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.
37
,
No
.
1
,
J
an
u
ar
y
20
25
,
p
p
.
2
8
8
~
2
9
9
I
SS
N:
2
502
-
4
7
52
,
DOI
: 1
0
.
1
1
5
9
1
/ijee
cs
.v
37.
i
1
.
p
p
288
-
2
9
9
288
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ee
cs
.
ia
esco
r
e.
co
m
A
co
ncep
tual a
pp
ro
a
ch of optimiza
tion in
federat
ed l
ea
rning
F
a
rha
nn
a
M
a
r’
i
1
,
Ahm
a
d Af
if
Su
pia
nto
2,
3
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
c
s
,
F
a
c
u
l
t
y
o
f
En
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
a
s
N
e
g
e
r
i
S
u
r
a
b
a
y
a
,
S
u
r
a
b
a
y
a
,
I
n
d
o
n
e
si
a
2
D
e
p
a
r
t
me
n
t
o
f
I
C
T
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s,
N
o
r
w
e
g
i
a
n
U
n
i
v
e
r
s
i
t
y
o
f
S
c
i
e
n
c
e
a
n
d
Te
c
h
n
o
l
o
g
y
,
Tr
o
n
d
h
e
i
m,
N
o
r
w
a
y
3
R
e
s
e
a
r
c
h
C
e
n
t
e
r
f
o
r
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
C
y
b
e
r
S
e
c
u
r
i
t
y
,
N
a
t
i
o
n
a
l
R
e
sea
r
c
h
a
n
d
I
n
n
o
v
a
t
i
o
n
A
g
e
n
c
y
(
B
R
I
N
)
,
J
a
k
a
r
t
a
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
May
23
,
2
0
2
4
R
ev
is
ed
Sep
3
,
202
4
Acc
ep
ted
Sep
7
,
2
0
2
4
F
e
d
e
r
a
te
d
l
e
a
r
n
i
n
g
(
F
L
)
i
s
a
n
e
m
e
r
g
i
n
g
a
p
p
r
o
a
c
h
t
o
d
i
s
t
r
i
b
u
t
e
d
l
e
a
rn
i
n
g
f
r
o
m
d
e
c
e
n
t
r
a
l
iz
e
d
d
a
ta
,
d
e
s
i
g
n
e
d
w
i
th
p
r
i
v
a
c
y
c
o
n
c
e
r
n
s
i
n
m
i
n
d
.
F
L
h
a
s
b
e
e
n
s
u
c
c
e
s
s
f
u
l
l
y
a
p
p
l
i
e
d
i
n
s
e
v
e
ra
l
f
ie
l
d
s
,
s
u
c
h
a
s
t
h
e
i
n
t
e
r
n
e
t
o
f
t
h
i
n
g
s
(
I
o
T
)
,
h
u
m
a
n
a
c
t
i
v
i
t
y
r
e
c
o
g
n
i
t
i
o
n
(
H
A
R
)
,
a
n
d
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
s
s
i
n
g
(
N
L
P
)
,
s
h
o
w
i
n
g
r
e
m
a
r
k
a
b
l
e
re
s
u
l
t
s
.
H
o
we
v
e
r
,
t
h
e
d
e
v
e
l
o
p
m
e
n
t
o
f
F
L
i
n
r
e
a
l
-
w
o
r
l
d
a
p
p
l
i
c
a
t
i
o
n
s
s
t
i
l
l
f
a
c
e
s
s
e
v
e
r
a
l
c
h
a
l
l
e
n
g
e
s
.
Re
c
e
n
t
o
p
t
i
m
i
z
a
t
i
o
n
s
o
f
F
L
h
a
v
e
b
e
e
n
m
a
d
e
t
o
a
d
d
r
e
s
s
t
h
e
s
e
i
ss
u
e
s
a
n
d
e
n
h
a
n
c
e
t
h
e
F
L
s
e
t
t
i
n
g
s
.
I
n
th
i
s
p
a
p
e
r
,
w
e
c
a
te
g
o
r
i
z
e
t
h
e
o
p
t
i
m
i
z
a
t
i
o
n
o
f
F
L
i
n
t
o
f
i
v
e
m
a
i
n
c
h
a
l
l
e
n
g
e
s
:
c
o
m
m
u
n
i
c
a
t
i
o
n
e
f
f
ic
i
e
n
c
y
,
h
e
te
r
o
g
e
n
e
i
t
y
,
p
r
i
v
a
c
y
a
n
d
s
e
c
u
r
i
t
y
,
s
c
a
la
b
i
l
i
t
y
,
a
n
d
c
o
n
v
e
r
g
e
n
c
e
r
a
t
e
.
W
e
p
r
o
v
i
d
e
a
n
o
v
e
r
v
i
e
w
o
f
v
a
r
i
o
u
s
o
p
t
i
m
i
z
a
t
i
o
n
f
r
a
m
e
w
o
r
k
s
f
o
r
F
L
p
r
o
p
o
s
e
d
i
n
p
r
e
v
i
o
u
s
r
e
s
e
a
r
c
h
,
i
l
l
u
s
t
ra
t
e
d
w
i
th
c
o
n
c
r
e
t
e
e
x
a
m
p
l
e
s
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s
b
a
s
e
d
o
n
t
h
e
s
e
f
i
v
e
o
p
t
i
m
i
z
a
t
io
n
g
o
a
l
s
.
A
d
d
i
t
i
o
n
a
l
l
y
,
w
e
p
r
o
p
o
s
e
t
w
o
o
p
t
i
o
n
a
l
i
n
t
e
g
r
a
t
e
d
c
o
n
c
e
p
t
u
a
l
f
r
a
m
e
w
o
r
k
s
(
C
F
s
)
f
o
r
o
p
t
i
m
i
z
i
n
g
F
L
b
y
c
o
m
b
i
n
i
n
g
s
e
v
e
r
a
l
o
p
t
i
m
i
z
a
t
i
o
n
m
e
t
h
o
d
s
t
o
a
c
h
i
e
v
e
t
h
e
b
e
s
t
i
m
p
l
e
m
e
n
t
a
t
i
o
n
o
f
F
L
t
h
a
t
a
d
d
r
e
s
s
e
s
t
h
e
f
i
v
e
c
h
a
l
le
n
g
e
s
.
K
ey
w
o
r
d
s
:
C
o
n
ce
p
tu
al
f
r
am
ewo
r
k
Dec
en
tr
alize
d
d
ata
Fed
er
ated
lear
n
in
g
Ma
ch
in
e
lear
n
in
g
Op
tim
izatio
n
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
:
Far
h
an
n
a
Ma
r
’
i
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atics,
Facu
lty
o
f
E
n
g
in
ee
r
i
n
g
,
Un
i
v
er
s
itas
Neg
er
i Su
r
ab
ay
a
Su
r
ab
ay
a
6
1
2
5
6
,
E
ast
J
av
a,
I
n
d
o
n
esia
E
m
ail: f
ar
h
an
n
a
m
ar
i@
u
n
esa.
ac
.
id
1.
I
NT
RO
D
UCT
I
O
N
M
a
c
h
i
n
e
l
ea
r
n
i
n
g
(
M
L
)
is
r
ev
o
l
u
t
i
o
n
i
z
i
n
g
v
a
r
i
o
u
s
d
o
m
ai
n
s
,
d
r
i
v
e
n
b
y
i
ts
a
b
il
i
t
y
t
o
a
n
a
l
y
z
e
v
a
s
t
a
m
o
u
n
t
s
o
f
d
a
t
a
a
n
d
d
e
r
i
v
e
a
ct
i
o
n
a
b
l
e
i
n
s
i
g
h
ts
.
As
a
k
e
y
a
r
e
a
w
it
h
i
n
a
r
ti
f
i
c
i
al
i
n
te
l
li
g
e
n
c
e
(
A
I
)
,
M
L
r
e
q
u
i
r
es
s
u
b
s
t
a
n
t
i
a
l
d
a
t
a
i
n
p
u
t
s
t
o
e
f
f
e
c
ti
v
e
l
y
t
r
ai
n
m
o
d
e
ls
a
n
d
m
a
k
e
p
r
e
d
i
c
t
i
o
n
s
.
I
n
t
h
e
p
r
a
t
i
ca
l
a
p
p
l
i
ca
t
i
o
n
s
o
f
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
(
F
L
)
,
p
a
r
t
i
c
u
l
a
r
l
y
w
ith
i
n
t
h
e
r
e
t
ai
l
i
n
d
u
s
t
r
y
,
M
L
l
ev
e
r
a
g
e
s
c
o
n
s
u
m
e
r
d
a
t
a
t
o
u
n
d
e
r
s
t
a
n
d
p
r
e
f
e
r
e
n
c
es
,
b
e
h
a
v
i
o
r
s
,
a
n
d
n
e
e
d
s
[
1
]
–
[
5
]
.
T
r
a
d
i
t
i
o
n
a
l
l
y
,
d
at
a
h
a
s
b
e
e
n
c
o
l
l
e
c
t
e
d
f
r
o
m
t
a
r
g
et
e
d
d
e
v
ic
e
s
a
n
d
p
r
o
c
e
s
s
e
d
c
e
n
t
r
a
l
l
y
o
n
c
l
o
u
d
-
b
a
s
e
d
s
e
r
v
e
r
s
o
r
d
a
t
a
c
e
n
t
e
r
s
.
T
h
is
d
a
t
a
i
s
t
h
e
n
u
s
e
d
t
o
d
e
v
e
l
o
p
k
n
o
w
l
e
d
g
e
a
n
d
i
n
f
e
r
e
n
c
e
m
o
d
e
l
s
,
s
e
r
v
i
n
g
as
t
r
a
i
n
i
n
g
f
o
r
ML
a
l
g
o
r
i
t
h
m
s
.
Ne
v
e
r
t
h
e
l
ess
,
th
o
s
e
a
p
p
r
o
a
c
h
e
s
w
e
r
e
n
o
l
o
n
g
e
r
e
f
f
e
c
t
i
v
e
b
e
c
a
u
s
e
d
a
t
a
i
s
c
e
r
t
a
i
n
l
y
s
e
n
s
it
i
v
e
f
o
r
t
h
e
o
w
n
e
r
,
a
n
d
s
e
v
e
r
a
l
p
o
l
i
ci
es
h
a
d
r
e
s
t
r
ic
t
e
d
a
c
c
ess
t
o
c
o
n
s
u
m
e
r
d
a
t
a
w
it
h
o
u
t
c
o
n
s
e
n
t
,
es
p
e
c
i
al
l
y
i
f
i
t
is
s
h
ar
e
d
w
i
t
h
t
h
i
r
d
p
a
r
t
i
es
.
O
n
e
e
x
a
m
p
l
e
o
f
t
h
e
r
es
t
r
i
ct
i
o
n
is
t
h
e
p
u
b
l
i
c
a
t
i
o
n
o
f
d
a
ta
p
r
i
v
a
c
y
a
n
d
s
e
c
u
r
it
y
p
o
l
i
c
y
i
n
it
i
a
t
e
d
b
y
T
h
e
E
u
r
o
p
e
a
n
U
n
i
o
n
i
n
t
h
e
g
e
n
e
r
al
d
a
t
a
p
r
o
t
e
c
ti
o
n
r
e
g
u
l
a
t
i
o
n
(
G
DPR
)
i
n
M
a
y
2
0
1
8
[
6
]
.
T
h
i
s
p
o
l
i
c
y
s
e
t
s
b
o
u
n
d
a
r
i
e
s
f
o
r
ML
p
r
o
c
e
s
s
t
o
g
e
t
d
a
t
a
i
n
p
u
t
w
h
i
c
h
is
a
n
im
p
o
r
t
a
n
t
p
o
i
n
t
i
n
i
ts
t
e
c
h
n
i
c
a
l
p
e
r
f
o
r
m
a
n
c
e
.
I
n
a
d
d
it
i
o
n
,
c
e
n
t
r
al
i
z
e
d
d
at
a
c
o
l
l
e
ct
i
o
n
i
n
t
h
e
d
a
t
a
c
e
n
t
e
r
w
i
ll
al
s
o
b
e
v
e
r
y
b
u
r
d
e
n
s
o
m
e
f
o
r
t
h
e
l
e
a
r
n
i
n
g
p
r
o
c
ess
d
u
e
t
o
l
i
m
it
e
d
c
e
n
te
r
r
es
o
u
r
c
e
s
.
I
t
wo
u
l
d
b
e
b
e
t
t
e
r
i
f
c
o
n
s
u
m
e
r
d
e
v
i
c
e
s
c
o
u
l
d
r
u
n
t
h
ei
r
l
e
a
r
n
i
n
g
p
r
o
c
e
s
s
,
b
e
c
a
u
s
e
,
i
n
to
d
a
y
'
s
m
o
d
e
r
n
e
r
a
,
t
h
e
d
e
v
i
c
es
w
e
r
e
e
q
u
i
p
p
e
d
wi
t
h
r
es
o
u
r
c
e
s
f
o
r
h
i
g
h
c
o
m
p
u
t
i
n
g
,
a
d
v
a
n
c
e
d
s
e
n
s
o
r
s
,
a
n
d
e
x
c
e
l
l
en
t
c
o
m
m
u
n
i
c
a
t
i
o
n
c
a
p
a
b
i
l
i
ti
e
s
[
7
]
.
M
u
c
h
r
e
s
e
a
r
c
h
h
a
s
b
e
e
n
d
o
n
e
t
o
o
v
e
r
c
o
m
e
t
h
o
s
e
p
r
o
b
l
e
m
s
a
n
d
t
u
r
n
t
h
e
m
i
n
t
o
o
p
p
o
r
t
u
n
i
t
i
e
s
f
o
r
a
n
e
w
e
r
a
o
f
ML
d
e
v
e
l
o
p
m
e
n
t
u
s
i
n
g
a
d
i
s
t
r
i
b
u
t
e
d
c
o
m
p
u
t
i
n
g
a
p
p
r
o
a
c
h
c
a
l
l
e
d
FL
[
8
]
–
[
1
0
]
.
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:
2
5
0
2
-
4
7
52
A
co
n
ce
p
tu
a
l a
p
p
r
o
a
ch
o
f o
p
timiz
a
tio
n
in
fed
era
ted
lea
r
n
in
g
(
F
a
r
h
a
n
n
a
Ma
r
’
i
)
289
FL
is
an
em
er
g
in
g
an
d
r
e
ce
n
t
ap
p
r
o
ac
h
in
ML
wh
ic
h
u
s
es
a
s
ettin
g
wh
e
r
e
m
a
n
y
clien
ts
co
llab
o
r
ativ
ely
tr
ain
a
m
o
d
el
u
s
in
g
d
ec
en
tr
aliza
tio
n
d
ata
[
1
1
]
.
T
h
ese
a
p
p
r
o
ac
h
es
also
s
u
p
p
o
r
t
in
f
o
r
m
atio
n
p
r
iv
ac
y
an
d
s
ec
u
r
ity
r
eg
a
r
d
in
g
its
p
o
lic
y
to
tr
ain
cu
s
to
m
er
s
’
d
ata
with
in
th
ei
r
m
o
b
ile
d
ev
ices,
n
o
t
s
to
r
e
d
o
n
th
e
s
er
v
er
[
1
2
]
–
[
1
5
]
.
Mo
r
eo
v
er
,
FL
p
r
o
v
id
es
t
h
e
ab
ilit
y
to
p
r
o
tect
d
ev
ice
p
r
iv
ac
y
wh
ile
en
s
u
r
in
g
h
ig
h
lear
n
in
g
p
er
f
o
r
m
an
ce
an
d
al
s
o
p
lay
s
a
n
im
p
o
r
tan
t
r
o
le
i
n
5
G
m
o
b
ile
ap
p
licatio
n
s
th
at
ar
e
s
en
s
itiv
e
to
co
m
p
u
tin
g
p
r
i
v
ac
y
,
s
u
ch
as
ed
g
e
an
d
ca
tch
in
g
co
m
p
u
tin
g
,
n
etwo
r
k
in
g
,
an
d
s
p
ec
tr
u
m
m
an
ag
em
e
n
t
[
1
6
]
.
B
ased
o
n
all
th
e
ad
v
an
tag
es
o
f
th
is
n
ew
ap
p
r
o
ac
h
,
FL
h
as
g
ain
ed
wid
esp
r
ea
d
atten
tio
n
f
r
o
m
p
r
ac
titi
o
n
er
s
an
d
r
esear
ch
er
s
in
v
ast
in
d
u
s
tr
ies.
I
n
co
n
tr
ast,
FL
h
as
b
ee
n
ap
p
lied
to
s
ev
er
al
to
o
ls
,
s
u
ch
as
Go
o
g
le
Key
b
o
ar
d
(
GB
o
ar
d
)
[
1
7
]
,
Hey
Go
o
g
le
[
1
8
]
,
an
d
Hey
Siri
[
1
9
]
.
I
n
ad
d
itio
n
,
FL
h
as
also
b
ee
n
ap
p
lie
d
to
in
s
titu
tio
n
s
th
at
h
o
ld
e
x
tr
a
s
en
s
itiv
e
d
ata,
s
u
ch
as
m
ed
ical
im
a
g
in
g
[
2
0
]
,
F
in
an
ce
Sp
ac
e
b
y
W
eBan
k
f
o
r
m
o
n
e
y
lau
n
d
er
in
g
d
etec
tio
n
[
2
1
]
,
also
f
i
n
an
cial
f
r
au
d
d
etec
tio
n
b
y
in
tel
an
d
co
n
s
ilien
t [
2
2
]
.
FL
co
n
d
u
cts
th
e
m
o
d
el
tr
ain
i
n
g
p
r
o
ce
s
s
lo
ca
lly
an
d
th
en
c
o
m
m
u
n
icate
s
with
th
e
s
er
v
er
to
p
er
f
o
r
m
ML
th
r
o
u
g
h
m
o
d
el
ag
g
r
e
g
atio
n
.
B
ec
au
s
e,
wh
at
is
s
en
t
to
th
e
s
er
v
er
is
o
n
ly
th
e
ag
g
r
eg
atio
n
r
esu
lt,
n
o
t
th
e
r
aw
d
ata
f
r
o
m
th
e
u
s
er
.
