I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
878
~
887
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
8
7
8
-
8
8
7
878
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
G
ra
dien
t
desce
nt
o
ptimiza
tion ba
se
d weighte
d f
edera
ted
lea
rning
f
o
r priva
cy
-
preserv
ing
f
ra
mewo
rk
G
urura
j
P
ra
ka
s
h M
urt
hy
,
C
ha
nd
ra
s
hek
ha
r
P
o
m
u Cha
v
a
n
D
e
p
a
r
t
me
n
t
o
f
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
r
i
n
g
,
P
ES
U
n
i
v
e
r
si
t
y
,
B
a
n
g
a
l
o
r
e
,
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
J
u
n
9
,
2
0
2
5
R
ev
is
ed
Dec
3
0
,
2
0
2
5
Acc
ep
ted
J
an
2
2
,
2
0
2
6
F
e
d
e
ra
ted
lea
rn
in
g
(F
L)
is
a
d
iss
e
m
in
a
ted
m
a
c
h
in
e
lea
rn
in
g
(
M
L)
p
a
ra
d
ig
m
th
a
t
g
a
i
n
e
d
si
g
n
ifi
c
a
n
t
c
o
n
sid
e
r
a
ti
o
n
in
m
o
d
e
rn
d
a
y
s,
p
a
rti
c
u
l
a
rly
in
a
d
o
m
a
in
o
f
t
h
e
in
ter
n
e
t
o
f
th
i
n
g
s
(Io
T).
F
L
sa
v
e
s
c
o
m
m
u
n
ica
ti
o
n
b
a
n
d
wi
d
th
wh
e
n
c
o
m
p
a
re
d
to
c
e
n
tralize
d
M
L
p
ro
c
e
ss
e
s
b
y
e
li
m
in
a
ti
n
g
th
e
n
e
e
d
to
tran
sm
it
ra
w
c
li
e
n
t
d
a
ta
to
a
c
e
n
t
ra
l
se
rv
e
r,
th
e
re
b
y
e
n
h
a
n
c
i
n
g
d
a
t
a
p
riv
a
c
y
.
Ne
v
e
rth
e
les
s,
p
a
rti
c
ip
a
n
t
p
r
iv
a
c
y
is
stil
l
c
o
m
p
r
o
m
ise
d
th
ro
u
g
h
in
fe
re
n
c
e
a
tt
a
c
k
s
a
n
d
sim
il
a
r
th
re
a
ts.
Ad
d
it
io
n
a
ll
y
,
a
d
a
ta
e
x
c
e
ll
e
n
c
e
p
ro
v
i
d
e
d
th
r
o
u
g
h
c
li
e
n
ts
c
a
n
d
iffers
si
g
n
if
ica
n
tl
y
,
a
n
d
e
x
c
e
ss
iv
e
i
n
c
lu
s
i
o
n
o
f
l
o
w
-
q
u
a
li
ty
d
a
ta
d
u
ri
n
g
train
in
g
m
a
y
d
e
g
ra
d
e
t
h
e
o
v
e
ra
ll
p
e
rfo
rm
a
n
c
e
o
f
t
h
e
g
lo
b
a
l
m
o
d
e
l.
He
n
c
e
,
th
is
re
se
a
rc
h
in
tr
o
d
u
c
e
s
a
g
ra
d
ien
t
d
e
sc
e
n
t
o
p
ti
m
iza
ti
o
n
a
ss
isted
we
ig
h
ted
fe
d
e
ra
ted
lea
rn
i
n
g
(G
DO
-
WF
L)
m
e
th
o
d
f
o
r
p
riv
a
c
y
p
re
se
rv
a
ti
o
n
.
Th
e
p
ro
p
o
se
d
G
DO
-
WF
L
a
p
p
r
o
a
c
h
is
sig
n
ifi
c
a
n
tl
y
e
fficie
n
t
a
s
it
st
re
n
g
th
e
n
s
p
riv
a
c
y
p
re
se
rv
a
ti
o
n
th
r
o
u
g
h
re
d
u
c
in
g
e
x
p
o
su
re
t
o
in
fe
re
n
c
e
a
t
tac
k
s
a
n
d
o
p
ti
m
ise
s
g
ra
d
ien
t
u
p
d
a
tes
fo
r
s
e
c
u
re
lea
rn
in
g
.
T
h
ro
u
g
h
we
ig
h
ti
n
g
c
li
e
n
t
c
o
n
tri
b
u
ti
o
n
s b
a
se
d
o
n
d
a
ta
q
u
a
li
t
y
,
a
n
u
n
d
e
sira
b
le
e
ffe
c
t
o
f
lo
w
-
q
u
a
li
ty
d
a
ta
c
a
n
b
e
m
in
imis
e
d
,
h
e
lp
i
n
g
t
o
m
a
in
tain
a
stre
n
g
t
h
a
s
we
ll
a
s
a
c
c
u
ra
c
y
o
f
th
e
g
lo
b
a
l
m
o
d
e
l
.
