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.
40
,
No
.
2
,
N
o
v
em
b
e
r
2
0
2
5
,
p
p
.
9
26
~
94
0
I
SS
N:
2
5
0
2
-
4
7
5
2
,
DOI
: 1
0
.
1
1
5
9
1
/ijeecs.v
40
.i
2
.
pp
926
-
94
0
926
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs.ia
esco
r
e.
co
m
Federate
d
lea
rnin
g
in
edg
e
AI
:
a
sy
stema
tic
r
ev
iew o
f
a
pplica
tions
,
priv
a
cy
challeng
es, an
d preserv
a
tion
tec
hniques
Chris
t
ina
T
ha
nk
a
m
Sa
j
a
n,
H
ela
nm
a
ry
M
.
Su
nn
y
,
Anj
u P
ra
t
a
p
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
S
a
i
n
t
g
i
t
s
c
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
(
A
u
t
o
n
o
m
o
u
s)
,
A
P
J Ab
d
u
l
K
a
l
a
m
Te
c
h
n
o
l
o
g
i
c
a
l
U
n
i
v
e
r
si
t
y
,
T
h
i
r
u
v
a
n
a
n
t
h
a
p
u
r
a
m
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
3
,
2
0
2
4
R
ev
is
ed
J
u
l
25
,
2
0
2
5
Acc
ep
ted
Oct
14
,
2
0
2
5
Ed
g
e
a
rti
ficia
l
i
n
telli
g
e
n
c
e
(E
d
g
e
AI)
i
n
v
o
lv
e
s
th
e
imp
lem
e
n
tati
o
n
o
f
AI
a
lg
o
rit
h
m
s
a
n
d
m
o
d
e
ls
d
irec
tl
y
o
n
l
o
c
a
l
e
d
g
e
d
e
v
ice
s,
su
c
h
a
s
se
n
so
rs
o
r
in
tern
e
t
o
f
t
h
in
g
s
(Io
T)
d
e
v
ice
s.
Th
is
a
ll
o
ws
f
o
r
imm
e
d
iate
p
r
o
c
e
ss
in
g
a
n
d
a
n
a
ly
sis
o
f
d
a
ta
wit
h
o
u
t
th
e
n
e
e
d
f
o
r
c
o
n
ti
n
u
o
u
s
d
e
p
e
n
d
e
n
c
e
o
n
c
l
o
u
d
in
fra
stru
c
tu
re
.
Co
n
c
e
rn
s
a
b
o
u
t
p
riv
a
c
y
h
a
v
e
g
r
o
wn
imp
o
rtan
c
e
in
re
c
e
n
t
y
e
a
rs
fo
r
b
u
si
n
e
ss
e
s
lo
o
k
i
n
g
t
o
u
p
h
o
ld
e
n
d
-
u
se
r
e
x
p
e
c
tati
o
n
s
a
n
d
sa
fe
g
u
a
rd
b
u
sin
e
ss
m
o
d
e
ls.
F
e
d
e
ra
ted
l
e
a
rn
in
g
(F
L)
h
a
s
e
m
e
rg
e
d
a
s
a
n
o
v
e
l
a
p
p
r
o
a
c
h
to
e
n
h
a
n
c
e
p
ri
v
a
c
y
.
To
imp
ro
v
e
g
e
n
e
ra
li
z
a
ti
o
n
q
u
a
li
ti
e
s,
F
L
tr
a
in
s
lo
c
a
l
m
o
d
e
ls
o
n
l
o
c
a
l
d
a
ta.
Th
e
se
m
o
d
e
ls
th
e
n
c
o
l
lab
o
ra
te
to
u
p
d
a
te
a
g
lo
b
a
l
m
o
d
e
l.
Eac
h
e
d
g
e
d
e
v
ice
(
li
k
e
sm
a
rtp
h
o
n
e
s,
Io
T
se
n
so
rs,
o
r
a
u
to
n
o
m
o
u
s
v
e
h
icle
s)
train
s
a
lo
c
a
l
m
o
d
e
l
o
n
it
s
o
wn
d
a
t
a
.
T
h
is
lo
c
a
l
train
i
n
g
h
e
lp
s
i
n
c
a
p
tu
rin
g
d
a
ta
p
a
tt
e
rn
s
s
p
e
c
ifi
c
to
e
a
c
h
d
e
v
ice
o
r
n
o
d
e
.
P
o
iso
n
in
g
,
b
a
c
k
d
o
o
rs,
a
n
d
g
e
n
e
ra
ti
v
e
a
d
v
e
rsa
rial
n
e
two
rk
(G
AN
)
-
b
a
se
d
a
tt
a
c
k
s
a
re
c
u
rre
n
tl
y
t
h
e
m
a
in
se
c
u
rit
y
risk
.
Ne
v
e
rth
e
les
s,
th
e
b
ig
g
e
st
th
re
a
t
to
F
L
’
s
p
riv
a
c
y
is
fro
m
i
n
fe
re
n
c
e
-
b
a
se
d
a
ss
a
u
lt
s
su
c
h
a
s
m
o
d
e
l
in
v
e
rsio
n
a
tt
a
c
k
s,
d
iffere
n
ti
a
l
p
ri
v
a
c
y
sh
o
rtco
m
i
n
g
s
a
n
d
FL
u
ti
li
z
e
s
b
lo
c
k
c
h
a
in
a
n
d
c
ry
p
to
g
ra
p
h
y
tec
h
n
o
l
o
g
ies
to
im
p
ro
v
e
p
riv
a
c
y
o
n
e
d
g
e
d
e
v
ice
s.
Th
is
p
a
p
e
r
p
re
se
n
ts
a
t
h
o
r
o
u
g
h
e
x
a
m
in
a
ti
o
n
o
f
t
h
e
c
u
rre
n
t
l
it
e
ra
tu
re
o
n
th
is
su
b
jec
t.
In
m
o
re
d
e
tail,
we
st
u
d
y
th
e
b
a
c
k
g
ro
u
n
d
o
f
F
L
a
n
d
it
s
d
iffere
n
t
e
x
isti
n
g
a
p
p
li
c
a
ti
o
n
s,
ty
p
e
s,
p
ri
v
a
c
y
th
re
a
ts
a
n
d
it
s
tec
h
n
iq
u
e
s
fo
r
p
riv
a
c
y
p
re
se
rv
a
ti
o
n
.
K
ey
w
o
r
d
s
:
Ar
tific
ial
in
tellig
en
ce
E
d
g
e
AI
E
d
g
e
co
m
p
u
tin
g
Fed
er
ated
lear
n
in
g
Priv
ac
y
-
p
r
eser
v
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
:
An
ju
Pra
tap
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
a
n
d
E
n
g
in
ee
r
in
g
,
Sain
tg
its
co
lleg
e
o
f
E
n
g
i
n
ee
r
in
g
(
Au
to
n
o
m
o
u
s
)
APJ
Ab
d
u
l K
alam
T
ec
h
n
o
lo
g
ical
Un
iv
er
s
ity
T
h
ir
u
v
a
n
an
th
ap
u
r
am
,
Ke
r
ala,
I
n
d
ia
E
m
ail:
an
ju
.
p
r
atap
@
s
ain
tg
its
.
o
r
g
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
p
h
r
ase
“
E
d
g
e
Ar
tific
ial
I
n
tellig
en
ce
(
E
d
g
e
AI
)
”
r
e
f
er
s
to
th
e
in
s
tallatio
n
o
f
AI
ap
p
licatio
n
s
o
n
h
ar
d
war
e
f
o
u
n
d
in
th
e
r
ea
l
w
o
r
ld
.
B
ec
au
s
e
th
e
AI
co
m
p
u
ta
tio
n
is
ca
r
r
ie
d
o
u
t
cl
o
s
e
to
th
e
u
s
er
at
th
e
ed
g
e
o
f
th
e
n
etwo
r
k
,
n
ea
r
er
th
e
d
ata
’
s
lo
ca
tio
n
,
as
o
p
p
o
s
ed
to
ce
n
tr
ally
at
a
cl
o
u
d
co
m
p
u
tin
g
f
ac
i
lity
o
r
p
r
iv
ate
d
ata
ce
n
ter
,
it
is
k
n
o
wn
as
“
E
d
g
e
AI
”
.
I
t
’
s
p
ar
ticu
lar
ly
u
s
ef
u
l
in
s
ce
n
ar
io
s
wh
er
e
r
ea
l
-
ti
m
e
p
r
o
ce
s
s
in
g
an
d
d
ec
is
io
n
-
m
ak
in
g
ar
e
cr
itical,
s
u
ch
as
au
to
n
o
m
o
u
s
v
eh
icles,
h
ea
lth
ca
r
e
m
o
n
ito
r
in
g
s
y
s
te
m
s
,
an
d
in
d
u
s
tr
ial
au
to
m
atio
n
.
Fed
er
ated
l
ea
r
n
i
n
g
(
FL)
in
ed
g
e
AI
is
a
p
ar
ad
ig
m
th
at
u
s
es
a
v
ar
iety
o
f
d
ec
en
tr
alize
d
ed
g
e
d
ev
ices
to
tr
ain
m
ac
h
in
e
lear
n
in
g
(
ML
)
m
o
d
els
with
o
u
t
tr
an
s
f
er
r
in
g
r
aw
d
ata.
T
h
is
m
eth
o
d
o
lo
g
y
f
ac
ilit
ates
o
n
g
o
in
g
ed
u
ca
tio
n
an
d
ad
ju
s
t
m
en
t
to
r
eg
io
n
al
cir
cu
m
s
tan
c
es.
E
ac
h
ed
g
e
d
e
v
ice
(
s
u
ch
a
s
a
s
m
ar
tp
h
o
n
e
o
r
in
ter
n
et
o
f
th
in
g
s
(
I
o
T
)
d
ev
ic
e
)
in
th
is
c
o
n
f
ig
u
r
atio
n
h
as
a
lo
ca
l
d
ataset
an
d
ta
k
es
p
ar
t
in
th
e
lear
n
in
g
p
r
o
ce
s
s
.
A
ce
n
tr
al
s
er
v
er
r
ec
eiv
es
th
e
lear
n
in
g
s
(
m
o
d
el
p
ar
am
eter
s
)
th
at
th
e
lear
n
in
g
m
o
d
el
h
as
ac
q
u
ir
ed
f
r
o
m
its
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
F
ed
era
ted
lea
r
n
in
g
i
n
ed
g
e
A
I
:
a
s
ystema
tic
r
ev
iew
o
f a
p
p
lica
tio
n
s
…
(
C
h
r
is
tin
a
Th
a
n
ka
m
S
a
ja
n
)
927
tr
ain
in
g
o
n
th
ese
lo
ca
l
d
ataset
s
.
T
h
e
s
er
v
er
th
e
n
co
m
p
iles
th
ese
m
o
d
if
icatio
n
s
to
e
n
h
an
c
e
th
e
g
lo
b
al
m
o
d
el.
Un
til th
e
m
o
d
el
ac
h
ie
v
es a
n
a
cc
u
r
ac
y
lev
el
th
at
is
ac
ce
p
tab
le,
th
is
p
r
o
ce
s
s
is
r
ep
ea
ted
m
u
lt
ip
le
tim
es.
FL
is
a
co
llab
o
r
ativ
e
p
r
o
ce
s
s
th
at
u
s
es
r
em
o
te
d
ata
s
h
ar
in
g
am
o
n
g
m
u
ltip
le
p
ar
ticip
a
n
ts
to
tr
ain
a
s
in
g
le
d
ee
p
lear
n
in
g
m
o
d
el
an
d
im
p
r
o
v
e
iter
ativ
ely
,
m
u
ch
lik
e
in
a
team
r
ep
o
r
t
o
r
p
r
esen
tatio
n
.
E
ac
h
in
d
iv
id
u
al
d
o
wn
lo
a
d
s
th
e
m
o
d
el,
wh
ich
is
o
f
ten
a
f
o
u
n
d
atio
n
m
o
d
el
th
at
h
as
alr
ea
d
y
b
ee
n
tr
ain
ed
,
f
r
o
m
a
clo
u
d
d
ata
ce
n
ter
.
T
h
ey
tr
ain
it
o
n
th
ei
r
o
w
n
p
r
iv
ate
d
ata,
th
en
s
u
m
m
ar
ize
an
d
en
c
r
y
p
t
th
e
m
o
d
el
’
s
n
ew
co
n
f
ig
u
r
atio
n
.
T
h
e
m
o
d
el
u
p
d
ates
ar
e
s
en
t
t
o
th
e
ce
n
tr
aliz
ed
m
o
d
el,
wh
ic
h
d
ec
o
d
es,
n
o
r
m
s
,
an
d
co
m
b
in
es
th
em
b
ef
o
r
e
s
en
d
in
g
th
em
b
a
ck
to
th
e
clo
u
d
.
T
h
e
co
llab
o
r
ativ
e
tr
ain
in
g
p
r
o
ce
s
s
ca
r
r
ie
s
o
n
iter
atio
n
af
ter
iter
at
io
n
u
n
til
th
e
m
o
d
el
is
co
m
p
letely
tr
ain
ed
.
Dis
tr
ib
u
ted
ML
ca
n
b
e
s
ec
u
r
ed
u
s
in
g
FL
as
an
ap
p
r
o
ac
h
to
co
llab
o
r
atin
g
to
ex
ec
u
te
FL
a
lg
o
r
ith
m
s
o
n
m
u
ltip
le
d
ev
ice
s
.
T
h
e
c
o
n
d
itio
n
is
th
at
th
er
e
ar
e
s
ca
tter
ed
ed
g
e
d
ev
ices
o
r
s
er
v
er
s
w
h
er
e
t
h
e
p
r
iv
ate
in
f
o
r
m
atio
n
is
n
o
t
lef
t
lo
ca
l.
A
d
ec
en
t
r
alize
d
ML
te
ch
n
iq
u
e
ca
lled
FL
u
s
es
s
ev
er
al
d
ev
ices
o
r
s
er
v
er
s
with
lo
ca
l
d
ata
s
am
p
les
to
tr
ain
m
o
d
els
with
o
u
t
tr
an
s
f
er
r
i
n
g
th
em
.
Fig
u
r
e
1
s
h
o
ws th
e
b
asic a
r
ch
itectu
r
e
o
f
FL
.
Fig
u
r
e
1
.
B
asic
ar
ch
itectu
r
e
o
f
f
ed
er
ated
lear
n
in
g
FL
in
E
d
g
e
AI
le
v
er
ag
es
th
e
c
o
m
p
u
tatio
n
al
p
o
wer
o
f
ed
g
e
d
ev
ices
lik
e
s
m
ar
tp
h
o
n
es
d
e
v
ices,
o
r
e
d
g
e
s
er
v
er
to
tr
ain
ML
m
o
d
els
lo
ca
lly
with
o
u
t
n
ee
d
i
n
g
to
tr
an
s
m
it
s
en
s
itiv
e
d
ata
to
a
ce
n
tr
al
s
er
v
er
.