Desp
ite
its
g
o
als,
th
ese
s
ettin
g
s
h
ad
s
e
v
er
al
ch
allen
g
es,
an
d
th
e
m
ai
n
ch
allen
g
e
was
th
e
co
s
t
o
f
co
m
m
u
n
icatio
n
r
o
u
n
d
b
etwe
en
a
s
in
g
le
d
ev
ice
an
d
th
e
s
er
v
er
to
co
n
d
u
ct
th
e
m
o
d
el’
s
ag
g
r
eg
ates.
Pra
ctica
lly
,
th
e
co
m
m
u
n
icati
o
n
r
o
u
n
d
c
o
s
t
is
ar
g
u
a
b
ly
m
o
r
e
tim
e
-
c
o
n
s
u
m
in
g
th
a
n
an
iter
atio
n
o
f
th
e
alg
o
r
ith
m
its
elf
[
2
3
]
.
I
n
a
d
d
i
tio
n
,
an
o
t
h
er
ch
allen
g
e
is
th
at
h
eter
o
g
e
n
eo
u
s
d
e
v
ices
ca
u
s
e
im
b
alan
ce
d
an
d
u
n
s
tab
le
lear
n
in
g
an
d
co
m
m
u
n
icatio
n
p
r
o
ce
s
s
es.
T
h
e
s
y
s
tem
'
s
co
m
p
lex
ity
is
also
o
n
e
o
f
th
e
im
p
o
r
ta
n
t
ch
allen
g
es
FL
f
ac
es.
FL
wa
s
f
ir
s
t
in
tr
o
d
u
ce
d
to
s
u
p
p
o
r
t
lar
g
e
-
s
ca
le
d
is
tr
ib
u
ted
tr
ain
in
g
an
d
s
ec
u
r
ity
an
d
p
r
iv
ac
y
c
o
n
ce
r
n
s
.
T
h
e
f
ir
s
t
a
lg
o
r
ith
m
th
at
h
as
b
ee
n
u
s
ed
f
o
r
FL
is
f
ed
er
ated
a
v
er
ag
in
g
(
Fed
Av
g
)
,
wh
ich
co
m
b
in
es
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
SGD)
o
n
ea
ch
cu
s
to
m
er
with
a
s
er
v
er
th
at
ex
ec
u
t
es
m
o
d
el
av
er
a
g
in
g
[
2
4
]
.
T
h
o
s
e
r
esear
ch
h
as
s
u
cc
ee
d
ed
i
n
r
e
d
u
cin
g
th
e
r
o
u
n
d
o
f
co
m
m
u
n
icatio
n
n
ee
d
e
d
to
tr
ain
ML
o
n
d
is
tr
ib
u
ted
d
ata
an
d
d
ea
l
with
u
n
b
alan
ce
d
an
d
h
eter
o
g
en
e
o
u
s
d
ata
an
d
d
ev
ices.
FL
d
if
f
er
s
f
r
o
m
co
m
m
o
n
d
is
tr
ib
u
ted
lear
n
in
g
b
ec
au
s
e
t
h
e
d
is
tr
ib
u
tio
n
o
f
th
e
d
ata
is
n
o
t
u
n
if
o
r
m
an
d
in
d
e
p
en
d
e
n
t
(
n
o
n
-
I
I
D)
.
T
h
e
u
s
e
o
f
s
tan
d
ar
d
FL
m
eth
o
d
s
s
u
ch
as
Fed
AVG
was
o
f
ten
d
if
f
icu
lt
to
tu
n
e
an
d
d
e
m
o
n
s
tr
ated
its
in
ab
ilit
y
to
ac
h
iev
e
co
n
v
er
g
en
ce
[
2
5
]
,
[
2
6
]
.
So
it i
s
n
ec
ess
ar
y
to
o
p
tim
ize
th
e
s
ta
n
d
ar
d
m
eth
o
d
to
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
an
ce
o
f
FL.
Ho
wev
er
,
i
n
th
e
p
r
ac
tice
o
f
i
m
p
lem
en
tin
g
FL
in
m
a
n
y
ca
s
es,
th
er
e
we
r
e
s
ev
er
al
th
in
g
s
th
at
co
u
l
d
h
av
e
b
ee
n
im
p
r
o
v
ed
.
So
m
e
o
f
th
ese
p
r
o
b
lem
s
h
av
e
b
ee
n
tr
ied
to
b
e
s
o
lv
ed
b
y
p
er
f
o
r
m
in
g
o
p
tim
izatio
n
tech
n
iq
u
es
u
s
in
g
s
ev
er
al
m
et
h
o
d
s
b
ased
o
n
t
h
e
g
o
als
o
r
o
b
jectiv
es
to
b
e
ac
h
iev
ed
.
I
n
s
u
ch
a
way
,
s
ev
er
al
s
tu
d
ies
h
av
e
ca
r
r
ied
o
u
t
th
is
o
p
tim
izatio
n
tech
n
i
q
u
e
with
t
h
e
in
ten
t
th
at
FL
h
as
s
h
o
wn
b
etter
p
er
f
o
r
m
an
ce
.
Sev
er
al
ty
p
es
o
f
r
esear
ch
h
av
e
b
ee
n
co
n
d
u
cted
to
o
p
tim
ize
FL
in
ac
tu
al
ca
s
es.
On
e
o
f
th
o
s
e
is
th
e
u
s
e
o
f
r
ein
f
o
r
ce
m
e
n
t
lear
n
i
n
g
to
o
v
er
co
m
e
th
e
h
eter
o
g
en
eity
o
f
th
e
d
ata
s
am
p
le
t
h
at
is
n
o
t
in
d
ep
en
d
en
t
an
d
id
en
tically
d
is
tr
ib
u
ted
(
I
I
D)
alo
n
g
with
a
f
r
am
ew
o
r
k
,
n
a
m
ely
FAVOR
.
T
h
ese
f
r
a
m
ewo
r
k
s
co
u
ld
in
tellig
en
tly
ch
o
o
s
e
th
e
c
u
s
to
m
er
’
s
d
e
v
ice
to
p
ar
ticip
ate
in
ea
c
h
s
tag
e
o
f
th
e
d
is
tr
ib
u
tio
n
p
r
o
ce
s
s
an
d
h
av
e
b
ee
n
p
r
o
v
en
t
o
s
p
ee
d
u
p
co
n
v
er
g
en
ce
[
2
7
]
.
I
n
ad
d
itio
n
,
t
h
er
e
is
a
n
ad
a
p
tiv
e
o
p
tim
izer
v
er
s
io
n
ca
lled
A
DAGRA
D,
ADAM
,
an
d
YOGI
to
o
v
er
c
o
m
e
th
e
w
ea
k
n
ess
o
f
Fed
AVG.
Ad
ap
tiv
e
o
p
tim
izer
s
h
a
v
e
s
u
cc
ee
d
ed
i
n
s
o
lv
in
g
p
r
o
b
lem
s
in
n
o
n
-
f
e
d
er
ated
s
ettin
g
s
,
b
u
t
th
ey
also
co
u
ld
b
e
u
s
ed
in
FL
[
2
5
]
.
An
o
th
er
wo
r
k
o
f
o
p
tim
izatio
n
in
FL
is
Fed
NL
,
th
e
Fed
er
ated
New
t
o
n
L
ea
r
n
th
e
m
eth
o
d
th
at
a
ls
o
p
r
o
v
ed
to
r
ed
u
ce
co
m
m
u
n
icatio
n
co
s
t
an
d
co
m
p
lex
ity
c
o
m
p
ar
e
d
to
th
e
cr
itical
b
aselin
es [
2
8
]
.
T
h
e
o
p
tim
izatio
n
o
f
FL
its
elf
is
th
e
p
r
in
cip
al
r
esear
ch
t
o
s
u
p
p
o
r
t
th
e
d
e
v
elo
p
m
e
n
t
an
d
a
d
v
an
ce
m
en
t
o
f
FL
tech
n
iq
u
es
an
d
m
o
d
e
ls
.
B
ased
o
n
t
h
e
n
ew
ap
p
r
o
ac
h
in
tr
o
d
u
ce
d
f
o
r
o
p
tim
izat
io
n
in
FL
,
m
an
y
f
r
am
ewo
r
k
s
h
a
v
e
b
ee
n
p
r
o
p
o
s
ed
to
o
v
er
co
m
e
s
ev
e
r
al
p
r
o
b
lem
s
,
s
u
ch
as
C
o
m
m
u
n
icatio
n
E
f
f
icien
c
y
[
2
7
]
–
[
3
1
]
,
Hete
r
o
g
en
eity
[
3
2
]
,
Scalab
ilit
y
[
2
4
]
,
[
2
8
]
,
[
2
9
]
,
Priv
ac
y
an
d
Secu
r
ity
[
3
0
]
,
[
3
2
]
,
[
3
3
]
,
an
d
C
o
n
v
er
g
e
n
ce
s
r
ate
[
2
5
]
,
[
3
1
]
,
[
3
4
]
,
m
ix
p
r
o
b
lem
s
[
3
5
]
–
[
3
7
]
,
s
o
t
h
at
it
n
e
ed
s
to
b
e
s
p
ec
if
ied
an
d
ca
teg
o
r
ized
.
I
n
p
r
e
v
io
u
s
r
esear
ch
,
th
er
e
is
n
o
d
is
cu
s
s
io
n
o
f
th
e
v
ar
io
u
s
o
p
tim
izatio
n
m
eth
o
d
s
in
FL,
in
s
tead
,
t
h
e
f
o
cu
s
is
o
n
ly
o
n
o
n
e
o
r
two
p
r
o
b
lem
s
.
Ou
r
r
esear
c
h
will
f
o
cu
s
o
n
th
o
s
e
g
ap
s
,
a
im
ed
to
class
if
y
th
e
f
r
am
ewo
r
k
,
an
d
p
r
o
p
o
s
e
an
in
teg
r
ated
co
n
ce
p
tu
al
f
r
am
ew
o
r
k
b
ased
o
n
all
o
f
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
s
o
f
th
e
o
p
tim
i
za
tio
n
in
FL
to
g
et
th
e
b
est
im
p
lem
en
tatio
n
o
f
F
L
.
I
n
s
u
m
m
ar
y
,
we
p
r
esen
t
th
e
f
o
llo
win
g
co
n
t
r
ib
u
tio
n
s
to
t
h
is
r
esear
ch
,
(
1
)
we
d
ef
in
e
th
e
c
o
n
ce
p
t
an
d
i
n
tr
o
d
u
ce
a
ty
p
o
l
o
g
y
o
f
FL,
(
2
)
we
class
if
ied
th
e
p
r
o
p
o
s
ed
f
r
am
ew
o
r
k
o
f
o
p
tim
izatio
n
in
FL,
an
d
(
3
)
we
p
r
o
p
o
s
ed
an
in
teg
r
ate
d
co
n
ce
p
tu
al
f
r
a
m
ewo
r
k
f
o
r
o
p
tim
izatio
n
in
FL.
T
h
is
r
esear
ch
is
s
tr
u
ctu
r
ed
in
to
f
o
u
r
s
ec
tio
n
s
.
Sectio
n
1
is
ab
o
u
t
th
e
b
ac
k
g
r
o
u
n
d
a
n
d
th
e
r
elate
d
r
esear
ch
.
Sectio
n
2
illu
s
tr
ates
th
e
m
eth
o
d
s
,
s
ec
tio
n
3
d
ef
in
es
th
e
r
esu
lts
an
d
d
is
cu
s
s
io
n
,
w
e
class
if
ie
s
th
e
f
r
am
ewo
r
k
o
f
o
p
tim
izatio
n
in
FL,
an
d
we
p
r
o
p
o
s
ed
an
in
teg
r
ated
co
n
ce
p
tu
al
f
r
am
ew
o
r
k
f
o
r
Op
tim
izatio
n
in
FL.
Fin
ally
,
t
h
e
m
ain
f
in
d
in
g
s
a
r
e
o
u
tlin
ed
in
s
ec
tio
n
4
.
2.
M
E
T
H
O
D
Th
e
d
ev
elo
p
m
en
t
o
f
co
n
ce
p
tu
al
f
r
am
ewo
r
k
s
(
C
Fs
)
in
C
o
m
p
u
ter
Scien
ce
r
ep
r
esen
ts
a
s
et
o
f
co
n
ce
p
ts
f
o
r
ar
tic
u
latin
g
r
e
q
u
ir
em
e
n
ts
an
d
e
x
am
in
in
g
th
e
s
y
s
tem
ar
ch
itectu
r
es
[
3
8
]
.
T
h
e
C
FS
d
ev
elo
p
ed
in
th
is
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
288
-
2
9
9
290
r
esear
ch
is
an
o
p
tim
izatio
n
f
o
r
FL
b
ased
o
n
m
o
d
els
d
ev
elo
p
m
en
t.
Hen
ce
,
a
m
u
lti
-
ca
s
e
s
tu
d
y
ap
p
r
o
ac
h
is
u
s
ed
to
ev
alu
ate
th
e
o
p
tim
izatio
n
o
f
FL
.
T
h
e
r
esear
ch
m
eth
o
d
o
lo
g
y
c
o
n
s
is
ts
o
f
s
ev
er
al
k
e
y
s
tep
s
to
ad
d
r
ess
an
d
o
p
tim
ize
FL
b
y
s
y
s
tem
atica
lly
r
ev
iewin
g
an
d
an
aly
zin
g
ex
is
t
in
g
liter
atu
r
e
an
d
f
r
am
ewo
r
k
s
.
T
h
e
a
p
p
r
o
ac
h
ca
n
b
e
d
etailed
as f
o
llo
ws
an
d
v
is
u
alize
d
in
Fig
u
r
e
1
.
Fig
u
r
e
1
.
R
esear
ch
m
eth
o
d
s
A.
C
o
llectio
n
an
d
an
aly
s
is
o
f
f
o
u
n
d
atio
n
al
liter
atu
r
e
Ar
ticle
co
llectio
n
:
Gath
er
a
c
o
m
p
r
eh
e
n
s
iv
e
s
et
o
f
ar
ticles
t
h
at
d
ef
in
e
FL
an
d
o
u
tlin
e
its
ass
o
ciate
d
wea
k
n
ess
es.
T
h
is
in
v
o
lv
es
s
el
ec
tin
g
r
esear
ch
p
ap
e
r
s
,
r
ev
iew
s
,
an
d
s
u
r
v
ey
s
th
at
s
p
ec
if
ically
f
o
cu
s
o
n
t
h
e
co
r
e
co
n
ce
p
ts
an
d
c
h
allen
g
es o
f
FL.
B.
E
x
p
lo
r
atio
n
o
f
o
p
tim
izatio
n
te
ch
n
iq
u
es
Op
tim
izatio
n
r
ev
iew
:
C
o
n
d
u
c
t
an
e
x
ten
s
iv
e
r
ev
iew
o
f
p
u
b
l
icatio
n
s
d
is
cu
s
s
in
g
v
ar
i
o
u
s
o
p
tim
izatio
n
tech
n
iq
u
es
aim
ed
at
ad
d
r
ess
in
g
th
e
wea
k
n
ess
es
id
en
tifie
d
in
FL.
T
h
is
in
v
o
lv
es
s
tu
d
y
in
g
d
if
f
e
r
en
t
o
p
tim
izatio
n
s
tr
ateg
ies an
d
m
e
th
o
d
s
p
r
o
p
o
s
ed
in
th
e
liter
atu
r
e
to
s
o
lv
e
FL'
s
s
p
ec
if
ic
is
s
u
es.
C.
I
d
en
tific
atio
n
o
f
co
r
e
p
r
o
b
lem
s
Pro
b
lem
id
en
tific
atio
n
:
An
aly
ze
th
e
g
ath
er
ed
r
esear
ch
t
o
id
en
tify
an
d
ca
teg
o
r
ize
th
e
p
r
im
ar
y
p
r
o
b
lem
s
ass
o
ciate
d
with
FL.
D.
C
ateg
o
r
izatio
n
an
d
i
n
teg
r
atio
n
-
C
ateg
o
r
izatio
n
o
f
o
p
tim
izatio
n
tech
n
iq
u
es
:
C
lass
if
y
th
e
v
ar
io
u
s
o
p
tim
izatio
n
m
et
h
o
d
s
ac
co
r
d
in
g
to
th
e
f
iv
e
id
en
tifie
d
wea
k
n
ess
es.
T
h
is
s
tep
in
v
o
lv
es
m
ap
p
in
g
ea
ch
o
p
tim
izatio
n
tech
n
iq
u
e
to
th
e
s
p
ec
if
ic
p
r
o
b
lem
s
it a
d
d
r
ess
es.
-
Fra
m
ewo
r
k
i
n
teg
r
atio
n
:
Dev
el
o
p
a
co
m
p
r
eh
e
n
s
iv
e
f
r
a
m
ewo
r
k
b
y
co
m
b
in
in
g
th
e
id
en
tifie
d
o
p
tim
izatio
n
m
eth
o
d
s
.
T
h
e
g
o
al
is
to
cr
ea
te
a
u
n
if
ied
a
p
p
r
o
ac
h
t
h
at
a
d
d
r
ess
es
all
f
iv
e
p
r
o
b
lem
s
with
in
a
s
in
g
l
e
f
r
am
ewo
r
k
,
th
er
e
b
y
en
h
an
cin
g
FL's o
v
er
all
p
er
f
o
r
m
an
ce
.
B
y
f
o
llo
win
g
th
ese
s
tep
s
,
th
e
r
esear
ch
aim
s
t
o
d
e
v
elo
p
a
n
in
teg
r
ate
d
f
r
am
ewo
r
k
th
at
ef
f
ec
tiv
ely
r
eso
lv
es th
e
co
r
e
is
s
u
es o
f
FL
th
r
o
u
g
h
a
s
y
s
tem
atic
co
m
b
in
at
io
n
o
f
e
x
is
tin
g
o
p
tim
izatio
n
tech
n
iq
u
es
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
r
esear
ch
b
eg
in
s
with
a
d
o
m
ain
an
aly
s
is
to
b
u
ild
a
co
n
ce
p
t
b
ased
o
n
liter
atu
r
e
r
ev
iews
in
s
ev
er
al
d
ig
ital
p
u
b
licatio
n
s
,
s
u
ch
a
s
Scien
ce
Dir
ec
t,
Sp
r
in
g
er
,
Ar
x
iv
.
o
r
g
,
AC
M,
an
d
I
E
E
E
.