T
h
e
e
x
p
e
rime
n
tal
re
su
lt
s
il
l
u
stra
te
a
p
r
o
p
o
se
d
G
DO
-
WF
L
a
p
p
ro
a
c
h
m
a
in
tain
s
a
n
o
v
e
ra
ll
a
c
c
u
ra
c
y
o
f
9
9
.
3
a
n
d
9
1
.
5
%
o
n
M
NIST
a
n
d
CIF
AR
-
1
0
d
a
tas
e
ts as
c
o
m
p
a
re
d
t
o
th
e
e
x
isti
n
g
m
e
th
o
d
o
f
F
e
d
lab
X
m
e
th
o
d
.
K
ey
w
o
r
d
s
:
C
en
tr
alize
d
m
ac
h
in
e
lear
n
i
n
g
C
o
m
m
u
n
icatio
n
b
a
n
d
wid
th
Fed
er
ated
lear
n
in
g
Gr
ad
ien
t d
escen
t o
p
tim
izatio
n
I
n
ter
n
et
o
f
th
in
g
s
Priv
ac
y
p
r
eser
v
i
n
g
W
eig
h
ted
f
ed
er
ated
lea
r
n
in
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
:
Gu
r
u
r
aj
Pra
k
ash
M
u
r
th
y
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
,
PES U
n
i
v
er
s
ity
B
an
g
alo
r
e,
I
n
d
ia
E
m
ail: g
u
r
u
r
aj
p
@
p
es.e
d
u
1.
I
NT
RO
D
UCT
I
O
N
T
h
r
o
u
g
h
a
n
ad
v
an
ce
m
en
t
o
f
a
r
tific
ial
in
tellig
en
ce
(
AI
)
a
p
p
r
o
ac
h
es
an
d
tec
h
n
iq
u
es,
in
ter
n
et
o
f
th
in
g
s
(
I
o
T
)
s
etu
p
s
ar
e
b
r
o
a
d
ly
o
r
g
a
n
is
ed
in
v
ar
io
u
s
f
ield
s
s
u
ch
as
wir
eles
s
n
etwo
r
k
s
[
1
]
,
h
ea
lth
ca
r
e,
s
m
ar
t
cities,
an
d
th
e
m
ilit
ar
y
[
2
]
.
Nev
e
r
th
e
less
,
co
n
v
en
tio
n
al
clo
u
d
co
m
p
u
tin
g
f
r
am
ew
o
r
k
s
f
ac
e
ch
allen
g
es
in
m
ee
tin
g
th
e
d
ata
p
r
o
ce
s
s
in
g
r
e
q
u
ir
em
e
n
ts
o
f
r
ea
l
-
wo
r
ld
ap
p
licatio
n
s
d
u
e
to
lim
itatio
n
s
in
n
etwo
r
k
b
a
n
d
wid
th
an
d
g
r
o
win
g
co
n
ce
r
n
s
o
v
e
r
d
ata
p
r
iv
ac
y
[
3
]
.
T
o
tack
le
th
e
c
o
n
s
tr
ain
t
s
,
ed
g
e
co
m
p
u
tin
g
[
4
]
is
d
ev
elo
p
ed
with
in
a
co
m
p
u
tatio
n
f
r
am
ewo
r
k
o
f
I
o
T
.
I
t
d
ev
elo
p
s
an
ed
g
e
s
er
v
er
f
o
r
lo
ca
l
p
r
o
ce
s
s
in
g
th
at
em
p
lo
y
s
ac
tu
al
in
f
o
r
m
atio
n
th
r
o
u
g
h
ag
g
r
eg
at
io
n
,
m
in
in
g
,
o
r
co
m
m
u
n
icati
o
n
o
p
er
atio
n
s
[
5
]
.
I
n
ed
g
e
c
o
m
p
u
tin
g
,
th
e
ed
g
e
s
er
v
er
b
ec
o
m
es
a
s
ig
n
if
ican
t
p
ar
t
as
a
p
r
im
ar
y
d
is
p
en
s
atio
n
to
o
l
th
at
g
iv
es
ap
p
r
o
p
r
iate
lo
ca
l
s
er
v
ices
b
y
th
e
en
tire
clo
u
d
s
er
v
ice
ar
c
h
itectu
r
e
[
6
]
.
T
h
u
s
,
a
b
lo
c
k
ag
e
o
f
co
m
p
u
tatio
n
as
well
as
co
m
m
u
n
icatio
n
f
o
r
co
n
v
en
tio
n
al
clo
u
d
-
b
ased
f
r
a
m
ewo
r
k
s
h
as
to
b
e
ad
d
r
ess
ed
.