E
d
g
e
d
ev
ices
p
er
f
o
r
m
m
o
d
el
tr
ain
in
g
u
s
in
g
lo
ca
l
d
ata.
T
h
is
c
o
u
ld
in
clu
d
e
d
ata
c
o
llected
f
r
o
m
s
en
s
o
r
s
,
u
s
er
in
ter
ac
tio
n
s
o
r
o
th
er
r
eso
u
r
ce
s
.
On
ly
m
o
d
if
i
ca
tio
n
s
to
m
o
d
els
ar
e
u
p
lo
ad
ed
to
a
ce
n
tr
al
s
er
v
e
r
;
r
aw
d
ata
ca
n
n
o
t
b
e
s
en
t.
User
p
r
iv
ac
y
is
u
p
h
el
d
in
th
is
way
.
FL
en
ab
les
p
er
s
o
n
ali
ze
d
an
d
co
n
tex
t
-
awa
r
e
in
tel
li
g
en
ce
d
i
r
ec
tly
o
n
d
ev
ices w
ith
o
u
t r
ely
in
g
h
ea
v
ily
o
n
cl
o
u
d
s
er
v
ices.
C
lo
u
d
co
m
p
u
tin
g
em
er
g
e
d
a
s
a
u
n
iq
u
e
co
m
p
u
ter
ar
ch
ite
ctu
r
e
f
o
r
th
e
I
n
ter
n
et
b
ased
o
n
h
ig
h
l
y
r
eso
u
r
ce
d
d
ata
ce
n
ter
s
as
I
n
f
o
r
m
atio
n
tech
n
o
lo
g
y
ad
v
an
ce
d
af
ter
2
0
0
0
.
Dev
elo
p
m
e
n
t
an
d
in
ter
est
in
clo
u
d
co
m
p
u
tin
g
h
as
g
r
o
wn
to
th
e
p
o
in
t
wh
er
e,
b
y
2
0
2
0
,
m
o
r
e
t
h
an
9
0
%
o
f
all
d
ata
ce
n
ter
tr
af
f
ic
will
o
r
ig
in
ate
f
r
o
m
s
o
u
r
ce
s
in
th
e
clo
u
d
[
1
]
.
T
h
e
im
m
en
s
e
p
o
ten
tial
o
f
ed
g
e
AI
h
as
f
in
ally
b
ee
n
r
ea
lized
to
th
e
m
o
s
t
r
ec
en
t
s
tr
id
es
in
AI
ef
f
icien
cy
,
th
e
g
r
o
wth
o
f
ed
g
e
co
m
p
u
tin
g
,
an
d
th
e
e
x
p
lo
s
io
n
o
f
I
o
T
d
ev
ices.
W
h
en
AI
co
m
p
u
tatio
n
s
a
r
e
p
er
f
o
r
m
ed
i
n
p
r
o
x
im
ity
to
c
o
n
s
u
m
er
s
o
n
a
n
etwo
r
k
ed
g
e,
th
e
y
a
r
e
r
e
f
er
r
ed
to
as
ed
g
e
-
b
ased
AI
.
T
h
is
is
i
n
c
o
n
tr
ast
to
ce
n
tr
alize
d
d
ata
s
to
r
ag
e,
s
u
ch
a
s
clo
u
d
s
er
v
ice
p
r
o
v
id
er
s
o
r
p
r
iv
ately
h
eld
d
ata
war
eh
o
u
s
es
[
2
]
.
T
h
e
s
u
cc
ess
f
u
l
o
p
er
atio
n
o
f
t
ask
s
is
en
h
an
ce
d
th
r
o
u
g
h
t
h
e
6
G
s
er
v
ices
p
r
o
v
id
e
d
f
o
r
ed
g
e
co
m
p
u
tin
g
an
d
a
u
to
n
o
m
o
u
s
v
eh
icu
lar
d
r
iv
i
n
g
ap
p
licatio
n
s
.
T
h
e
s
ig
n
if
ican
t
am
o
u
n
t
o
f
d
at
a
g
en
er
ated
b
y
th
ese
ap
p
licatio
n
s
ca
n
b
e
ad
v
an
tag
eo
u
s
f
o
r
th
e
AI
an
d
ML
in
d
u
s
tr
y
.
B
y
p
r
eser
v
in
g
th
e
ab
ilit
y
to
lear
n
f
r
o
m
d
ec
en
tr
alize
d
d
ata
s
ets,
FL
,
al
s
o
k
n
o
wn
as
FL,
is
an
e
s
s
en
tia
l
elem
en
t
in
an
in
teg
r
ated
s
o
lu
tio
n
to
p
r
iv
ac
y
an
d
tech
n
o
lo
g
ical
is
s
u
es.
T
r
ain
in
g
is
lim
ited
to
u
s
er
d
ev
ices,
an
d
th
e
s
er
v
er
r
ec
eiv
es
th
e
lo
ca
lly
co
m
p
u
ted
p
ar
am
eter
,
wh
ich
a
g
g
r
e
g
ates
t
h
e
u
p
d
ated
weig
h
ts
to
o
p
tim
ize
a
g
lo
b
al
m
o
d
el
[
3
]
.
T
h
e
e
m
er
g
en
ce
o
f
n
o
v
el
tech
n
o
lo
g
ies
lik
e
b
ig
d
ata,
ed
g
e
co
m
p
u
tin
g
,
f
o
g
co
m
p
u
tin
g
,
ar
tific
ial
in
tellig
en
ce
o
f
th
in
g
s
(
AI
o
T
)
,
an
d
f
o
g
co
m
p
u
tin
g
h
as
ca
u
s
ed
p
r
o
b
le
m
s
f
o
r
s
m
ar
t
city
ap
p
licatio
n
s
,
in
clu
d
in
g
th
e
d
is
clo
s
u
r
e
o
f
s
en
s
itiv
e
an
d
p
r
iv
ate
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
40
,
No
.
2
,
No
v
em
b
er
20
25
:
9
2
6
-
94
0
928
d
ata.
FL
ca
n
d
ea
l
with
p
len
ty
o
f
s
m
ar
t
city
co
n
ce
r
n
s
,
in
cl
u
d
in
g
en
o
r
m
o
u
s
am
o
u
n
ts
o
f
d
at
a
an
d
s
af
eg
u
ar
d
in
g
p
r
iv
ac
y
,
em
p
o
wer
in
g
d
ec
is
io
n
-
m
ak
er
s
to
ac
t
s
wif
tly
.
FL
i
s
b
en
ef
icial
in
tr
ain
in
g
s
h
ar
ed
s
tatis
t
ics
th
r
o
u
g
h
d
ec
en
tr
alize
d
d
e
v
ices
o
r
s
er
v
e
r
s
[
4
]
.
T
h
e
wid
esp
r
ea
d
d
e
p
lo
y
m
en
t
o
f
AI
in
h
ea
lth
ca
r
e
p
o
s
e
s
ch
allen
g
es
d
u
e
t
o
th
e
s
ca
tter
ed
n
atu
r
e
o
f
h
ea
lth
d
ata.
Priv
ac
y
co
n
ce
r
n
s
ca
n
b
e
ef
f
e
ctiv
ely
h
an
d
le
d
with
th
e
u
s
e
o
f
p
r
iv
ac
y
-
p
r
eser
v
in
g
al
g
o
r
ith
m
s
in
FL
,
wh
ich
was
d
ev
elo
p
e
d
t
o
ad
d
r
ess
d
ata
f
r
ag
m
en
tatio
n
.
T
o
e
n
h
an
ce
s
ec
u
r
ity
a
n
d
co
m
p
u
tatio
n
al
ef
f
ec
tiv
en
ess
,
FL
ca
n
b
e
p
air
e
d
with
o
t
h
er
tech
n
o
lo
g
ies
s
u
ch
as
ed
g
e
co
m
p
u
tin
g
an
d
b
lo
ck
ch
ain
[
5
]
.
T
h
e
s
in
g
le
p
o
in
t
o
f
f
ailu
r
e
th
at
is
th
e
b
o
ttlen
ec
k
o
f
b
o
th
tr
ad
itio
n
al
FL
an
d
HFL
s
y
s
tem
s
i
s
th
eir
r
elian
ce
o
n
a
ce
n
tr
alize
d
s
er
v
er
to
m
an
a
g
e
th
e
lear
n
in
g
p
r
o
ce
s
s
[
6
]
.
E
d
g
e
c
o
m
p
u
tin
g
is
o
n
e
way
o
f
AI
en
h
an
cin
g
cy
b
er
s
ec
u
r
it
y
.
E
d
g
e
co
m
p
u
tin
g
an
aly
ze
s
d
ata
at
th
e
n
etwo
r
k
’
s
e
d
g
es,
in
cl
u
d
in
g
in
d
iv
id
u
al
d
ev
ices,
r
o
u
ter
s
,
a
n
d
f
ir
ewa
lls
,
r
ath
er
th
an
f
o
r
war
d
in
g
it
to
a
ce
n
tr
al
p
lace
.
T
h
is
h
as
a
n
u
m
b
er
o
f
s
ec
u
r
ity
b
en
e
f
its
,
in
clu
d
in
g
th
r
ea
t
d
etec
tio
n
an
d
p
r
ev
e
n
tio
n
,
an
o
m
aly
d
etec
tio
n
,
en
h
an
ce
d
d
ata
p
r
iv
ac
y
,
ad
ap
tiv
e
s
ec
u
r
ity
p
o
licies,
an
d
f
r
au
d
d
etec
tio
n
.
Gain
in
g
u
s
er
tr
u
s
t
will
r
eq
u
ir
e
ad
d
r
ess
in
g
d
if
f
ic
u
lties
r
elate
d
to
p
er
f
o
r
m
an
ce
,
d
ata
p
r
o
ce
s
s
in
g
,
an
d
h
u
m
an
m
o
n
ito
r
in
g
.
As
ed
g
e
an
d
I
o
T
ad
o
p
tio
n
g
r
o
ws,
s
o
lid
lo
ca
lize
d
s
ec
u
r
ity
will
b
ec
o
m
e
m
o
r
e
v
ital.
I
n
cy
b
e
r
s
ec
u
r
ity
,
FL
al
lo
ws
b
u
s
in
ess
es
t
o
in
ter
ac
t
an
d
s
h
ar
e
in
s
ig
h
ts
f
r
o
m
th
eir
d
ata
with
o
u
t
d
is
clo
s
in
g
th
e
d
ata
its
elf
,
r
ed
u
cin
g
th
e
d
an
g
er
s
co
n
n
ec
ted
with
d
ata
b
r
e
ac
h
es
an
d
p
r
iv
a
cy
v
io
latio
n
s
.
T
h
is
d
ec
e
n
tr
alize
d
s
tr
ateg
y
also
aid
s
in
t
h
e
cr
ea
tio
n
o
f
m
o
r
e
r
esil
ien
t
an
d
ac
cu
r
ate
m
o
d
e
ls
b
y
in
teg
r
atin
g
s
ev
er
al
d
a
ta
s
o
u
r
ce
s
wh
ile
m
ain
tain
in
g
in
d
iv
id
u
al
d
ata
p
r
iv
ac
y
[
7
]
.
T
h
e
r
em
ai
n
d
er
o
f
th
is
p
ap
e
r
is
o
r
g
an
ized
as
f
o
llo
ws:
I
n
s
ec
tio
n
2
d
is
cu
s
s
es
th
e
o
v
er
v
iew,
im
p
o
r
tan
ce
,
d
i
f
f
er
en
t
ty
p
es
o
f
m
o
d
els,
ap
p
licatio
n
s
,
p
r
iv
ac
y
ch
allen
g
es
an
d
p
r
eser
v
atio
n
tech
n
iq
u
es
o
f
FL
in
E
d
g
e
AI
.
I
n
s
ec
tio
n
3
p
r
esen
t
s
th
e
r
esu
lts
an
d
d
is
cu
s
s
io
n
s
.
Fin
ally
,
s
ec
tio
n
4
c
o
n
clu
d
es
t
h
e
p
ap
e
r
with
f
u
tu
r
e
r
esear
ch
d
i
r
ec
tio
n
s
.
2.
O
VE
RVI
E
W
O
F
F
E
D
E
RA
T
E
D
L
E
ARNI
NG
I
N
E
DG
E
AI
FL
in
E
d
g
e
AI
co
m
b
in
es
th
e
p
r
in
cip
les
o
f
FL
with
ed
g
e
co
m
p
u
tin
g
,
f
ac
ilit
atin
g
th
e
im
m
ed
iate
tr
ain
in
g
o
f
ML
m
o
d
els
o
n
ed
g
e
g
ad
g
ets
wh
ile
p
r
eser
v
i
n
g
d
ata
p
r
iv
ac
y
an
d
r
e
d
u
ci
n
g
c
o
m
m
u
n
icatio
n
o
v
er
h
ea
d
.
FL
is
a
c
o
o
p
e
r
ativ
el
y
d
ec
e
n
tr
alize
d
s
o
lu
ti
o
n
t
h
at
p
r
o
tects
p
r
iv
ac
y
wh
ile
ad
d
r
ess
in
g
is
s
u
es
with
d
ata
s
en
s
itiv
ity
an
d
s
ilo
s
.
I
n
FL,
in
s
tead
o
f
ag
g
r
eg
atin
g
d
ata
in
a
ce
n
tr
alize
d
s
er
v
er
,
th
e
tr
ain
in
g
p
r
o
ce
s
s
o
cc
u
r
s
lo
ca
lly
o
n
p
er
ip
h
er
al
d
ev
ices
s
u
ch
as
ed
g
e
co
m
p
u
t
in
g
s
er
v
er
s
,
m
o
b
ile
d
ev
ices
an
d
I
o
T
d
e
v
ices.
B
y
r
etain
in
g
r
aw
d
ata
o
n
ed
g
e
d
ev
ices,
it
g
u
ar
an
tees
d
a
ta
p
r
iv
ac
y
.
T
h
e
o
n
ly
u
p
d
ates
to
th
e
m
o
d
el
th
at
ar
e
s
en
t
to
th
e
ce
n
tr
alize
d
s
er
v
er
f
o
r
ag
g
r
eg
atio
n
ar
e
th
e
weig
h
ts
o
r
g
r
ad
ien
ts
.
T
h
is
ap
p
r
o
ac
h
m
in
im
izes
th
e
r
is
k
o
f
d
ata
b
r
ea
ch
es
o
r
p
r
iv
ac
y
v
io
latio
n
s
.