Pre
v
io
u
s
ly
,
th
e
s
u
s
tain
ab
ilit
y
o
f
r
esear
ch
in
FL
will
b
e
s
h
o
wn
b
ase
d
o
n
th
e
n
u
m
b
er
o
f
ar
ticl
es
co
m
b
in
atio
n
o
n
r
ev
iews/s
u
r
v
ey
,
o
r
ig
in
al
r
ese
ar
ch
,
b
o
o
k
ch
ap
ter
s
,
an
d
p
r
o
ce
ed
in
g
s
th
at
wer
e
p
u
b
lis
h
ed
u
s
in
g
k
ey
wo
r
d
FL
s
tar
tin
g
in
th
e
y
ea
r
o
f
FL
was
in
tr
o
d
u
ce
d
f
r
o
m
2
0
1
6
to
2
0
2
3
to
s
h
o
w
an
in
cr
ea
s
in
g
i
n
ter
est
in
th
e
to
p
ic
o
f
FL
in
Fig
u
r
e
2
s
tar
tin
g
f
r
o
m
Au
g
u
s
t
2
,
2
0
2
4
.
B
ased
o
n
th
e
F
ig
u
r
e
2
,
a
s
ig
n
if
ican
t
in
cr
ea
s
e
o
cc
u
r
s
ev
er
y
y
ea
r
,
wh
ich
in
d
icate
s
th
at
FL
i
s
a
p
o
p
u
lar
an
d
h
ig
h
ly
d
e
v
elo
p
ed
to
p
ic.
I
n
th
e
d
o
m
ai
n
an
al
y
s
is
s
tag
e,
we
f
ir
s
t
s
ea
r
ch
ed
f
o
r
a
r
ticles
th
at
in
clu
d
ed
FL
in
th
e
titl
e
an
d
k
e
y
wo
r
d
s
.
T
h
en
,
we
s
elec
ted
p
a
p
er
s
th
at
th
o
r
o
u
g
h
ly
d
ef
in
ed
FL
an
d
c
o
n
clu
d
ed
th
e
d
ef
in
itio
n
o
f
FL
b
ased
o
n
it
s
ex
p
lan
atio
n
.
T
h
e
d
ef
in
itio
n
s
tu
d
y
s
h
o
wed
FL'
s
m
ain
p
er
s
p
ec
tiv
es a
n
d
co
n
ce
p
t
s
,
esp
ec
ially
in
th
e
in
ter
n
et
o
f
th
in
g
s
(
I
o
T
)
,
wir
eless
co
m
m
u
n
icatio
n
,
an
d
f
u
tu
r
e
g
en
er
atio
n
c
o
m
p
u
ter
s
y
s
tem
s
.
B
ased
o
n
th
is
an
aly
s
is
,
we
als
o
h
av
e
t
o
b
u
ild
a
ty
p
o
lo
g
y
o
f
o
p
tim
izatio
n
in
FL.
3
.
1.
Wha
t
is
f
eder
a
t
ed
lea
rn
ing
?
T
h
e
ter
m
FL
,
f
ir
s
t
i
n
tr
o
d
u
ce
d
b
y
Mc
Ma
h
an
et
a
l.
[
3
9
]
,
is
o
n
e
o
f
th
e
ML
s
ettin
g
s
in
wh
ich
s
ev
er
al
p
ar
ticip
ated
d
e
v
ices
o
r
clien
ts
s
o
lv
e
th
e
m
o
d
el
lear
n
in
g
task
an
d
ar
e
c
o
o
r
d
i
n
ated
b
y
a
s
er
v
er
.
C
lien
t'
s
d
ev
ices
p
r
o
ce
s
s
th
eir
o
wn
lo
ca
l
m
o
d
el
with
task
s
cr
ip
t
in
s
tr
u
ctio
n
f
r
o
m
th
e
ce
n
tr
al
s
er
v
er
an
d
th
e
n
s
en
d
it
b
ac
k
.
Nex
t,
th
e
ce
n
tr
al
s
er
v
er
d
o
es
th
e
a
g
g
r
eg
atio
n
p
r
o
ce
s
s
u
s
in
g
t
h
e
lo
ca
l
tr
ain
in
g
m
o
d
el.
Fin
ally
,
t
h
e
ce
n
tr
al
s
er
v
er
will
s
en
d
th
e
g
lo
b
al
m
o
d
el
in
f
o
r
m
atio
n
to
its
clien
t.
I
n
th
e
p
r
ac
tical
ap
p
licatio
n
,
th
er
e
ar
e
two
s
ettin
g
s
:
cr
o
s
s
-
d
ev
ice
FL,
wh
ich
tar
g
ets
ML
f
o
r
m
o
b
ile
d
ev
ices,
an
d
cr
o
s
s
-
s
ilo
FL,
wh
ich
tar
g
ets
ML
f
o
r
s
ev
er
al
o
r
g
an
izatio
n
s
[
1
4
]
.
An
o
v
er
v
iew
o
f
th
e
FL
f
r
am
ewo
r
k
is
p
r
esen
ted
in
Fig
u
r
e
3
.
T
h
is
s
et
was
m
ad
e
as
a
f
o
llo
w
-
u
p
ap
p
licatio
n
o
f
s
ev
er
al
p
o
licies
r
elate
d
to
co
n
s
u
m
e
r
d
ata
s
ec
u
r
ity
.
T
h
u
s
,
th
e
d
ata
tr
ain
in
g
p
r
o
ce
s
s
o
n
ML
will
b
e
ca
r
r
ied
o
u
t
lo
ca
lly
o
n
th
e
clien
t'
s
m
o
b
ile
d
ev
ice
r
ath
er
th
an
s
en
d
in
g
it
to
th
e
d
ata
ce
n
ter
f
o
r
tr
ain
in
g
[
4
0
]
.
T
h
e
u
s
e
o
f
lo
ca
l
d
ata
in
r
ed
u
cin
g
th
e
r
is
k
o
f
attac
k
s
o
n
th
e
s
er
v
er
an
d
th
e
u
s
e
o
f
f
o
cu
s
ed
an
d
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:
2
5
0
2
-
4
7
52
A
co
n
ce
p
tu
a
l a
p
p
r
o
a
ch
o
f o
p
timiz
a
tio
n
in
fed
era
ted
lea
r
n
in
g
(
F
a
r
h
a
n
n
a
Ma
r
’
i
)
291
tem
p
o
r
ar
y
d
ata
m
in
im
izatio
n
m
u
s
t
b
e
d
o
n
e
in
to
d
ay
'
s
er
a
ca
u
s
e
tech
n
o
l
o
g
ical
d
ev
elo
p
m
en
ts
g
o
h
an
d
in
h
a
n
d
with
in
cr
ea
s
in
g
ly
s
o
p
h
is
ticated
lev
els
o
f
cy
b
e
r
cr
im
e
[
4
1
]
,
u
p
d
ates
an
d
in
itial
ag
g
r
e
g
atio
n
f
o
llo
ws
th
e
p
r
in
cip
le
o
f
d
ata
m
in
im
izatio
n
[
1
4
]
,
wh
ich
was
p
r
o
p
o
s
ed
b
y
th
e
2
0
1
2
W
h
ite
Ho
u
s
e
r
ep
o
r
t
o
n
th
e
p
r
i
v
ac
y
o
f
co
n
s
u
m
er
d
ata
[
4
2
]
.
C
o
n
s
u
m
e
r
s
tr
u
s
t
co
m
p
an
ies
to
s
h
ar
e
ess
en
tial
d
ata
r
elate
d
to
p
u
b
lic
o
r
p
r
i
v
ate
m
atter
s
s
o
th
at
co
m
p
an
ies
ca
n
u
s
e
th
e
d
ata
to
cu
s
to
m
ize
co
n
s
u
m
er
s
'
p
r
ef
er
en
ce
s
.
T
h
u
s
,
th
e
lev
el
o
f
co
n
s
u
m
er
s
atis
f
ac
tio
n
is
m
et.
Hen
ce
,
co
m
p
an
ies
m
u
s
t
en
s
u
r
e
th
at
cu
s
to
m
er
d
ata
r
em
ain
s
s
af
e
an
d
av
o
id
th
ese
attac
k
s
.
FL
is
n
ee
d
ed
to
f
u
lf
ill
th
i
s
asp
ec
t
in
lin
e
with
th
e
p
o
p
u
lar
ity
o
f
im
p
lem
en
tin
g
th
e
I
o
T
,
wir
eless
co
m
m
u
n
icatio
n
,
an
d
f
u
tu
r
e
g
e
n
er
atio
n
c
o
m
p
u
te
r
s
y
s
tem
s
.
On
e
o
f
th
e
f
u
tu
r
e
ar
ea
s
o
f
FL
i
m
p
lem
en
tatio
n
is
th
e
I
o
T
[
4
3
]
.
I
o
T
r
eq
u
ir
es
an
in
cr
ed
ib
le
am
o
u
n
t
o
f
d
e
v
ices
to
co
m
p
lem
en
t
ea
ch
o
th
er
'
s
in
ten
d
ed
u
s
e
b
etwe
en
o
n
e
d
ev
ice
an
d
.
L
ik
ewise,
th
e
d
a
ta
g
en
er
ated
f
r
o
m
th
ese
d
ev
i
ce
s
will
al
s
o
b
e
o
v
er
lo
ad
ed
if
d
ep
o
s
ited
o
n
a
ce
n
tr
alize
d
s
er
v
er
.
A
r
ec
en
t
f
o
r
ec
ast
esti
m
ates
th
at
4
1
.
6
b
illi
o
n
co
n
n
ec
ted
I
o
T
d
e
v
ices
will
g
en
er
ate
7
9
.
4
ze
ttab
y
tes
(
Z
B
)
o
f
d
ata
in
2
0
2
5
[
4
4
]
.
As
a
tec
h
n
iq
u
e
to
o
v
er
co
m
e
th
e
p
h
en
o
m
en
a
m
en
tio
n
ed
ab
o
v
e,
FL
is
th
e
b
est
way
to
o
v
er
co
m
e
th
em
.
I
n
th
e
f
ir
s
t
s
tag
e,
th
e
u
s
er
’
s
d
ev
ice
in
d
ep
en
d
en
tly
p
r
o
ce
s
s
es
lo
ca
l
co
m
p
u
tatio
n
iter
ativ
ely
u
s
in
g
th
eir
d
atasets
.
T
h
e
r
esu
lt
o
f
its
p
r
o
ce
s
s
,
th
a
t
is
,
th
e
lo
ca
l
lear
n
in
g
m
o
d
el
u
p
d
ates,
wo
u
ld
b
e
s
en
t
to
th
e
ag
g
r
eg
atio
n
s
er
v
er
f
o
r
th
e
g
lo
b
al
s
tr
ateg
y
.
I
n
th
is
s
ettin
g
,
ag
g
r
eg
atio
n
ca
n
b
e
d
o
n
e
b
o
th
at
th
e
ed
g
e
o
r
clo
u
d
[
4
5
]
.
Fig
u
r
e
2
.
Nu
m
b
er
o
f
ar
ticles
p
u
b
lis
h
ed
in
2
0
1
6
-
2
0
2
3
Fig
u
r
e
3
.
An
o
v
er
v
iew
o
f
FL
3
.
2
.
O
pti
m
iza
t
io
n in
f
eder
a
t
ed
lea
rning
FL
p
r
o
b
lem
ca
n
b
e
d
en
o
ted
as
in
(
1
)
.
T
h
e
f
ir
s
t
alg
o
r
ith
m
u
s
ed
to
s
o
lv
e
th
e
FL
p
r
o
b
lem
is
i
n
s
p
ir
ed
by
SGD
,
wh
ich
is
th
e
f
av
o
r
e
d
m
e
th
o
d
f
o
r
o
p
tim
izatio
n
.
m
in
∈
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
288
-
2
9
9
292
{
(
)
≔
1
∑
(
)
=
1
}
(
1
)
I
n
f
o
r
m
atio
n
:
=
d
im
en
s
io
n
o
f
th
e
m
o
d
e
l
∈
=
to
tal
n
u
m
b
er
o
f
d
e
v
ice
s
(
)
=
th
e
lo
s
s
[
]
=
d
ata
s
to
r
ed
in
m
ac
h
in
e
(
)
=
th
e
em
p
ir
ical
lo
s
s
SGD
is
ap
p
lied
to
th
e
f
ed
er
ate
d
o
p
tim
izatio
n
p
r
o
b
lem
with
a
n
ad
ap
tatio
n
th
at
is
th
e
s
elec
ti
o
n
o
f
a
C
-
f
r
ac
tio
n
o
f
clien
ts
o
n
ea
ch
r
o
u
n
d
,
th
en
co
m
p
u
tin
g
th
e
g
r
a
d
ien
t
o
f
th
e
lo
s
s
o
v
er
all
th
e
d
ata
o
wn
ed
b
y
th
ese
clien
ts
,
s
o
C
co
n
tr
o
ls
th
e
g
l
o
b
al
b
atch
s
ize,
with
C
=1
s
u
itab
le
to
f
u
ll
-
b
atch
g
r
ad
ien
t
d
escen
t,
th
is
ca
lled
Fed
SGD
[
4
6
]
.
Ho
wev
er
,
o
p
tim
izatio
n
o
f
FL
is
n
ee
d
e
d
to
g
et
th
e
b
est
FL
m
o
d
el
th
at
ca
n
b
e
ap
p
lied
ac
c
o
r
d
in
g
to
th
e
m
ain
p
u
r
p
o
s
e
o
f
f
o
r
m
i
n
g
th
e
FL
m
o
d
el.
T
h
e
m
ai
n
o
b
jectiv
es
o
f
th
e
v
ar
io
u
s
o
p
tim
izatio
n
s
th
at
ex
is
t
v
ar
y
ac
co
r
d
i
n
g
to
t
h
e
n
ee
d
s
o
f
ea
ch
ap
p
licatio
n
.
Her
e
we
c
ateg
o
r
ize
s
ev
er
al
ty
p
es
o
f
o
p
ti
m
izatio
n
s
b
ased
o
n
th
e
m
ain
n
ee
d
s
to
b
e
ac
h
ie
v
ed
.
3
.
2
.
1
.
Co
mm
un
ica
t
io
n
ef
f
icie
ncy
FL
r
eq
u
ir
es
co
m
m
u
n
icatio
n
co
s
ts
b
etwe
en
lo
ca
l
cl
ien
ts
an
d
s
er
v
er
s
to
s
en
d
an
d
u
p
d
ate
th
e
ag
g
r
eg
atio
n
m
o
d
el
in
t
h
e
l
ea
r
n
in
g
p
r
o
ce
s
s
.
So
th
at
m
i
n
im
izin
g
co
m
m
u
n
icatio
n
co
s
ts
is
an
ess
en
tia
l
r
eq
u
ir
em
e
n
t
to
d
o
c
o
n
s
id
er
in
g
FL
d
ep
en
d
s
o
n
th
e
co
m
m
u
n
ic
atio
n
r
elatio
n
s
h
ip
.
Sev
er
al
ap
p
r
o
ac
h
es
h
a
v
e
b
ee
n
tak
en
to
m
in
im
ize
c
o
m
m
u
n
icatio
n
co
s
ts
,
in
clu
d
in
g
u
s
in
g
a
m
eth
o
d
o
f
u
p
lo
ad
in
g
th
e
g
r
ad
ien
t
lo
ca
ted
in
a
p
ar
ticu
lar
in
ter
v
al
clip
p
ed
b
y
s
o
m
e
th
r
esh
o
ld
v
alu
es
[
4
7
]
,
s
ca
lin
g
d
o
wn
th
e
u
s
ed
p
ar
am
e
ter
s
b
y
m
in
im
izin
g
th
e
co
m
p
lex
ity
o
f
th
e
n
eu
r
al
n
etwo
r
k
m
o
d
els
[
2
9
]
,
also
s
tr
u
c
tu
r
ed
an
d
s
k
etch
ed
u
p
d
ates
[
4
0
]
.
Mc
Ma
h
an
et
a
l.
[
3
9
]
p
r
o
p
o
s
ed
Fed
AVG
th
at
a
d
d
s
m
o
r
e
co
m
p
u
tatio
n
to
ea
ch
clien
t b
y
iter
atin
g
th
e
lo
ca
l u
p
d
ate
m
u
ltip
le
tim
es
b
ef
o
r
e
th
e
a
v
er
ag
in
g
s
tep
.
I
n
h
is
wo
r
k
,
Fed
AVG
co
m
p
ar
ed
with
SGD
an
d
Fed
SGD
u
s
ed
i
n
a
lar
g
e
-
s
ca
le
n
ex
t
-
wo
r
d
p
r
ed
ictio
n
an
d
s
h
o
ws
th
at
Fed
AVG
s
u
cc
ess
tr
ain
s
h
ig
h
-
q
u
ality
m
o
d
els
u
s
in
g
r
elativ
ely
f
ewe
r
r
o
u
n
d
s
o
f
co
m
m
u
n
icatio
n
th
an
o
th
er
al
g
o
r
ith
m
s
.
Ma
n
y
v
a
r
ian
ts
o
f
Fed
AVG
h
av
e
b
ee
n
u
s
ed
in
s
ev
er
al
ty
p
es
o
f
r
esear
ch
,
in
clu
d
i
n
g
[
4
8
]
–
[
5
2
]
.
Z
h
u
an
d
J
in
[
5
3
]
p
r
o
p
o
s
ed
m
o
d
if
ied
s
p
a
r
s
e
ev
o
lu
tio
n
ar
y
tr
ain
in
g
(
SET
)
alg
o
r
ith
m
to
r
ed
u
ce
th
e
co
m
m
u
n
icatio
n
co
s
t
with
o
u
t
r
ed
u
cin
g
g
lo
b
al
lear
n
in
g
ac
cu
r
ac
y
.