T
h
ese
d
ata
ar
e
o
f
ten
s
o
u
r
ce
d
f
r
o
m
ed
g
e
d
ev
ices
lik
e
s
m
ar
tp
h
o
n
es,
h
ea
lth
ca
r
e,
g
lo
b
al
p
o
s
itio
n
in
g
s
y
s
tem
(
GP
S
)
d
ev
ices,
an
d
s
o
o
n
[
7
]
.
Nev
er
th
eless
,
th
is
d
ata
m
o
s
tly
in
v
o
lv
es p
a
r
ticip
an
ts
’
d
ata,
m
ed
ical
r
ec
o
r
d
s
,
as we
ll a
s
tr
av
el
an
tiq
u
ity
[
8
]
.
Un
a
u
th
o
r
is
ed
ac
ce
s
s
to
th
is
p
er
s
o
n
al
d
ata
lead
s
to
an
im
p
o
r
tan
t
im
p
air
m
e
n
t.
Mo
r
e
o
v
er
,
i
n
p
ar
ticu
lar
d
ed
icate
d
o
r
g
an
is
atio
n
s
,
d
ata
s
h
ar
in
g
is
n
o
t
allo
wab
le
[
9
]
.
T
h
u
s
,
p
r
eser
v
in
g
th
e
p
r
iv
ac
y
o
f
p
ar
tic
ip
an
ts
’
in
f
o
r
m
atio
n
d
u
r
in
g
p
e
r
f
o
r
m
in
g
m
ac
h
in
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Gra
d
ien
t d
escen
t o
p
timiz
a
tio
n
b
a
s
ed
w
eig
h
ted
fed
era
ted
lea
r
n
in
g
fo
r
p
r
iva
cy
-
…
(
Gu
r
u
r
a
j
P
r
a
ka
s
h
Mu
r
th
y)
879
lear
n
in
g
(
ML
)
b
ec
o
m
es
o
f
s
u
p
r
em
e
s
ig
n
if
ican
ce
[
1
0
]
.
T
h
e
f
ed
er
ated
lear
n
in
g
(
FL)
is
th
e
m
o
s
t
im
p
o
r
tan
t
ad
v
an
ce
m
e
n
t,
wh
ich
b
ec
o
m
es
a
cr
u
cial
p
ar
t
in
th
e
j
o
in
t
lear
n
in
g
ap
p
r
o
ac
h
.
T
h
e
FL
h
as
b
e
en
b
r
o
a
d
ly
u
tili
s
ed
in
v
ar
io
u
s
ar
ea
s
lik
e
b
lo
ck
ch
a
in
,
im
ag
e
p
r
o
ce
s
s
in
g
,
co
m
p
u
t
er
v
is
io
n
,
m
ed
ical
im
ag
in
g
,
a
s
well
a
s
au
to
m
atic
s
y
s
tem
s
[
1
1
]
,
[
1
2
]
.
Fo
r
c
o
n
v
en
tio
n
al
ce
n
tr
alize
d
lear
n
in
g
,
it
n
ee
d
s
to
g
ath
er
a
m
ax
im
u
m
am
o
u
n
t
o
f
u
s
e
r
in
f
o
r
m
atio
n
f
o
r
tr
ain
i
n
g
a
n
et
wo
r
k
[
1
3
]
.
Nev
er
th
eless
,
u
s
er
in
f
o
r
m
atio
n
in
v
o
lv
es
p
e
r
s
o
n
al
d
ata,
wh
ich
r
esu
lts
in
u
s
er
in
f
o
r
m
atio
n
o
u
tf
lo
w
.
T
o
ig
n
o
r
e
p
r
iv
ac
y
o
u
tf
lo
w
an
d
in
ter
r
u
p
tio
n
o
f
th
e
d
ata
k
ey
s
,
FL
is
d
ev
elo
p
e
d
[
1
4
]
.
As
an
e
v
o
lv
i
n
g
d
is
tr
ib
u
ted
lear
n
in
g
m
o
d
el,
FL
h
as
tr
ain
ed
m
o
d
el
p
a
r
a
m
eter
s
f
r
o
m
v
ar
io
u
s
u
s
er
s
s
u
p
p
o
r
tiv
ely
th
r
o
u
g
h
a
lack
o
f
p
er
ce
iv
in
g
th
eir
ac
tu
al
in
f
o
r
m
atio
n
[
1
5
]
.
T
h
e
m
o
d
el
is
a
p
r
iv
ac
y
p
r
o
tectio
n
m
et
h
o
d
w
h
ich
at
tain
ed
th
e
tr
u
s
t
o
f
m
o
r
e
p
a
r
ticip
an
ts
.
T
h
u
s
,
FL
h
as
s
tim
u
lated
ex
ten
s
iv
e
ap
p
r
eh
e
n
s
io
n
in
v
ar
io
u
s
f
ield
s
[
1
6
]
.