B
y
p
e
r
f
o
r
m
in
g
tr
ai
n
in
g
lo
ca
ll
y
o
n
ed
g
e
d
ev
ices,
FL
r
ed
u
ce
s
th
e
n
ee
d
f
o
r
d
at
a
tr
an
s
m
is
s
io
n
to
a
ce
n
tr
al
s
er
v
er
.
T
h
is
is
p
ar
ticu
lar
ly
b
en
ef
icial
in
ed
g
e
co
m
p
u
tin
g
en
v
ir
o
n
m
en
ts
with
lim
ited
b
an
d
wid
th
o
r
in
ter
m
itten
t
co
n
n
ec
tiv
ity
.
E
d
g
e
d
ev
ices
co
llab
o
r
ativ
el
y
c
o
n
tr
ib
u
te
to
m
o
d
el
tr
ain
in
g
b
y
p
er
f
o
r
m
in
g
lo
ca
l
u
p
d
ates
b
ased
o
n
th
eir
r
esp
ec
tiv
e
d
atasets
.
T
h
ese
u
p
d
ates
a
r
e
th
en
ag
g
r
eg
ated
to
im
p
r
o
v
e
th
e
g
l
o
b
al
m
o
d
el,
lev
er
ag
in
g
in
s
ig
h
ts
f
r
o
m
d
iv
e
r
s
e
ed
g
e
d
ev
ices.
FL
e
n
ab
l
es
m
o
d
els
to
b
e
tr
ain
e
d
a
n
d
u
p
d
ated
in
r
ea
l
-
tim
e
o
n
ed
g
e
d
ev
ices,
f
ac
ilit
atin
g
q
u
i
ck
d
ec
is
io
n
-
m
ak
in
g
an
d
i
n
f
er
en
ce
with
o
u
t
r
ely
in
g
o
n
a
ce
n
tr
alize
d
s
er
v
er
[
8
]
.
I
t
is
in
h
er
en
tly
s
ca
lab
le
as
it
d
is
tr
ib
u
tes
co
m
p
u
tatio
n
ac
r
o
s
s
n
u
m
er
o
u
s
e
d
g
e
d
e
v
ices.
T
h
is
allo
ws
f
o
r
lar
g
e
-
s
ca
le
d
ep
lo
y
m
en
t
o
f
ed
g
e
A
I
s
y
s
tem
s
with
o
u
t
o
v
er
b
u
r
d
en
i
n
g
an
y
s
in
g
le
d
ev
ice
o
r
ce
n
t
r
al
s
er
v
er
.
Mo
d
els
tr
ain
ed
u
s
in
g
FL
ca
n
ad
a
p
t
to
ch
an
g
in
g
d
ata
d
is
tr
ib
u
ti
o
n
s
an
d
e
n
v
ir
o
n
m
en
tal
co
n
d
itio
n
s
in
r
ea
l
-
tim
e,
m
ak
in
g
th
em
well
-
s
u
ited
f
o
r
d
y
n
am
ic
ed
g
e
co
m
p
u
tin
g
en
v
ir
o
n
m
en
ts
.
Her
e
’
s
a
s
tep
-
by
-
s
tep
ex
p
la
n
atio
n
o
f
h
o
w
it wo
r
k
s
:
1.
I
n
itial
m
o
d
el
d
is
tr
ib
u
tio
n
:
a
ce
n
tr
al
s
er
v
er
in
itializes
a
g
lo
b
al
ML
m
o
d
el,
wh
ich
co
u
ld
b
e
a
n
eu
r
al
n
etwo
r
k
o
r
an
o
th
er
ty
p
e
o
f
m
o
d
el
(
g
lo
b
al
m
o
d
el
in
itializatio
n
)
.
T
h
e
in
itial
v
er
s
io
n
o
f
t
h
is
m
o
d
el
is
th
en
s
en
t
to
all
p
ar
ticip
atin
g
ed
g
e
d
e
v
ices.
E
ac
h
d
ev
ice
r
ec
ei
v
es a
co
p
y
o
f
t
h
e
s
am
e
m
o
d
el
(
m
o
d
el
d
is
tr
ib
u
tio
n
).
2.
L
o
ca
l
tr
ain
in
g
o
n
e
d
g
e
d
ev
ice
s
:
ea
ch
ed
g
e
d
e
v
ice
h
as
its
o
wn
lo
ca
l
d
ataset,
wh
ich
co
u
ld
b
e
u
s
er
-
s
p
ec
if
ic
d
ata
lik
e
tex
t
m
ess
ag
es,
im
ag
e
s
,
s
en
s
o
r
d
ata,
o
r
ap
p
licatio
n
u
s
ag
e
p
atter
n
s
.
E
ac
h
d
ev
ice
tr
ai
n
s
th
e
r
ec
ei
v
ed
g
lo
b
al
m
o
d
el
o
n
its
lo
ca
l
d
ata.
T
h
is
tr
ain
in
g
in
v
o
l
v
es
s
ev
er
al
iter
atio
n
s
o
f
an
o
p
tim
izatio
n
alg
o
r
ith
m
(
e.
g
.
,
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t)
t
o
u
p
d
ate
th
e
m
o
d
el
’
s
p
ar
a
m
eter
s
b
ased
o
n
t
h
e
lo
ca
l
d
ata
s
et.
Af
ter
lo
ca
l
tr
ain
in
g
,
ea
ch
d
ev
ice
co
m
p
u
t
es
th
e
ch
an
g
es
to
th
e
m
o
d
el
p
ar
am
eter
s
(
e.
g
.
,
g
r
ad
ien
ts
o
r
weig
h
t
u
p
d
ates)
b
ased
o
n
its
lo
ca
l d
ata.
3.
C
o
m
m
u
n
icatio
n
with
th
e
ce
n
t
r
al
s
er
v
er
:
in
s
tead
o
f
s
h
ar
in
g
th
e
ac
tu
al
d
ata,
ea
c
h
ed
g
e
d
e
v
ice
s
en
d
s
o
n
ly
th
e
m
o
d
el
u
p
d
ates
(
i.e
.
,
th
e
c
h
an
g
es
in
th
e
m
o
d
el
p
ar
am
eter
s
)
to
th
e
ce
n
tr
al
s
er
v
er
.
T
h
ese
u
p
d
ates
ca
n
b
e
en
cr
y
p
ted
o
r
p
r
o
ce
s
s
ed
u
s
in
g
t
ec
h
n
iq
u
es
lik
e
d
if
f
e
r
en
tial
p
r
i
v
ac
y
to
en
s
u
r
e
th
at
s
en
s
itiv
e
i
n
f
o
r
m
atio
n
f
r
o
m
th
e
lo
ca
l d
ata
i
s
n
o
t e
x
p
o
s
ed
.
4.
Ag
g
r
eg
atio
n
o
f
u
p
d
ates:
th
e
c
en
tr
al
s
er
v
er
co
llects
u
p
d
ates
f
r
o
m
m
u
ltip
le
d
ev
ices.
I
t
th
en
a
g
g
r
eg
ates
th
ese
u
p
d
ates
to
f
o
r
m
a
n
ew
g
lo
b
al
m
o
d
el.
A
co
m
m
o
n
m
eth
o
d
is
to
av
er
ag
e
th
e
u
p
d
ates,
b
u
t
m
o
r
e
s
o
p
h
is
ticated
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
F
ed
era
ted
lea
r
n
in
g
i
n
ed
g
e
A
I
:
a
s
ystema
tic
r
ev
iew
o
f a
p
p
lica
tio
n
s
…
(
C
h
r
is
tin
a
Th
a
n
ka
m
S
a
ja
n
)
929
tech
n
iq
u
es
ca
n
b
e
u
s
ed
to
e
n
s
u
r
e
r
o
b
u
s
tn
ess
ag
ain
s
t
o
u
tlier
s
o
r
m
alicio
u
s
u
p
d
ates.
T
h
e
ce
n
tr
al
s
er
v
er
u
p
d
ates
th
e
g
lo
b
al
m
o
d
el
b
a
s
ed
o
n
th
e
ag
g
r
eg
ated
in
f
o
r
m
atio
n
a
n
d
p
r
ep
a
r
es
it
f
o
r
t
h
e
n
e
x
t
r
o
u
n
d
o
f
tr
ain
in
g
.
5.
I
ter
ativ
e
p
r
o
ce
s
s
:
th
e
u
p
d
ate
d
g
lo
b
al
m
o
d
el
is
th
en
r
ed
is
tr
ib
u
ted
to
all
p
ar
ticip
atin
g
e
d
g
e
d
ev
ices.
Step
s
2
th
r
o
u
g
h
4
a
r
e
r
ep
ea
ted
f
o
r
s
e
v
er
al
iter
atio
n
s
(
o
r
r
o
u
n
d
s
)
u
n
til
th
e
m
o
d
el
co
n
v
er
g
es,
m
ea
n
in
g
th
at
f
u
r
th
er
u
p
d
ates r
esu
lt in
m
in
im
al
im
p
r
o
v
em
en
t.
6.
Fin
al
m
o
d
el
d
e
p
lo
y
m
e
n
t:
o
n
c
e
th
e
m
o
d
el
h
as
r
ea
ch
ed
a
s
atis
f
ac
to
r
y
lev
el
o
f
ac
c
u
r
ac
y
,
it
c
an
b
e
d
ep
lo
y
ed
f
o
r
in
f
e
r
en
ce
o
n
th
e
e
d
g
e
d
ev
i
ce
s
,
allo
win
g
th
em
to
m
a
k
e
p
r
ed
ictio
n
s
lo
ca
lly
b
ased
o
n
th
e
tr
ain
ed
m
o
d
el.
Ov
er
all,
FL
in
E
d
g
e
AI
o
f
f
er
s
a
d
ec
en
tr
alize
d
a
n
d
p
r
iv
ac
y
-
p
r
eser
v
in
g
a
p
p
r
o
ac
h
to
ML
th
at
is
well
-
s
u
ited
f
o
r
ed
g
e
co
m
p
u
tin
g
e
n
v
ir
o
n
m
en
ts
,
in
clu
d
i
n
g
I
o
T
,
s
m
ar
t
cities,
au
to
n
o
m
o
u
s
v
eh
icles,
an
d
m
o
r
e.
I
t
ad
d
r
ess
es
ch
allen
g
es
r
elate
d
to
d
ata
p
r
iv
ac
y
,
co
m
m
u
n
icatio
n
o
v
er
h
ea
d
,
s
ca
lab
ilit
y
,
an
d
ad
a
p
tab
ilit
y
,
m
ak
in
g
it a
p
o
wer
f
u
l p
ar
ad
i
g
m
f
o
r
d
ep
lo
y
in
g
AI
a
p
p
licatio
n
s
at
th
e
ed
g
e.
2
.
1
.
I
m
po
r
t
a
nce
o
f
f
eder
a
t
e
d lea
rning
FL
is
a
tr
ain
in
g
m
eth
o
d
f
o
r
d
e
ep
-
lear
n
in
g
AI
m
o
d
els
th
at
in
v
o
lv
es
co
llab
o
r
atio
n
.
FL
tak
es
m
o
d
els
t
o
u
s
er
’
s
d
ev
ices
f
o
r
tr
ain
in
g
with
lo
ca
l
d
ata
u
n
til
th
e
y
m
at
u
r
e,
in
s
tead
o
f
ce
n
tr
alizin
g
c
u
s
to
m
er
d
ata
in
a
s
in
g
le
r
ep
o
s
ito
r
y
.
T
h
e
f
u
lly
tr
ain
ed
m
o
d
els
ar
e
th
en
s
en
t
b
ac
k
to
th
e
p
r
o
v
id
e
r
o
r
b
u
s
in
ess
.
T
h
is
m
eth
o
d
en
s
u
r
es
th
at
th
e
AI
p
r
o
v
id
er
d
o
esn
’
t
ac
ce
s
s
an
y
en
d
-
u
s
er
d
ata
wh
ile
tr
ain
in
g
,
p
r
eser
v
in
g
d
ata
p
r
iv
ac
y
wh
ile
s
till
m
ak
in
g
cr
u
cial
u
s
e
o
f
en
d
-
u
s
er
d
ata
f
o
r
m
o
d
el
im
p
r
o
v
em
e
n
t.
E
d
g
e
AI
is
an
AI
s
y
s
tem
t
h
at
r
u
n
s
AI
-
d
r
iv
en
o
p
e
r
atio
n
s
clo
s
er
to
w
h
er
e
t
h
e
ac
tu
al
u
s
er
d
ata
is
lo
ca
ted
in
s
tead
o
f
o
n
a
ce
n
tr
alize
d
s
er
v
er
.
A
co
m
b
in
atio
n
o
f
ed
g
e
co
m
p
u
tin
g
an
d
AI
is
u
s
ed
to
en
ab
le
d
ig
ital
s
er
v
ices
to
u
s
e
A
I
ca
p
ab
ilit
ies
lo
ca
ll
y
with
o
u
t
t
h
e
n
ee
d
f
o
r
ce
n
tr
al
clo
u
d
c
o
n
n
ec
tiv
ity
.
W
h
en
FL
is
ap
p
lied
to
E
d
g
e
A
I
,
it
en
a
b
les
E
d
g
e
AI
a
p
p
licatio
n
s
to
c
o
n
tin
u
o
u
s
ly
ev
o
l
v
e
th
eir
u
n
d
er
s
tan
d
in
g
o
f
e
n
d
-
u
s
er
d
y
n
am
ics
with
o
u
t
th
e
n
ee
d
f
o
r
tak
i
n
g
e
n
d
-
u
s
er
d
ata
to
th
eir
ce
n
tr
al
clo
u
d
s
to
r
ag
e.
E
n
d
u
s
er
s
ca
n
tak
e
a
d
v
an
tag
e
o
f
th
is
b
y
n
o
t h
a
v
in
g
t
o
s
h
ar
e
s
en
s
itiv
e
d
ata
with
an
y
b
u
s
in
ess
.
T
h
e
co
m
b
in
atio
n
o
f
FL
a
n
d
E
d
g
e
AI
allo
ws
f
o
r
th
e
d
e
v
elo
p
m
en
t
o
f
m
o
r
e
r
o
b
u
s
t
a
n
d
p
r
iv
ac
y
-
p
r
eser
v
in
g
AI
s
y
s
tem
s
th
at
c
an
lear
n
a
n
d
ad
ap
t
i
n
r
ea
l
-
ti
m
e,
p
r
ec
is
ely
wh
e
r
e
t
h
e
d
ata
is
cr
ea
ted
,
at
th
e
n
etwo
r
k
’
s
ed
g
e
.
T
h
is
r
esu
lts
in
f
aster
r
esp
o
n
s
e
tim
es,
a
d
ec
r
ea
s
e
in
n
etwo
r
k
laten
cy
,
an
d
im
p
r
o
v
ed
d
ata
p
r
iv
ac
y
.