T
h
ese
ap
p
r
o
ac
h
o
p
ti
m
izes
d
ee
p
n
eu
r
al
n
etwo
r
k
m
o
d
els
b
y
m
o
d
if
y
in
g
th
e
o
p
e
r
ato
r
b
y
r
em
o
v
in
g
o
p
er
atio
n
s
at
th
e
last
tr
ain
in
g
ep
o
c
h
.
T
h
e
r
esu
lt
s
h
o
ws
th
at
th
e
m
o
d
if
ied
SET
Alg
o
r
ith
m
s
u
cc
ee
d
ed
i
n
m
ax
i
m
izin
g
th
e
lear
n
in
g
p
er
f
o
r
m
a
n
ce
an
d
m
in
im
izin
g
th
e
co
m
m
u
n
icatio
n
co
s
t
b
y
e
n
co
d
in
g
o
n
ly
two
h
y
p
er
p
a
r
a
m
eter
s
.
Z
h
u
an
d
J
in
[
5
4
]
p
r
o
p
o
s
ed
Fed
Net2
Net,
a
n
o
v
el
s
ch
em
e
b
ased
o
n
a
m
o
d
el
g
r
o
win
g
with
a
m
o
d
if
ied
tr
ain
in
g
s
ch
em
e.
T
h
is
ap
p
r
o
ac
h
u
s
ed
two
tr
an
s
f
o
r
m
atio
n
s
,
wh
ic
h
ar
e
c
alled
Net2
W
id
er
n
et
an
d
Net
2
Dee
p
er
Net.
T
h
e
n
,
ea
ch
s
te
p
,
th
e
m
o
d
el
was
d
ev
elo
p
e
d
af
ter
e
n
h
an
ci
n
g
u
t
ilizin
g
a
f
u
n
ctio
n
al
co
n
s
er
v
i
n
g
tr
an
s
f
o
r
m
atio
n
o
n
it.
T
h
e
r
esu
lt
s
h
o
ws
th
at
Fed
Net2
Net
h
ad
an
ex
ce
llen
t
p
er
f
o
r
m
an
ce
th
an
o
th
er
alg
o
r
i
th
m
s
li
k
e
Fed
AVG,
Fed
er
ated
Dr
o
p
o
u
t
[
5
5
]
,
an
d
Hete
r
o
FL
[
5
6
]
.
I
n
ad
d
itio
n
,
c
o
m
m
u
n
icatio
n
co
m
p
le
x
ity
also
co
u
ld
b
e
r
ed
u
ce
d
b
y
c
o
n
s
tr
u
ctin
g
th
e
tr
ain
in
g
p
h
ase
o
f
FL
m
o
d
els u
s
in
g
p
e
r
s
o
n
aliza
tio
n
.
Han
ze
l,
P.
an
d
R
ich
tar
ik
,
P.
p
r
o
p
o
s
ed
s
ev
er
al
ef
f
icien
t
v
a
r
ian
ts
o
f
SGD
with
an
d
with
o
u
t
p
a
r
tial
p
ar
ticip
atio
n
an
d
v
a
r
ian
ce
r
ed
u
ctio
n
[
5
7
]
.
T
h
o
s
e
a
p
p
r
o
ac
h
e
s
u
s
e
a
n
ew
o
p
tim
izatio
n
f
o
r
m
u
latio
n
to
lear
n
a
m
ix
tu
r
e
o
f
th
e
g
l
o
b
al
m
o
d
el,
a
s
s
h
o
wn
in
(
2
)
.
min
1
,
…
.
,
∈
ℝ
{
(
)
≔
(
)
+
⋋
(
)
}
(
2
)
(
)
≔
1
∑
(
)
,
=
1
(
)
∶
=
1
2
∑
‖
−
̅
‖
2
=
1
I
n
f
o
r
m
atio
n
:
⋋
≥
0
=
1
,
…
.
,
∈
ℝ
=
∶
=
(
1
,
2
,
…
,
)
,
̅
∶
=
1
∑
=
1
=
ℎ
ℎ
3
.
2
.
2
.
H
et
er
o
g
eneit
y
On
e
o
f
th
e
p
r
o
p
o
n
en
ts
o
f
th
e
em
er
g
en
ce
o
f
FL
is
th
at
clien
t
p
r
iv
ac
y
m
u
s
t
b
e
m
ain
tain
ed
o
n
th
e
d
ata
u
s
ed
as
m
o
d
elin
g
m
ater
ial
in
ML
.
Ho
wev
er
,
FL
s
tan
d
ar
d
s
ar
e
s
till
v
u
ln
er
ab
le
to
p
r
iv
ac
y
an
d
co
n
f
id
en
tiality
leak
ag
e
wh
en
h
an
d
lin
g
h
et
er
o
g
en
e
o
u
s
d
ata
f
r
o
m
m
u
lt
ip
le
s
o
u
r
ce
s
[
5
8
]
.
Ga
o
et
a
l.
[
5
8
]
p
r
o
p
o
s
ed
h
eter
o
g
en
e
o
u
s
f
ed
er
ated
tr
an
s
f
er
lear
n
in
g
(
HFTL
)
,
a
n
a
p
p
r
o
ac
h
to
elim
in
ate
co
v
ar
iate
s
h
if
ts
o
f
h
o
m
o
g
en
eo
u
s
f
ea
tu
r
e
s
p
ac
es
a
n
d
p
ass
o
v
er
d
if
f
er
e
n
t
d
ata
o
wn
e
r
s
'
h
eter
o
g
en
eo
u
s
f
ea
tu
r
e
s
p
ac
es
u
s
in
g
en
d
-
to
-
e
n
d
s
ec
u
r
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:
2
5
0
2
-
4
7
52
A
co
n
ce
p
tu
a
l a
p
p
r
o
a
ch
o
f o
p
timiz
a
tio
n
in
fed
era
ted
lea
r
n
in
g
(
F
a
r
h
a
n
n
a
Ma
r
’
i
)
293
m
u
lti
-
p
ar
ty
lear
n
in
g
p
r
o
to
co
ls
.
T
h
e
r
esu
lt
s
h
o
ws
th
at
HFTL
o
u
tp
er
f
o
r
m
s
th
e
lo
ca
l
m
o
d
el
an
d
h
o
m
o
g
e
n
eo
u
s
FL
u
n
d
er
ch
allen
g
in
g
c
o
n
d
iti
o
n
s
ettin
g
s
.
Ho
wev
er
,
h
eter
o
g
en
eity
d
ata
b
etwe
en
u
n
iq
u
e
clien
ts
also
af
f
ec
t
m
o
d
el
p
er
f
o
r
m
a
n
ce
lik
e
ac
c
u
r
ac
y
r
ed
u
ctio
n
[
1
1
]
,
[
5
1
]
,
[
5
9
]
.
Fo
r
in
s
tan
ce
,
m
u
ltip
le
b
r
an
c
h
es
o
f
a
b
an
k
wo
u
ld
lik
e
to
co
llab
o
r
ativ
ely
tr
ain
a
cu
s
to
m
er
'
s
b
eh
av
io
r
u
s
ed
FL.
E
ac
h
b
an
k
co
llects
cu
s
to
m
er
d
ata
in
d
ep
e
n
d
en
tly
,
r
esu
ltin
g
in
u
n
b
alan
ce
d
an
d
h
eter
o
g
en
e
o
u
s
d
atasets
(
i.e
.
,
n
o
n
I
d
e
n
tically
I
n
d
ep
en
d
en
tl
y
Dis
tr
ib
u
ted
(
No
n
-
I
I
D)
.
On
e
b
a
n
k
m
ig
h
t
n
o
t
h
av
e
d
ata
s
am
p
les
r
eg
ar
d
in
g
a
s
p
ec
if
ic
b
eh
av
io
r
ca
teg
o
r
ies.
W
h
en
ea
ch
b
an
k
tr
ai
n
s
a
m
o
d
el
lo
ca
lly
,
th
e
lo
ca
l
g
o
a
l
m
ay
d
if
f
er
f
r
o
m
th
e
g
l
o
b
al
g
o
al.
As
a
r
esu
lt,
th
e
g
lo
b
al
av
er
ag
e
m
o
d
el
is
n
o
t
eq
u
al
to
t
h
e
g
lo
b
al
o
b
jectiv
e
,
an
d
th
is
ca
s
e
is
ca
lled
m
o
d
el
d
r
if
t
w
h
ich
c
au
s
es
p
o
o
r
m
o
d
el
p
e
r
f
o
r
m
an
ce
[
6
0
]
–
[
6
3
]
.
I
n
ad
d
itio
n
,
d
ata
h
e
ter
o
g
en
eity
also
ca
u
s
es
ac
c
u
r
a
cy
p
a
r
ity
wh
ic
h
is
th
e
ca
s
e
w
h
en
a
m
o
d
el
b
iases
r
eg
ar
d
in
g
th
e
u
n
iq
u
e
co
n
d
itio
n
in
ea
ch
lo
ca
l
s
u
b
ject,
f
o
r
ex
am
p
le,
g
eo
g
r
ap
h
ical
an
d
r
ac
e
d
is
c
r
im
in
atio
n
[
6
4
]
.
Z
h
o
u
et
a
l.
[
6
4
]
p
r
o
p
o
s
ed
a
n
ew
d
esig
n
s
ettin
g
f
o
r
FL
u
s
in
g
g
lo
b
al
-
l
o
ca
l
Kn
o
wled
g
e
Fu
s
io
n
(
Fed
KF)
to
o
v
er
co
m
e
b
o
t
h
ch
allen
g
es
in
h
eter
o
g
en
eity
d
ata,
s
u
c
h
as
m
o
d
el
d
r
if
t
an
d
ac
c
u
r
ac
y
p
ar
ity
.
Fed
KF's
co
n
ce
p
t
is
to
ass
ig
n
th
e
s
er
v
er
th
e
task
o
f
r
etu
r
n
in
g
g
lo
b
al
k
n
o
wled
g
e
to
g
u
id
e
th
e
lo
ca
l
clien
t
tr
ain
i
n
g
,
s
o
th
at
th
e
lo
ca
l
m
o
d
el
is
r
eg
u
lar
ized
b
y
th
e
g
l
o
b
al
o
p
tim
a,
wh
ich
d
eg
r
a
d
es
th
e
clien
t
m
o
d
el
d
r
if
t
ca
s
e
i
n
e
ac
h
tr
ain
in
g
r
o
u
n
d
.
T
1
is
d
esig
n
e
d
as
an
ac
tiv
e
-
i
n
ac
tiv
e
m
o
d
el
ag
g
r
eg
atio
n
f
o
r
th
e
g
lo
b
al
m
o
d
el
o
n
th
e
s
er
v
er
s
id
e,
a
n
d
T
2
is
d
esig
n
ed
to
s
u
p
p
o
r
t
k
n
o
wled
g
e
f
u
s
io
n
o
n
th
e
clien
t
s
id
e.
I
n
th
e
Fed
KF
s
ettin
g
,
th
e
ce
n
tr
al
s
er
v
er
s
ets
K
d
if
f
er
en
t
ca
ch
e
s
lo
ts
f
o
r
s
av
in
g
th
e
latest
lo
ca
l
m
o
d
els.
I
n
ev
er
y
tr
ain
in
g
r
o
u
n
d
,
th
e
s
elec
ted
clien
ts
h
ad
to
u
p
lo
ad
t
h
eir
lo
ca
l m
o
d
els to
th
e
ce
n
tr
al
s
er
v
er
.
T
h
e
alg
o
r
ith
m
o
f
Fed
KF m
e
n
tio
n
as f
o
llo
ws
:
1.
ph
a
s
e
t
-
1
:
T
h
e
ce
n
t
r
al
s
er
v
er
ag
g
r
eg
ates
ac
tiv
e
clien
ts
b
y
u
p
lo
ad
in
g
lo
ca
l
m
o
d
els
to
o
b
tai
n
ac
tiv
e
clien
t
ag
g
r
eg
ates
(
AC
A)
,
wh
er
ea
s
Fed
KF
ag
g
r
eg
ates
all
clien
ts
ca
ch
ed
m
o
d
els
to
o
b
tain
o
v
er
all
clien
t
ag
g
r
eg
ates (
OC
A)
.
2.
ph
a
s
e
t
:
A
p
o
r
tio
n
o
f
th
e
g
e
n
er
al
clien
tele
is
ch
o
s
en
to
b
e
ac
tiv
e
clien
ts
{
1
,
…
,
}
.
τ
is
th
e
r
ate
o
f
s
elec
tio
n
th
at
f
o
llo
ws
t
h
at
=
.
All
ac
tiv
e
clien
ts
r
ec
eiv
e
b
o
th
th
e
AC
A
an
d
OC
A
m
o
d
els
f
r
o
m
th
e
ce
n
tr
al
s
er
v
er
.
3.
E
ac
h
ac
tiv
e
clien
t
d
eter
m
in
es
th
e
AC
A
m
o
d
els
as
(
s
tu
d
en
t
m
o
d
el)
an
d
OC
A
m
o
d
els
as
(
teac
h
e
r
m
o
d
el)
4.
Uses
d
ata
-
f
r
ee
Kn
o
wled
g
e
Di
s
till
atio
n
to
tr
an
s
f
er
k
n
o
wled
g
e
f
r
o
m
t
h
e
teac
h
e
r
to
th
e
s
tu
d
en
t
m
o
d
els.
Me
an
wh
ile,
th
e
s
tu
d
en
t
m
o
d
el
is
tr
ain
ed
u
s
in
g
th
e
lo
ca
l
d
ataset
o
f
ea
ch
ac
tiv
e
clie
n
t.
All
g
lo
b
al
k
n
o
wled
g
e
is
co
m
b
in
ed
an
d
t
r
an
s
m
itted
to
th
e
s
tu
d
en
t m
o
d
el
.
5.
All a
ctiv
e
clien
ts
u
p
lo
ad
th
eir
lo
ca
l stu
d
en
t m
o
d
el
as th
e
u
p
d
ated
lo
ca
l m
o
d
el
t
o
th
e
ce
n
t
r
al
s
er
v
er
.
6.
T
h
e
weig
h
t
i
n
th
e
ca
c
h
e
s
lo
ts
is
u
p
d
ated
b
y
t
h
e
ce
n
tr
al
s
er
v
e
r
.
I
f
th
e
m
o
d
el
h
as
b
ee
n
f
u
lly
tr
ain
ed
,
eith
e
r
ex
it o
r
p
r
o
ce
ed
t
o
s
tep
2
.
B
ased
o
n
th
e
s
tu
d
y
,
T
h
e
r
esu
lt
is
Fed
KF
h
ad
s
u
ce
ed
to
b
e
a
b
etter
s
o
lu
tio
n
in
h
eter
o
g
en
eo
u
s
ag
n
o
s
tic
FL
with
h
ig
h
m
o
d
el
p
er
f
o
r
m
a
n
ce
,
p
r
iv
ac
y
-
p
r
eser
v
i
n
g
,
a
n
d
f
air
n
ess
[
6
4
]
.
3
.
2
.
3
.
Sca
la
bil
it
y
T
h
e
s
ca
lab
ilit
y
o
f
FL
co
u
ld
b
e
ca
lcu
lated
u
s
in
g
s
ev
er
al
v
ital
f
ac
to
r
s
,
lik
e
th
e
n
u
m
b
e
r
o
f
o
v
er
lap
p
in
g
s
am
p
les,
also
th
e
d
im
en
s
io
n
o
f
h
id
d
en
r
ep
r
esen
tatio
n
,
a
n
d
a
to
tal
o
f
th
e
f
ea
tu
r
es
[
6
5
]
,
th
e
Sy
s
tem
’
s
ab
ilit
y
to
h
an
d
le
th
e
n
u
m
b
e
r
o
f
clien
t
d
ev
ices
[
6
6
]
,
th
e
am
o
u
n
t
o
f
l
ea
r
n
in
g
o
n
lar
g
e
v
o
lu
m
es
o
f
d
ata
[
6
7
]
,
ac
cu
r
ac
y
,
co
m
p
u
tatio
n
al,
s
to
r
ag
e,
an
d
n
etwo
r
k
o
v
er
h
ea
d
.
Ha
n
an
d
Han
[
3
5
]
p
r
o
p
o
s
ed
DeFL
:
D
ec
en
tr
alize
d
W
eig
h
t
Ag
g
r
eg
atio
n
f
o
r
C
r
o
s
s
-
Sil
o
FL
with
two
k
ey
i
d
ea
s
th
at
ar
e
r
ed
u
cin
g
lo
ca
l
u
p
d
ates
in
th
e
s
er
v
er
b
y
co
llectin
g
all
th
e
lo
ca
l
u
p
d
ates
it
s
elf
,
th
e
n
weig
h
t
th
at’
s
o
n
ly
in
th
e
cu
r
r
en
t
tr
ain
in
g
ar
e
m
ain
tain
ed
a
n
d
s
y
n
ch
r
o
n
ized
s
o
th
e
n
etwo
r
k
a
n
d
th
e
s
to
r
a
g
e
o
v
e
r
h
ea
d
co
u
ld
b
e
r
ed
u
ce
d
[
3
5
]
.
I
n
th
e
DeFL
ap
p
r
o
ac
h
ab
s
tr
ac
ts
,
ea
c
h
p
ar
ticip
atin
g
n
o
d
e
h
as d
if
f
er
e
n
t
r
o
les:
clien
t
an
d
r
e
p
lica.
A
clien
t
ag
g
r
eg
ates
co
r
r
ec
t
weig
h
t
f
r
o
m
ac
tu
al
n
o
d
es
with
a
M
u
lti
-
Kr
u
m
f
ilter
[
6
8
]
.