T
h
e
FL
p
r
o
ce
s
s
also
im
p
r
o
v
es
th
e
g
en
er
aliza
tio
n
ac
cu
r
ac
y
as
ea
ch
clien
t
is
eq
u
ip
p
ed
to
h
a
n
d
le
th
e
d
a
ta
n
o
t
ex
p
o
s
ed
to
it
p
r
ev
io
u
s
ly
,
d
u
e
to
co
llab
o
r
ativ
e
lear
n
in
g
with
th
e
o
th
er
clien
ts
,
wh
ich
ar
e
ex
p
o
s
ed
to
t
h
e
d
ata
[
1
7
]
.
Ultim
ately
,
t
h
e
e
f
f
ec
tiv
en
ess
o
f
th
e
g
lo
b
al
m
o
d
el
is
en
h
an
ce
d
d
u
e
to
th
e
m
o
d
el
ag
g
r
e
g
atio
n
at
t
h
e
s
er
v
er
.
FL
ar
ch
itectu
r
e
s
h
o
u
ld
im
p
r
o
v
e
th
e
p
er
s
o
n
aliza
tio
n
,
g
en
er
aliza
tio
n
,
an
d
g
l
o
b
al
ac
c
u
r
ac
y
[
1
8
]
,
[
1
9
]
.
FL
a
r
ch
itectu
r
e
s
av
es
co
m
m
u
n
icatio
n
b
an
d
wid
th
wh
e
n
co
m
p
ar
ed
to
t
h
e
ce
n
tr
alis
ed
ML
p
r
o
ce
s
s
an
d
is
also
co
n
s
id
er
ed
t
o
b
e
p
r
iv
ac
y
-
p
r
eser
v
in
g
as
th
e
r
aw
d
ata
at
th
e
clien
ts
n
ee
d
n
o
t
b
e
tr
an
s
m
itted
to
th
e
s
er
v
er
f
o
r
th
e
FL
lear
n
in
g
[
2
0
]
.
T
h
e
p
r
ev
io
u
s
wo
r
k
s
b
ased
o
n
th
e
FL
-
b
ased
p
r
iv
ac
y
p
r
eser
v
in
g
ar
e
d
is
cu
s
s
ed
h
er
e,
alo
n
g
with
th
eir
ad
v
an
tag
es
an
d
lim
itatio
n
s
lim
itatio
n
s
.
Yan
et
a
l.
[
2
1
]
in
tr
o
d
u
ce
d
th
e
v
a
r
io
u
s
co
m
m
u
n
i
ca
tio
n
ef
f
ec
tiv
en
ess
m
ec
h
an
is
m
s
as
well
as
p
r
iv
ac
y
-
p
r
eser
v
in
g
cr
y
p
to
g
r
ap
h
ic
ap
p
r
o
ac
h
es.
A
p
r
iv
ac
y
-
p
r
eser
v
in
g
m
eth
o
d
in
teg
r
ates
th
e
cr
y
p
to
g
r
a
p
h
ic
a
p
p
r
o
ac
h
es
an
d
co
m
m
u
n
icatio
n
n
etwo
r
k
in
g
s
o
lu
tio
n
s
f
o
r
s
ec
u
r
in
g
s
e
n
s
itiv
e
d
ata.
I
n
th
at
in
tr
o
d
u
ce
d
ap
p
r
o
ac
h
,
Kaf
k
a
was
in
tr
o
d
u
ce
d
f
o
r
co
m
m
u
n
i
ca
tio
n
d
is
tr
ib
u
tio
n
,
a
Dif
f
ie
-
Hellm
an
m
eth
o
d
f
o
r
s
ec
u
r
e
s
er
v
er
ag
g
r
eg
atio
n
,
as
well
as
g
r
ad
ien
t
d
is
cr
ep
an
c
y
p
r
iv
ac
y
f
o
r
i
n
tr
u
s
io
n
attac
k
an
ticip
atio
n
.
An
in
tr
o
d
u
ce
d
m
eth
o
d
p
r
eser
v
ed
t
r
ain
in
g
ef
f
ec
tiv
en
ess
wh
ile
b
e
in
g
ca
p
ab
le
o
f
s
o
lv
in
g
g
r
a
d
ien
t
o
u
tf
lo
w
is
s
u
es
as
well
as
in
ter
f
er
en
ce
attac
k
s
.
R
ec
en
tly
,
th
e
d
ev
elo
p
m
en
t
o
f
Kaf
k
a
-
Z
o
o
k
ee
p
e
r
h
as
en
a
b
led
asy
n
ch
r
o
n
o
u
s
co
m
m
u
n
icatio
n
a
n
d
s
ec
u
r
e,
r
o
le
-
b
ased
ac
ce
s
s
co
n
tr
o
l,
p
r
o
v
id
in
g
an
o
n
y
m
o
u
s
,
an
d
r
eliab
le
d
ata
p
r
o
ce
s
s
in
g
ca
p
ab
ilit
ies.
