s
ince
the
I
S
P
c
ompl
e
tely
knows
t
he
types
of
a
ll
I
oT
de
vice
s
.
As
a
r
e
s
ult
,
the
I
S
P
c
a
n
maximi
z
e
it
s
uti
li
ty
a
t
the
e
xpe
ns
e
of
z
e
r
o
uti
li
ti
e
s
f
or
a
ll
the
pa
r
ti
c
ipating
I
oT
de
vice
s
,
a
s
il
lu
s
tr
a
ted
in
F
igur
e
3(
b)
.
Ne
xt
,
it
c
a
n
be
obs
e
r
ve
d
in
F
igur
e
s
3(
a
)
a
nd
3(
b)
that
the
pr
opos
e
d
c
ontr
a
c
t
-
ba
s
e
d
s
ys
tem
c
a
n
a
c
hieve
uti
li
ty
of
the
I
S
P
a
nd
uti
l
it
y
of
I
oT
de
vice
s
up
t
o
44
a
nd
572%
highe
r
than
thos
e
of
the
ba
s
e
li
ne
mec
ha
nis
m,
r
e
s
pe
c
ti
ve
ly.
T
his
is
due
to
the
non
-
c
ontr
a
c
t
m
e
c
ha
nis
m
in
whic
h
the
pa
r
ti
c
ipating
I
o
T
de
vice
s
will
r
e
c
e
ive
li
ne
a
r
/pr
opo
r
ti
ona
l
ince
nti
ve
s
f
or
their
c
ontr
ibut
io
ns
in
the
F
L
pr
oc
e
s
s
.
F
r
om
F
igur
e
3,
i
t
c
a
n
be
s
umm
a
r
ize
d
that
the
tot
a
l
uti
li
ty
of
the
I
S
P
a
nd
I
oT
de
vice
s
f
or
the
p
r
opos
e
d
f
r
a
mew
or
k
is
c
los
e
to
that
of
the
I
S
P
a
n
d
I
oT
de
vice
s
f
or
the
inf
o
r
mation
-
s
ymm
e
tr
y
s
c
he
me
a
s
the
uppe
r
bound
s
olut
ion.
T
his
indi
c
a
tes
that
the
pr
opos
e
d
c
ontr
a
c
t
-
ba
s
e
d
f
r
a
mew
or
k
is
s
uit
a
ble
f
or
the
F
L
pr
oc
e
s
s
thr
ough
ba
lanc
ing
the
uti
li
ty
pe
r
f
or
manc
e
of
the
I
S
P
a
nd
pa
r
ti
c
ipating
I
oT
de
vice
s
e
f
f
e
c
ti
ve
ly
[
31]
.
(
a
)
(
b)
F
igur
e
3.
Nor
malize
d
u
ti
li
ty
pe
r
f
or
manc
e
f
or
(
a
)
th
e
I
S
P
a
nd
(
b
)
pa
r
ti
c
ipating
I
o
T
de
vice
s
3.
2.
L
e
ar
n
in
g
p
e
r
f
or
m
an
c
e
Ac
c
or
ding
to
the
opti
mal
c
ontr
a
c
ts
that
maximi
z
e
uti
li
ty
o
f
the
I
S
P
a
nd
I
oT
de
vice
s
in
s
e
c
ti
on
3
.
1,
the
a
c
c
ur
a
c
y
a
nd
los
s
pe
r
f
or
manc
e
c
ompar
is
ons
c
a
n
then
be
e
va
luate
d
whe
n
10
I
oT
de
vice
s
pa
r
ti
c
ipate
in
the
F
L
pr
oc
e
s
s
.
S
pe
c
if
ica
ll
y
,
whe
n
va
r
ious
number
s
o
f
labe
ls
a
r
e
us
e
d
a
s
s
hown
in
T
a
ble
1
,
the
2
-
labe
l
s
c
e
na
r
io
outper
f
or
ms
a
ll
the
pe
r
f
o
r
manc
e
s
of
the
12
-
labe
l
s
c
e
na
r
io.
He
r
e
,
both
tr
a
i
ning
a
nd
va
li
da
ti
on
a
c
c
ur
a
c
y
of
2
-
labe
l
s
c
e
n
a
r
io
a
c
hieve
s
mor
e
than
2%
be
tt
e
r
th
a
n
thos
e
of
12
-
labe
l
s
c
e
na
r
io.
T
his
is
be
c
a
us
e
the
12
-
labe
l
s
c
e
na
r
io
may
s
uf
f
e
r
f
r
om
mi
s
c
las
s
if
ica
ti
on
due
to
many
c
las
s
e
s
.
T
his
r
e
s
ult
a
ls
o
a
li
gns
with
the
los
s
pe
r
f
or
manc
e
w
he
r
e
the
tr
a
ini
ng
a
nd
va
li
da
ti
on
los
s
e
s
of
the
2
-
labe
l
s
c
e
na
r
io
r
e
a
c
h
7
ti
mes
a
nd
3
ti
m
e
s
be
tt
e
r
than
thos
e
of
the
12
-
labe
l
s
c
e
na
r
io,
r
e
s
pe
c
ti
ve
ly.
W
he
n
dif
f
e
r
e
nt
a
ppr
oa
c
he
s
a
r
e
us
e
d
a
s
obs
e
r
ve
d
in
T
a
ble
2
,
the
a
c
c
ur
a
c
y
pe
r
f
o
r
manc
e
of
the
pr
opos
e
d
F
L
f
r
a
me
wor
k
,
i.
e
.
,
DN
N
F
L
,
a
r
e
be
twe
e
n
the
DN
N
g
lobal
a
nd
DN
N
loca
l.
I
n
pa
r
ti
c
ular
,
t
he
DN
N
global
c
a
n
a
c
hieve
the
a
c
c
ur
a
c
y
that
is
s
li
ghtl
y
higher
than
that
of
the
DN
N
F
L
by
1%
.
T
he
r
e
a
s
on
is
that
the
DN
N
global
a
c
ts
a
s
the
uppe
r
bound
whe
r
e
a
ll
t
he
ne
twor
k
tr
a
f
f
ic
da
ta
is
tr
a
ined
a
t
the
c
loud
of
the
I
S
P
.
None
thele
s
s
,
thi
s
method
may
lea
d
to
the
pr
ivac
y
lea
ka
ge
of
the
I
o
T
de
vice
s
whe
n
the
c
loud
c
oll
e
c
ts
their
loca
l
ne
twor
k
tr
a
f
f
ic
da
ta.
I
n
c
ontr
a
s
t,
the
DN
N
F
L
c
a
n
outper
f
o
r
m
the
a
c
c
ur
a
c
y
pe
r
f
or
manc
e
o
f
the
loca
l
lea
r
ning
or
DN
N
loca
l
by
a
ppr
oxim
a
tely
3%
.