T
h
e
r
esu
lt
s
h
o
ws
th
at
DeF
L
h
ad
s
u
cc
ess
f
u
lly
d
ef
en
d
ed
th
e
s
tan
d
ar
d
th
r
ea
t
m
o
d
els with
th
e
b
est p
er
f
o
r
m
a
n
ce
an
d
p
r
o
d
u
ce
d
e
f
f
icien
t sto
r
ag
e
an
d
n
etwo
r
k
u
s
ag
e.
3
.
2
.
4
.
P
riv
a
c
y
a
nd
s
ec
urit
y
Priv
ac
y
an
d
Secu
r
ity
ar
e
th
e
a
d
v
an
tag
es
h
ig
h
lig
h
ted
i
n
FL
b
ec
au
s
e
th
e
tr
ain
in
g
d
ata
is
n
o
t sen
t
to
th
e
s
er
v
er
b
u
t
r
em
ai
n
s
o
n
ea
ch
cl
ien
t
d
ev
ice
an
d
r
u
n
s
th
e
tr
ain
in
g
p
r
o
ce
s
s
in
d
ep
en
d
e
n
tly
.
H
o
wev
er
,
i
n
r
ea
lity
,
p
r
iv
ac
y
s
till
ca
n
n
o
t
b
e
g
u
a
r
a
n
teed
s
in
ce
tr
ain
in
g
m
o
d
els
f
r
o
m
u
s
er
s
will
s
till
b
e
s
en
t
t
o
s
er
v
er
s
f
o
r
g
l
o
b
al
ag
g
r
eg
atio
n
[
6
9
]
.
T
h
e
tr
ain
e
d
m
o
d
els
s
h
ar
ed
o
n
th
e
s
er
v
er
co
u
ld
lead
to
p
r
iv
ac
y
atta
ck
s
th
at
ar
e
m
o
d
el
in
v
er
s
io
n
[
7
0
]
.
T
o
o
v
er
co
m
e
th
is
p
r
o
b
lem
,
an
o
p
tim
izatio
n
f
r
am
ewo
r
k
o
n
FL
was
d
e
v
elo
p
ed
u
s
in
g
s
ec
u
r
e
ag
g
r
eg
atio
n
(
SA)
s
o
th
at
th
e
s
er
v
er
o
n
ly
lear
n
th
e
g
lo
b
al
m
o
d
el
u
p
d
ate
n
o
t
th
e
i
n
d
iv
id
u
al
m
o
d
el
u
p
d
ate
[
6
9
].
SA
is
a
p
ar
t
o
f
s
ec
u
r
e
m
u
l
ti
-
p
ar
ty
c
o
m
p
u
tatio
n
alg
o
r
ith
m
s
th
at
h
o
ld
ea
c
h
p
r
iv
ate
v
alu
e
an
d
co
llab
o
r
ate
to
co
m
p
u
te
an
ag
g
r
eg
ate
v
al
u
e
with
o
u
t
k
n
o
win
g
o
th
er
in
f
o
r
m
atio
n
ab
o
u
t
its
v
alu
e
b
u
t
o
n
l
y
wh
at
is
lear
n
ab
le
f
r
o
m
t
h
e
ag
g
r
eg
at
e
v
alu
e
its
elf
[
7
1
].
B
o
n
awitz
e
t
a
l.
[
7
1
]
p
r
o
p
o
s
ed
a
SA
p
r
o
to
co
l
co
n
s
is
tin
g
o
f
f
o
u
r
r
o
u
n
d
s
:
Sh
ar
e
Ke
y
s
,
m
ask
ed
in
p
u
t
co
llectio
n
,
c
o
n
s
i
s
ten
cy
ch
ec
k
,
an
d
u
n
m
ask
i
n
g
in
a
s
y
n
ch
r
o
n
o
u
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
288
-
2
9
9
294
n
etw
o
r
k
.
A
v
ar
ia
n
t
o
f
SA
ca
l
led
s
ec
u
r
e
in
d
ex
in
g
also
h
ad
b
ee
n
p
r
o
p
o
s
ed
to
g
u
a
r
an
tee
p
r
iv
ac
y
a
n
d
s
ec
u
r
it
y
[
7
2
].
I
n
a
d
d
i
t
i
o
n
,
s
i
n
c
e
t
h
e
t
r
ai
n
e
d
m
o
d
e
l
w
a
s
s
e
n
t
t
o
t
h
e
s
e
r
v
e
r
,
t
h
e
s
y
s
t
e
m
c
o
u
l
d
g
et
p
o
i
s
o
n
in
g
m
o
d
e
l
a
t
t
a
c
k
s
b
y
m
a
l
i
c
i
o
u
s
c
l
ie
n
t
s
.
I
t
w
a
s
c
h
a
l
le
n
g
i
n
g
t
o
i
d
e
n
t
i
f
y
t
h
e
m
a
l
i
c
i
o
u
s
c
li
e
n
t
s
b
e
c
a
u
s
e
n
o
l
o
c
a
l
c
l
i
e
n
t
d
a
t
a
w
as
a
v
a
i
l
a
b
l
e
o
n
t
h
e
s
e
r
v
e
r
[
3
6
]
.
M
a
e
t
a
l
.
[
3
6
]
p
r
o
p
o
s
e
d
a
m
o
d
e
l
u
p
d
a
t
e
a
g
g
r
e
g
a
t
i
o
n
(
MU
B
)
t
o
d
e
f
e
n
d
a
g
a
i
n
s
t
t
h
e
t
h
r
e
a
t
o
f
b
y
z
a
n
ti
n
e
a
tt
a
c
k
s
,
wh
i
c
h
a
r
e
a
d
d
it
i
v
e
n
o
is
e
a
n
d
s
i
g
n
-
f
l
i
p
p
i
n
g
a
t
t
ac
k
s
.
T
h
i
s
r
e
s
ea
r
c
h
p
r
o
v
e
d
c
o
u
l
d
e
n
h
a
n
c
e
b
o
t
h
p
r
i
v
a
c
y
a
n
d
s
e
c
u
r
i
t
y
u
s
i
n
g
a
c
o
m
b
i
n
a
ti
o
n
o
f
M
U
B
a
n
d
i
n
it
i
al
c
l
i
e
n
t
m
o
d
e
l
i
n
iti
a
l
i
z
at
i
o
n
(
I
C
M
I
)
.
3
.
2
.
5
.
Co
nv
er
g
ence
s
ra
t
e
T
r
ain
in
g
p
r
o
ce
s
s
es
in
FL
n
ee
d
to
b
e
f
ast
an
d
ac
cu
r
ate
b
ec
au
s
e
th
er
e
ar
e
s
ev
er
al
ty
p
es
o
f
tr
ain
in
g
b
atch
es
o
n
-
d
e
v
ice
th
at
s
h
o
u
l
d
b
e
f
in
is
h
ed
to
g
et
th
e
b
est
r
esu
lt.
Mo
s
t
o
f
th
e
r
esear
ch
h
ad
f
o
cu
s
ed
o
n
d
esig
n
in
g
t
h
e
b
est
ag
g
r
e
g
atio
n
s
tr
ateg
ies
f
o
r
im
p
r
o
v
i
n
g
c
o
n
v
er
g
e
n
ce
r
ates,
b
u
t
Qiu
et
a
l
.
[
7
3
]
h
ad
d
if
f
e
r
en
t
v
iews
an
d
p
r
o
p
o
s
ed
Z
e
r
o
FL,
wh
ich
is
th
e
f
r
am
ewo
r
k
th
at
is
d
ev
o
ted
to
ac
ce
le
r
atin
g
o
n
-
d
ev
ice
tr
ain
i
n
g
b
y
u
s
in
g
h
ig
h
ly
s
p
a
r
s
e
o
p
er
atio
n
s
.
T
h
er
e
ar
e
th
r
ee
s
tr
ateg
ies
u
s
ed
in
Z
er
o
FL
f
o
r
d
o
i
n
g
lo
c
al
s
p
ar
s
if
icatio
n
to
p
r
o
d
u
ce
b
o
t
h
p
er
f
o
r
m
a
n
ce
im
p
r
o
v
is
atio
n
an
d
r
ed
u
ce
co
m
m
u
n
icatio
n
co
s
ts
.
T
h
is
ap
p
r
o
ac
h
b
eliev
es th
at
n
o
t a
ll
o
f
th
e
weig
h
ts
n
ee
d
to
b
e
tr
a
n
s
f
er
r
ed
to
th
e
ce
n
t
r
al
s
er
v
er
f
o
r
th
e
ag
g
r
e
g
atio
n
a
n
d
u
s
es
lo
ca
l
s
p
ar
s
if
icatio
n
in
s
tead
.
T
h
r
ee
lo
ca
l
s
p
a
r
s
if
icatio
n
s
ar
e
u
s
ed
:
T
o
p
-
K
-
W
eig
h
ts
,
Dif
f
o
n
T
o
p
-
K
-
W
eig
h
t,
an
d
T
o
p
-
KW
eig
h
ts
Dif
f
.
T
h
e
r
esu
lt
s
h
o
ws
th
at
Z
e
r
o
FL
en
h
an
ce
s
th
e
ac
c
u
r
ac
y
p
er
f
o
r
m
a
n
ce
b
y
1
.
5
%
wh
ile
r
e
d
u
cin
g
1
.
9
x
u
p
lin
k
co
m
m
u
n
icatio
n
[
7
3
]
.
Z
er
o
FL
also
s
h
o
ws
th
e
b
es
t
p
er
f
o
r
m
a
n
ce
in
No
n
-
I
I
D
Data
,
ev
en
th
o
u
g
h
th
ese
ty
p
es
o
f
d
ata
ca
u
s
e
s
lo
w
m
o
d
el
tr
ain
in
g
an
d
en
f
o
r
ce
a
d
d
itio
n
al
co
m
m
u
n
icati
o
n
r
o
u
n
d
s
f
o
r
FL
t
o
co
n
v
er
g
e
[
7
4
]
.
W
u
,
Ho
n
g
,
W
an
g
,
a
n
d
Pin
g
p
r
o
p
o
s
ed
f
ed
e
r
ated
ad
ap
tiv
e
weig
h
tin
g
(
Fed
Ad
p
)
th
at
also
ac
ce
ler
ates
m
o
d
el
co
n
v
er
g
en
ce
s
u
n
d
er
th
e
p
r
ess
u
r
e
o
f
n
o
n
-
I
I
D
d
ata
[
7
4
]
.
T
h
e
m
ain
id
ea
o
f
Fed
Ad
p
is
to
ass
ig
n
a
d
if
f
er
en
t
weig
h
t f
o
r
d
if
f
e
r
en
t m
o
d
els at
ea
ch
co
m
m
u
n
icatio
n
r
o
u
n
d
b
a
s
ed
o
n
s
m
o
o
th
ed
an
g
le
ad
ap
ti
v
ely
.
T
h
er
e
ar
e
two
s
tep
s
u
s
ed
in
t
h
is
s
ettin
g
:
A
n
o
n
-
lin
ea
r
m
ap
p
in
g
f
u
n
ctio
n
u
s
in
g
a
v
ar
i
an
t
o
f
th
e
Go
m
p
er
t
z
f
u
n
cti
o
n
[
7
5
]
an
d
W
eig
h
tin
g
u
s
in
g
th
e
So
f
tm
ax
f
u
n
ctio
n
[
7
4
]
.
Fed
A
d
p
s
u
cc
ee
d
ed
in
g
ettin
g
b
etter
p
er
f
o
r
m
a
n
ce
co
m
p
ar
e
d
with
Fed
Av
g
b
y
r
ed
u
cin
g
th
e
co
m
m
u
n
icatio
n
r
o
u
n
d
s
b
y
u
p
t
o
5
4
.
1
%
o
n
th
e
MN
I
ST
d
ataset
a
n
d
u
p
to
4
5
.
4
%
o
n
th
e
Fas
h
io
n
MN
I
ST
d
ataset.
3.
2
.
6.
A
co
ncept
ua
l f
ra
m
ewo
rk
o
f
o
ptim
iza
t
io
n in f
eder
a
t
ed
lea
rning
B
ased
o
n
th
e
o
p
tim
izatio
n
ca
r
r
ied
o
u
t
in
p
r
ev
i
o
u
s
s
tu
d
ies,
it
is
p
r
o
v
en
to
b
e
ab
le
to
im
p
r
o
v
e
FL
p
er
f
o
r
m
an
ce
co
m
p
ar
e
d
to
p
r
e
-
ex
is
tin
g
s
tan
d
ar
d
al
g
o
r
ith
m
s
.
So
if
th
e
o
p
tim
izatio
n
m
eth
o
d
s
b
ased
o
n
th
ese
d
if
f
er
en
t
o
b
jectiv
es
ar
e
p
u
t
to
g
eth
er
,
it
ca
n
b
e
g
u
ar
a
n
teed
th
at
FL's
p
er
f
o
r
m
an
ce
will
b
e
e
v
en
b
etter
an
d
m
o
r
e
p
er
f
ec
t.
I
n
o
r
d
e
r
to
p
r
o
v
e
t
h
at
o
p
tim
izatio
n
in
FL
ca
n
also
b
e
ca
r
r
ied
o
u
t
o
n
m
o
r
e
th
an
o
n
e
ap
p
r
o
ac
h
,
s
ev
er
a
l
o
th
er
s
tu
d
ies
h
av
e
also
b
ee
n
co
n
d
u
cted
s
u
c
h
as
o
p
tim
izatio
n
in
f
ed
er
ated
p
r
in
cip
al
c
o
m
p
o
n
en
ts
an
aly
s
is
(
PC
A)
b
ased
o
n
Gr
ass
m
an
n
to
co
v
e
r
p
r
i
v
ac
y
c
o
n
ce
r
n
an
d
th
e
s
ca
lab
ilit
y
o
f
th
e
lim
ited
d
ev
ice
f
o
r
an
o
m
aly
d
etec
tio
n
in
I
o
T
Netwo
r
k
s
[
7
1
]
.
T
h
o
s
e
r
esear
c
h
y
ield
s
d
r
as
tically
r
ed
u
cin
g
th
e
an
al
y
s
is
tim
e
o
f
th
e
s
y
s
tem
.
No
t
o
n
ly
th
at,
p
r
iv
ac
y
co
n
ce
r
n
also
b
e
o
p
tim
ized
with
co
n
v
er
g
en
ce
s
r
ate
wh
e
n
th
e
clien
ts
h
ad
b
ee
n
cl
u
s
ter
ed
in
to
a
p
e
r
v
asiv
e
s
o
cial
c
o
n
n
ec
tio
n
b
e
twee
n
u
s
er
s
.
Ph
o
n
g
et
a
l
.
[
7
6
]
p
r
o
p
o
s
ed
a
n
o
v
el
s
o
ci
al
-
awa
r
e
clu
s
ter
ed
f
ed
er
ated
lea
r
n
in
g
(
SC
FL)
to
ac
h
iev
e
th
e
b
est
p
er
f
o
r
m
a
n
ce
o
f
FL
with
o
u
t
s
ac
r
if
icin
g
th
e
m
o
d
el
b
ec
a
u
s
e
th
e
n
o
is
e
g
iv
en
f
o
r
p
r
iv
ac
y
co
n
ce
r
n
i
n
d
i
f
f
er
en
tial
p
r
iv
ac
y
.
E
f
f
icien
t
FL
co
u
ld
b
e
ac
h
iev
ab
le
wh
e
n
th
e
h
eter
o
g
en
eity
an
d
t
h
e
co
n
v
er
g
en
ce
r
ate
h
ad
b
ee
n
p
r
o
v
id
ed
with
th
e
b
est
p
e
r
f
o
r
m
an
ce
s
.
On
e
o
f
th
e
ch
ar
ac
ter
is
tics
o
f
n
o
n
-
I
I
D
d
a
ta
is
s
tati
s
tical
h
eter
o
g
en
eity
th
at
co
u
ld
b
e
s
o
lv
e
d
with
SGD
in
tr
ad
itio
n
al
ce
n
tr
alize
d
lear
n
i
n
g
[
7
7
]
,
B
u
t
in
th
e
FL
s
ettin
g
s
tatis
tical
h
eter
o
g
en
eity
h
a
d
b
ec
o
m
e
a
s
ev
er
e
p
r
o
b
lem
th
at
SGD
co
u
ld
n
o
t
s
o
lv
e
an
d
y
ield
s
d
eg
r
ad
atio
n
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
s
[
7
8
]
.
So
it
ca
n
n
o
t
b
e
d
en
ied
th
at
s
o
lv
in
g
h
eter
o
g
en
eity
p
r
o
b
lem
s
will si
m
u
ltan
eo
u
s
ly
in
c
r
ea
s
e
th
e
p
er
f
o
r
m
an
ce
m
o
d
el
as th
e
c
o
n
v
er
g
en
ce
r
ate.
Fed
C
G
[
7
9
]
is
also
o
n
e
o
f
th
e
f
r
am
ewo
r
k
s
th
at
p
r
o
v
e
E
f
f
i
cien
t
FL
co
u
ld
b
e
d
ev
elo
p
ed
with
m
o
r
e
th
an
o
n
e
g
o
al
o
f
o
p
tim
izatio
n
s
.
Fed
C
G
u
s
in
g
ad
ap
tiv
e
c
lien
t
s
elec
tio
n
an
d
g
r
ad
ien
t
co
m
p
r
ess
io
n
.
T
h
is
ap
p
r
o
ac
h
s
elec
ts
a
r
ep
r
esen
t
ativ
e
clien
t
r
eg
a
r
d
in
g
s
tatis
t
ical
h
eter
o
g
e
n
eity
,
an
d
af
te
r
tr
ain
in
g
lo
ca
lly
,
co
m
p
r
ess
ed
m
o
d
el
u
p
d
ates
wo
u
ld
b
e
u
p
lo
ad
e
d
with
m
atch
in
g
th
eir
ca
p
a
b
ilit
ies
to
th
e
p
ar
a
m
eter
s
er
v
er
f
o
r
t
h
e
ag
g
r
eg
atio
n
p
h
ase.