Ho
wev
er
,
th
e
in
tr
o
d
u
ce
d
m
eth
o
d
d
id
n
o
t
ad
d
r
e
s
s
r
o
b
u
s
tn
ess
to
h
eter
o
g
en
e
o
u
s
d
ata
d
is
tr
ib
u
tio
n
s
o
r
B
y
za
n
tin
e
th
r
ea
ts
,
an
d
Ka
f
k
a
in
teg
r
atio
n
ad
d
ed
co
m
p
le
x
ity
.
C
h
en
et
a
l
.
[
2
2
]
p
r
esen
t
ed
a
p
r
ac
tical
an
d
ef
f
icien
t
p
r
iv
ac
y
-
p
r
eser
v
in
g
f
ed
er
ated
lear
n
in
g
(
PEPFL)
m
o
d
el.
I
n
itially
,
a
b
o
o
s
ted
d
is
s
em
in
ated
E
lGam
a
l
cr
y
p
to
s
y
s
tem
was
d
ev
elo
p
ed
to
ad
d
r
ess
a
m
u
lti
-
k
ey
is
s
u
e
in
FL.
T
h
en
,
th
e
p
r
ac
tical
p
ar
tially
s
in
g
le
in
s
tr
u
ctio
n
m
u
ltip
le
d
ata
(
PS
I
MD
)
s
tr
u
ctu
r
e
was
in
tr
o
d
u
ce
d
to
en
co
d
e
a
p
lain
tex
t
m
atr
i
x
with
in
in
d
iv
id
u
al
p
lain
tex
t
f
o
r
en
cr
y
p
tio
n
,
e
n
h
a
n
cin
g
en
c
r
y
p
tio
n
p
er
f
o
r
m
a
n
ce
,
as
well
as
m
in
im
ized
co
m
m
u
n
icatio
n
co
s
t
in
th
e
in
co
m
p
letely
h
o
m
o
m
o
r
p
h
ic
cr
y
p
to
s
y
s
tem
.
Mo
r
eo
v
er
,
ac
co
r
d
in
g
to
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
an
d
a
d
ev
elo
p
ed
cr
y
p
to
s
y
s
tem
m
o
d
el,
a
n
ew
p
r
iv
ac
y
-
p
r
ese
r
v
in
g
FL
s
tr
u
ctu
r
e
was
in
tr
o
d
u
ce
d
th
r
o
u
g
h
th
e
u
tili
s
atio
n
o
f
m
o
m
e
n
tu
m
g
r
a
d
ien
t
d
escen
t
(
MG
D)
.
Ho
we
v
er
,
wh
ile
PEPFL
with
en
h
a
n
ce
d
E
lGam
al
an
d
PS
I
MD
o
f
f
er
ed
im
p
r
o
v
ed
s
ec
u
r
ity
,
th
e
m
o
d
el
f
ac
e
d
ch
allen
g
es
in
r
ea
l
-
tim
e
ex
ec
u
tio
n
d
u
e
to
th
e
co
m
p
u
tatio
n
al
o
v
e
r
h
ea
d
in
tr
o
d
u
ce
d
b
y
h
o
m
o
m
o
r
p
h
ic
en
cr
y
p
tio
n
an
d
m
atr
ix
-
b
ased
en
c
o
d
in
g
.
Su
m
itra
et
a
l
.
[
2
3
]
d
ev
elo
p
ed
a
HAFed
L
,
a
n
im
p
r
o
v
ed
n
o
v
el
Hess
ian
-
awa
r
e
ad
ap
tiv
e
p
r
i
v
ac
y
p
r
eser
v
i
n
g
FL
s
ch
em
e
.
T
h
e
ar
ch
itectu
r
e
in
tr
o
d
u
ce
s
s
p
ec
if
ic
en
h
an
ce
m
e
n
ts
to
s
tr
en
g
th
en
p
r
iv
ac
y
ag
ai
n
s
t
g
r
ad
ien
t
leak
ag
e
attac
k
s
(
GL
A)
.
T
h
ese
m
o
d
i
f
icatio
n
s
en
s
u
r
e
th
at
th
e
m
o
d
el’
s
ef
f
ec
tiv
en
ess
is
n
o
t
s
ig
n
if
ican
tly
co
m
p
r
o
m
is
e
d
wh
ile
i
m
p
r
o
v
in
g
o
v
er
all
s
ec
u
r
ity
with
in
th
e
FL
f
r
am
ewo
r
k
.
T
h
e
HAFed
L
is
also
r
o
b
u
s
t
to
th
e
d
ata
h
ete
r
o
g
en
eity
an
d
d
ev
ice
h
eter
o
g
en
eity
(
p
ar
ticu
lar
ly
th
e
s
tr
ag
g
ler
ef
f
ec
t)
,
wh
ich
m
ay
b
e
p
r
esen
t
in
th
e
clien
ts
p
ar
ticip
atin
g
in
th
e
FL.