FL
in
E
d
g
e
A
I
o
f
f
er
s
s
ig
n
if
ican
t
ad
v
an
tag
es
th
at
ca
ter
to
th
e
u
n
iq
u
e
d
em
an
d
s
o
f
ed
g
e
co
m
p
u
tin
g
en
v
ir
o
n
m
en
ts
.
Her
e
’
s
a
d
ee
p
er
d
iv
e
in
to
its
im
p
o
r
tan
ce
:
1.
Priv
ac
y
p
r
eser
v
atio
n
:
FL
en
a
b
les
AI
m
er
ely
s
en
d
in
g
r
aw
d
ata
to
a
d
is
tan
t
s
er
v
er
,
m
o
d
el
s
ca
n
b
e
tr
ain
ed
lo
ca
lly
o
n
co
m
m
o
n
d
e
v
ices
l
ik
e
s
m
ar
tp
h
o
n
es
o
r
I
o
T
s
s
en
s
o
r
s
.
T
h
is
p
r
o
tects
u
s
er
p
r
iv
a
cy
b
y
k
ee
p
in
g
s
en
s
itiv
e
d
ata
d
ec
en
tr
alize
d
an
d
lo
wer
in
g
t
h
e
lik
elih
o
o
d
o
f
d
ata
b
r
ea
ch
es o
r
p
r
iv
ac
y
v
io
lati
o
n
s
[
9
]
.
2.
R
ed
u
ce
d
l
aten
cy
:
FL
r
ed
u
ce
s
th
e
r
eq
u
ir
em
e
n
t
to
s
en
d
d
ata
to
a
ce
n
tr
al
s
er
v
er
f
o
r
p
r
o
ce
s
s
in
g
b
y
e
x
ec
u
tin
g
m
o
d
el
tr
ain
in
g
an
d
in
f
er
e
n
ce
o
n
ed
g
e
d
ev
ices.
T
h
is
d
ec
r
ea
s
es
laten
cy
a
n
d
p
r
o
v
id
es
r
ea
l
-
tim
e
r
esp
o
n
s
iv
en
ess
,
m
ak
in
g
it
p
er
f
ec
t
f
o
r
lo
w
-
laten
cy
in
ter
ac
ti
o
n
s
in
ap
p
licatio
n
s
lik
e
s
elf
-
d
r
iv
in
g
ca
r
s
an
d
au
g
m
en
ted
r
ea
lity
[
1
0
]
.
3.
B
an
d
wid
th
co
n
s
er
v
ati
o
n
:
h
u
g
e
am
o
u
n
ts
o
f
d
ata
tr
an
s
m
itte
d
f
r
o
m
ed
g
e
d
ev
ices
to
a
ce
n
tr
al
s
er
v
er
ca
n
s
tr
ai
n
n
etwo
r
k
b
an
d
wid
th
a
n
d
r
esu
lt
in
s
ig
n
if
ican
t
e
x
p
en
d
itu
r
es,
p
ar
ticu
lar
ly
in
s
ettin
g
s
with
r
estricte
d
co
n
n
ec
tio
n
.
FL
allev
iates
th
is
s
tr
ain
b
y
co
n
d
u
ctin
g
m
o
d
el
ch
an
g
es
lo
ca
lly
,
wh
ich
s
av
es
b
an
d
wid
th
an
d
r
ed
u
ce
s
n
etwo
r
k
co
n
g
esti
o
n
.
4.
R
o
b
u
s
tn
ess
to
co
n
n
ec
tiv
ity
is
s
u
es:
ed
g
e
d
ev
ices
f
r
e
q
u
e
n
tly
o
p
e
r
ate
in
ar
ea
s
with
in
co
n
s
is
ten
t
o
r
u
n
p
r
e
d
ictab
le
n
etwo
r
k
ac
ce
s
s
.
FL
is
r
esis
tan
t to
s
u
ch
o
b
s
tacl
es b
ec
au
s
e
it a
llo
ws d
ev
ices to
lear
n
an
d
d
r
aw
co
n
clu
s
io
n
s
in
d
e
p
en
d
e
n
tly
ev
e
n
wh
en
th
e
y
ar
e
r
em
o
v
e
d
f
r
o
m
th
e
n
etwo
r
k
.
5.
Ad
ap
ta
b
ilit
y
an
d
pe
r
s
o
n
aliza
tio
n
:
FL
allo
ws
AI
m
o
d
els
t
o
b
e
tailo
r
ed
an
d
ad
ju
s
ted
t
o
s
p
ec
if
ic
ed
g
e
d
ev
ices
o
r
p
e
o
p
le
with
o
u
t
s
ac
r
if
icin
g
p
r
i
v
ac
y
.
C
u
s
to
m
ized
ad
v
ice
an
d
s
er
v
ices
ar
e
m
ad
e
p
o
s
s
ib
le
b
y
th
is
in
d
iv
id
u
alize
d
a
p
p
r
o
ac
h
,
wh
ic
h
also
im
p
r
o
v
es u
s
er
ex
p
e
r
ien
c
e.
FL
in
E
d
g
e
AI
o
f
f
er
s
a
p
r
i
v
ac
y
-
p
r
eser
v
in
g
,
lo
w
-
laten
cy
,
an
d
b
an
d
wid
t
h
-
ef
f
icien
t
ap
p
r
o
ac
h
to
tr
ain
in
g
AI
m
o
d
els
d
ir
ec
tl
y
o
n
ed
g
e
d
ev
ices,
m
a
k
in
g
it
in
d
is
p
en
s
ab
le
f
o
r
m
u
ltip
le
ap
p
licatio
n
s
in
t
h
e
I
o
T
s
,
h
ea
lth
ca
r
e,
s
m
ar
t c
ities
,
an
d
v
ar
io
u
s
o
th
er
f
ield
s
.
2
.
2
.
Va
ri
o
us
f
eder
a
t
ed
lea
rning
m
o
del
I
n
th
is
s
ec
tio
n
,
we
ex
p
lain
an
d
co
m
p
ar
e
d
if
f
e
r
en
t
t
y
p
es
o
f
FL
,
s
u
ch
as
h
o
r
iz
o
n
tal
FL
(
H
FL)
,
v
er
tical
FL
,
f
ed
er
ated
tr
a
n
s
f
er
lear
n
i
n
g
(
FTL
)
,
an
d
cr
o
s
s
-
s
ilo
FL
,
b
ased
o
n
t
h
eir
f
ea
tu
r
es a
s
s
h
o
wn
i
n
T
ab
le
1
.
2
.
2
.
1
.
H
o
ri
zo
nt
a
l
f
eder
a
t
ed
l
ea
rning
HFL
is
a
f
o
r
m
o
f
FL
in
wh
ich
d
atasets
f
r
o
m
m
u
ltip
le
n
o
d
es
s
h
ar
e
th
e
s
am
e
f
ea
tu
r
e
s
p
ac
e
b
u
t
u
tili
ze
v
ar
io
u
s
s
am
p
les.
I
t
ca
n
also
b
e
r
ef
er
r
ed
to
as
s
am
p
le
-
b
ased
FL
o
r
h
o
m
o
g
en
e
o
u
s
FL
.
I
t
wo
r
k
s
well
wh
en
th
er
e
is
s
ig
n
if
ican
t
o
v
er
lap
i
n
th
e
u
s
er
f
ea
tu
r
es
o
f
two
d
atasets
b
u
t
n
o
t
in
th
e
to
tal
n
u
m
b
e
r
o
f
u
s
er
s
.
I
n
o
r
d
er
to
ex
tr
ac
t
th
e
p
o
r
tio
n
o
f
th
e
d
at
a
wh
er
e
u
s
er
attr
ib
u
tes
a
r
e
s
im
ilar
b
u
t
u
s
er
s
ar
e
n
o
t
p
r
ec
is
ely
th
e
s
am
e
f
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
40
,
No
.
2
,
No
v
em
b
er
20
25
:
9
2
6
-
94
0
930
tr
ain
in
g
,
we
d
iv
id
e
th
e
d
atase
ts
h
o
r
izo
n
tally
(
b
y
th
e
u
s
er
d
i
m
e
n
s
io
n
)
in
th
is
lear
n
in
g
p
r
o
ce
s
s
.
Mu
lti
-
task
FL
r
ed
u
ce
s
co
m
m
u
n
icatio
n
co
s
ts
an
d
im
p
r
o
v
es
f
au
lt
to
ler
an
c
e
co
m
p
ar
ed
to
d
is
tr
ib
u
ted
m
u
lti
-
task
lear
n
in
g
.
s
en
s
itiv
e
in
f
o
r
m
atio
n
is
p
r
eser
v
ed
o
f
f
t
h
e
s
er
v
er
b
y
u
s
in
g
c
lien
t
-
s
p
ec
if
ic
d
ata
d
iv
is
io
n
.
A
f
ter
ca
lcu
latin
g
t
h
e
lo
ca
l
g
r
ad
ien
t
a
n
d
u
p
lo
ad
i
n
g
it
to
th
e
s
er
v
er
,
ea
ch
clien
t m
o
d
if
ies
th
e
g
lo
b
al
m
o
d
el
to
ac
c
o
u
n
t
f
o
r
th
e
g
r
ad
ien
t
ch
an
g
es
[
1
1
]
.
T
h
e
F
ig
u
r
e
2
s
h
o
ws th
e
ar
ch
itectu
r
e
o
f
HF
L
.
T
h
e
wo
r
k
in
g
o
f
HFL
co
n
s
is
ts
th
e
f
o
llo
win
g
s
tep
s
:
a.
T
h
e
r
em
o
te
s
er
v
er
r
ec
eiv
es a
n
e
n
co
d
ed
g
r
ad
ie
n
t f
r
o
m
th
e
lo
c
al
m
o
d
el.
b.
T
h
e
s
er
v
er
h
an
d
les th
e
s
af
e
co
m
b
in
atio
n
.
c.
T
h
e
m
o
d
el
r
ec
eiv
es u
p
d
ates f
r
o
m
th
e
s
er
v
er
.
d.
T
h
e
m
o
d
els ar
e
u
p
d
ate
d
b
ased
o
n
th
e
i
n
f
o
r
m
atio
n
f
r
o
m
th
e
s
er
v
er
.
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
HFL
2
.
2
.
2
.
Ver
t
ica
l
f
eder
a
t
ed
lea
r
nin
g
Ver
tical
f
ed
er
ated
lear
n
i
n
g
(
VFL)
is
a
s
p
ec
ialized
f
o
r
m
o
f
FL
,
d
esig
n
ed
to
en
ab
le
m
u
ltip
le
o
r
g
an
izatio
n
s
o
r
e
n
titi
es
to
co
llab
o
r
ativ
ely
tr
ain
ML
m
o
d
els
with
o
u
t
s
h
ar
in
g
th
eir
r
aw
d
at
a.
I
t
’
s
p
ar
ticu
lar
l
y
u
s
ef
u
l
wh
en
th
ese
o
r
g
a
n
izatio
n
s
p
o
s
s
ess
d
if
f
er
en
t
ty
p
es
o
f
d
ata
ab
o
u
t
t
h
e
s
am
e
s
et
o
f
u
s
er
s
o
r
en
titi
es.
I
n
VFL,
th
e
d
ata
is
p
ar
titi
o
n
e
d
v
er
tically
,
m
ea
n
in
g
d
if
f
er
en
t
o
r
g
an
izatio
n
s
h
o
ld
d
i
f
f
er
en
t
f
e
atu
r
es
o
r
attr
ib
u
tes
o
f
th
e
s
am
e
u
s
er
s
.
Fo
r
ex
am
p
le,
a
b
an
k
m
ig
h
t
h
av
e
f
in
an
cial
d
ata
ab
o
u
t
its
cu
s
to
m
er
s
,
wh
ile
a
h
ea
lth
ca
r
e
p
r
o
v
id
e
r
m
ig
h
t
h
av
e
m
e
d
ical
d
ata
ab
o
u
t
th
e
s
am
e
in
d
iv
id
u
als.
T
h
ese
o
r
g
an
izatio
n
s
wan
t
to
co
llab
o
r
ate
to
b
u
ild
a
b
etter
m
o
d
el,
b
u
t
th
ey
ca
n
n
o
t
s
h
ar
e
t
h
eir
r
aw
d
ata
d
u
e
to
p
r
i
v
ac
y
co
n
ce
r
n
s
.
T
h
e
o
r
g
a
n
izatio
n
s
co
llab
o
r
ate
to
t
r
ain
a
ML
m
o
d
el
b
y
s
h
ar
in
g
en
cr
y
p
ted
in
te
r
m
ed
iate
co
m
p
u
tatio
n
s
in
s
tea
d
o
f
r
aw
d
ata.
E
ac
h
o
r
g
an
izatio
n
co
n
tr
i
b
u
tes
to
th
e
m
o
d
el
b
y
u
s
in
g
its
lo
ca
l
d
a
ta
to
co
m
p
u
te
c
er
tain
asp
ec
ts
(
e.
g
.
,
g
r
ad
ien
ts
o
r
m
o
d
el
u
p
d
ates)
a
n
d
s
h
ar
es
th
ese
with
th
e
o
th
er
p
ar
ties
in
a
s
ec
u
r
e
m
an
n
er
.
VFL
em
p
lo
y
s
v
ar
io
u
s
cr
y
p
to
g
r
ap
h
ic
tech
n
iq
u
es,
s
u
c
h
as
s
ec
u
r
e
m
u
lti
-
p
ar
t
y
co
m
p
u
tatio
n
(
SMPC
)
o
r
h
o
m
o
m
o
r
p
h
ic
en
cr
y
p
tio
n
(
HE
)
,
to
en
s
u
r
e
th
at
wh
ile
th
e
c
o
m
p
u
tatio
n
s
ar
e
s
h
a
r
ed
,
th
e
ac
tu
al
d
ata
r
e
m
ain
s
p
r
iv
at
e.
T
h
is
allo
ws
t
h
e
o
r
g
an
izatio
n
s
t
o
lear
n
f
r
o
m
ea
ch
o
th
e
r
’
s
d
ata
with
o
u
t
ac
tu
all
y
s
ee
in
g
it.
A
cr
u
cial
s
tep
in
VFL
is
alig
n
in
g
th
e
d
ata
ac
r
o
s
s
th
e
d
if
f
er
en
t
p
ar
ti
es.
Sin
ce
ea
ch
o
r
g
an
izatio
n
h
as
d
ata
o
n
th
e
s
am
e
u
s
er
s
b
u
t
in
d
if
f
er
en
t
f
o
r
m
s
,
th
ey
n
ee
d
to
e
n
s
u
r
e
th
e
y
ar
e
wo
r
k
in
g
with
th
e
s
am
e
u
s
er
s
with
o
u
t
d
ir
ec
tly
s
h
a
r
in
g
i
d
en
t
if
iab
le
in
f
o
r
m
atio
n
.