I
n
a
m
o
b
il
e
wir
eless
d
ev
ice,
a
s
p
ec
tr
u
m
allo
ca
tio
n
o
p
tim
izatio
n
was
d
o
n
e
to
en
h
a
n
ce
FL,
m
in
im
ize
tim
e
co
n
s
u
m
p
tio
n
,
an
d
en
s
u
r
e
f
ast
c
o
n
v
er
g
en
ce
s
b
y
im
p
lem
en
ti
n
g
a
r
o
b
u
s
t
d
e
v
ice
s
elec
tio
n
[
8
0
]
.
Fed
HP
is
also
p
r
o
p
o
s
ed
to
o
v
er
co
m
e
h
eter
o
g
en
eity
as
th
e
cr
itical
ch
allen
g
e
o
n
FL,
wh
ich
in
teg
r
ates
an
ad
ap
tiv
e
co
n
tr
o
l
o
f
lo
ca
l
u
p
d
a
tin
g
f
r
eq
u
e
n
cy
an
d
th
e
n
etwo
r
k
to
p
o
l
o
g
y
[
8
1
]
.
Fed
HP
s
u
cc
e
s
s
f
u
lly
r
ed
u
ce
d
th
e
tim
e
co
m
p
etitio
n
b
y
ab
o
u
t
5
1
%
an
d
i
n
cr
ea
s
ed
m
o
d
el
ac
cu
r
ac
y
u
p
to
5
%
i
n
h
ete
r
o
g
en
eo
u
s
co
n
d
itio
n
s
.
Fu
r
th
er
m
o
r
e
,
th
e
p
r
ev
io
u
s
r
esear
ch
,
wh
ich
r
eq
u
ir
es
r
e
-
weig
h
ts
lo
ca
l
u
p
d
ates,
ca
u
s
es
o
th
e
r
ch
allen
g
es
th
at
as
p
o
o
r
er
o
p
tim
a
o
f
th
e
clie
n
t
m
o
d
el
wh
en
it
is
in
m
o
r
e
h
eter
o
g
en
eo
u
s
c
o
n
d
itio
n
s
[
8
2
]
.
Fed
S
k
ip
is
th
e
p
r
o
p
o
s
ed
m
eth
o
d
b
y
Fan
et
a
l.
[
8
2
]
w
h
ich
co
u
ld
im
p
r
o
v
e
th
e
clien
t
o
p
tim
a
b
y
s
y
s
tem
atica
lly
s
k
ip
p
in
g
Fed
Av
g
an
d
s
p
r
ea
d
in
g
lo
ca
l
m
o
d
els
to
th
e
cr
o
s
s
d
ev
ices.
I
n
ad
d
itio
n
t
o
s
o
lv
in
g
s
tatis
tical
h
eter
o
g
en
eity
p
r
o
b
lem
s
an
d
im
p
r
o
v
in
g
m
o
d
el
p
er
f
o
r
m
an
c
e,
Fed
Sk
ip
h
as
also
r
ed
u
ce
d
c
o
m
m
u
n
icatio
n
r
o
u
n
d
s
in
ca
s
es
with
m
an
y
clien
ts
,
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:
2
5
0
2
-
4
7
52
A
co
n
ce
p
tu
a
l a
p
p
r
o
a
ch
o
f o
p
timiz
a
tio
n
in
fed
era
ted
lea
r
n
in
g
(
F
a
r
h
a
n
n
a
Ma
r
’
i
)
295
f
o
r
e
x
am
p
le,
o
n
C
I
FAR
-
1
0
0
d
ata.
T
h
u
s
,
to
f
in
d
an
ef
f
ec
tiv
e
an
d
ef
f
icien
t
c
o
m
b
in
atio
n
o
f
o
p
tim
izatio
n
s
in
FL,
th
o
r
o
u
g
h
test
in
g
o
f
all
ty
p
es
o
f
o
p
tim
izatio
n
s
f
o
u
n
d
i
n
p
r
ev
i
o
u
s
s
tu
d
ies
is
r
eq
u
ir
ed
,
s
o
th
at
o
p
tim
izatio
n
in
FL
will b
e
an
u
p
-
a
n
d
-
c
o
m
in
g
r
esear
ch
.
B
ased
o
n
th
e
liter
atu
r
e
s
tu
d
ies
th
at
h
av
e
b
ee
n
ca
r
r
ied
o
u
t
o
n
s
ev
er
al
o
p
tim
izatio
n
ap
p
r
o
ac
h
es
to
FL
,
we
h
av
e
ca
teg
o
r
ized
in
to
s
ev
er
al
s
u
b
-
ca
teg
o
r
ies
b
ased
o
n
th
e
o
p
tim
izatio
n
o
b
jectiv
es
ac
h
iev
ed
,
wh
ich
ar
e
C
o
m
m
u
n
icati
o
n
ef
f
icien
cy
,
h
eter
o
g
en
eity
,
s
ca
lab
ilit
y
,
p
r
iv
ac
y
an
d
s
ec
u
r
ity
,
also
co
n
v
er
g
en
ce
s
,
r
ate.
I
n
th
is
s
ec
tio
n
,
we
p
r
o
p
o
s
e
a
co
n
ce
p
tu
al
f
r
am
ewo
r
k
b
ased
o
n
t
h
o
s
e
o
p
tim
izatio
n
s
in
th
e
p
r
ev
io
u
s
s
tu
d
ies.
T
h
e
p
r
o
p
o
s
ed
co
n
ce
p
t
u
al
f
r
am
ewo
r
k
is
b
ased
o
n
th
e
im
p
lem
e
n
tatio
n
o
f
o
p
tim
izatio
n
in
two
o
b
jects,
th
at
is
,
th
e
s
er
v
er
an
d
clien
t
s
id
es.
I
n
d
etail,
th
e
p
r
o
p
o
s
ed
co
n
ce
p
tu
al
f
r
am
ewo
r
k
o
f
o
p
tim
izatio
n
FL
in
th
is
s
tu
d
y
is
d
ep
icted
in
Fig
u
r
e
4
.
T
h
e
n
ew
f
o
r
m
in
o
p
tim
izatio
n
f
o
r
FL
to
o
p
tim
ize
p
er
f
o
r
m
an
ce
u
s
in
g
s
ev
er
al
ap
p
r
o
ac
h
es
f
r
o
m
t
h
e
f
iv
e
asp
ec
ts
o
f
FL
o
p
tim
izatio
n
n
e
ed
s
.
B
esid
es
th
at,
th
e
f
r
am
ewo
r
k
also
d
iv
id
es
th
e
o
p
tim
izatio
n
in
to
two
s
id
es:
clien
ts
a
n
d
s
e
r
v
er
s
,
b
ased
o
n
th
e
im
p
lem
e
n
tatio
n
o
f
o
p
tim
izatio
n
m
et
h
o
d
s
.
I
n
p
r
ac
tice,
t
h
is
f
r
am
ewo
r
k
is
p
o
s
s
ib
le
to
r
u
n
,
co
n
s
id
er
in
g
th
at
ea
ch
o
p
ti
m
izatio
n
ap
p
r
o
ac
h
h
as
its
al
g
o
r
ith
m
to
im
p
r
o
v
e
d
if
f
er
en
t
o
p
tim
izatio
n
asp
ec
ts
s
o
th
at
if
all
o
f
t
h
ese
o
p
tim
iza
tio
n
s
tr
ateg
ies
ar
e
im
p
lem
en
te
d
,
it
will
p
r
o
d
u
ce
a
p
er
f
ec
t
FL
s
ettin
g
.
T
h
e
r
e
ar
e
two
o
p
tio
n
al
f
r
am
ewo
r
k
s
o
f
o
p
tim
izatio
n
in
FL,
as
d
e
p
ic
ted
in
Fig
u
r
e
4
.
I
n
co
m
m
u
n
icatio
n
ef
f
icien
c
y
,
t
h
e
SET
alg
o
r
ith
m
is
ex
ec
u
te
d
f
o
r
th
e
f
ir
s
t
tim
e,
wh
ic
h
is
a
s
ettin
g
f
o
r
th
e
d
ee
p
n
eu
r
al
n
etwo
r
k
u
s
ed
in
th
e
lea
r
n
in
g
p
r
o
ce
s
s
,
th
en
Fed
Net2
Net
o
p
tim
izatio
n
is
also
ex
ec
u
ted
in
th
e
clien
t
ar
ea
as
well
as
in
th
e
lear
n
in
g
s
tag
e
b
y
p
r
o
v
id
in
g
a
co
n
d
itio
n
f
o
r
th
e
p
r
eser
v
in
g
t
r
an
s
f
o
r
m
ati
o
n
f
u
n
ctio
n
.
Af
te
r
th
at,
to
o
v
er
co
m
e
th
e
p
r
o
b
le
m
o
f
h
eter
o
g
e
n
eity
,
th
er
e
ar
e
two
o
p
tio
n
s
th
at
we
co
u
ld
ch
o
o
s
e
b
etwe
en
HFTL
u
s
in
g
an
en
d
-
to
-
e
n
d
s
ec
u
r
e
m
u
lti
-
p
ar
ty
lea
r
n
in
g
p
r
o
to
co
l
an
d
Fed
KF
o
n
two
s
id
es,
n
am
el
y
s
er
v
er
s
a
n
d
clien
ts
u
s
in
g
k
n
o
wled
g
e
f
u
s
io
n
.
Fig
u
r
e
4
.
C
o
n
ce
p
tu
al
f
r
a
m
ewo
r
k
o
f
o
p
tim
izatio
n
i
n
FL
FL
was
d
esig
n
ed
to
ad
d
r
ess
th
e
p
r
iv
ac
y
an
d
s
ec
u
r
ity
o
f
d
ata
u
s
ed
d
u
r
in
g
th
e
ML
p
r
o
ce
s
s
.
Ho
wev
er
,
in
p
r
ac
tice,
FL
s
till
f
ac
es
s
ec
u
r
ity
th
r
ea
ts
b
ec
a
u
s
e,
af
ter
all,
th
e
lo
ca
l
tr
ain
i
n
g
r
esu
lts
m
u
s
t
s
till
b
e
s
en
t
to
th
e
s
er
v
er
f
o
r
f
u
r
t
h
er
g
lo
b
al
ag
g
r
e
g
atio
n
,
in
h
an
d
lin
g
FL
as
well
ca
n
b
e
o
p
tim
ized
to
h
an
d
le
p
r
iv
ac
y
an
d
s
ec
u
r
ity
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
5
0
2
-
4
7
52
I
n
d
o
n
esian
J
E
lec
E
n
g
&
C
o
m
p
Sci
,
Vo
l.
37
,
No
.
1
,
J
an
u
ar
y
20
25
:
288
-
2
9
9
296
u
s
in
g
a
s
ec
u
r
e
ag
g
r
eg
atio
n
p
r
o
to
co
ls
m
o
d
el.
I
n
ad
d
itio
n
,
th
e
f
u
n
d
a
m
en
tal
p
r
o
b
lem
f
o
r
FL
is
s
ca
lab
ilit
y
wh
ich
ca
n
also
b
e
o
p
tim
ized
t
o
g
et
t
h
e
b
est
p
er
f
o
r
m
a
n
ce
u
s
in
g
d
e
ce
n
tr
alize
d
weig
h
t
a
g
g
r
e
g
atio
n
(
DeFL
)
.
T
h
en
,
n
o
less
im
p
o
r
tan
t
is
th
e
is
s
u
e
o
f
co
n
v
er
g
en
ce
r
ate
b
ec
a
u
s
e
FL
ca
r
r
ies
o
u
t
th
e
tr
ain
in
g
p
r
o
ce
s
s
s
ep
ar
ately
.
I
n
g
en
er
al,
th
e
r
esu
ltin
g
co
n
v
er
g
en
ce
r
esu
lts
ar
e
d
e
f
in
itely
lo
w
er
,
s
o
th
is
is
also
an
im
p
o
r
ta
n
t
ch
allen
g
e
f
o
r
FL,
o
n
e
o
f
th
e
o
p
tim
izatio
n
s
tr
ate
g
ies
th
at
ca
n
b
e
im
p
lem
en
ted
is
b
y
im
p
lem
en
tin
g
Z
er
o
FL
u
s
in
g
a
h
ig
h
ly
s
p
ar
s
e
o
p
er
atio
n
to
ac
ce
ler
ate
o
n
-
d
ev
ice
tr
ain
in
g
.
Ap
ar
t
f
r
o
m
th
at,
i
n
o
r
d
er
to
o
p
tim
ize
th
e
c
o
n
v
e
r
g
en
ce
lev
el
b
etter
,
Fed
Ad
p
o
p
er
atio
n
s
ca
n
also
b
e
ad
d
ed
u
s
in
g
Ad
ap
tiv
e
W
eig
h
tin
g
b
y
ass
ig
n
in
g
d
if
f
er
e
n
t
weig
h
ts
to
ea
ch
d
if
f
er
en
t
m
o
d
el
.
B
ased
o
n
t
h
e
p
r
o
p
o
s
ed
f
r
am
ew
o
r
k
o
f
o
p
t
im
izatio
n
in
FL
,
we
ca
n
co
n
d
u
ct
ex
p
e
r
im
en
ts
to
p
r
o
v
e
h
o
w
th
e
p
er
f
o
r
m
an
ce
o
f
v
ar
i
o
u
s
o
p
tim
izatio
n
m
eth
o
d
s
is
ca
r
r
ied
o
u
t
s
im
u
ltan
eo
u
s
ly
.
So
h
o
p
e
f
u
lly
,
we
ca
n
f
in
d
t
h
e
b
est f
o
r
m
o
f
o
p
ti
m
izatio
n
f
r
o
m
FL.
4.
CO
NCLU
SI
O
N
I
n
m
an
y
p
r
ac
tical
ap
p
licatio
n
s
o
f
FL
,
v
ar
io
u
s
o
p
tim
izati
o
n
tech
n
iq
u
es
h
a
v
e
b
e
en
em
p
lo
y
ed
to
ad
d
r
ess
th
ese
is
s
u
es,
u
tili
zin
g
d
if
f
er
en
t
m
eth
o
d
s
d
ep
e
n
d
in
g
o
n
th
e
s
p
ec
if
ic
g
o
als o
r
o
b
jecti
v
es.
Op
tim
izin
g
FL
is
a
k
ey
ar
ea
o
f
r
esear
ch
th
at
d
r
iv
es
th
e
ad
v
an
ce
m
e
n
t
an
d
e
n
h
an
ce
m
e
n
t
o
f
FL
tech
n
iq
u
es
an
d
m
o
d
els.
T
h
is
r
esear
ch
h
as
an
aly
ze
d
h
o
w
o
p
tim
izatio
n
ca
n
b
e
m
a
d
e
FL
p
er
f
o
r
m
s
b
etter
.
O
u
r
r
esear
ch
in
d
icate
s
th
at,
ac
co
r
d
in
g
to
cu
r
r
en
t
s
tu
d
ies
,
th
e
o
p
tim
izatio
n
o
f
FL
is
d
iv
id
ed
in
to
f
i
v
e
f
u
n
ctio
n
s
o
f
o
p
tim
izatio
n
,
th
at
ar
e
r
ed
u
cin
g
co
m
m
u
n
icatio
n
c
o
s
t
b
etwe
en
cu
s
to
m
er
a
n
d
d
ata
ce
n
ter
,
s
o
lv
in
g
h
eter
o
g
en
eity
,
p
r
iv
ac
y
an
d
s
ec
u
r
ity
co
n
ce
r
n
s
,
o
v
er
c
o
m
in
g
s
ca
lab
il
ity
,
an
d
in
cr
ea
s
in
g
co
n
v
er
g
en
ce
s
r
ate.
T
h
is
r
esear
ch
h
as
p
r
o
p
o
s
ed
a
co
n
ce
p
tu
al
f
r
am
ewo
r
k
f
o
r
d
esig
n
i
n
g
an
d
im
p
lem
en
tin
g
o
p
tim
izatio
n
in
FL
b
ased
o
n
th
e
s
u
cc
ess
f
u
l
o
p
tim
izatio
n
u
s
ed
in
p
r
ev
io
u
s
s
tu
d
ies
w
ith
t
h
e
h
o
p
e
th
at
it
co
u
l
d
b
e
u
s
ed
to
o
p
t
im
ize
FL
wh
ile
ad
d
r
ess
in
g
f
i
v
e
m
ain
p
r
o
b
lem
s
.
T
h
er
e
ar
e
two
o
p
tio
n
al
f
r
am
e
wo
r
k
s
f
o
r
o
p
tim
izatio
n
in
FL.
I
n
ter
m
s
o
f
co
m
m
u
n
icatio
n
e
f
f
icien
cy
,
th
e
SET
alg
o
r
ith
m
is
f
ir
s
t
ex
ec
u
ted
,
wh
ich
s
ets
u
p
th
e
d
ee
p
n
e
u
r
al
n
etwo
r
k
u
s
ed
in
t
h
e
lear
n
in
g
p
r
o
ce
s
s
.
T
h
e
n
,
Fed
Net2
Net
o
p
tim
izatio
n
is
p
er
f
o
r
m
e
d
b
o
th
o
n
th
e
clien
t
s
id
e
an
d
d
u
r
in
g
th
e
lear
n
in
g
s
tag
e,
with
a
co
n
d
itio
n
f
o
r
m
ain
tain
i
n
g
th
e
t
r
an
s
f
o
r
m
atio
n
f
u
n
ctio
n
.
T
o
ad
d
r
ess
th
e
is
s
u
e
o
f
h
eter
o
g
en
eity
,
th
e
r
e
ar
e
two
o
p
tio
n
s
to
ch
o
o
s
e
f
r
o
m
:
HFTL
u
s
in
g
a
n
en
d
-
to
-
en
d
s
ec
u
r
e
m
u
lti
-
p
ar
t
y
lear
n
in
g
p
r
o
to
c
o
l,
o
r
Fed
K
F,
wh
ich
in
v
o
lv
es
Kn
o
wled
g
e
Fu
s
io
n
o
n
b
o
th
s
er
v
er
s
an
d
clien
ts
.