T
h
is
is
due
to
the
ins
uf
f
icie
nt
ne
twor
k
tr
a
f
f
ic
da
ta
whic
h
is
tr
a
ined
a
t
the
I
oT
de
vice
loca
ll
y
withou
t
a
ny
c
oll
a
b
or
a
ti
on
with
the
other
I
oT
de
vice
s
.
T
o
f
ur
the
r
s
how
a
mor
e
pr
a
c
ti
c
a
l
s
c
e
na
r
io,
di
f
f
e
r
e
n
t
number
of
pa
r
ti
c
ipating
I
o
T
de
vice
s
is
e
xe
c
uted.
I
n
thi
s
c
a
s
e
,
the
number
of
I
oT
de
vice
s
va
r
ies
f
r
o
m
5
to
20
de
vice
s
.
As
s
hown
in
T
a
ble
3,
the
pr
opos
e
d
F
L
c
a
n
s
li
ghtl
y
p
r
oduc
e
a
higher
va
li
da
ti
on
a
c
c
ur
a
c
y
a
nd
a
lowe
r
los
s
whe
n
10,
15
,
a
nd
20
numbe
r
o
f
I
oT
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
332
4
-
3333
3330
de
vice
s
a
r
e
de
ployed.
T
his
im
pli
e
s
that
mor
e
pa
r
ti
c
ipating
I
oT
de
vice
s
with
mor
e
ne
twor
k
t
r
a
f
f
ic
da
ta
a
nd
s
uf
f
icie
nt
da
ta
qua
li
ty
c
a
n
im
pr
ove
the
a
c
c
ur
a
c
y
a
n
d
los
s
pe
r
f
or
manc
e
s
.
T
a
ble
1.
T
he
F
L
pe
r
f
or
manc
e
s
f
or
di
f
f
e
r
e
nt
numbe
r
of
labe
ls
with
10
I
oT
de
vice
s
N
umbe
r
of
l
a
be
ls
T
r
a
in
in
g a
c
c
ur
a
c
y (
%
)
V
a
li
da
ti
on a
c
c
ur
a
c
y (
%
)
T
r
a
in
in
g l
os
s
V
a
li
da
ti
on l
os
s
2
98.6
98.4
0.043
0.039
12
97.13
97.5
0.091
0.125
T
a
ble
2.
T
he
F
L
pe
r
f
or
manc
e
s
f
or
p
r
opos
e
d
a
nd
ot
he
r
a
ppr
oa
c
he
s
with
2
labe
ls
a
nd
10
I
o
T
de
vice
s
M
e
th
od
T
r
a
in
in
g a
c
c
ur
a
c
y (
%
)
V
a
li
da
ti
on a
c
c
ur
a
c
y (
%
)
T
r
a
in
in
g l
os
s
V
a
li
da
ti
on l
os
s
D
N
N
gl
oba
l
99.61
99.64
0.011
0.016
D
N
N
F
L
98.6
98.4
0.043
0.039
D
N
N
l
oc
a
l
98.4
95.4
0.067
0.372
T
a
ble
3.
T
he
F
L
pe
r
f
or
manc
e
s
f
or
va
r
ious
number
of
I
o
T
de
vice
s
with
2
labe
ls
N
umbe
r
of
I
oT
de
vi
c
e
s
T
r
a
in
in
g a
c
c
ur
a
c
y (
%
)
V
a
li
da
ti
on a
c
c
ur
a
c
y (
%
)
T
r
a
in
in
g l
os
s
V
a
li
da
ti
on l
os
s
5
98.5
98.37
0.043
0.038
10
98.6
98.4
0.043
0.039
15
98.6
98.4
0.042
0.038
20
98.43
98.42
0.045
0.043
T
o
s
how
the
pe
r
f
o
r
manc
e
s
of
the
a
bove
s
c
e
na
r
ios
in
mor
e
de
tail,
the
va
li
da
ti
on
a
c
c
ur
a
c
y
a
nd
los
s
f
o
r
50
lea
r
ning
r
ounds
a
r
e
s
tudi
e
d.
T
his
c
a
n
be
obs
e
r
ve
d
c
lea
r
ly
in
F
igur
e
s
4
to
6
that
a
lt
hough
the
a
c
c
ur
a
c
y
a
nd
los
s
ga
ps
a
r
e
high
a
t
the
be
ginni
ng
of
the
F
L
pr
oc
e
s
s
,
the
dif
f
e
r
e
nc
e
ge
ts
lo
we
r
whe
n
mor
e
lea
r
nin
g
r
ounds
a
r
e
c
onduc
ted.
P
a
r
ti
c
ular
ly
,
both
2
-
labe
l
a
nd
12
-
labe
l
s
c
e
n
a
r
ios
in
F
igur
e
4
a
nd
va
r
ious
numbe
r
of
I
o
T
de
vice
s
s
c
e
na
r
ios
in
F
igur
e
6
c
a
n
a
c
hieve
the
a
c
c
ur
a
c
y
c
onve
r
ge
nc
e
a
f
ter
30
a
nd
25
lea
r
ning
r
ounds
,
r
e
s
pe
c
ti
ve
ly.
Additi
ona
ll
y,
ther
e
e
xis
ts
a
pe
r
f
or
manc
e
a
nomaly
f
or
the
DN
N
loca
l
in
F
igur
e
5
whe
r
e
it
s
va
li
da
ti
on
los
s
s
uf
f
e
r
s
f
r
om
ove
r
f
it
ti
ng,
i.
e
.
,
the
va
li
da
ti
on
los
s
ke
e
ps
incr
e
a
s
ing.
T
his
is
be
c
a
us
e
the
tr
a
ini
ng
pr
oc
e
s
s
ge
ne
r
a
tes
a
s
im
ple
tr
a
ined
model
(
with
li
mi
ted
loca
l
ne
twor
k
t
r
a
f
f
ic
da
ta
due
to
inher
e
nt
r
e
s
tr
icte
d
s
tor
a
ge
a
nd
r
e
s
our
c
e
s
a
t
the
I
o
T
de
vice
)
,
ther
e
by
le
a
ding
to
the
ove
r
f
it
ti
ng
f
o
r
the
tes
ti
ng/validation
pr
oc
e
s
s
.
F
igur
e
4.
T
he
va
li
da
ti
on
a
c
c
ur
a
c
y
a
nd
los
s
pe
r
f
or
m
a
nc
e
s
f
or
va
r
ious
number
of
labe
ls
F
igur
e
5.
T
he
va
li
da
ti
on
a
c
c
ur
a
c
y
a
nd
los
s
pe
r
f
or
m
a
nc
e
s
f
or
va
r
ious
lea
r
ning
methods
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
C
ontr
ac
t
-
bas
e
d
fede
r
ated
lear
ning
fr
ame
w
or
k
for
i
ntr
us
ion
de
tec
ti
on
s
y
s
tem
in
…
(
Y
ur
is
M
ulya
Saput
r
a
)
3331
F
igur
e
6.