T
h
e
p
er
f
o
r
m
an
ce
o
f
HAFed
L
is
test
ed
f
o
r
two
ap
p
licatio
n
s
-
I
o
T
d
ev
ice
id
en
tific
atio
n
an
d
d
ig
it
class
if
icatio
n
.
Ho
wev
er
,
th
e
d
ev
elo
p
ed
HAFed
L
r
esis
ted
g
r
ad
ien
t
leak
ag
e,
b
u
t
it
lack
ed
ev
alu
atio
n
u
n
d
er
ad
v
e
r
s
ar
ial
s
ce
n
ar
io
s
an
d
d
y
n
am
ic
p
ar
ti
cip
an
t
d
r
o
p
o
u
t
d
u
r
in
g
tr
ain
i
n
g
.
W
an
g
et
a
l
.
[
2
4
]
im
p
le
m
en
ted
a
p
r
iv
ac
y
p
r
eser
v
in
g
f
ed
er
ate
d
lear
n
i
n
g
m
ec
h
an
is
m
th
r
o
u
g
h
p
a
r
tial
lo
w
-
q
u
ality
d
ata
(
PP
FL
-
L
QDP)
.
T
h
e
im
p
lem
en
ted
ap
p
r
o
ac
h
attain
ed
b
etter
tr
ain
in
g
r
esu
lts
b
y
p
er
m
itti
n
g
co
n
t
r
ib
u
to
r
s
to
ac
ce
s
s
p
ar
tial,
lo
w
-
q
u
ality
d
ata,
th
u
s
im
p
r
o
v
in
g
a
p
r
iv
ac
y
as
well
a
s
r
ef
u
g
e
o
f
FL
a
p
p
r
o
ac
h
.
Par
ti
cu
lar
ly
,
a
d
is
p
er
s
ed
Pailli
er
cr
y
p
to
g
r
a
p
h
ic
s
ch
em
e
was
u
tili
ze
d
f
o
r
th
e
p
r
o
tectio
n
o
f
p
r
iv
ac
y
an
d
s
ec
u
r
ity
o
f
m
em
b
er
s
’
in
f
o
r
m
atio
n
at
th
e
f
ed
er
ated
tr
ain
in
g
p
r
o
ce
d
u
r
e.
Ho
we
v
er
,
t
h
e
im
p
l
em
en
ted
s
ch
em
e
d
id
n
o
t
c
o
n
s
id
er
p
e
r
f
o
r
m
an
ce
d
eg
r
ad
ati
o
n
d
u
e
to
th
e
f
r
eq
u
e
n
t
in
clu
s
io
n
o
f
n
o
is
y
d
ata.
Z
h
o
n
g
et
a
l
.
[
2
5
]
d
ev
el
o
p
ed
a
lig
h
tweig
h
t
p
r
i
v
ac
y
-
p
r
eser
v
in
g
FL
s
ch
em
e
b
ased
o
n
a
d
u
al
-
s
er
v
er
ar
ch
itectu
r
e.
Ou
r
s
ch
em
e
in
v
o
lv
es
o
n
l
y
lig
h
tweig
h
t
cr
y
p
to
g
r
ap
h
ic
o
p
er
atio
n
s
,
i.e
.
,
h
ash
an
d
s
y
m
m
etr
ic
en
cr
y
p
tio
n
o
p
er
atio
n
s
,
a
n
d
it
h
as
lo
w
co
m
m
u
n
icatio
n
o
v
er
h
ea
d
.
T
h
u
s
,
it
is
co
m
p
u
tatio
n
ally
lig
h
tweig
h
t
an
d
r
o
u
n
d
-
ef
f
icien
t.
Fu
r
t
h
er
,
it
allo
ws
u
s
er
s
to
jo
in
/q
u
it
an
FL
task
,
an
d
it
is
ac
cu
r
ac
y
-
lo
s
s
less
.
Ho
wev
er
,
th
e
d
esig
n
ed
lig
h
tweig
h
t
s
ch
em
e
with
d
u
al
-
s
er
v
er
a
r
ch
itectu
r
e
f
ailed
to
tack
le
p
o
is
o
n
in
g
attac
k
s
o
r
in
co
r
p
o
r
ate
ad
ap
tiv
e
m
o
d
el
u
p
d
ate
s
tr
a
teg
ies.
W
an
g
et
a
l
.
[
2
6
]
p
r
esen
ted
a
p
r
iv
ac
y
-
im
p
r
o
v
e
d
an
d
d
ep
en
d
ab
le
d
ec
en
tr
alize
d
f
e
d
er
ated
lear
n
in
g
m
ec
h
an
is
m
(
PTDFL
)
.