Fo
r
f
u
t
u
r
e
wo
r
k
s
,
it
is
n
e
ce
s
s
ar
y
to
p
r
o
v
e
wh
eth
er
th
e
co
m
b
in
atio
n
o
f
o
p
tim
izatio
n
s
p
r
o
p
o
s
ed
in
th
is
r
esear
ch
ca
n
i
m
p
r
o
v
e
th
e
p
e
r
f
o
r
m
an
ce
o
f
FL
.
RE
F
E
R
E
NC
E
S
[
1
]
P
.
C
h
o
u
,
H
.
H
.
C
.
C
h
u
a
n
g
,
Y
.
C
.
C
h
o
u
,
a
n
d
T
.
P
.
Li
a
n
g
,
“
P
r
e
d
i
c
t
i
v
e
a
n
a
l
y
t
i
c
s
f
o
r
c
u
st
o
mer
r
e
p
u
r
c
h
a
se
:
I
n
t
e
r
d
i
sc
i
p
l
i
n
a
r
y
i
n
t
e
g
r
a
t
i
o
n
o
f
b
u
y
t
i
l
l
y
o
u
d
i
e
mo
d
e
l
i
n
g
a
n
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
”
E
u
ro
p
e
a
n
J
o
u
rn
a
l
o
f
O
p
e
r
a
t
i
o
n
a
l
R
e
se
a
rc
h
,
v
o
l
.
2
9
6
,
n
o
.
2
,
p
p
.
6
3
5
–
6
5
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
j
o
r
.
2
0
2
1
.
0
4
.
0
2
1
.
[
2
]
E.
S
a
m
u
n
d
e
r
u
a
n
d
M
.
F
a
r
r
u
g
i
a
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
w
i
t
h
a
p
p
l
i
c
a
t
i
o
n
s
p
r
e
d
i
c
t
i
n
g
c
u
s
t
o
mer
p
u
r
p
o
s
e
o
f
t
r
a
v
e
l
i
n
a
l
o
w
-
c
o
s
t
t
r
a
v
e
l
e
n
v
i
r
o
n
m
e
n
t
—
a
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,
”
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
w
i
t
h
Ap
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
9
,
n
o
.
A
p
r
i
l
2
0
2
1
,
p
.
1
0
0
3
7
9
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
ml
w
a
.
2
0
2
2
.
1
0
0
3
7
9
.
[
3
]
Z.
X
u
,
G
.
Zh
u
,
N
.
M
e
t
a
w
a
,
a
n
d
Q
.
Z
h
o
u
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
b
a
se
d
c
u
st
o
mer
me
t
a
-
c
o
m
b
i
n
a
t
i
o
n
b
r
a
n
d
e
q
u
i
t
y
a
n
a
l
y
s
i
s
f
o
r
mark
e
t
i
n
g
b
e
h
a
v
i
o
r
e
v
a
l
u
a
t
i
o
n
,
”
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
a
n
d
Ma
n
a
g
e
m
e
n
t
,
v
o
l
.
5
9
,
n
o
.
1
,
p
.
1
0
2
8
0
0
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
p
m.
2
0
2
1
.
1
0
2
8
0
0
.
[
4
]
G
.
P
a
n
g
e
st
u
,
F
.
M
a
r
’
I
,
F
.
A
.
B
a
c
h
t
i
a
r
,
a
n
d
F
.
I
.
M
a
u
l
a
n
a
,
“
T
h
e
h
e
r
o
n
f
o
r
m
u
l
a
a
p
p
r
o
a
c
h
f
o
r
h
e
a
d
m
o
v
e
me
n
t
s
d
e
t
e
c
t
i
o
n
f
o
r
st
u
d
e
n
t
f
o
c
u
s
d
e
t
e
c
t
i
o
n
d
u
r
i
n
g
p
a
n
d
e
mi
c
o
n
l
i
n
e
c
l
a
ss
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
2
0
2
1
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
Ma
n
a
g
e
m
e
n
t
a
n
d
T
e
c
h
n
o
l
o
g
y
,
I
C
I
MT
e
c
h
2
0
2
1
,
n
o
.
A
u
g
u
st
,
p
p
.
6
7
8
–
6
8
3
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
I
M
Te
c
h
5
3
0
8
0
.
2
0
2
1
.
9
5
3
5
0
6
5
.
[
5
]
F
.
M
a
r
’
i
a
n
d
A
.
A
.
S
u
p
i
a
n
t
o
,
“
C
l
u
st
e
r
i
n
g
c
r
e
d
i
t
c
a
r
d
h
o
l
d
e
r
s
b
a
se
d
o
n
b
i
l
l
p
a
y
m
e
n
t
s u
s
i
n
g
i
m
p
r
o
v
e
d
K
-
mea
n
s
w
i
t
h
p
a
r
t
i
c
l
e
sw
a
r
m
o
p
t
i
m
i
z
a
t
i
o
n
,
”
J
u
r
n
a
l
T
e
k
n
o
l
o
g
i
I
n
f
o
r
m
a
si
d
a
n
I
l
m
u
K
o
m
p
u
t
e
r
,
v
o
l
.
5
,
n
o
.
6
,
p
p
.
7
3
7
–
7
4
4
,
2
0
1
8
,
d
o
i
:
1
0
.
2
5
1
2
6
/
j
t
i
i
k
.
2
0
1
8
5
8
5
8
.
[
6
]
G
D
P
R
,
“
G
e
n
e
r
a
l
d
a
t
a
p
r
o
t
e
c
t
i
o
n
r
e
g
u
l
a
t
i
o
n
(
G
D
P
R
)
–
o
f
f
i
c
i
a
l
l
e
g
a
l
t
e
x
t
,
”
G
e
n
e
ra
l
D
a
t
a
Pr
o
t
e
c
t
i
o
n
R
e
g
u
l
a
t
i
o
n
.
p
p
.
1
–
9
9
,
2
0
1
8
.
[
7
]
W
.
Y
.
B
.
Li
m
e
t
a
l
.
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
i
n
mo
b
i
l
e
e
d
g
e
n
e
t
w
o
r
k
s:
a
c
o
m
p
r
e
h
e
n
s
i
v
e
s
u
r
v
e
y
,
”
I
E
EE
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
u
rv
e
y
s
a
n
d
T
u
t
o
ri
a
l
s
,
v
o
l
.
2
2
,
n
o
.
3
,
p
p
.
2
0
3
1
–
2
0
6
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
C
O
M
S
T.
2
0
2
0
.
2
9
8
6
0
2
4
.
[
8
]
D
.
U
p
r
e
t
i
,
H
.
K
i
m,
E.
Y
a
n
g
,
a
n
d
C
.
S
e
o
,
“
D
e
f
e
n
d
i
n
g
a
g
a
i
n
s
t
l
a
b
e
l
-
f
l
i
p
p
i
n
g
a
t
t
a
c
k
s
i
n
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
s
y
st
e
ms
u
s
i
n
g
u
n
i
f
o
r
m
man
i
f
o
l
d
a
p
p
r
o
x
i
ma
t
i
o
n
a
n
d
p
r
o
j
e
c
t
i
o
n
,
”
I
A
ES
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
3
,
n
o
.
1
,
p
p
.
4
5
9
–
4
6
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
3
.
i
1
.
p
p
4
5
9
-
4
6
6
.
[
9
]
M
.
C
h
e
n
a
n
d
D
.
K
.
H
a
l
i
m,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
sc
a
m
c
l
a
ssi
f
i
c
a
t
i
o
n
i
n
sma
l
l
I
n
d
o
n
e
si
a
n
l
a
n
g
u
a
g
e
d
a
t
a
se
t
:
a
n
i
n
i
t
i
a
l
st
u
d
y
,
”
I
n
d
o
n
e
si
a
n
J
o
u
r
n
a
l
o
f
El
e
c
t
ri
c
a
l
E
n
g
i
n
e
e
r
i
n
g
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
3
0
,
n
o
.
1
,
p
p
.
3
2
5
–
3
3
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
e
c
s.
v
3
0
.
i
1
.
p
p
3
2
5
-
3
3
1
.
[
1
0
]
M
.
N
.
M
e
q
d
a
d
,
A
.
H
.
H
u
sse
i
n
,
S
.
O
.
H
u
sa
i
n
,
a
n
d
A
.
M
.
J
a
w
a
d
,
“
C
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
e
l
e
c
t
r
o
c
a
r
d
i
o
g
r
a
m
si
g
n
a
l
s
b
a
sed
o
n
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
a
n
d
a
g
a
u
ssi
a
n
mu
l
t
i
v
a
r
i
a
t
e
a
g
g
r
e
g
a
t
i
o
n
mo
d
u
l
e
,
”
I
n
d
o
n
e
s
i
a
n
J
o
u
rn
a
l
o
f
E
l
e
c
t
r
i
c
a
l
En
g
i
n
e
e
r
i
n
g
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
3
0
,
n
o
.
2
,
p
p
.
9
3
6
–
9
4
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
e
c
s.
v
3
0
.
i
2
.
p
p
9
3
6
-
9
4
3
.
[
1
1
]
P
.
K
a
i
r
o
u
z
e
t
a
l
.
,
“
A
d
v
a
n
c
e
s
a
n
d
o
p
e
n
p
r
o
b
l
e
ms
i
n
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
F
o
u
n
d
a
t
i
o
n
s
a
n
d
T
ren
d
s
i
n
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
,
v
o
l
.
1
4
,
n
o
.
1
–
2
,
p
p
.
1
–
2
1
0
,
2
0
2
1
,
d
o
i
:
1
0
.
1
5
6
1
/
2
2
0
0
0
0
0
0
8
3
.
[
1
2
]
D
.
L
i
,
Z.
L
u
o
,
a
n
d
B
.
C
a
o
,
“
B
l
o
c
k
c
h
a
i
n
-
b
a
se
d
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
me
t
h
o
d
o
l
o
g
i
e
s
i
n
smar
t
e
n
v
i
r
o
n
me
n
t
s
,
”
C
l
u
s
t
e
r
C
o
m
p
u
t
i
n
g
,
v
o
l
.
0
1
2
3
4
5
6
7
8
9
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
5
8
6
-
0
2
1
-
0
3
4
2
4
-
y.
[
1
3
]
R
.
C
.
G
e
y
e
r
,
T
.
K
l
e
i
n
,
a
n
d
M
.
N
a
b
i
,
“
D
i
f
f
e
r
e
n
t
i
a
l
l
y
p
r
i
v
a
t
e
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
:
a
c
l
i
e
n
t
l
e
v
e
l
p
e
r
s
p
e
c
t
i
v
e
,
”
n
o
.
N
i
p
s,
p
p
.
1
–
7
,
2
0
1
7
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
7
1
2
.
0
7
5
5
7
.
[
1
4
]
J.
W
a
n
g
e
t
a
l
.
,
“
A
f
i
e
l
d
g
u
i
d
e
t
o
f
e
d
e
r
a
t
e
d
o
p
t
i
m
i
z
a
t
i
o
n
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
2
1
0
7
.
0
6
9
1
7
,
p
p
.
1
–
8
8
,
2
0
2
1
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
1
0
7
.
0
6
9
1
7
.
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:
2
5
0
2
-
4
7
52
A
co
n
ce
p
tu
a
l a
p
p
r
o
a
ch
o
f o
p
timiz
a
tio
n
in
fed
era
ted
lea
r
n
in
g
(
F
a
r
h
a
n
n
a
Ma
r
’
i
)
297
[
1
5
]
Y
.
M
.
S
a
p
u
t
r
a
a
n
d
G
.
A
l
f
i
a
n
,
“
P
r
i
v
a
c
y
a
w
a
r
e
-
b
a
se
d
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
r
a
mew
o
r
k
f
o
r
d
a
t
a
s
h
a
r
i
n
g
p
r
o
t
e
c
t
i
o
n
o
f
i
n
t
e
r
n
e
t
o
f
t
h
i
n
g
s
d
e
v
i
c
e
s,
”
I
n
d
o
n
e
s
i
a
n
J
o
u
rn
a
l
o
f
E
l
e
c
t
r
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
3
1
,
n
o
.
2
,
p
p
.
9
7
9
–
9
8
5
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
e
c
s.
v
3
1
.
i
2
.
p
p
9
7
9
-
9
8
5
.
[
1
6
]
D
.
C
.
N
g
u
y
e
n
,
P
.
N
.
P
a
t
h
i
r
a
n
a
,
M
.
D
i
n
g
,
a
n
d
A
.
S
e
n
e
v
i
r
a
t
n
e
,
“
B
l
o
c
k
c
h
a
i
n
f
o
r
5
G
a
n
d
b
e
y
o
n
d
n
e
t
w
o
r
k
s:
A
st
a
t
e
o
f
t
h
e
a
r
t
s
u
r
v
e
y
,
”
J
o
u
rn
a
l
o
f
N
e
t
w
o
rk
a
n
d
C
o
m
p
u
t
e
r
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
6
6
,
n
o
.
A
p
r
i
l
,
p
.
1
0
2
6
9
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
n
c
a
.
2
0
2
0
.
1
0
2
6
9
3
.
[
1
7
]
A
.
H
a
r
d
e
t
a
l
.
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
mo
b
i
l
e
k
e
y
b
o
a
r
d
p
r
e
d
i
c
t
i
o
n
,
”
2
0
1
8
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
8
1
1
.
0
3
6
0
4
.
[
1
8
]
G
o
o
g
l
e
,
“
Y
o
u
r
v
o
i
c
e
&
a
u
d
i
o
d
a
t
a
s
t
a
y
s
p
r
i
v
a
t
e
w
h
i
l
e
G
o
o
g
l
e
A
ssi
s
t
a
n
t
i
m
p
r
o
v
e
s
-
G
o
o
g
l
e
A
ss
i
st
a
n
t
H
e
l
p
,
”
2
0
2
1
.
[
1
9
]
A
p
p
l
e
,
“
D
e
s
i
g
n
i
n
g
f
o
r
P
r
i
v
a
c
y
-
W
W
D
C
2
0
1
9
-
V
i
d
e
o
s
-
A
p
p
l
e
D
e
v
e
l
o
p
e
r
,
”
2
0
1
9
.
[
2
0
]
E.
D
a
r
z
i
d
e
h
k
a
l
a
n
i
,
M
.
G
h
a
sem
i
-
r
a
d
,
a
n
d
P
.
m
.
a
.
v
a
n
O
o
i
j
e
n
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
i
n
m
e
d
i
c
a
l
i
ma
g
i
n
g
:
P
a
r
t
I
I
:
met
h
o
d
s
,
c
h
a
l
l
e
n
g
e
s,
a
n
d
c
o
n
s
i
d
e
r
a
t
i
o
n
s
,
”
J
o
u
r
n
a
l
o
f
t
h
e
Am
e
r
i
c
a
n
C
o
l
l
e
g
e
o
f
R
a
d
i
o
l
o
g
y
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
a
c
r
.
2
0
2
2
.
0
3
.
0
1
6
.
[
2
1
]
W
e
B
a
n
k
,
“
U
t
i
l
i
z
a
t
i
o
n
o
f
F
A
T
E
i
n
a
n
t
i
mo
n
e
y
l
a
u
n
d
e
r
i
n
g
t
h
r
o
u
g
h
mu
l
t
i
p
l
e
b
a
n
k
s
,
”
p
.
2
0
2
0
,
2
0
2
0
[
o
n
l
i
n
e
]
:
h
t
t
p
s
:
/
/
w
w
w
.
f
e
d
a
i
.
o
r
g
/
c
a
s
e
s/
u
t
i
l
i
z
a
t
i
o
n
-
of
-
f
a
t
e
-
in
-
a
n
t
i
-
mo
n
e
y
-
l
a
u
n
d
e
r
i
n
g
-
t
h
r
o
u
g
h
-
m
u
l
t
i
p
l
e
-
b
a
n
k
s
/
[
2
2
]
I
n
t
e
l
,
“
I
n
t
e
l
a
n
d
c
o
n
s
i
l
i
e
n
t
j
o
i
n
f
o
r
c
e
s
t
o
f
i
g
h
t
f
i
n
a
n
c
i
a
l
f
r
a
u
d
w
i
t
h
A
I
,
”
[
o
n
l
i
n
e
]
:
h
t
t
p
s
:
/
/
w
w
w
.
i
n
t
e
l
.
c
o
m/
c
o
n
t
e
n
t
/
w
w
w
/
u
s/
e
n
/
n
e
w
sr
o
o
m/
n
e
w
s/
f
i
g
h
t
-
f
i
n
a
n
c
i
a
l
-
f
r
a
u
d
-
a
i
.
h
t
ml
#
g
s
.
f
4
p
o
7
2
[
2
3
]
J.
K
o
n
e
č
n
ý
,
B
.
M
c
M
a
h
a
n
,
a
n
d
D
.
R
a
ma
g
e
,
“
F
e
d
e
r
a
t
e
d
o
p
t
i
mi
z
a
t
i
o
n
:
d
i
st
r
i
b
u
t
e
d
o
p
t
i
mi
z
a
t
i
o
n
b
e
y
o
n
d
t
h
e
d
a
t
a
c
e
n
t
e
r
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
1
5
1
1
.
0
3
5
7
5
,
n
o
.
1
,
p
p
.
1
–
5
,
2
0
1
5
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
5
1
1
.
0
3
5
7
5
.
[
2
4
]
E.
Je
o
n
g
,
S
.
O
h
,
H
.
K
i
m,
J.
P
a
r
k
,
M
.
B
e
n
n
i
s,
a
n
d
S
.
-
L.
K
i
m,
“
C
o
mm
u
n
i
c
a
t
i
o
n
-
e
f
f
i
c
i
e
n
t
o
n
-
d
e
v
i
c
e
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
:
f
e
d
e
r
a
t
e
d
d
i
s
t
i
l
l
a
t
i
o
n
a
n
d
a
u
g
me
n
t
a
t
i
o
n
u
n
d
e
r
n
o
n
-
I
I
D
p
r
i
v
a
t
e
d
a
t
a
,
”
n
o
.
N
i
p
s
,
2
0
1
8
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
8
1
1
.
1
1
4
7
9
.
[
2
5
]
S
.
R
e
d
d
i
e
t
a
l
.