T
he
va
li
da
ti
on
a
c
c
ur
a
c
y
a
nd
los
s
pe
r
f
or
m
a
nc
e
s
f
or
dif
f
e
r
e
nt
number
s
o
f
I
oT
de
vice
s
4.
CONC
L
USI
ON
T
his
pa
pe
r
pr
e
s
e
nts
a
c
ontr
a
c
t
-
ba
s
e
d
F
L
f
r
a
mew
or
k
f
o
r
I
DS
in
the
I
o
T
ne
twor
k
to
im
p
r
ove
the
int
r
us
ion
de
tec
ti
on
a
c
c
ur
a
c
y
without
p
r
ivac
y
lea
ka
ge
f
r
om
the
I
o
T
de
vice
s
.
P
a
r
ti
c
ular
ly
,
the
I
S
P
c
a
n
f
ir
s
t
dr
ive
the
I
oT
de
vice
s
in
the
c
ons
ider
e
d
ne
twor
k
to
joi
n
the
F
L
pr
oc
e
s
s
thr
ough
e
mpl
oying
the
c
ont
r
a
c
t
-
ba
s
e
d
ince
nti
ve
mec
ha
nis
m
ba
s
e
d
on
their
loca
l
da
ta
qua
li
ty
a
nd
qua
nti
ty.
I
n
th
is
c
a
s
e
,
the
I
S
P
is
r
e
quir
e
d
to
a
ddr
e
s
s
the
c
ontr
a
c
t
op
ti
mi
z
a
ti
on
pr
oblem
a
im
ing
a
t
obtai
ning
the
opti
mal
c
ontr
a
c
ts
.
Us
ing
the
opti
mal
c
ontr
a
c
ts
f
or
joi
ning
I
o
T
de
vice
s
,
th
e
F
L
pr
oc
e
s
s
is
then
im
p
leme
nted
to
p
r
e
dict
I
o
T
ne
twor
k
tr
a
f
f
ic
types
.
He
r
e
,
e
a
c
h
I
o
T
de
vice
c
onduc
ts
the
lea
r
ning
p
r
oc
e
s
s
loca
ll
y
a
nd
only
s
e
nds
the
t
r
a
ined
model
to
the
c
loud
f
or
t
he
model
upda
te.
T
hr
ough
e
xpe
r
im
e
ntal
r
e
s
ult
s
us
ing
the
r
e
a
l
-
wor
ld
I
oT
ne
t
wor
k
t
r
a
f
f
ic
da
tas
e
t,
the
pr
opos
e
d
F
L
-
ba
s
e
d
s
ys
tem
c
a
n
pr
oduc
e
44%
h
igher
uti
li
ty
than
that
o
f
the
ba
s
e
li
ne
method
a
nd
p
r
e
diction
a
c
c
ur
a
c
y
by
98.
4%
,
a
3%
higher
than
that
of
the
loca
l
lea
r
ning
method.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
T
he
a
uthor
s
thank
the
P
r
og
r
a
m
P
e
n
ingkata
n
Ac
a
de
mi
c
E
xc
e
ll
e
nc
e
S
ke
ma
B
Unive
r
s
it
a
s
Ga
djah
M
a
da
2024
that
pr
ovides
s
pons
or
a
nd
f
inanc
ial
s
uppor
t
unde
r
gr
a
nt
numbe
r
6529/UN1
.
P
1/P
T
.
01.
03
/2
024.
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ogn
ize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Yur
is
M
ulya
S
a
putr
a
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
D
i
v
i
G
a
l
i
h
P
r
a
s
e
ty
o
P
u
t
r
i
✓
✓
✓
✓
✓
✓
J
im
my
T
r
io
P
utr
a
✓
✓
✓
✓
✓
B
udi
B
a
yu
M
ur
ti
✓
✓
✓
✓
W
a
hyono
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
Author
s
s
tate
no
c
onf
li
c
t
of
int
e
r
e
s
t.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
T
he
da
ta
that
s
uppor
t
the
f
ind
ings
of
thi
s
s
tudy
a
r
e
ope
nly
a
va
il
a
ble
in
[
UC
I
M
a
c
hine
L
e
a
r
ning
R
e
pos
it
or
y
]
a
t
htt
p:/
/doi
.
or
g/10
.
24432/C
5P
338,
r
e
f
e
r
e
nc
e
[
30]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
Ar
ti
f
I
ntell
,
Vol.
14
,
No.
4
,
Augus
t
2025
:
332
4
-
3333
3332
RE
F
E
RE
NC
E
S
[
1
]
M
.
A
bo
uba
ka
r
,
M
.
K
e
ll
i
l,
a
nd
P
.
R
ou
x,
“
A
r
e
vi
e
w
o
f
I
oT
ne
t
w
or
k
ma
n
a
ge
me
nt
:
c
u
r
r
e
n
t
s
ta
tu
s
a
n
d
p
e
r
s
pe
c
t
iv
e
s
,”
J
o
ur
n
al
o
f
K
in
g
S
au
d
U
ni
v
e
r
s
i
ty
-
C
om
pu
te
r
a
nd
I
nf
or
m
at
io
n
S
c
i
e
nc
e
s
,
vo
l.
34
,
n
o.
7
,
p
p.
41
63
–
4
17
6
,
20
22
,
d
o
i:
1
0
.1
01
6/
j.
jk
s
uc
i.
20
21
.0
3.
00
6.
[
2
]
H
.
H
a
d
da
d
P
a
jo
uh
,
A
.
D
e
hg
ha
nt
a
n
ha
,
R
.
M
.
P
a
r
iz
i,
M
.
A
le
dha
r
i,
a
nd
H
.
K
a
r
i
m
ip
ou
r
,
“
A
s
u
r
ve
y
o
n
i
n
te
r
ne
t
of
th
i
ngs
s
e
c
u
r
i
t
y:
r
e
qu
i
r
e
m
e
n
ts
,
c
ha
l
le
ng
e
s
,
a
n
d s
o
lu
ti
ons
,”
I
n
te
r
n
e
t
o
f
T
hi
ngs
, v
o
l.
1
4,
J
u
n.
2
02
1,
d
oi
:
10
.1
01
6
/j
.i
ot
.2
01
9.
10
01
29
.
[
3
]
I
.
H
.
S
a
r
ke
r
,
M
.
M
.
H
oq
ue
,
M
.
K
.
U
dd
in
,
a
n
d
T
.