T
h
is
is
o
f
ten
d
o
n
e
th
r
o
u
g
h
s
ec
u
r
e
alig
n
m
en
t
p
r
o
to
c
o
ls
th
at
m
atch
u
s
er
s
ac
r
o
s
s
d
atasets
b
ased
o
n
en
cr
y
p
ted
id
en
tifie
r
s
.
Fig
u
r
e
3
s
h
o
ws th
e
ar
ch
itectu
r
e
o
f
v
er
tical
FL
.
A
ty
p
ical
VFL
p
r
o
ce
d
u
r
e
f
o
r
e
ac
h
lear
n
in
g
tim
e
f
r
am
e
h
as sev
en
im
p
o
r
tan
t step
s
[
1
2
]
:
a.
Priv
ate
s
et
in
ter
s
ec
tio
n
(
PS
I
)
:
t
o
alig
n
tr
ain
in
g
d
ata
s
am
p
les,
th
e
s
y
s
tem
u
s
es
PS
I
o
r
s
ec
u
r
e
e
n
tity
alig
n
m
en
t
to
id
e
n
tify
c
o
m
m
o
n
i
d
en
tifie
r
s
s
h
ar
e
d
b
y
all
p
ar
ticip
a
n
ts
,
in
clu
d
in
g
g
u
est
a
n
d
h
o
s
t
o
r
g
an
izatio
n
s
.
PS
I
is
a
s
ec
u
r
e
s
y
s
tem
th
at
id
en
tifie
s
co
m
m
o
n
I
Ds
am
o
n
g
m
u
ltip
le
p
a
r
ticip
an
ts
’
d
ata.
C
o
m
m
o
n
ly
u
s
ed
PS
I
ap
p
r
o
ac
h
es
in
clu
d
e
n
aïv
e
h
ash
in
g
,
o
b
liv
io
u
s
p
o
ly
n
o
m
ial
ev
alu
atio
n
,
an
d
o
b
liv
io
u
s
tr
an
s
f
er
.
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
F
ed
era
ted
lea
r
n
in
g
i
n
ed
g
e
A
I
:
a
s
ystema
tic
r
ev
iew
o
f a
p
p
lica
tio
n
s
…
(
C
h
r
is
tin
a
Th
a
n
ka
m
S
a
ja
n
)
931
b.
B
o
tto
m
m
o
d
el
f
o
r
war
d
p
r
o
p
a
g
atio
n
:
a
f
ter
alig
n
i
n
g
d
ata
s
a
m
p
les,
p
ar
ticip
an
ts
will
u
s
e
l
o
ca
l
d
ata
to
d
o
f
o
r
war
d
p
r
o
p
ag
atio
n
b
ased
o
n
th
eir
b
o
tto
m
m
o
d
el.
T
h
e
f
o
r
war
d
p
r
o
p
ag
atio
n
p
r
o
ce
d
u
r
e
is
s
im
ilar
to
co
n
v
en
tio
n
al
tr
ain
in
g
,
with
th
e
ex
ce
p
tio
n
o
f
d
ete
r
m
in
in
g
th
e
lo
s
s
v
alu
e.
c.
Fo
r
war
d
o
u
tp
u
t
tr
a
n
s
m
is
s
io
n
:
e
ac
h
p
ar
ticip
a
n
t
m
u
s
t
p
r
o
v
i
d
e
th
eir
f
o
r
war
d
o
u
tp
u
t
to
t
h
e
lab
el
o
wn
e
r
.
T
h
e
f
o
r
war
d
o
u
tp
u
t
r
ep
r
esen
ts
th
e
in
ter
m
ed
iate
o
u
tco
m
es
o
f
lo
ca
l
n
eu
r
al
n
etwo
r
k
s
th
at
tr
an
s
late
th
e
o
r
ig
in
al
p
r
o
p
er
ties
in
to
f
ea
tu
r
e
s
.
d.
B
ac
k
war
d
o
u
tp
u
t
tr
an
s
m
is
s
io
n
:
e
ac
h
g
u
est
p
ar
ticip
a
n
t
r
ec
eiv
es
th
e
g
r
a
d
ien
ts
o
f
th
eir
f
o
r
war
d
o
u
tp
u
t.
T
h
e
co
m
m
u
n
icatio
n
c
o
s
t (
tr
an
s
m
is
s
io
n
b
its
)
f
o
r
g
r
ad
ien
ts
is
ty
p
ically
lo
wer
th
an
f
o
r
in
ter
m
ed
iate
o
u
tp
u
ts
.
e.
B
o
tto
m
m
o
d
el
b
ac
k
war
d
p
r
o
p
ag
atio
n
:
p
ar
ticip
an
ts
ch
an
g
e
t
h
eir
b
o
tto
m
m
o
d
el
p
ar
am
eter
s
d
ep
en
d
in
g
o
n
lo
ca
l d
ata
an
d
t
h
e
lab
el
o
wn
e
r
’
s
f
o
r
war
d
o
u
tp
u
ts
.
Fig
u
r
e
3
.
Ar
c
h
itectu
r
e
o
f
VFL
[
1
2
]
2
.
2
.
3
.
F
eder
a
t
ed
t
ra
ns
f
er
lea
rning
FTL
,
s
im
ilar
to
s
tan
d
ar
d
ML
,
in
v
o
lv
es
ad
d
i
n
g
a
n
ew
f
ea
t
u
r
e
to
a
p
r
e
-
tr
ain
ed
m
o
d
el.
E
x
ten
d
in
g
v
er
tical
FL
to
in
cl
u
d
e
s
am
p
le
in
s
tan
ce
s
f
r
o
m
n
o
n
-
co
llab
o
r
atin
g
o
r
g
a
n
izatio
n
s
is
a
g
o
o
d
e
x
am
p
le
t
h
at
p
r
o
m
o
tes
c
o
m
p
lem
e
n
tar
y
ex
c
h
an
g
e
o
f
in
f
o
r
m
atio
n
b
etwe
en
th
e
d
o
m
ain
s
in
a
d
ata
f
e
d
er
at
io
n
an
d
k
n
o
wled
g
e
s
h
ar
in
g
with
o
u
t
jeo
p
ar
d
izin
g
u
s
er
p
r
iv
ac
y
.
T
h
is
allo
ws
a
tar
g
et
-
d
o
m
ain
p
ar
ty
to
le
v
er
ag
e
r
ich
lab
els
f
r
o
m
th
e
s
o
u
r
ce
d
o
m
ai
n
to
d
ev
el
o
p
ad
ap
tab
le
an
d
p
o
wer
f
u
l
m
o
d
els.
T
h
is
m
eth
o
d
o
lo
g
y
d
eliv
er
s
th
e
s
am
e
d
eg
r
ee
o
f
ac
cu
r
ac
y
as
n
o
n
-
p
r
iv
ac
y
-
p
r
eser
v
in
g
tr
an
s
f
e
r
lear
n
in
g
m
et
h
o
d
s
with
v
er
y
litt
le
ad
ju
s
tm
en
t
to
th
e
p
r
ev
ailin
g
s
tr
u
ctu
r
e.
I
t
ad
ap
ts
well
to
s
ec
u
r
e
m
u
lti
-
p
ar
ty
ML
wo
r
k
l
o
ad
s
[
1
3
]
.
Fig
u
r
e
4
s
h
o
ws
ar
ch
itectu
r
e
o
f
FTL
.
Her
e
ar
e
th
e
ty
p
ical
s
tep
s
in
v
o
lv
ed
in
FTL
:
a.
I
n
itial
s
etu
p
an
d
d
ata
p
r
ep
ar
at
io
n
:
id
en
tify
th
e
p
ar
ticip
atin
g
en
titi
es
(
clien
ts
)
an
d
th
eir
r
esp
ec
tiv
e
d
atasets
.
Pre
p
r
o
ce
s
s
an
d
s
tan
d
ar
d
ize
d
a
ta
ac
r
o
s
s
all
clien
ts
to
en
s
u
r
e
co
n
s
is
ten
cy
.
b.
Pre
-
tr
ain
ed
m
o
d
el
s
elec
tio
n
:
s
elec
t
a
p
r
e
-
tr
ain
ed
m
o
d
el
th
at
will
b
e
u
s
ed
as
th
e
b
ase
m
o
d
el
f
o
r
tr
an
s
f
e
r
lear
n
in
g
.
T
h
is
m
o
d
el
is
ty
p
ical
ly
tr
ain
ed
o
n
a
lar
g
e,
d
iv
e
r
s
e
d
ataset
an
d
p
r
o
v
id
es a
g
o
o
d
s
ta
r
tin
g
p
o
i
n
t.
c.
L
o
ca
l
m
o
d
el
cu
s
to
m
izatio
n
:
ea
ch
clien
t
f
in
e
-
tu
n
es
th
e
p
r
e
-
tr
ain
ed
m
o
d
el
o
n
th
eir
lo
ca
l
d
ata.
T
h
is
s
tep
in
v
o
lv
es:
d
o
w
n
lo
ad
in
g
th
e
p
r
e
-
tr
ain
ed
m
o
d
el,
T
r
ain
in
g
t
h
e
m
o
d
el
o
n
lo
ca
l
d
ata
b
y
ad
ju
s
tin
g
m
o
d
el
weig
h
ts
b
ased
o
n
lo
ca
l
d
ataset
ch
ar
ac
ter
is
tics
.
d.
L
o
ca
l
m
o
d
el
u
p
d
ates:
af
ter
lo
ca
l
tr
ain
in
g
,
ea
ch
clien
t
c
o
m
p
u
tes
th
e
m
o
d
el
u
p
d
ates
(
g
r
a
d
ien
ts
o
r
m
o
d
el
p
ar
am
eter
s
)
b
ased
o
n
th
eir
lo
c
al
d
ataset.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
40
,
No
.
2
,
No
v
em
b
er
20
25
:
9
2
6
-
94
0
932
e.
Secu
r
e
ag
g
r
e
g
atio
n
:
to
p
r
o
tect
d
ata
p
r
iv
ac
y
,
th
e
m
o
d
el
u
p
d
a
tes
ar
e
s
ec
u
r
ely
ag
g
r
eg
ate
d
.
T
h
is
ca
n
b
e
d
o
n
e
u
s
in
g
tech
n
iq
u
es
lik
e
s
ec
u
r
e
m
u
ltip
ar
ty
co
m
p
u
tatio
n
(
S
MC)
o
r
d
if
f
er
en
tial
p
r
i
v
ac
y
to
en
s
u
r
e
th
at
in
d
iv
id
u
al
u
p
d
ates
r
em
ain
co
n
f
id
en
tial.
C
lien
ts
s
en
d
th
eir
en
cr
y
p
t
ed
m
o
d
el
u
p
d
ates
to
a
ce
n
tr
al
s
er
v
er
o
r
an
ag
g
r
e
g
ato
r
.
f.
Glo
b
al
m
o
d
el
u
p
d
ate:
t
h
e
ce
n
t
r
al
s
er
v
er
o
r
ag
g
r
eg
ato
r
d
ec
r
y
p
ts
an
d
ag
g
r
eg
ates
th
e
l
o
ca
l
u
p
d
ates
to
u
p
d
ate
th
e
g
lo
b
al
m
o
d
el.
T
h
is
s
tep
in
v
o
lv
es:
i.
C
o
m
b
in
in
g
th
e
u
p
d
ates f
r
o
m
a
ll c
lien
ts
.
ii.
Ap
p
ly
in
g
th
e
ag
g
r
eg
ated
u
p
d
a
tes to
th
e
g
lo
b
al
m
o
d
el.
g.
Glo
b
al
m
o
d
el
d
is
tr
ib
u
tio
n
:
th
e
u
p
d
ated
g
lo
b
al
m
o
d
el
is
th
en
d
is
tr
ib
u
ted
b
ac
k
to
all
clien
ts
.
h.
I
ter
ativ
e
p
r
o
ce
s
s
:
s
tep
s
3
to
7
ar
e
r
ep
ea
ted
iter
ativ
ely
.
I
n
ea
ch
iter
atio
n
,
th
e
g
lo
b
al
m
o
d
el
b
ec
o
m
es
m
o
r
e
r
ef
in
ed
as it
lear
n
s
f
r
o
m
th
e
d
i
v
er
s
e
lo
ca
l d
atasets
o
f
all
clien
ts
.
i.
C
o
n
v
er
g
en
ce
a
n
d
f
i
n
al
m
o
d
el
:
th
e
p
r
o
ce
s
s
co
n
tin
u
es
u
n
til
th
e
g
lo
b
al
m
o
d
el
co
n
v
er
g
es
t
o
a
s
atis
f
ac
to
r
y
p
er
f
o
r
m
an
ce
le
v
el
o
r
a
p
r
ed
e
f
in
ed
n
u
m
b
e
r
o
f
iter
atio
n
s
is
r
ea
ch
ed
.
T
h
e
f
in
al
m
o
d
el
is
t
h
en
u
s
ed
b
y
all
clien
ts
f
o
r
in
f
er
e
n
ce
o
n
th
eir
l
o
ca
l d
ata.
j.
E
v
alu
atio
n
a
n
d
d
ep
lo
y
m
en
t:
e
v
alu
ate
th
e
f
in
al
m
o
d
el
’
s
p
er
f
o
r
m
an
ce
o
n
a
v
alid
atio
n
d
ata
s
et.
Dep
lo
y
th
e
f
in
al
m
o
d
el
f
o
r
r
ea
l
-
wo
r
ld
u
s
e.
Fig
u
r
e
4
.
I
ll
u
s
tr
atio
n
o
f
f
ed
er
a
ted
tr
an
s
f
er
lear
n
i
n
g
ar
c
h
itectu
r
e
[
1
3
]
2
.
2
.
4
.
Cro
s
s
-
s
ilo
f
eder
a
t
ed
lea
rning
C
r
o
s
s
-
s
ilo
FL
is
u
tili
ze
d
wh
e
n
th
er
e
ar
e
f
ewe
r
p
ar
ticip
atin
g
d
ev
ices
an
d
th
ey
ar
e
av
ailab
le
f
o
r
all
r
o
u
n
d
s
.
T
h
e
tr
ai
n
in
g
d
ata
m
ay
b
e
i
n
h
o
r
izo
n
tal
o
r
v
er
tic
al
FL
f
o
r
m
at.
C
r
o
s
s
-
s
ilo
is
m
o
s
tly
u
tili
ze
d
f
o
r
o
r
g
an
izatio
n
al
ca
s
es
[
1
4
]
.