Par
ticu
lar
ly
,
an
e
f
f
icien
t
g
r
ad
ien
t
e
n
cr
y
p
tio
n
alg
o
r
ith
m
was
in
itially
d
ev
el
o
p
ed
f
o
r
th
e
p
r
o
tectio
n
o
f
d
ata
p
r
iv
ac
y
,
an
d
af
ter
in
v
e
n
t
ed
a
c
o
n
cise
p
r
o
o
f
th
r
o
u
g
h
lac
k
o
f
tr
a
p
d
o
o
r
s
to
m
ak
e
s
u
r
e
an
ef
f
ec
tiv
e
n
ess
o
f
in
clin
es.
T
em
p
o
r
a
r
ily
,
a
n
ew
lo
ca
l
c
o
m
b
in
atio
n
m
ec
h
an
is
m
was
d
esig
n
ed
to
o
p
er
ate
with
o
u
t
r
ely
in
g
o
n
a
tr
u
s
ted
th
ir
d
p
a
r
ty
,
en
s
u
r
i
n
g
t
h
at
a
co
m
b
in
atio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
878
-
8
8
7
880
o
u
tco
m
e
r
em
ain
s
s
ec
u
r
e
a
n
d
r
esp
o
n
s
ib
le.
Fu
r
th
er
m
o
r
e,
P
T
DFL
was
also
h
elp
f
u
l
o
f
t
h
e
d
ata
o
wn
er
s
f
o
r
lo
g
g
in
g
in
an
d
lo
g
g
i
n
g
o
u
t
at
an
en
tire
DFL
o
p
er
atio
n
.
Ho
wev
er
,
th
e
in
tr
o
d
u
ce
d
PTDFL
with
d
ec
en
tr
alize
d
ag
g
r
eg
atio
n
r
elied
h
ea
v
ily
o
n
tr
u
s
tles
s
s
y
s
tem
s
,
wh
ich
in
cr
ea
s
es
co
m
p
lex
ity
an
d
r
is
k
s
laten
cy
in
lar
g
e
-
s
ca
le
s
y
s
tem
s
.
Su
n
et
a
l
.
[
2
7
]
d
ev
e
lo
p
ed
a
n
o
v
el
d
if
f
er
e
n
tially
p
r
iv
ate
f
ed
er
ate
d
lear
n
in
g
(
DP
FL)
s
ch
em
e
n
am
ed
Ad
ap
-
Fed
I
T
K,
wh
ich
aim
e
d
to
ac
h
iev
e
lo
w
-
co
m
m
u
n
ica
tio
n
o
v
er
h
ea
d
an
d
h
ig
h
-
m
o
d
el
ac
cu
r
ac
y
wh
ile
g
u
ar
an
teein
g
clien
t
-
lev
el
DP.
Sp
ec
if
ically
,
th
is
d
y
n
am
ical
ly
ad
ju
s
ts
th
e
g
r
a
d
ien
t
clip
p
in
g
th
r
esh
o
l
d
f
o
r
d
if
f
er
en
t
clien
ts
in
ea
ch
r
o
u
n
d
,
b
ased
o
n
th
e
h
ete
r
o
g
e
n
eity
o
f
g
r
ad
ien
ts
.
T
h
is
ap
p
r
o
ac
h
aim
s
to
m
itig
ate
th
e
n
eg
ativ
e
im
p
ac
t
o
f
DP
a
n
d
ac
h
iev
e
a
p
r
i
v
ac
y
u
tili
ty
tr
ad
e
o
f
f
.
T
o
allev
iate
t
h
e
h
ig
h
-
c
o
m
m
u
n
icatio
n
o
v
er
h
ea
d
p
r
o
b
lem
in
FL,
an
im
p
r
o
v
e
d
to
p
-
k
alg
o
r
ith
m
was
in
tr
o
d
u
ce
d
,
wh
ich
u
tili
s
ed
s
p
ar
s
ity
a
n
d
q
u
an
tis
atio
n
t
o
co
m
p
r
ess
th
e
m
o
d
el,
elim
in
at
e
co
m
m
u
n
icatio
n
r
ed
u
n
d
an
c
y
,
an
d
also
in
teg
r
ates
co
d
in
g
te
ch
n
iq
u
es
to
f
u
r
th
er
r
ed
u
ce
co
m
m
u
n
icatio
n
.
Ho
w
ev
er
,
Ad
ap
-
Fed
I
T
K,
d
esp
ite
in
co
r
p
o
r
atin
g
d
y
n
am
ic
g
r
a
d
ien
t
clip
p
in
g
,
f
ac
ed
d
if
f
icu
lties
in
b
alan
cin
g
p
r
iv
a
cy
an
d
u
tili
ty
d
u
e
to
th
e
u
s
e
o
f
f
ix
ed
clip
p
in
g
t
h
r
esh
o
ld
s
,
wh
ich
co
u
ld
lead
to
p
o
ten
tial
ac
cu
r
ac
y
d
eg
r
a
d
atio
n
.