A
ls
a
n
oo
s
y,
“
M
ob
il
e
d
a
t
a
s
c
ie
nc
e
a
n
d
in
te
ll
ig
e
n
t
a
pps
:
c
o
nc
e
pt
s
,
A
I
-
ba
s
e
d
m
o
de
l
i
ng
a
nd
r
e
s
e
a
r
c
h
d
i
r
e
c
t
io
ns
,”
M
ob
i
le
N
e
tw
or
k
s
an
d
A
pp
li
c
a
ti
ons
,
v
o
l.
26
,
n
o.
1,
p
p.
28
5
–
30
3,
20
21
,
d
o
i:
1
0
.1
00
7/
s
11
03
6
-
0
20
-
0
16
50
-
z.
[
4
]
R
.
V
in
a
ya
ku
ma
r
,
M
.
A
la
z
a
b,
K
.
P
.
S
o
ma
n,
P
.
P
o
o
r
na
c
ha
nd
r
a
n
,
A
.
A
l
-
N
e
m
r
a
t
,
a
nd
S
.
V
e
nk
a
t
r
a
ma
n,
“
D
e
e
p
le
a
r
ni
ng
a
p
p
r
oa
c
h
f
o
r
in
te
ll
i
ge
n
t
in
t
r
us
io
n
d
e
te
c
t
io
n
s
ys
te
m
,”
I
E
E
E
A
c
c
e
s
s
,
vo
l
.
7
,
p
p.
41
52
5
–
41
55
0,
20
19
,
d
o
i:
1
0
.1
10
9/
A
C
C
E
S
S
.2
01
9.
28
95
33
4.
[
5
]
S
.
G
u
r
u
ng
,
M
.
K
.
G
h
os
e
,
a
nd
A
.
S
u
be
d
i,
“
D
e
e
p
le
a
r
n
in
g
a
pp
r
oa
c
h
o
n
ne
tw
or
k
i
nt
r
us
i
on
de
t
e
c
ti
on
s
y
s
te
m
us
in
g
N
S
L
-
K
D
D
d
a
t
a
s
e
t,
”
I
n
te
r
na
ti
on
a
l
J
our
na
l
o
f
C
o
m
p
ut
e
r
N
e
tw
or
k
an
d
I
n
fo
r
m
a
ti
on
S
e
c
ur
i
ty
,
vo
l.
1
1,
n
o.
3,
p
p.
8
–
1
4,
2
01
9
,
d
o
i:
1
0.
58
15
/i
jc
ni
s
.
20
19
.0
3.
02
.
[
6
]
S
.
N
.
M
i
gha
n
a
nd
M
.
K
a
ha
ni
,
“
A
n
ov
e
l
s
c
a
la
b
le
in
t
r
us
io
n
de
t
e
c
ti
on
s
ys
te
m
ba
s
e
d
on
de
e
p
le
a
r
ni
ng
,”
I
n
te
r
na
t
io
na
l
J
our
nal
o
f
I
n
fo
r
m
a
t
io
n
Se
c
ur
i
ty
,
vo
l.
2
0,
n
o.
3
,
pp
.
38
7
–
40
3,
2
02
1,
d
oi
:
1
0
.1
00
7/
s
10
20
7
-
0
20
-
0
05
08
-
5.
[
7
]
L
.
A
s
hi
ku
a
nd
C
.
D
a
g
li
,
“
N
e
tw
or
k
i
nt
r
us
i
on
de
te
c
t
io
n
s
ys
te
m
us
i
ng
d
e
e
p
l
e
a
r
n
in
g,
”
P
r
oc
e
di
a
C
om
pu
te
r
Sc
ie
nc
e
,
vo
l.
18
5
,
p
p
.
23
9
–
24
7,
2
02
1,
d
oi
:
10
.1
01
6/
j.
p
r
o
c
s
.
20
21
.0
5.
02
5.
[
8
]
E
.
U
.
Q
a
z
i,
M
.
I
m
r
a
n
,
N
.
H
a
id
e
r
,
M
.
S
hoa
ib
,
a
nd
I
.
R
a
z
z
a
k,
“
A
n
in
te
ll
ig
e
n
t
a
n
d
e
f
f
i
c
ie
nt
ne
tw
or
k
i
n
tr
us
io
n
de
te
c
t
io
n
s
ys
te
m
u
s
i
ng
de
e
p
l
e
a
r
n
in
g,
”
C
om
pu
te
r
s
a
nd
E
le
c
tr
ic
al
E
n
gi
ne
e
r
i
ng
, v
o
l.
9
9,
2
02
2,
d
oi
:
10
.1
01
6/
j
.c
o
mp
e
l
e
c
e
ng
.2
02
2.
10
77
64
.
[
9
]
N
.
S
ho
ne
,
T
.
N
.
N
goc
,
V
.
D
.
P
ha
i
,
a
nd
Q
.
S
h
i,
“
A
de
e
p
le
a
r
n
in
g
a
p
pr
oa
c
h
to
n
e
t
w
o
r
k
i
n
tr
us
io
n
d
e
t
e
c
t
i
o
n
,”
I
E
E
E
T
r
ans
ac
t
io
ns
o
n
E
m
e
r
gi
ng
T
o
pi
c
s
in
C
o
m
p
ut
at
io
na
l
I
n
te
l
l
ig
e
nc
e
,
vo
l.
2
,
no.
1,
p
p.
4
1
–
50
,
20
18
,
do
i:
1
0.
11
09
/T
E
T
C
I
.
20
17
.2
77
27
92
.
[
1
0
]
W
.
Z
h
on
g,
N
.
Y
u
,
a
nd
C
.
A
i
,
“
A
p
pl
yi
ng
b
ig
da
ta
ba
s
e
d
d
e
e
p
l
e
a
r
n
in
g
s
y
s
te
m
to
i
nt
r
u
s
i
on
de
te
c
ti
on
,”
B
ig
D
a
t
a
M
i
ni
ng
a
n
d
A
n
a
ly
t
ic
s
,
vo
l.
3
,
no
.
3,
p
p.
1
81
–
1
95
,
20
20
,
do
i
:
10
.2
65
99
/B
D
M
A
.2
02
0.
90
20
00
3.
[
1
1
]
M
.
G
e
,
X
.
F
u
,
N
.
S
ye
d,
Z
.
B
a
ig
,
G
.
T
e
o
,
a
nd
A
.
R
ob
le
s
-
K
e
l
ly
,
“
D
e
e
p
le
a
r
n
in
g
-
ba
s
e
d
i
nt
r
us
i
on
d
e
te
c
t
io
n
f
or
I
oT
ne
tw
o
r
ks
,
”
in
2
0
19
I
E
E
E
24
th
P
ac
i
fi
c
R
i
m
I
n
te
r
na
ti
on
a
l
S
y
m
p
os
iu
m
o
n
D
e
p
e
n
da
bl
e
C
om
p
u
ti
ng
(
P
R
D
C
)
,
20
19
,
pp
.