C
r
o
s
s
-
s
ilo
FL
i
s
a
co
llab
o
r
ativ
e
ML
st
r
ateg
y
in
wh
ich
m
u
ltip
le
co
m
p
an
ies,
ea
ch
with
its
o
wn
d
ata
s
i
lo
,
wo
r
k
to
g
eth
er
to
tr
ain
a
co
m
m
o
n
m
o
d
el
with
o
u
t
s
h
ar
in
g
th
eir
r
a
w
d
ata.
I
n
s
tead
,
th
ey
ex
ch
an
g
e
m
o
d
el
u
p
d
ates
o
r
g
r
ad
ien
ts
wh
ile
k
ee
p
in
g
th
eir
d
ata
lo
ca
l,
wh
ich
p
r
o
tects
p
r
i
v
ac
y
an
d
s
ec
u
r
ity
.
T
h
is
m
eth
o
d
allo
ws
e
n
ter
p
r
is
e
s
to
h
ar
n
ess
co
llectiv
e
ex
p
er
tis
e
wh
ile
p
r
o
tectin
g
s
en
s
itiv
e
d
a
ta.
Fig
u
r
e
5
s
h
o
ws
th
e
ar
ch
itectu
r
e
o
f
cr
o
s
s
-
s
ilo
FL
.
A
ty
p
ical
cr
o
s
s
-
s
ilo
FL
m
eth
o
d
co
m
p
r
is
es m
u
ltip
le
r
o
u
n
d
s
.
E
ac
h
r
o
u
n
d
h
as f
o
u
r
s
tep
s
:
a.
T
h
e
ce
n
tr
al
s
er
v
er
p
r
o
v
id
e
s
t
h
e
clien
ts
with
t
h
e
g
l
o
b
al
m
o
d
el
f
r
o
m
t
h
e
p
r
ev
io
u
s
cy
cle.
T
h
e
d
o
wn
lo
ad
e
d
m
o
d
el
is
in
itialized
at
r
an
d
o
m
in
th
e
f
ir
s
t r
o
u
n
d
.
b.
User
s
tr
ain
ac
q
u
ir
ed
m
o
d
els o
n
p
r
iv
ate
lo
ca
l
d
ata
s
ets,
r
esu
ltin
g
in
u
p
d
ated
l
o
ca
l m
o
d
els.
c.
L
o
ca
l m
o
d
el
u
p
d
ates a
r
e
u
p
lo
a
d
ed
b
y
cl
ien
ts
to
th
e
ce
n
tr
al
s
e
r
v
er
.
d.
I
n
th
e
s
u
b
s
eq
u
en
t
r
o
u
n
d
,
clien
ts
will
r
ec
eiv
e
a
n
ew
g
lo
b
al
m
o
d
el
th
at
th
e
s
er
v
er
h
as
cr
ea
te
d
b
y
co
m
b
in
in
g
th
e
s
u
b
m
itted
m
o
d
el
r
ev
is
io
n
s
.
I
n
th
e
b
elo
w
T
a
b
le
1
s
h
o
ws
th
e
co
m
p
ar
is
o
n
b
etwe
en
d
if
f
er
e
n
t
ty
p
es
o
f
FL
s
u
ch
as
HFL,
VFL,
FTL
,
an
d
c
r
o
s
s
-
s
ilo
FL
b
ased
o
n
th
eir
f
ea
tu
r
es
lik
e
d
ata
d
is
tr
ib
u
ti
o
n
,
m
o
d
el
ar
ch
itectu
r
e,
c
o
m
m
u
n
icatio
n
o
v
er
h
ea
d
,
m
o
d
el
ac
cu
r
ac
y
,
clien
t
h
eter
o
g
en
eity
an
d
u
s
e
ca
s
e.
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
F
ed
era
ted
lea
r
n
in
g
i
n
ed
g
e
A
I
:
a
s
ystema
tic
r
ev
iew
o
f a
p
p
lica
tio
n
s
…
(
C
h
r
is
tin
a
Th
a
n
ka
m
S
a
ja
n
)
933
Fig
u
r
e
5
.
Ar
c
h
itectu
r
e
o
f
c
r
o
s
s
-
s
ilo
f
ed
er
ated
lear
n
in
g
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
b
etwe
en
d
if
f
er
en
t ty
p
es o
f
f
ed
er
ated
lea
r
n
in
g
2
.
3
.
Appl
ica
t
io
ns
2
.
3
.
1
.
H
ea
lt
hca
re
Dis
tr
ib
u
ted
in
tellig
en
ce
u
s
in
g
ed
g
e
d
ev
ices
is
m
ad
e
p
o
s
s
ib
le
b
y
ed
g
e
AI
,
an
in
ter
d
is
cip
lin
ar
y
tech
n
o
lo
g
y
.
I
t
in
v
o
lv
es
m
o
d
el
s
f
o
r
FL
,
DL
,
an
d
ML
th
at
ar
e
d
ev
elo
p
ed
an
d
r
u
n
at
th
e
ed
g
e
o
f
th
e
n
etwo
r
k
,
in
d
ep
en
d
en
t
o
f
ce
n
tr
alize
d
d
a
ta
ce
n
ter
s
.
I
t
also
in
clu
d
es
d
ata
an
aly
tics
an
d
AI
.
E
d
g
e
co
m
p
u
tin
g
allo
ws
f
o
r
lo
ca
l d
ata
p
r
o
ce
s
s
in
g
a
n
d
an
al
y
s
is
,
wh
ich
lo
wer
s
laten
cy
an
d
f
ac
ilit
ates q
u
ick
d
ec
is
io
n
-
m
ak
in
g
[
1
5
]
.
A
d
is
tr
ib
u
ted
ML
tech
n
iq
u
e
c
alled
FL
u
s
es
in
f
o
r
m
atio
n
f
r
o
m
s
ev
er
al
d
ec
e
n
tr
alize
d
ed
g
e
clien
ts
to
b
u
ild
a
g
l
o
b
al
m
o
d
el.
Data
p
r
o
tectio
n
a
n
d
s
ca
lab
ilit
y
ar
e
o
n
ly
two
o
f
its
m
an
y
b
en
e
f
its
.
W
h
en
d
ea
lin
g
with
h
eter
o
g
en
e
o
u
s
d
ev
ice
s
,
FL
tech
n
iq
u
es
also
co
m
e
with
s
o
m
e
d
an
g
e
r
s
an
d
co
m
p
u
tatio
n
al
co
m
p
lex
ity
lim
itatio
n
s
.
Hea
l
th
ca
r
e
p
r
o
v
id
e
r
s
,
f
o
r
e
x
am
p
le,
ca
n
u
s
e
FL
to
co
llectiv
ely
tr
ain
a
m
o
d
el,
lev
er
ag
in
g
th
e
p
o
o
le
d
ex
p
er
t
is
e
o
f
all
p
ar
ticip
atin
g
en
titi
es
wh
ile
en
s
u
r
in
g
th
at
ea
ch
en
tity
’
s
d
ata
s
tay
s
p
r
iv
ate
[
1
6
]
.
F
e
a
t
u
r
e
s
H
F
L
V
F
L
F
TL
C
r
o
ss
-
s
i
l
o
F
L
D
a
t
a
d
i
s
t
r
i
b
u
t
i
o
n
D
a
t
a
i
s
d
i
st
r
i
b
u
t
e
d
h
o
r
i
z
o
n
t
a
l
l
y
,
w
i
t
h
e
a
c
h
c
l
i
e
n
t
h
a
v
i
n
g
a
s
u
b
s
e
t
o
f
t
h
e
t
o
t
a
l
d
a
t
a
D
a
t
a
i
s
d
i
st
r
i
b
u
t
e
d
v
e
r
t
i
c
a
l
l
y
,
w
i
t
h
e
a
c
h
c
l
i
e
n
t
h
a
v
i
n
g
a
su
b
s
e
t
o
f
t
h
e
f
e
a
t
u
r
e
s
D
a
t
a
a
c
r
o
ss
c
l
i
e
n
t
s
i
s
o
f
t
e
n
n
o
n
-
I
I
D
(
n
o
n
-
I
n
d
e
p
e
n
d
e
n
t
a
n
d
I
d
e
n
t
i
c
a
l
l
y
D
i
st
r
i
b
u
t
e
d
)
,
u
n
b
a
l
a
n
c
e
d
a
n
d
d
i
f
f
e
r
e
n
t
f
e
a
t
u
r
e
s
p
a
c
e
s
I
I
D
o
r
n
o
n
-
I
I
D
,
o
v
e
r
l
a
p
p
i
n
g
f
e
a
t
u
r
e
sp
a
c
e
s
M
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
Ty
p
i
c
a
l
l
y
u
ses
a
s
i
n
g
l
e
,
sh
a
r
e
d
m
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
a
c
r
o
ss
a
l
l
c
l
i
e
n
t
s
R
e
q
u
i
r
e
s
a
c
u
s
t
o
m
i
z
e
d
m
o
d
e
l
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
e
a
c
h
c
l
i
e
n
t
a
s
e
a
c
h
c
l
i
e
n
t
h
a
s a
u
n
i
q
u
e
set
o
f
f
e
a
t
u
r
e
s
D
i
f
f
e
r
e
n
t
a
r
c
h
i
t
e
c
t
u
r
e
s
,
f
i
n
e
-
t
u
n
i
n
g
S
a
me
a
r
c
h
i
t
e
c
t
u
r
e
a
c
r
o
ss
c
l
i
e
n
t
s
C
o
mm
u
n
i
c
a
t
i
o
n
o
v
e
r
h
e
a
d
H
i
g
h
c
o
mm
u
n
i
c
a
t
i
o
n
o
v
e
r
h
e
a
d
R
e
d
u
c
e
c
o
mm
u
n
i
c
a
t
i
o
n
o
v
e
r
h
e
a
d
Lo
w
e
r
,
b
u
t
v
a
r
i
a
b
l
e
H
i
g
h
e
r
,
b
u
t
p
r
e
d
i
c
t
a
b
l
e
M
o
d
e
l
a
c
c
u
r
a
c
y
H
i
g
h
m
o
d
e
l
A
c
c
u
r
a
c
y
R
e
q
u
i
r
e
a
d
d
i
t
i
o
n
a
l
t
e
c
h
n
i
q
u
e
t
o
a
c
h
i
e
v
e
h
i
g
h
m
o
d
e
l
a
c
c
u
r
a
c
y
V
a
r
i
a
b
l
e
,
d
e
p
e
n
d
s
o
n
t
r
a
n
sf
e
r
e
f
f
e
c
t
i
v
e
n
e
ss
G
e
n
e
r
a
l
l
y
h
i
g
h
,
c
o
n
si
s
t
e
n
t
i
mp
r
o
v
e
me
n
t
s
C
l
i
e
n
t
h
e
t
e
r
o
g
e
n
e
i
t
y
H
o
mo
g
e
n
o
u
s
c
l
i
e
n
t
w
i
t
h
si
mi
l
a
r
d
a
t
a
d
i
st
r
i
b
u
t
i
o
n
H
e
t
e
r
o
g
e
n
e
o
u
s
c
l
i
e
n
t
w
i
t
h
d
i
f
f
e
r
e
n
t
d
a
t
a
d
i
s
t
r
i
b
u
t
i
o
n
H
i
g
h
,
d
i
f
f
e
r
e
n
t
d
a
t
a
a
n
d
c
a
p
a
b
i
l
i
t
i
e
s
Lo
w
t
o
mo
d
e
r
a
t
e
,
si
mi
l
a
r
d
a
t
a
a
n
d
r
e
so
u
r
c
e
s
U
se
c
a
se
S
u
i
t
a
b
l
e
f
o
r
a
p
p
l
i
c
a
t
i
o
n
s
w
i
t
h
h
o
r
i
z
o
n
t
a
l
l
y
d
i
s
t
r
i
b
u
t
e
d
d
a
t
a
s
u
c
h
a
s
i
ma
g
e
s,
v
i
d
e
o
s
S
u
i
t
a
b
l
e
f
o
r
a
p
p
l
i
c
a
t
i
o
n
w
i
t
h
v
e
r
t
i
c
a
l
l
y
d
i
st
r
i
b
u
t
e
d
d
a
t
a
su
c
h
a
s fi
n
a
n
c
i
a
l
t
r
a
n
s
a
c
t
i
o
n
s,
m
e
d
i
c
a
l
r
e
c
o
r
d
s
P
e
r
so
n
a
l
i
z
e
d
,
d
o
m
a
i
n
-
sp
e
c
i
f
i
c
a
d
a
p
t
a
t
i
o
n
s
C
o
l
l
a
b
o
r
a
t
i
v
e
l
e
a
r
n
i
n
g
a
c
r
o
ss
o
r
g
a
n
i
z
a
t
i
o
n
s wi
t
h
si
mi
l
a
r
d
a
t
a
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
5
2
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
40
,
No
.
2
,
No
v
em
b
er
20
25
:
9
2
6
-
94
0
934
2
.
3
.
2
.
F
ina
nce
E
d
g
e
AI
is
a
tech
n
iq
u
e
th
at
m
ak
es
ed
g
e
d
e
v
ices
ca
p
ab
le
o
f
d
is
tr
ib
u
ted
in
tellig
e
n
ce
.
I
t
en
ab
les
th
e
d
ep
lo
y
m
e
n
t a
n
d
ex
ec
u
tio
n
o
f
AI
an
d
d
ata
an
aly
tics
at
th
e
n
etwo
r
k
’
s
ed
g
e,
awa
y
f
r
o
m
ce
n
tr
alize
d
d
ata
ce
n
ter
s
.
T
h
is
allo
ws
f
o
r
f
aster
d
ec
is
io
n
-
m
ak
in
g
an
d
lo
wer
laten
c
y
[
1
7
]
.
A
ML
m
eth
o
d
ca
lled
FL
u
s
es
a
n
etwo
r
k
o
f
d
ec
en
tr
alize
d
e
d
g
e
clien
ts
to
lear
n
f
r
o
m
in
o
r
d
er
to
cr
ea
te
a
g
lo
b
al
m
o
d
el.
Scalab
ilit
y
a
n
d
d
ata
p
r
iv
ac
y
ar
e
am
o
n
g
th
e
b
e
n
ef
its
it
p
r
o
v
id
e
s
.
B
u
t
wh
en
it
co
m
es
to
p
r
o
ce
s
s
in
g
co
m
p
lex
ity
,
it
d
o
es
ca
u
s
e
s
o
m
e
p
r
o
b
lem
s
wh
en
d
ea
lin
g
with
h
eter
o
g
en
e
o
u
s
d
ev
ices.
Fu
r
th
er
m
o
r
e
,
FL
en
ab
les
E
d
g
e
AI
ap
p
s
to
co
n
tin
u
o
u
s
ly
im
p
r
o
v
e
th
eir
u
n
d
er
s
tan
d
in
g
o
f
en
d
-
u
s
er
d
y
n
am
ics
with
o
u
t
h
a
v
in
g
to
t
r
an
s
f
er
e
n
d
-
u
s
er
d
ata
to
a
ce
n
tr
al
clo
u
d
s
to
r
e.