,
“
A
d
a
p
t
i
v
e
f
e
d
e
r
a
t
e
d
o
p
t
i
mi
z
a
t
i
o
n
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
r
Xi
v
:
2
0
0
3
.
0
0
2
9
5
,
n
o
.
2
,
p
p
.
1
–
3
8
,
2
0
2
0
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
0
0
3
.
0
0
2
9
5
.
[
2
6
]
F
.
M
a
r
’
i
,
A
.
A
.
S
u
p
i
a
n
t
o
,
a
n
d
F
.
A
.
B
a
c
h
t
i
a
r
,
“
C
o
mp
a
r
i
s
o
n
o
f
f
e
d
e
r
a
t
e
d
a
n
d
c
e
n
t
r
a
l
i
z
e
d
l
e
a
r
n
i
n
g
f
o
r
i
m
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
Re
s
e
a
r
c
h
,
Em
b
e
d
d
e
d
S
y
st
e
m
s
a
n
d
L
o
g
i
c
,
v
o
l
.
1
1
,
n
o
.
2
2
5
,
p
p
.
3
9
3
–
4
0
0
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
5
5
8
/
p
i
k
s
e
l
.
v
1
1
i
2
.
7
3
6
7
.
[
2
7
]
H
.
W
a
n
g
,
Z
.
K
a
p
l
a
n
,
D
.
N
i
u
,
a
n
d
B
.
Li
,
“
O
p
t
i
m
i
z
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
o
n
N
o
n
-
I
I
D
d
a
t
a
w
i
t
h
r
e
i
n
f
o
r
c
e
me
n
t
l
e
a
r
n
i
n
g
,
”
Pro
c
e
e
d
i
n
g
s
-
I
E
EE
I
N
FO
C
O
M
,
v
o
l
.
2
0
2
0
-
Ju
l
y
,
p
p
.
1
6
9
8
–
1
7
0
7
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
I
N
F
O
C
O
M
4
1
0
4
3
.
2
0
2
0
.
9
1
5
5
4
9
4
.
[
2
8
]
M
.
S
a
f
a
r
y
a
n
,
R
.
I
sl
a
m
o
v
,
X
.
Q
i
a
n
,
a
n
d
P
.
R
i
c
h
t
á
r
i
k
,
“
F
e
d
N
L:
ma
k
i
n
g
n
e
w
t
o
n
-
t
y
p
e
me
t
h
o
d
s
a
p
p
l
i
c
a
b
l
e
t
o
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
a
rXi
v
p
re
p
r
i
n
t
a
rXi
v
:
2
1
0
6
.
0
2
9
6
9
,
2
0
2
1
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
1
0
6
.
0
2
9
6
9
.
[
2
9
]
H
.
Zh
u
a
n
d
Y
.
J
i
n
,
“
M
u
l
t
i
-
o
b
j
e
c
t
i
v
e
e
v
o
l
u
t
i
o
n
a
r
y
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
rks
a
n
d
L
e
a
r
n
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
1
,
n
o
.
4
,
p
p
.
1
3
1
0
–
1
3
2
2
,
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
LS
.
2
0
1
9
.
2
9
1
9
6
9
9
.
[
3
0
]
R
.
K
a
n
a
g
a
v
e
l
u
e
t
a
l
.
,
“
C
E
-
F
e
d
:
c
o
m
mu
n
i
c
a
t
i
o
n
e
f
f
i
c
i
e
n
t
m
u
l
t
i
-
p
a
r
t
y
c
o
m
p
u
t
a
t
i
o
n
e
n
a
b
l
e
d
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
Arra
y
,
v
o
l
.
1
5
,
n
o
.
Ju
n
e
,
p
.
1
0
0
2
0
7
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
a
r
r
a
y
.
2
0
2
2
.
1
0
0
2
0
7
.
[
3
1
]
Z.
Li
n
,
G
.
S
.
M
e
m
b
e
r
,
a
n
d
H
.
L
i
u
,
“
C
F
LI
T
:
c
o
e
x
i
st
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
a
n
d
i
n
f
o
r
ma
t
i
o
n
t
r
a
n
sf
e
r
,
”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
Wi
re
l
e
ss
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
2
2
,
n
o
.
1
1
,
p
p
.
8
4
3
6
–
8
4
5
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
TW
C
.
2
0
2
3
.
3
2
6
3
1
4
8
.
[
3
2
]
X
.
D
o
n
g
,
S
.
Q
.
Z
h
a
n
g
,
A
.
L
i
,
a
n
d
H
.
T.
K
u
n
g
,
“
S
p
h
e
r
e
F
e
d
:
h
y
p
e
r
s
p
h
e
r
i
c
a
l
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
I
n
Eu
r
o
p
e
a
n
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
,
n
o
.
i
,
p
p
.
1
–
3
0
,
2
0
2
2
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
2
0
7
.
0
9
4
1
3
.
[
3
3
]
D
.
Jh
u
n
j
h
u
n
w
a
l
a
,
“
F
e
d
V
A
R
P
:
t
a
c
k
l
i
n
g
t
h
e
v
a
r
i
a
n
c
e
d
u
e
t
o
p
a
r
t
i
a
l
c
l
i
e
n
t
p
a
r
t
i
c
i
p
a
t
i
o
n
,
”
I
n
U
n
c
e
r
t
a
i
n
t
y
i
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
n
o
.
U
a
i
,
2
0
2
2
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
2
0
7
.
1
4
1
3
0
.
[
3
4
]
J.
K
o
n
e
č
n
ý
,
H
.
B
.
M
c
M
a
h
a
n
,
D
.
R
a
mag
e
,
a
n
d
P
.
R
i
c
h
t
á
r
i
k
,
“
F
e
d
e
r
a
t
e
d
o
p
t
i
mi
z
a
t
i
o
n
:
d
i
s
t
r
i
b
u
t
e
d
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
f
o
r
o
n
-
d
e
v
i
c
e
i
n
t
e
l
l
i
g
e
n
c
e
,
”
a
rX
i
v
p
re
p
r
i
n
t
a
rX
i
v
:
1
6
1
0
.
0
2
5
2
7
,
p
p
.
1
–
3
8
,
2
0
1
6
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
6
1
0
.
0
2
5
2
7
.
[
3
5
]
J.
H
a
n
a
n
d
Y
.
H
a
n
,
“
D
e
F
L
:
d
e
c
e
n
t
r
a
l
i
z
e
d
w
e
i
g
h
t
a
g
g
r
e
g
a
t
i
o
n
f
o
r
c
r
o
ss
-
s
i
l
o
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
a
rXi
v
p
r
e
p
r
i
n
t
a
rXi
v
:
2
2
0
8
.
0
0
8
4
8
,
v
o
l
.
A
u
g
u
s
t
,
2
0
2
2
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
2
0
8
.
0
0
8
4
8
.
[
3
6
]
X
.
M
a
,
H
.
S
u
n
,
R
.
Q
.
H
u
,
a
n
d
Y
.
Q
i
a
n
,
“
A
n
e
w
i
m
p
l
e
m
e
n
t
a
t
i
o
n
o
f
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
p
r
i
v
a
c
y
a
n
d
s
e
c
u
r
i
t
y
e
n
h
a
n
c
e
m
e
n
t
,
”
I
n
G
L
O
BEC
O
M
2
0
2
2
-
2
0
2
2
I
EEE
G
l
o
b
a
l
C
o
m
m
u
n
i
c
a
t
i
o
n
s
C
o
n
f
e
re
n
c
e
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
2
0
8
.
0
1
8
2
6
.
[
3
7
]
C
.
X
i
e
,
P
.
-
Y
.
C
h
e
n
,
C
.
Zh
a
n
g
,
a
n
d
B
.
Li
,
“
I
mp
r
o
v
i
n
g
p
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
v
e
r
t
i
c
a
l
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
b
y
e
f
f
i
c
i
e
n
t
c
o
mm
u
n
i
c
a
t
i
o
n
w
i
t
h
A
D
M
M
,
”
I
n
2
0
2
4
I
EEE
C
o
n
f
e
re
n
c
e
o
n
S
e
c
u
re
a
n
d
T
r
u
st
w
o
rt
h
y
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
(
S
a
T
ML)
,
p
p
.
1
–
2
2
,
2
0
2
2
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
2
0
7
.
1
0
2
2
6
.
[
3
8
]
P
.
A
n
t
u
n
e
s,
N
.
H
.
T
h
u
a
n
,
D
.
J
o
h
n
st
o
n
e
,
a
n
d
N
.
H
o
a
n
g
,
“
N
a
t
u
r
e
a
n
d
p
u
r
p
o
s
e
o
f
c
o
n
c
e
p
t
u
a
l
f
r
a
m
e
w
o
r
k
s
i
n
d
e
s
i
g
n
sc
i
e
n
c
e
,
”
S
c
a
n
d
i
n
a
v
i
a
n
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
,
v
o
l
.
3
3
,
n
o
.
2
,
p
p
.
5
9
–
9
6
,
2
0
2
1
.
[
3
9
]
H
.
B
.
M
c
M
a
h
a
n
,
E
.
M
o
o
r
e
,
D
.
R
a
m
a
g
e
,
S
.
H
a
mp
s
o
n
,
a
n
d
B
.
A
.
y
A
r
c
a
s
,
“
C
o
mm
u
n
i
c
a
t
i
o
n
-
e
f
f
i
c
i
e
n
t
l
e
a
r
n
i
n
g
o
f
d
e
e
p
n
e
t
w
o
r
k
f
r
o
m
d
e
c
e
n
t
r
a
l
i
z
e
d
d
a
t
a
,
”
i
n
2
0
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
S
t
a
t
i
s
t
i
c
s
(
AI
S
T
AT
S
)
,
2
0
1
7
,
v
o
l
.
5
4
,
p
.
1
0
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
6
0
2
.
0
5
6
2
9
.
[
4
0
]
J.
K
o
n
e
č
n
ý
,
H
.
B
.
M
c
M
a
h
a
n
,
F
.
X
.
Y
u
,
P
.
R
i
c
h
t
á
r
i
k
,
A
.
T
.
S
u
r
e
s
h
,
a
n
d
D
.
B
a
c
o
n
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
:
s
t
r
a
t
e
g
i
e
s
f
o
r
i
mp
r
o
v
i
n
g
c
o
mm
u
n
i
c
a
t
i
o
n
e
f
f
i
c
i
e
n
c
y
,
”
a
rX
i
v
p
r
e
p
ri
n
t
a
r
Xi
v
:
1
6
1
0
.
0
5
4
9
2
,
p
p
.
1
–
1
0
,
2
0
1
6
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
6
1
0
.
0
5
4
9
2
.
[
4
1
]
M
.
L
u
b
i
s
a
n
d
D
.
O
.
D
.
H
a
n
d
a
y
a
n
i
,
“
Th
e
r
e
l
a
t
i
o
n
s
h
i
p
o
f
p
e
r
s
o
n
a
l
d
a
t
a
p
r
o
t
e
c
t
i
o
n
t
o
w
a
r
d
s
i
n
t
e
r
n
e
t
a
d
d
i
c
t
i
o
n
:
C
y
b
e
r
c
r
i
m
e
s,
p
o
r
n
o
g
r
a
p
h
y
a
n
d
r
e
d
u
c
e
d
p
h
y
si
c
a
l
a
c
t
i
v
i
t
y
,
”
Pr
o
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
9
7
,
n
o
.
2
0
2
1
,
p
p
.
1
5
1
–
1
6
1
,
2
0
2
1
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s
.
2
0
2
1
.
1
2
.
1
2
9
.
[
4
2
]
W
h
i
t
e
H
o
u
se,
“
C
o
n
su
mer d
a
t
a
p
r
i
v
a
c
y
i
n
a
n
e
t
w
o
r
k
e
d
w
o
r
l
d
:
a
f
r
a
m
e
w
o
r
k
f
o
r
p
r
o
t
e
c
t
i
n
g
p
r
i
v
a
c
y
a
n
d
p
r
o
m
o
t
i
n
g
i
n
n
o
v
a
t
i
o
n
i
n
t
h
e
g
l
o
b
a
l
d
i
g
i
t
a
l
e
c
o
n
o
my
w
h
i
t
e
h
o
u
se
r
e
p
o
r
t
*
f
o
r
e
w
o
r
d
,
”
J
o
u
r
n
a
l
o
f
Pr
i
v
a
c
y
a
n
d
C
o
n
f
i
d
e
n
t
i
a
l
i
t
y
,
v
o
l
.
4
,
n
o
.
2
,
p
p
.
9
5
–
1
4
2
,
2
0
1
2
.
[
4
3
]
K
.
B
o
g
a
c
k
a
,
K
.
W
.
M
i
c
h
i
n
i
e
w
s
k
a
,
M
.
P
a
p
r
z
y
c
k
i
,
M
.
G
.
A
n
a
t
a
s
y
i
a
D
a
n
i
l
e
n
k
a
,
L.
T
a
ssa
k
o
s
,
a
n
d
E.
G
a
r
r
o
,
“
I
n
t
r
o
d
u
c
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
i
n
t
o
i
n
t
e
r
n
e
t
o
f
t
h
i
n
g
s
e
c
o
sy
s
t
e
m
s
–
p
r
e
l
i
mi
n
a
r
y
c
o
n
s
i
d
e
r
a
t
i
o
n
s
,
”
a
rX
i
v
,
n
o
.
i
,
p
.
7
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
W
F
-
I
o
T5
4
3
8
2
.
2
0
2
2
.
1
0
1
5
2
1
4
2
.
[
4
4
]
G
l
o
b
a
l
d
o
t
s
,
“
4
1
.
6
B
i
l
l
i
o
n
I
o
T
d
e
v
i
c
e
s
w
i
l
l
b
e
g
e
n
e
r
a
t
i
n
g
7
9
.
4
Ze
t
t
a
b
y
t
e
s
o
f
d
a
t
a
i
n
2
0
2
5
,
”
2
0
1
9
,
[
o
n
l
i
n
e
]
:
h
t
t
p
s
:
/
/
w
w
w
.
g
l
o
b
a
l
d
o
t
s.
c
o
m
/
r
e
s
o
u
r
c
e
s/
b
l
o
g
/
4
1
-
6
-
b
i
l
l
i
o
n
-
i
o
t
-
d
e
v
i
c
e
s
-
w
i
l
l
-
be
-
g
e
n
e
r
a
t
i
n
g
-
79
-
4
-
z
e
t
t
a
b
y
t
e
s
-
of
-
d
a
t
a
-
in
-
2
0
2
5
/
[
4
5
]
L.
U
.
K
h
a
n
,
W
.
S
a
a
d
,
Z
.
H
a
n
,
E
.
H
o
ssa
i
n
,
C
.
S
e
o
n
H
o
n
g
,
a
n
d
S
.
M
e
m
b
e
r
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
i
n
t
e
r
n
e
t
o
f
t
h
i
n
g
s
:
r
e
c
e
n
t
a
d
v
a
n
c
e
s,
t
a
x
o
n
o
m
y
,
a
n
d
o
p
e
n
c
h
a
l
l
e
n
g
e
s,
”
I
EE
E
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
u
rv
e
y
s
a
n
d
T
u
t
o
ri
a
l
s
,
p
p
.
1
–
4
1
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
0
0
9
.
1
3
0
1
2
.
[
4
6
]
H
.
B
.
M
c
M
a
h
a
n
,
E.
M
o
o
r
e
,
D
.
R
a
m
a
g
e
,
S
.
H
a
m
p
s
o
n
,
a
n
d
B
.
A
.
y
A
r
c
a
s
,
“
C
o
mm
u
n
i
c
a
t
i
o
n
-
e
f
f
i
c
i
e
n
t
l
e
a
r
n
i
n
g
o
f
d
e
e
p
n
e
t
w
o
r
k
s
f
r
o
m
d
e
c
e
n
t
r
a
l
i
z
e
d
d
a
t
a
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
S
t
a
t
i
st
i
c
s,
AI
S
T
ATS
2
0
1
7
,
v
o
l
.
5
4
,
2
0
1
7
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
1
6
0
2
.
0
5
6
2
9
.
[
4
7
]
R
.
S
h
o
k
r
i
a
n
d
V
.
S
h
mat
i
k
o
v
,
“
P
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
,
”
2
0
1
5
5
3
r
d
A
n
n
u
a
l
Al
l
e
r
t
o
n
C
o
n
f
e
re
n
c
e
o
n
C
o
m
m
u
n
i
c
a
t
i
o
n
,
C
o
n
t
r
o
l
,
a
n
d
C
o
m
p
u
t
i
n
g
,
A
l
l
e
rt
o
n
2
0
1
5
,
p
p
.
9
0
9
–
9
1
0
,
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
A
LLER
TO
N
.
2
0
1
5
.
7
4
4
7
1
0
3
.
[
4
8
]
H
.
Z
h
a
o
,
Z.
Li
,
a
n
d
P
.
R
i
c
h
t
á
r
i
k
,
“
F
e
d
P
A
G
E:
A
f
a
st
l
o
c
a
l
st
o
c
h
a
s
t
i
c
g
r
a
d
i
e
n
t
met
h
o
d
f
o
r
c
o
mm
u
n
i
c
a
t
i
o
n
-
e
f
f
i
c
i
e
n
t
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
r
Xi
v
:
2
1
0
8
.
0
4
7
5
5
,
2
0
2
1
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
1
0
8
.
0
4
7
5
5
.
[
4
9
]
H
.
Y
u
a
n
a
n
d
T.
M
a
,
“
F
e
d
e
r
a
t
e
d
a
c
c
e
l
e
r
a
t
e
d
st
o
c
h
a
st
i
c
g
r
a
d
i
e
n
t
d
e
sce
n
t
,
”
Ad
v
a
n
c
e
s
i
n
N
e
u
r
a
l
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
s
t
e
m
s
,
v
o
l
.
2
0
2
0
-
D
e
c
e
m,
n
o
.
N
e
u
r
I
P
S
,
p
p
.
1
–
1
3
,
2
0
2
0
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
0
0
6
.
0
8
9
5
0
.
Evaluation Warning : The document was created with Spire.PDF for Python.