2
56
–
2
560
9
,
d
o
i:
1
0.
11
09
/
P
R
D
C
47
00
2.
20
19
.0
00
56
.
[
1
2
]
Y
.
O
t
ou
m,
D
.
L
i
u,
a
n
d
A
.
N
a
y
a
k
,
“
D
L
-
I
D
S
:
a
de
e
p
le
a
r
n
i
ng
–
ba
s
e
d
i
nt
r
us
i
on
de
te
c
t
io
n
f
r
a
m
e
w
or
k
f
o
r
s
e
c
ur
i
ng
I
oT
,
”
T
r
a
ns
a
c
t
io
ns
on
E
m
e
r
g
in
g
T
e
le
c
om
m
u
ni
c
a
ti
ons
T
e
c
hn
ol
og
ie
s
,
vo
l.
3
3,
n
o.
3
,
20
22
,
do
i:
1
0.
10
02
/e
tt
.3
80
3.
[
1
3
]
P
.
J
it
hu
, J
.
S
ha
r
e
e
na
,
A
.
R
a
mda
s
,
a
n
d
A
.
P
.
H
a
r
ip
r
iy
a
,
“
I
n
t
r
us
i
o
n
de
te
c
ti
on
s
ys
te
m
f
o
r
I
oT
B
o
tn
e
t
a
t
ta
c
ks
us
in
g
de
e
p
l
e
a
r
n
in
g
,”
S
N
C
om
pu
te
r
Sc
ie
nc
e
,
vo
l.
2
,
no
.
3,
2
02
1,
d
o
i:
1
0.
10
07
/s
4
29
79
-
0
21
-
0
05
16
-
9.
[
1
4
]
T
.
S
a
ba
,
A
.
R
e
h
ma
n,
T
.
S
a
da
d
,
H
.
K
o
li
va
nd
,
a
nd
S
.
A
.
B
a
h
a
j
,
“
A
n
oma
l
y
-
ba
s
e
d
in
t
r
us
io
n
de
te
c
ti
on
s
ys
te
m
f
or
I
oT
ne
two
r
ks
t
h
r
o
ug
h
d
e
e
p
le
a
r
ni
ng
mo
de
l,
”
C
o
m
pu
te
r
s
an
d
E
le
c
tr
i
c
a
l
E
n
gi
ne
e
r
in
g
,
vo
l.
99
,
2
02
2,
d
o
i:
1
0
.1
01
6/
j.
c
o
mpe
le
c
e
n
g.
20
22
.1
07
81
0.
[
1
5
]
O
.
E
l
na
ki
b,
E
.
S
ha
a
ba
n,
M
.
M
a
hm
ou
d,
a
n
d
K
.
E
m
a
r
a
,
“
E
I
D
M
:
de
e
p
l
e
a
r
n
in
g
m
od
e
l
f
or
I
oT
i
nt
r
us
i
on
de
te
c
t
io
n
s
ys
te
m
s
,”
J
our
na
l
o
f
Su
pe
r
c
om
pu
ti
ng
,
vo
l.
7
9,
n
o.
1
2,
p
p.
1
32
41
–
1
32
61,
20
23
,
do
i:
1
0.
10
07
/s
1
12
27
-
02
3
-
0
51
97
-
0.
[
16]
Y
.
S
u
n,
M
.
P
e
ng
,
Y
.
Z
ho
u,
Y
.
H
ua
ng
,
a
nd
S
.
M
a
o,
“
A
pp
l
ic
a
ti
on
o
f
ma
c
hi
ne
le
a
r
ni
ng
i
n
w
i
r
e
le
s
s
ne
two
r
ks
:
k
e
y
t
e
c
hn
iq
ue
s
a
n
d
o
pe
n
is
s
u
e
s
,”
I
E
E
E
C
om
m
u
ni
c
a
ti
ons
Sur
v
e
y
s
a
nd
T
u
to
r
ia
ls
,
vo
l.
2
1,
no
.
4,
p
p.
30
2
–
31
08
,
201
9
,
d
o
i:
1
0.
11
09
/C
O
M
S
T
.2
01
9.
29
24
24
3.
[1
7]
C
.
Z
h
a
n
g,
P
.
P
a
t
r
a
s
,
a
n
d
H
.
H
a
d
da
d
i
,
“
D
e
e
p
l
e
a
r
n
in
g
in
mob
i
le
a
nd
w
i
r
e
l
e
s
s
ne
t
w
o
r
ki
ng
:
a
s
u
r
ve
y
,”
I
E
E
E
C
om
m
u
ni
c
a
ti
o
ns
S
ur
v
e
y
s
an
d
T
ut
or
i
a
ls
,
v
ol
.
21
,
no
.
3,
p
p.
2
22
4
–
22
87
,
20
19
,
do
i
:
10
.1
10
9/
C
O
M
S
T
.
20
19
.2
90
48
97
.
[
1
8
]
W
.
Y
.
B
.
L
i
m
e
t
a
l.
,
“
F
e
de
r
a
te
d
le
a
r
n
i
ng
i
n
mo
bi
le
e
dge
n
e
tw
o
r
ks
:
a
c
om
p
r
e
he
ns
iv
e
s
u
r
ve
y,
”
I
E
E
E
C
om
m
u
ni
c
a
ti
ons
S
ur
v
e
y
s
a
n
d
T
u
to
r
i
al
s
,
vo
l.
2
2,
n
o.
3
,
pp
.
20
31
–
2
06
3,
2
02
0,
d
oi
:
1
0.
110
9
/C
O
M
S
T
.
20
20
.2
98
60
24
.
[
1
9
]
Z
.
C
he
n,
N
.
L
v,
P
.
L
i
u,
Y
.
F
a
ng
,
K
.
C
h
e
n
,
a
n
d
W
.
P
a
n
,
“
I
nt
r
us
io
n
de
te
c
t
io
n
f
o
r
w
i
r
e
l
e
s
s
e
dg
e
ne
tw
o
r
ks
ba
s
e
d
on
f
e
de
r
a
te
d
l
e
a
r
n
i
ng
,”
I
E
E
E
A
c
c
e
s
s
,
vo
l.
8
,
pp
.
21
74
63
–
2
17
47
2,
2
02
0,
d
oi
:
1
0.
11
09
/AC
C
E
S
S
.2
02
0.
30
41
79
3.
[
2
0
]
Z
.
T
a
ng
,
H
.
H
u,
a
n
d
C
.
X
u,
“
A
f
e
de
r
a
t
e
d
l
e
a
r
n
in
g
me
th
od
f
o
r
n
e
t
w
o
r
k
in
t
r
us
io
n
de
te
c
ti
on
,”
C
o
nc
ur
r
e
nc
y
a
nd
C
o
m
p
ut
a
ti
o
n:
P
r
ac
t
ic
e
a
nd
E
x
pe
r
ie
n
c
e
,
v
ol
.