T
h
is
g
iv
es
en
d
u
s
er
s
an
ed
g
e
b
ec
au
s
e
th
ey
d
o
n
o
t
h
a
v
e
to
d
is
clo
s
e
s
en
s
itiv
e
in
f
o
r
m
atio
n
with
an
y
f
ir
m
.
T
h
is
en
s
u
r
es
th
at
n
o
p
er
s
o
n
al
d
ata
leav
es
th
e
d
ev
ice.
Fo
r
ex
am
p
le,
f
in
an
cial
o
r
g
a
n
izatio
n
s
m
i
g
h
t
tr
ain
a
m
o
d
el
co
o
p
er
ativ
e
ly
u
s
in
g
FL
,
wh
ich
allo
ws
th
em
to
ex
p
lo
it
t
h
e
a
g
g
r
eg
ate
i
n
tellig
e
n
ce
o
f
all
p
ar
ticip
atin
g
in
s
titu
tio
n
s
wh
ile
en
s
u
r
in
g
t
h
at
ea
ch
in
s
titu
tio
n
’
s
d
ata
s
tay
s
p
r
iv
ate
[
1
8
]
.
2
.
3
.
3
.
Sm
a
rt
h
o
m
es
E
d
g
e
AI
is
a
r
ev
o
lu
tio
n
ar
y
d
e
v
elo
p
m
en
t
in
AI
th
at
f
u
n
d
am
e
n
tally
ch
an
g
es
h
o
w
we
th
i
n
k
ab
o
u
t
d
ata
p
r
o
ce
s
s
in
g
an
d
d
e
v
ice
in
ter
ac
tio
n
.
W
h
e
th
er
it
’
s
an
ed
g
e
s
er
v
er
,
s
m
ar
tp
h
o
n
e,
o
r
o
th
er
I
o
T
d
ev
ice,
th
e
m
ag
ic
h
ap
p
en
s
r
ig
h
t
th
e
r
e
o
r
v
er
y
c
lo
s
e
to
th
e
d
ata
s
o
u
r
ce
.
T
h
is
ch
an
g
e
f
u
n
d
am
en
tally
af
f
ec
ts
h
o
w
q
u
ick
ly
an
d
ef
f
icien
tly
d
e
v
ices
ca
n
f
u
n
cti
o
n
,
n
o
t
ju
s
t
wh
er
e
th
ey
a
r
e
l
o
ca
ted
.
R
ea
l
-
tim
e
d
ec
is
i
on
-
m
ak
in
g
an
d
q
u
ick
er
r
ep
lies
ar
e
m
a
d
e
p
o
s
s
ib
le
b
y
E
d
g
e
AI
,
wh
ich
r
ed
u
ce
s
th
e
laten
cy
o
f
tr
an
s
f
er
r
i
n
g
d
ata
b
ac
k
a
n
d
f
o
r
t
h
t
o
r
em
o
te
co
m
p
u
ter
s
.
Fu
r
t
h
er
m
o
r
e,
d
ata
p
r
iv
ac
y
b
e
n
ef
its
g
r
ea
tly
f
r
o
m
th
is
s
p
ec
ialized
p
r
o
ce
s
s
in
g
.
E
d
g
e
AI
m
ax
im
izes
u
s
er
co
n
tr
o
l
b
y
r
ed
u
cin
g
th
e
i
n
h
er
en
t
d
an
g
e
r
s
ass
o
ciate
d
with
f
r
eq
u
e
n
t
d
ata
tr
an
s
f
er
s
t
o
ex
ter
n
al
s
er
v
er
s
b
y
s
to
r
in
g
cr
itical
in
f
o
r
m
atio
n
clo
s
er
to
h
o
m
e
[
1
9
]
.
C
o
n
v
e
r
s
ely
,
FL
allo
ws
f
o
r
jo
in
t
AI
tr
ain
in
g
with
o
u
t
jeo
p
ar
d
izin
g
th
e
p
r
iv
ac
y
o
f
p
er
s
o
n
al
in
f
o
r
m
atio
n
.
W
ith
th
is
M
L
tec
h
n
iq
u
e,
all
tr
ain
in
g
d
ata
is
r
etain
ed
o
n
th
e
o
r
i
g
in
al
d
ev
ice
an
d
a
s
h
ar
e
d
g
lo
b
al
m
o
d
el
is
lear
n
t
ac
r
o
s
s
s
ev
er
al
d
ev
ices.
T
h
is
p
r
o
m
o
tes
p
r
iv
ac
y
b
e
ca
u
s
e
n
o
r
aw
d
ata
is
s
h
ar
ed
o
r
k
ep
t
in
a
ce
n
tr
al
lo
ca
tio
n
.
On
e
ex
am
p
le
o
f
th
is
is
th
e
Fed
Ho
m
e
f
r
am
ewo
r
k
,
wh
ic
h
is
an
ar
ch
itectu
r
e
f
o
r
in
-
h
o
m
e
h
ea
lth
m
o
n
ito
r
in
g
b
ased
o
n
th
e
clo
u
d
e
d
g
e.
I
t
cr
ea
tes
a
s
h
ar
ed
g
l
o
b
al
m
o
d
el
in
t
h
e
clo
u
d
b
y
u
tili
zin
g
s
ev
e
r
al
h
o
u
s
es
at
th
e
n
etwo
r
k
e
d
g
es,
an
d
it
p
r
eser
v
es
u
s
er
p
r
iv
ac
y
b
y
s
to
r
i
n
g
u
s
er
d
ata
lo
ca
lly
.
2
.
3
.
4
.
T
elec
o
m
m
un
ica
t
io
n
I
n
th
e
co
n
te
x
t
o
f
telec
o
m
m
u
n
i
ca
tio
n
s
,
ed
g
e
AI
r
ef
e
r
s
to
th
e
ap
p
licatio
n
o
f
AI
alg
o
r
ith
m
s
t
o
n
etwo
r
k
ed
g
e
d
e
v
ices
lik
e
s
witch
es,
r
o
u
ter
s
,
an
d
o
th
e
r
d
ev
ices.
T
h
is
f
ac
ilit
ates
q
u
ick
er
an
s
wer
s
an
d
r
ea
l
-
tim
e
d
ec
is
io
n
-
m
ak
in
g
,
b
o
th
o
f
wh
i
ch
ar
e
ess
en
tial
in
a
co
m
m
u
n
icatio
n
s
n
etwo
r
k
.
Ad
d
itio
n
ally
,
E
d
g
e
AI
ca
n
h
elp
s
af
eg
u
ar
d
d
ata
p
r
iv
ac
y
b
ec
au
s
e
it
p
r
o
ce
s
s
es
d
ata
lo
ca
lly
r
a
th
er
th
an
tr
an
s
f
er
r
i
n
g
it
t
o
a
ce
n
tr
al
s
er
v
er
[
2
0
]
.
FL
is
b
ec
o
m
in
g
in
cr
ea
s
in
g
l
y
p
o
p
u
lar
in
th
e
telec
o
m
s
ec
t
o
r
as
co
m
m
u
n
icatio
n
s
er
v
ice
p
r
o
v
id
er
s
(
C
SP
s
)
co
n
s
id
er
h
o
w
t
o
lev
e
r
ag
e
th
eir
d
ata
ass
ets
wh
ile
u
p
h
o
ld
in
g
p
r
iv
ac
y
r
eg
u
latio
n
s
.
Ov
er
f
iv
e
b
illi
o
n
co
n
s
u
m
er
s
’
d
ata
ar
e
s
to
r
ed
b
y
th
e
to
p
5
0
ca
r
r
ier
s
wo
r
ld
wid
e.
T
h
e
u
t
ilizatio
n
o
f
FL
is
in
cr
ea
s
in
g
ly
im
p
o
r
ta
n
t
in
th
e
d
ev
elo
p
m
e
n
t
o
f
ce
n
tr
alize
d
m
o
d
els
with
d
is
tr
ib
u
ted
tr
ain
in
g
d
ata,
as
telec
o
m
m
u
n
icatio
n
co
m
p
an
ies
em
p
l
o
y
AI
/ML
tech
n
o
lo
g
y
to
ex
t
r
ac
t
an
aly
tical
an
d
p
r
ed
ictiv
e
ca
p
ab
ilit
ies.
Sig
n
if
ican
tly
m
o
r
e
n
etwo
r
k
ca
p
ac
ity
,
r
ed
u
ce
d
laten
cy
,
f
aster
s
p
ee
d
s
,
an
d
g
r
ea
ter
ef
f
icien
c
y
ar
e
m
a
d
e
p
o
s
s
ib
le
b
y
5
G
an
d
ed
g
e
c
o
m
p
u
tin
g
.
5
G
E
d
g
e
co
m
p
u
tin
g
will
d
is
p
er
s
e
d
ata
an
d
AI
m
o
d
els
am
o
n
g
n
u
m
er
o
u
s
n
o
d
es,
h
o
wev
er
s
h
ar
in
g
th
e
d
ata
ca
n
b
e
d
if
f
icu
lt d
u
e
to
s
ec
u
r
ity
,
b
an
d
wid
th
,
s
to
r
ag
e,
a
n
d
o
t
h
er
lim
itati
o
n
s
.
FL
is
p
er
f
ec
t f
o
r
th
is
k
i
n
d
o
f
s
ettin
g
.
2
.
3
.
5
.
Sm
a
rt
f
a
r
m
FL
-
b
ased
m
o
n
ito
r
in
g
s
y
s
tem
s
f
o
r
s
m
ar
t
f
ar
m
s
d
etec
t
a
n
im
al
d
is
ea
s
es.
Un
lik
e
p
r
io
r
s
tu
d
ies,
wh
ich
d
id
n
o
t
u
s
e
FL
f
o
r
an
im
al
d
is
ea
s
e
d
iag
n
o
s
is
,
th
is
tech
n
iq
u
e
is
b
ased
o
n
ex
ten
s
iv
e
e
x
p
er
im
en
tatio
n
with
in
f
o
r
m
atio
n
f
r
o
m
th
e
in
ter
n
et
o
f
a
n
im
al
h
ea
lth
t
h
in
g
s
(
I
o
A
HT
)
.
T
h
ese
s
tu
d
ies
o
n
clin
ica
l
m
asti
tis
in
co
ws
p
r
o
v
id
e
a
s
tr
o
n
g
f
o
u
n
d
atio
n
f
o
r
ass
ess
in
g
FL
’
s
ef
f
icac
y
in
ac
tu
al
ag
r
icu
ltu
r
al
s
ettin
g
s
.
Sm
a
r
t
f
ar
m
i
n
g
in
v
o
l
v
es
s
o
lar
-
p
o
wer
ed
s
en
s
o
r
s
attac
h
e
d
to
ea
c
h
an
im
al
t
o
m
o
n
ito
r
t
h
eir
h
ea
lth
.
T
h
e
i
n
f
o
r
m
atio
n
i
s
p
er
io
d
ically
s
en
t
v
ia
lo
n
g
-
r
a
n
g
e
(
L
o
R
a)
tr
an
s
m
is
s
io
n
to
ed
g
e
d
ev
ices,
s
u
ch
g
atew
ay
s
,
an
d
s
u
b
s
eq
u
en
tl
y
to
a
clo
u
d
s
er
v
er
.
Far
m
er
s
ar
e
ab
le
to
ef
f
ec
tiv
ely
o
v
er
s
ee
f
ar
m
o
p
er
atio
n
s
an
d
k
ee
p
an
ey
e
o
n
l
iv
esto
ck
th
an
k
s
to
th
is
in
f
r
astru
ctu
r
e.
W
h
ile
s
o
lar
-
p
o
wer
ed
s
en
s
o
r
s
o
f
f
e
r
lab
o
r
-
s
av
in
g
an
d
en
v
ir
o
n
m
e
n
tal
b
en
e
f
its
,
th
ey
also
p
o
s
e
s
u
b
s
tan
tial
p
r
o
b
lem
s
.
T
o
e
n
h
an
ce
s
m
ar
t
f
ar
m
in
g
au
to
m
atio
n
,
a
clo
u
d
s
er
v
e
r
ca
n
u
s
e
a
d
ee
p
lear
n
in
g
(
DL
)
m
o
d
el
to
an
aly
ze
d
ata
f
r
o
m
g
atew
ay
s
an
d
d
etec
t
d
is
ea
s
es
lik
e
m
asti
tis
in
ca
ttle.
T
h
is
ap
p
r
o
ac
h
p
r
o
tects
d
ata
p
r
i
v
ac
y
a
n
d
en
co
u
r
ag
es
s
u
s
tain
ab
le
ag
r
ic
u
ltu
r
al
p
r
ac
tices
b
y
en
h
an
cin
g
d
is
ea
s
e
p
r
ed
ictio
n
i
n
s
m
ar
t
f
ar
m
s
th
r
o
u
g
h
th
e
u
s
e
o
f
FL
.
L
o
ca
l
elem
e
n
ts
,
in
clu
d
i
n
g
s
en
s
o
r
s
,
ar
e
c
r
u
cial
to
im
p
r
o
v
in
g
FL
’
s
f
o
r
ec
ast
ac
cu
r
ac
y
[
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
F
ed
era
ted
lea
r
n
in
g
i
n
ed
g
e
A
I
:
a
s
ystema
tic
r
ev
iew
o
f a
p
p
lica
tio
n
s
…
(
C
h
r
is
tin
a
Th
a
n
ka
m
S
a
ja
n
)
935
2
.
4
.
P
ri
v
a
cy
t
hrea
t
s
in f
eder
a
t
ed
lea
rning
Priv
ac
y
f
laws
ar
e
am
o
n
g
th
e
m
o
s
t
co
m
m
o
n
wo
r
r
ies
ab
o
u
t
tr
ad
itio
n
al
ML
.
T
o
p
r
eser
v
e
th
eir
p
r
iv
ac
y
,
FL
r
eq
u
ests
th
at
p
ar
ticip
an
ts
c
o
n
tr
ib
u
te
th
e
lo
ca
l
tr
ain
in
g
m
o
d
el
p
ar
am
eter
s
r
ath
er
th
an
th
eir
ac
tu
al
d
ata.
T
h
e
d
an
g
er
s
p
r
esen
t
in
FL
ca
n
b
e
b
r
o
ad
ly
class
if
ied
in
to
m
a
n
y
ty
p
es
o
f
in
f
er
en
ce
-
b
ased
ass
au
lts
.
T
h
e
m
ain
p
r
iv
ac
y
p
r
o
tectio
n
is
s
u
es a
n
d
d
ata
s
ec
u
r
ity
r
is
k
s
th
at
FL
f
ac
ed
wh
ile
wo
r
k
in
g
at
E
C
in
clu
d
e
[
2
2
]
:
a.