3
4,
n
o.
1
0,
2
02
2,
d
oi
:
10
.1
00
2/
c
pe
.6
81
2.
[
2
1
]
O
.
F
r
i
ha
,
M
.
A
.
F
e
r
r
a
g
,
L
.
S
hu
,
L
.
M
a
gl
a
r
a
s
,
K
.
K
.
R
.
C
h
oo
,
a
n
d
M
.
N
a
f
a
a
,
“
F
E
L
I
D
S
:
F
e
de
r
a
t
e
d
le
a
r
n
in
g
-
ba
s
e
d
in
t
r
us
i
o
n
d
e
t
e
c
t
i
on
s
ys
te
m
f
o
r
a
g
r
i
c
u
lt
u
r
a
l
i
n
te
r
ne
t
o
f
t
h
in
gs
,”
J
o
ur
na
l
of
P
ar
al
le
l
an
d
D
is
tr
i
bu
te
d
C
om
p
u
ti
ng
,
v
o
l.
16
5,
p
p.
17
–
3
1,
2
0
22
,
do
i:
1
0.
10
16
/j
.
jp
dc
.
20
22
.0
3.
00
3.
[
2
2
]
P
.
R
.
-
A
lc
a
z
a
r
e
t
al
.
,
“
I
nt
r
u
s
i
on
de
te
c
t
io
n
ba
s
e
d
on
p
r
iv
a
c
y
-
p
r
e
s
e
r
vi
ng
f
e
d
e
r
a
t
e
d
l
e
a
r
n
in
g
f
o
r
t
he
i
nd
us
t
r
ia
l
I
oT
,”
I
E
E
E
T
r
a
ns
a
c
t
io
ns
on
I
n
dus
tr
ia
l
I
nf
or
m
a
t
ic
s
,
vo
l.
1
9,
n
o.
2
,
pp
.
114
5
–
1
15
4,
2
02
3,
d
oi
:
10
.1
10
9/
T
I
I
.
20
21
.3
12
67
28
.
[
2
3
]
M
.
J
.
I
d
r
is
s
i
e
t
a
l.
,
“
F
e
d
-
A
N
I
D
S
:
f
e
de
r
a
te
d
le
a
r
ni
ng
f
o
r
a
nom
a
l
y
-
b
a
s
e
d
ne
tw
o
r
k
i
nt
r
us
i
on
de
te
c
t
io
n
s
ys
t
e
ms
,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
it
h
A
pp
l
ic
a
t
io
ns
,
v
ol
.
23
4,
2
02
3,
d
oi
:
10
.1
01
6
/j
.e
s
w
a
.
20
23
.12
1
00
0.
[
2
4
]
J
.
A
.
D.
O
li
ve
i
r
a
e
t
a
l
.
,
“
F
-
N
I
D
S
—
A
ne
tw
o
r
k
i
nt
r
u
s
i
on
de
te
c
t
io
n
s
y
s
te
m
ba
s
e
d
o
n
f
e
de
r
a
te
d
le
a
r
ni
ng
,”
C
om
pu
te
r
N
e
tw
or
k
s
,
v
o
l.
2
36
,
20
23
,
do
i:
1
0.
10
16
/
j.
c
om
ne
t.
20
23
.1
10
01
0.
[
2
5
]
P
.
B
o
l
to
n a
nd
M
.
D
e
w
a
t
r
i
po
nt
,
C
o
nt
r
ac
t
th
e
or
y
.
C
a
m
b
r
i
dg
e
,
M
a
s
s
a
c
h
us
e
t
ts
:
M
I
T
P
r
e
s
s
,
2
00
5.
[
2
6
]
J
.
K
a
n
g,
Z
.
X
io
ng
,
D
.
N
i
ya
to
,
S
.
X
ie
,
a
nd
J
.
Z
h
a
n
g,
“
I
nc
e
n
ti
ve
m
e
c
h
a
n
is
m
f
o
r
r
e
l
ia
bl
e
f
e
d
e
r
a
t
e
d
le
a
r
n
in
g:
a
j
oi
nt
o
pt
i
mi
z
a
ti
o
n
a
pp
r
oa
c
h
t
o
c
om
bi
ni
ng
r
e
p
ut
a
t
io
n
a
nd
c
on
t
r
a
c
t
th
e
o
r
y
,”
I
E
E
E
I
n
te
r
ne
t
o
f
T
h
in
gs
J
our
na
l
,
vo
l
.
6
,
n
o.
6,
pp
.
1
07
00
–
1
07
1
4,
2
0
19
,
do
i:
1
0.
11
09
/J
I
O
T
.2
01
9.
29
40
82
0.
[
2
7
]
Y
.
Z
h
a
n
g,
L
.
S
on
g,
W
.
S
a
a
d
,
Z
.
D
a
w
y,
a
nd
Z
.
H
a
n
,
“
C
on
t
r
a
c
t
-
b
a
s
e
d
i
nc
e
nt
iv
e
m
e
c
h
a
n
is
ms
f
o
r
de
v
ic
e
-
to
-
d
e
v
ic
e
c
om
mu
n
ic
a
ti
ons
i
n
c
e
ll
ul
a
r
ne
tw
o
r
ks
,”
I
E
E
E
J
o
ur
n
a
l
o
n
Se
le
c
te
d
A
r
e
as
i
n
C
om
m
u
n
ic
at
io
ns
,
v
ol
.
3
3,
no
.
1
0,
pp
.
2
14
4
–
21
5
5,
2
0
15
,
do
i:
1
0.
11
09
/J
S
A
C
.2
01
5.
24
35
35
6.
[
2
8
]
Y
.
M
.
S
a
p
ut
r
a
,
D
.
T
.
H
oa
ng
,
D
.
N
.
N
g
uye
n,
L
.
N
.
T
r
a
n
,
S
.
G
o
ng
,
a
n
d
E
.
D
u
t
ki
e
w
i
c
z
,
“
D
y
na
mi
c
f
e
de
r
a
te
d
l
e
a
r
n
in
g
-
b
a
s
e
d
e
c
o
no
mi
c
f
r
a
me
w
o
r
k
f
o
r
in
te
r
ne
t
-
of
-
ve
h
ic
le
s
,”
I
E
E
E
T
r
a
ns
a
c
t
i
ons
on
M
o
bi
le
C
om
p
u
ti
ng
,
vo
l
.
22
,
n
o.
4
,
p
p.
2
10
0
–
21
15
,
2
02
3
,
d
o
i:
1
0.
11
09
/T
M
C
.2
02
1.
31
22
43
6.
[
2
9
]
W
.
L
iu
,
Z
.
W
a
ng
,
X
.
L
i
u,
N
.
Z
e
n
g,
Y
.
L
iu
,
a
nd
F
.
E
.
A
ls
a
a
di
,
“
A
s
u
r
ve
y
o
f
d
e
e
p
n
e
u
r
a
l
ne
two
r
k
a
r
c
h
it
e
c
t
u
r
e
s
a
n
d
th
e
i
r
a
pp
li
c
a
ti
ons
,”
N
e
u
r
oc
om
pu
ti
ng
,
vo
l
.
23
4,
p
p.
1
1
–
26
,
20
17
.
[
3
0
]
B
.
S
h
a
r
m
i
a
nd
R
.
N
a
g
a
pa
dm
a
,
“
R
T
-
I
oT
20
22
,”
U
C
I
M
ac
h
in
e
L
e
ar
n
in
g
R
e
pos
i
to
r
y
,
20
23
,
do
i
:
10
.2
44
32
/C
5P
33
8.
[
3
1
]
Y
.
M
.
S
a
pu
tr
a
,
D
.
N
.
N
g
uy
e
n
,
D
.
T
.
H
oa
ng
,
Q
.
V
.
P
h
a
m
,
E
.
D
ut
ki
e
w
ic
z
,
a
n
d
W
.
J
.
H
w
a
ng
,
“
F
e
de
r
a
t
e
d
le
a
r
n
i
ng
f
r
a
m
e
w
o
r
k
w
i
th
s
t
r
a
gg
li
ng
m
i
ti
ga
ti
on
a
n
d
pr
i
va
c
y
-
a
w
a
r
e
ne
s
s
f
o
r
A
I
-
ba
s
e
d
mo
bi
le
a
p
pl
ic
a
t
io
n
s
e
r
vi
c
e
s
,
”
I
E
E
E
T
r
a
ns
a
c
t
io
ns
on
M
ob
i
le
C
om
pu
ti
ng
,
vo
l.
2
2,
n
o.
9
,
pp
.
52
96
–
5
31
2,
2
02
3,
d
o
i:
1
0.
11
09/
T
M
C
.
20
22
.3
17
89
49
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
Ar
ti
f
I
ntell
I
S
S
N:
2252
-
8938
C
ontr
ac
t
-
bas
e
d
fede
r
ated
lear
ning
fr
ame
w
or
k
for
i
ntr
us
ion
de
tec
ti
on
s
y
s
tem
in
…
(
Y
ur
is
M
ulya
Saput
r
a
)
3333
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
Y
uri
s
M
ul
y
a
Sa
putr
a
i
s
cu
rren
t
l
y
an
A
s
s
i
s
t
an
t
Pro
f
es
s
o
r
as
w
el
l
as
t
h
e
D
ep
u
t
y
H
ead
at
D
ep
ar
t
men
t
o
f
E
l
ec
t
ri
ca
l
E
n
g
i
n
eeri
n
g
a
n
d
In
fo
rma
t
i
c
s
,
V
o
cat
i
o
n
al
Co
l
l
e
g
e,
U
n
i
v
er
s
i
t
as
G
ad
j
ah
Mad
a,
In
d
o
n
e
s
i
a
an
d
an
A
d
j
u
n
c
t
Fel
l
o
w
at
Sch
o
o
l
o
f
E
l
ec
t
ri
ca
l
an
d
D
at
a
E
n
g
i
n
eeri
n
g
,
U
n
i
v
ers
i
t
y
o
f
T
ec
h
n
o
l
o
g
y
Sy
d
n
e
y
.
H
e
o
b
t
a
i
n
e
d
Ph
.
D
.
i
n
E
l
ec
t
ri
ca
l
an
d
D
a
t
a
E
n
g
i
n
eeri
n
g
fro
m
t
h
e
U
n
i
v
ers
i
t
y
o
f
T
ech
n
o
l
o
g
y
Sy
d
n
ey
(A
u
s
t
ra
l
i
a)
i
n
2
0
2
2
.
H
i
s
res
earc
h
i
n
t
eres
t
s
i
n
c
l
u
d
e
m
o
b
i
l
e
c
o
mp
u
t
i
n
g
,
en
er
g
y
an
d
eco
n
o
mi
c
effi
c
i
en
c
y
,
mach
i
n
e
l
ear
n
i
n
g
,
I
o
T
,
an
d
o
p
t
i
m
i
za
t
i
o
n
p
ro
b
l
em
s
fo
r
w
i
re
l
es
s
co
mmu
n
i
ca
t
i
o
n
n
et
w
o
r
k
s
.
H
e
i
s
affi
l
i
at
e
d
w
i
t
h
IE
E
E
as
a
s
e
n
i
o
r
memb
er
.
H
e
i
s
cu
rre
n
t
l
y
an
ac
t
i
v
e
rev
i
ew
er
f
o
r
v
ar
i
o
u
s
Q
1
j
o
u
rn
a
l
s
i
n
c
l
u
d
i
n
g
I
E
E
E
T
MC,
IE
E
E
W
CM,
IE
E
E
Io
T
J
o
u
rn
a
l
,
IE
E
E
T
N
N
L
S,
IE
E
E
T
BD
,
IE
E
E
T
CCN
,
an
d
IE
E
E
T
CE
.
H
e
can
b
e
co
n
t
act
e
d
at
ema
i
l
:
y
m.
s
a
p
u
t
ra@
u
g
m.
ac
.
i
d
.
D
i
v
i
Ga
l
i
h
P
ra
s
et
y
o
P
utri
recei
v
e
d
h
er
b
ac
h
el
o
r
d
e
g
ree
an
d
mas
t
er’s
d
eg
ree
i
n
In
fo
rma
t
i
c
s
E
n
g
i
n
eer
i
n
g
fro
m
In
s
t
i
t
u
t
T
ek
n
o
l
o
g
i
Se
p
u
l
u
h
N
o
p
emb
er,
I
n
d
o
n
e
s
i
a
i
n
2
0
1
4
an
d
2
0
1
6
re
s
p
ec
t
i
v
el
y
.
Sh
e
o
b
t
ai
n
ed
h
er
Ph
.
D
.
o
f
I
n
fo
rma
t
i
cs
i
n
2
0
2
1
fro
m
U
n
i
v
ers
i
t
y
o
f
M
i
l
a
n
o
Bi
co
cca.
Cu
rre
n
t
l
y
,
s
h
e
i
s
an
A
s
s
i
s
t
an
t
Pro
fes
s
o
r
a
t
t
h
e
D
ep
art
me
n
t
o
f
E
l
ec
t
ri
ca
l
E
n
g
i
n
eeri
n
g
an
d
I
n
fo
rma
t
i
c
s
,
U
n
i
v
ers
i
t
a
s
G
ad
j
a
h
Mad
a.