Mo
d
el
in
v
er
s
io
n
attac
k
:
b
y
d
e
ce
n
tr
alizin
g
th
e
tr
ain
in
g
p
r
o
ce
s
s
an
d
en
ab
lin
g
d
e
v
ices
to
co
o
p
er
ativ
ely
lear
n
a
s
h
ar
ed
m
o
d
el
wh
ile
m
ain
tain
in
g
lo
ca
l
ac
ce
s
s
to
th
e
r
a
w
d
ata,
FL
s
ee
k
s
to
s
af
eg
u
ar
d
d
ata
p
r
iv
ac
y
.
Ho
wev
er
,
th
e
m
o
d
el
p
ar
am
ete
r
s
an
d
u
p
d
ates
ex
c
h
an
g
e
d
d
u
r
in
g
FL
ca
n
s
till
leak
in
f
o
r
m
atio
n
.
T
h
e
s
er
v
er
g
ath
er
s
u
p
d
ates
f
r
o
m
clien
t
m
o
d
els
in
FL
.
An
attac
k
e
r
c
an
d
ed
u
ce
d
etails
ab
o
u
t
t
h
e
tr
ain
in
g
s
et
b
y
ex
am
in
in
g
th
e
p
ar
am
eter
s
if
t
h
ey
m
an
a
g
e
to
o
b
tain
ac
ce
s
s
to
eith
er
th
e
g
lo
b
al
m
o
d
el
o
r
th
ese
u
p
d
ates.
T
h
e
ass
ailan
t
r
ec
o
n
s
tr
u
cts
th
e
in
p
u
t
d
ata
u
s
in
g
th
e
m
o
d
e
l
p
ar
am
eter
s
.
T
h
is
ca
n
b
e
ac
co
m
p
lis
h
ed
b
y
o
p
tim
izin
g
an
in
p
u
t
s
o
th
at
th
e
o
b
s
er
v
ed
o
u
t
p
u
ts
f
r
o
m
th
e
v
alid
m
o
d
e
co
in
cid
e
with
th
e
o
u
tp
u
ts
f
r
o
m
th
e
m
o
d
el
(
o
r
i
n
ter
m
ed
iate
la
y
er
o
u
tp
u
ts
)
.
A
n
attac
k
er
c
an
ex
p
l
o
it
th
ese
g
r
a
d
ien
ts
to
a
p
p
r
o
x
i
m
ate
th
e
tr
ain
i
n
g
d
ata
b
ec
au
s
e
FL
r
eq
u
ir
es e
x
ch
an
g
in
g
g
r
ad
ien
ts
o
r
m
o
d
el
u
p
d
ates
[
2
3
]
.
b.
Mo
d
el
p
o
is
o
n
i
n
g
:
m
o
d
el
p
o
is
o
n
in
g
attac
k
is
also
k
n
o
wn
as
ad
v
er
s
ar
ial
attac
k
.
A
m
alicio
u
s
clien
t
m
ig
h
t
s
en
d
m
an
ip
u
lated
m
o
d
el
u
p
d
ates
to
th
e
ce
n
tr
al
s
er
v
er
.
T
h
ese
u
p
d
ates
ca
n
b
e
cr
af
ted
to
d
eg
r
ad
e
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
g
lo
b
al
m
o
d
el.
T
h
is
co
u
ld
b
e
d
o
n
e
b
y
tr
ain
in
g
o
n
p
o
is
o
n
ed
d
ata
o
r
b
y
d
elib
er
ately
in
tr
o
d
u
cin
g
er
r
o
r
s
in
to
th
e
m
o
d
el
u
p
d
ates.
c.
B
ac
k
d
o
o
r
attac
k
s
:
in
th
is
s
ce
n
ar
io
,
a
n
attac
k
e
r
in
jects
a
h
id
d
e
n
b
ac
k
d
o
o
r
in
to
th
e
g
lo
b
al
m
o
d
el.
T
h
is
b
ac
k
d
o
o
r
ac
tiv
ates
wh
en
th
e
m
o
d
el
en
co
u
n
ter
s
s
p
ec
if
ic
tr
ig
g
er
in
p
u
ts
,
ca
u
s
in
g
it
to
b
eh
a
v
e
in
co
r
r
ec
tly
wh
ile
p
er
f
o
r
m
in
g
n
o
r
m
ally
o
n
r
e
g
u
lar
d
ata.
T
h
e
b
ac
k
d
o
o
r
ca
n
b
e
im
p
la
n
ted
b
y
s
u
b
tly
m
o
d
if
y
in
g
th
e
m
o
d
el
u
p
d
ates.
d.
Data
p
o
is
o
n
in
g
:
alth
o
u
g
h
th
i
s
is
n
o
t
a
r
e
f
ac
to
r
in
g
o
f
t
h
e
m
o
d
el
u
p
d
ates
th
em
s
elv
es,
d
ata
p
o
is
o
n
in
g
in
v
o
lv
es
in
jectin
g
m
alicio
u
s
d
ata
in
to
th
e
tr
ain
in
g
d
ataset.
T
h
is
ca
n
in
d
ir
ec
tly
ca
u
s
e
th
e
m
o
d
el
to
lear
n
in
co
r
r
ec
t o
r
h
ar
m
f
u
l p
atter
n
s
,
af
f
ec
tin
g
its
p
er
f
o
r
m
an
ce
.
e.
Mo
d
el
ex
tr
ac
tio
n
attac
k
:
in
FL
,
a
m
o
d
el
ex
tr
ac
tio
n
attac
k
o
cc
u
r
s
wh
en
a
h
o
s
tile
p
ar
t
y
u
s
es
th
eir
lo
ca
l
d
ata
to
tr
y
an
d
r
eb
u
ild
o
r
e
x
tr
ac
t
s
en
s
itiv
e
in
f
o
r
m
atio
n
ab
o
u
t
th
e
g
l
o
b
al
m
o
d
el
th
at
is
b
ein
g
tr
ai
n
ed
.
T
h
e
f
ac
t
th
at
FL
f
r
eq
u
e
n
tly
u
s
es
p
r
o
p
r
ietar
y
o
r
s
en
s
itiv
e
m
o
d
els
an
d
th
at
its
o
b
jectiv
e
is
to
p
r
eser
v
e
in
d
iv
id
u
al
d
ata
p
r
iv
ac
y
wh
ile
l
ea
r
n
in
g
a
g
lo
b
al
m
o
d
el
m
a
k
es th
is
ass
au
lt e
s
p
ec
ially
wo
r
r
is
o
m
e.
2
.
5
.
P
ri
v
a
cy
preserv
a
t
io
n t
e
chniq
ues
Glo
b
al
an
d
lo
ca
l
p
r
iv
ac
y
ar
e
th
e
two
ca
teg
o
r
ies
in
to
w
h
ich
FL
p
r
iv
ac
y
f
alls
.
E
v
e
r
y
iter
atio
n
,
all
u
n
r
eliab
le
th
ir
d
p
ar
ties
’
p
r
iv
ac
y
is
s
af
eg
u
ar
d
e
d
,
with
t
h
e
ex
ce
p
tio
n
o
f
th
e
tr
u
s
ted
c
en
tr
al
ag
g
r
e
g
atio
n
s
er
v
er
,
u
s
in
g
g
lo
b
al
p
r
iv
ac
y
r
eg
u
latio
n
s
an
d
lo
ca
lly
cr
ea
t
ed
m
o
d
el
u
p
d
ates.
Fo
r
th
e
s
er
v
e
r
p
r
iv
ac
y
to
b
e
p
r
o
tecte
d
,
m
o
d
el
ch
an
g
es
ar
e
n
ee
d
ed
f
o
r
lo
ca
l
p
r
iv
ac
y
.
C
u
r
r
en
tly
,
ad
v
er
s
ar
ial
tr
ain
in
g
(
AT
)
,
b
lo
ck
ch
ai
n
,
d
is
tu
r
b
an
ce
,
cr
y
p
to
g
r
ap
h
y
,
a
n
d
KD
ar
e
co
m
m
o
n
tech
n
o
lo
g
ies f
o
r
en
h
a
n
cin
g
FL
p
r
iv
ac
y
s
ec
u
r
ity
.
2
.
5
.
1
.
Dif
f
er
ent
ia
l priv
a
cy
t
e
chno
lo
g
y
Fu
zz
y
p
r
o
ce
s
s
in
g
is
co
m
m
o
n
ly
em
p
lo
y
ed
in
DM
L
to
s
af
eg
u
ar
d
tr
ain
in
g
d
ataset
’
s
p
r
iv
ac
y
an
d
s
ec
u
r
ity
.
T
h
is
in
clu
d
es
u
s
in
g
g
en
er
aliza
tio
n
,
n
o
is
e
d
is
tu
r
b
an
ce
,
r
an
d
o
m
izatio
n
an
d
co
m
p
r
ess
io
n
to
co
n
ce
al
tr
ain
in
g
d
ata
a
n
d
im
p
r
o
v
e
p
r
i
v
ac
y
p
er
f
o
r
m
an
ce
to
s
o
m
e
e
x
ten
t.
I
n
FL,
DP
is
ty
p
ically
e
m
p
lo
y
ed
to
d
is
g
u
is
e
s
ig
n
if
ican
t
f
ea
tu
r
es
b
y
a
d
d
in
g
n
o
is
e
to
tr
ain
in
g
d
ata,
m
o
d
el
p
ar
am
eter
s
,
o
r
g
r
a
d
ien
t
in
f
o
r
m
atio
n
.
DP
ca
n
en
s
u
r
e
d
ata
p
r
iv
ac
y
.
DP
im
p
r
o
v
es
d
ata
p
r
iv
ac
y
an
d
s
ec
u
r
it
y
b
y
ad
d
in
g
n
o
is
e
in
to
s
en
s
itiv
e
d
ata.
T
h
e
u
s
e
o
f
D
P
in
FL
to
in
tr
o
d
u
ce
n
o
is
e
d
is
tu
r
b
an
ce
s
to
m
o
d
el
p
a
r
a
m
eter
s
p
r
o
v
id
e
d
b
y
FL
p
ar
ti
cip
an
ts
,
o
r
to
ap
p
l
y
g
en
er
aliza
tio
n
m
eth
o
d
s
to
d
is
g
u
is
e
cr
itical
d
ata
f
ea
tu
r
es,
p
r
ev
en
ts
r
ev
er
s
e
d
ata
r
etr
ie
v
al,
a
llo
win
g
ML
m
o
d
els
to
to
ler
ate
ad
v
e
r
s
ar
ial
ex
am
p
l
es [
2
4
]
.
T
h
e
c
o
m
m
u
n
icatio
n
o
v
er
h
ea
d
o
f
SMPC
is
m
u
ch
h
ig
h
er
th
a
n
th
at
o
f
DP.
DP
alg
o
r
ith
m
s
h
av
e
b
ee
n
d
ev
elo
p
e
d
in
ex
is
tin
g
s
tu
d
ies.
An
alg
o
r
ith
m
f
o
r
DP
-
p
r
o
tecti
o
n
FL
o
p
tim
izatio
n
o
n
th
e
clien
t
s
id
e.
T
o
p
r
o
v
i
d
e
a
u
s
er
-
lev
el
DP
tr
ain
in
g
p
r
o
c
ess
f
o
r
lar
g
e
n
eu
r
al
n
etwo
r
k
s
,
a
u
s
er
-
lev
el
p
r
i
v
ac
y
p
r
o
tect
io
n
is
ad
d
itio
n
ally
in
co
r
p
o
r
ated
to
th
e
FL
a
v
er
a
g
in
g
m
eth
o
d
.
B
o
th
s
o
u
g
h
t
to
p
r
o
tect
p
r
i
v
ate
d
ata
b
y
m
ask
in
g
u
s
er
-
u
p
lo
ad
e
d
m
o
d
el
p
ar
a
m
eter
s
d
u
r
in
g
tr
ai
n
in
g
,
weig
h
i
n
g
m
o
d
el
p
er
f
o
r
m
an
ce
ag
ain
s
t
p
r
iv
ac
y
lo
s
s
.
T
h
ese
m
eth
o
d
s
wer
e
test
e
d
o
n
g
en
u
in
e
d
atasets
.
T
h
is
d
em
o
n
s
tr
ated
th
at
with
en
o
u
g
h
d
e
v
ices
p
ar
ticip
atin
g
in
f
ed
er
ated
tr
ain
in
g
,
p
r
iv
ac
y
p
r
o
tectio
n
ca
n
b
e
ac
h
i
ev
ed
with
lo
w
o
v
er
h
ea
d
.
B
o
th
ap
p
r
o
ac
h
es e
n
s
u
r
ed
h
ig
h
m
o
d
el
co
r
r
ec
tn
ess
.
Ho
wev
er
,
th
is
tech
n
iq
u
e
n
e
g
le
cted
to
co
n
s
id
er
th
e
p
o
s
s
ib
ilit
y
th
at
in
co
r
p
o
r
atin
g
DP in
FL
with
f
ewe
r
p
ar
ticip
an
ts
m
ay
im
p
air
o
v
e
r
a
ll
m
o
d
el
ac
cu
r
ac
y
.
DP
n
o
is
e
was
s
u
b
s
titu
ted
in
to
a
n
eu
r
al
n
etwo
r
k
b
y
p
r
u
n
in
g
a
s
p
ec
if
ic
lay
er
,
with
th
e
p
u
r
p
o
s
e
o
f
s
af
eg
u
ar
d
in
g
s
en
s
itiv
e
d
ata
f
r
o
m
lea
k
in
g
wh
ile
m
ain
tain
in
g
m
o
d
el
ac
cu
r
ac
y
.
T
h
e
r
e
is
a
n
o
v
el
p
r
i
v
ac
y
-
p
r
eser
v
in
g
lear
n
in
g
f
r
a
m
ewo
r
k
b
ased
o
n
g
r
ap
h
n
eu
r
al
n
etwo
r
k
s
(
GNNs)
.
T
h
e
f
r
am
ewo
r
k
p
r
o
v
id
es
a
f
o
r
m
al
p
r
iv
ac
y
g
u
ar
an
tee
b
y
u
tili
zin
g
ed
g
e
-
l
o
ca
l
DP
to
p
r
o
tect
n
o
d
e
f
ea
tu
r
es
an
d
ed
g
e
p
r
i
v
ac
y
.
T
h
e
s
y
s
tem
co
m
b
in
es a
GNN
wit
h
a
p
r
iv
ac
y
u
t
ilit
y
to
s
ec
u
r
e
u
s
er
d
ata
p
r
iv
ac
y
with
in
a
b
u
d
g
et.
Evaluation Warning : The document was created with Spire.PDF for Python.