H
er
re
s
ear
ch
i
n
t
ere
s
t
s
i
n
cl
u
d
e
s
e
v
eral
t
o
p
i
c
s
i
n
i
n
f
o
rmat
i
o
n
ret
ri
e
v
al
,
s
en
t
i
m
en
t
a
n
al
y
s
i
s
,
d
at
a
s
c
i
en
c
e,
an
d
s
o
ft
w
are
en
g
i
n
eer
i
n
g
.
Sh
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
d
i
v
i
.
g
a
l
i
h
@
u
g
m.
ac.
i
d
.
J
i
m
m
y
Tri
o
P
utra
recei
v
ed
h
i
s
b
ac
h
el
o
r
d
eg
ree
i
n
E
l
ec
t
ri
ca
l
E
n
g
i
n
e
eri
n
g
fro
m
U
n
i
v
er
s
i
t
as
Be
n
g
k
u
l
u
,
In
d
o
n
es
i
a,
i
n
2
0
1
3
.
H
e
g
rad
u
a
t
ed
w
i
t
h
a
Ma
s
t
er
o
f
E
n
g
i
n
e
eri
n
g
a
t
U
n
i
v
er
s
i
t
as
G
ad
j
ah
Mad
a
i
n
2
0
1
5
.
H
e
i
s
an
A
s
s
i
s
t
an
t
Pro
fes
s
o
r
i
n
t
h
e
D
ep
ar
t
men
t
o
f
E
l
ect
r
i
ca
l
E
n
g
i
n
eeri
n
g
a
n
d
I
n
fo
rma
t
i
c
s
,
V
o
cat
i
o
n
al
Co
l
l
e
g
e,
U
n
i
v
ers
i
t
a
s
G
ad
j
a
h
Mad
a.
H
e
i
s
p
u
r
s
u
i
n
g
h
i
s
D
o
c
t
o
ra
l
D
eg
ree
i
n
t
h
e
D
e
p
art
me
n
t
o
f
E
l
ect
r
i
cal
a
n
d
In
fo
rma
t
i
o
n
E
n
g
i
n
eer
i
n
g
,
Facu
l
t
y
o
f
E
n
g
i
n
eeri
n
g
,
U
n
i
v
ers
i
t
a
s
G
ad
j
a
h
Mad
a,
I
n
d
o
n
e
s
i
a.
H
i
s
res
earch
i
n
t
eres
t
s
are
p
o
w
er
s
y
s
t
ems
,
d
i
s
t
ri
b
u
t
e
d
g
e
n
erat
i
o
n
,
an
d
v
e
h
i
c
l
e
t
o
g
r
i
d
.
H
e
can
b
e
co
n
t
ac
t
ed
a
t
ema
i
l
:
j
i
mmy
t
ri
o
p
u
t
ra@
u
g
m.
ac.
i
d
.
Budi
Ba
y
u
M
urti
recei
v
ed
t
h
e
B.
S.
d
eg
ree
i
n
E
l
ec
t
ri
ca
l
E
n
g
i
n
eeri
n
g
fro
m
U
n
i
v
er
s
i
t
as
G
ad
j
ah
Mad
a,
Y
o
g
y
ak
ar
t
a,
In
d
o
n
es
i
a,
i
n
1
9
9
6
an
d
t
h
e
M.
E
n
g
.
d
eg
ree
fro
m
t
h
e
s
ame
u
n
i
v
ers
i
t
y
i
n
2
0
0
5
.
C
u
rren
t
l
y
,
h
e
i
s
a
l
ec
t
u
re
r,
res
e
arch
er
i
n
t
h
e
D
ep
ar
t
men
t
o
f
E
l
ect
r
i
ca
l
E
n
g
i
n
eeri
n
g
an
d
In
f
o
rmat
i
cs
,
V
o
cat
i
o
n
al
Co
l
l
e
g
e,
U
n
i
v
ers
i
t
a
s
G
ad
j
ah
Mad
a,
an
d
p
u
r
s
u
i
n
g
a
d
o
c
t
o
ra
l
d
eg
ree
i
n
E
l
ect
r
i
ca
l
E
n
g
i
n
e
eri
n
g
at
U
n
i
v
er
s
i
t
a
s
G
ad
j
ah
Mad
a.
H
i
s
res
earch
i
n
t
ere
s
t
s
i
n
c
l
u
d
e
w
i
rel
e
s
s
co
mm
u
n
i
cat
i
o
n
,
s
i
g
n
al
p
ro
ce
s
s
i
n
g
,
o
p
t
i
ca
l
co
mm
u
n
i
cat
i
o
n
,
co
g
n
i
t
i
v
e
rad
i
o
,
an
d
b
i
o
me
d
i
ca
l
i
n
s
t
ru
me
n
t
a
t
i
o
n
.
H
e
can
b
e
co
n
t
ac
t
ed
at
emai
l
:
b
u
d
i
b
m@
u
g
m.
ac.
i
d
.
Wa
hy
o
n
o
recei
v
e
d
t
h
e
B.
Sc.
d
e
g
ree
i
n
C
o
mp
u
t
e
r
Sci
en
ce
fro
m
U
n
i
v
er
s
i
t
as
G
ad
j
ah
Mad
a,
I
n
d
o
n
e
s
i
a,
an
d
t
h
e
Ph
.
D
.
d
e
g
ree
fro
m
T
h
e
U
n
i
v
er
s
i
t
y
o
f
U
l
s
a
n
,
So
u
t
h
K
o
rea.
H
e
i
s
an
A
s
s
o
c
i
at
e
Pro
fe
s
s
o
r
at
t
h
e
D
e
p
art
me
n
t
o
f
Co
mp
u
t
er
Sci
en
ce
a
n
d
E
l
ect
r
o
n
i
cs
,
U
n
i
v
er
s
i
t
as
G
ad
j
ah
Ma
d
a,
In
d
o
n
es
i
a.
H
i
s
re
s
earch
i
n
t
eres
t
s
i
n
c
l
u
d
e
mac
h
i
n
e
l
ear
n
i
n
g
,
co
mp
u
t
er
v
i
s
i
o
n
,
an
d
p
at
t
ern
reco
g
n
i
t
i
o
n
.
H
e
act
i
v
e
l
y
p
a
rt
i
c
i
p
a
t
es
as
a
memb
er
o
f
t
h
e
s
o
c
i
et
i
es
,
a
rev
i
ew
er
i
n
rep
u
t
ab
l
e
i
n
t
er
n
at
i
o
n
al
j
o
u
r
n
al
s
,
an
d
an
ed
i
t
o
r
i
n
s
ev
era
l
j
o
u
rn
a
l
s
.
H
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
w
a
h
y
o
@
u
g
m.
ac.
i
d
.
Evaluation Warning : The document was created with Spire.PDF for Python.