Sh
an
et
a
l
.
[
2
8
]
p
r
esen
ted
an
FL
m
ec
h
an
is
m
th
r
o
u
g
h
D
P
p
r
o
tectio
n
,
wh
ich
was
r
o
b
u
s
t
f
o
r
GL
A
.
T
h
e
d
is
cr
im
in
ato
r
y
u
p
d
ates
wer
e
u
til
ized
f
o
r
a
s
elec
tio
n
o
f
m
o
d
el
p
ar
am
eter
s
th
r
o
u
g
h
g
r
ea
ter
e
f
f
ec
tiv
en
ess
,
th
u
s
en
h
an
cin
g
a
m
o
d
el’
s
u
tili
ty
.
T
h
en
,
d
esig
n
ed
a
d
ee
p
lear
n
in
g
(
DL
)
ap
p
r
o
ac
h
was
d
esig
n
ed
th
r
o
u
g
h
an
au
to
m
at
ic
clip
p
in
g
a
n
d
n
o
is
e
atten
u
a
tio
n
s
ch
em
e
to
e
n
s
u
r
e
DP
a
n
d
o
p
tim
is
e
u
tili
ty
attain
m
en
t,
s
o
lv
in
g
an
in
tr
in
s
ic
d
r
awb
ac
k
o
f
co
n
v
en
tio
n
a
l
DL
m
eth
o
d
s
th
r
o
u
g
h
s
ec
u
r
e
DP
p
ar
am
eter
s
.
T
h
e
p
r
esen
ted
a
p
p
r
o
ac
h
es
e
x
h
ib
ited
th
e
r
ap
id
co
n
v
er
g
e
n
ce
r
ates
a
n
d
attain
ed
s
ig
n
if
ican
t
p
e
r
f
o
r
m
an
ce
.
Ho
wev
er
,
th
e
p
r
esen
ted
m
eth
o
d
ig
n
o
r
ed
m
o
d
el
co
n
v
er
g
e
n
c
e
d
elay
s
an
d
th
e
s
ca
lab
ilit
y
c
o
n
ce
r
n
s
u
n
d
er
lar
g
e
clien
t
p
ar
ticip
atio
n
.
T
h
e
r
ev
i
ewe
d
FL
liter
atu
r
e
e
x
h
ib
its
k
ey
lim
itatio
n
s
s
u
ch
as
v
u
l
n
er
ab
ilit
y
to
g
r
ad
ien
t
leak
ag
e,
in
e
f
f
icien
cy
with
h
et
er
o
g
en
e
o
u
s
d
ata,
h
ig
h
co
m
m
u
n
icatio
n
o
v
er
h
ea
d
,
a
n
d
r
elian
ce
o
n
tr
u
s
ted
t
h
ir
d
p
ar
ties
.
Ma
n
y
m
o
d
els
tr
ad
e
o
f
f
ac
cu
r
ac
y
f
o
r
p
r
iv
ac
y
an
d
lack
s
u
p
p
o
r
t
f
o
r
asy
n
ch
r
o
n
o
u
s
,
s
ec
u
r
e
co
m
m
u
n
icatio
n
.
T
o
o
v
e
r
co
m
e
th
ese,
th
e
p
r
o
p
o
s
ed
g
r
a
d
ien
t
d
escen
t
o
p
tim
izatio
n
ass
is
ted
weig
h
ted
f
ed
e
r
ated
lear
n
in
g
(
GDO
-
W
FL
)
in
teg
r
a
tes
Kaf
k
a
-
Z
o
o
k
ee
p
er
f
o
r
an
o
n
y
m
o
u
s
co
m
m
u
n
icatio
n
,
ap
p
lies
DP
,
an
d
u
s
es
weig
h
ted
m
o
d
el
u
p
d
ates
b
ase
d
o
n
d
ata
q
u
ality
an
d
tr
ai
n
in
g
ef
f
ec
tiv
en
ess
.
T
h
is
en
s
u
r
es
r
o
b
u
s
t,
s
ca
lab
le,
an
d
p
r
iv
ac
y
-
p
r
eser
v
in
g
lear
n
in
g
w
ith
o
u
t h
ea
v
y
cr
y
p
to
g
r
ap
h
ic
o
v
er
h
ea
d
.
T
h
e
k
ey
i
n
n
o
v
atio
n
s
o
f
t
h
is
r
esear
ch
ar
e
ar
r
a
n
g
ed
as f
o
llo
ws:
i)
An
in
teg
r
ated
f
r
am
ew
o
r
k
o
f
GDO
-
W
FL
i
s
in
tr
o
d
u
ce
d
in
th
is
r
esear
ch
to
r
ed
u
ce
th
e
s
ig
n
if
ican
t
th
r
ea
ts
in
FL.
ii)
T
h
e
p
r
ac
tical
GDO
-
W
FL
ap
p
r
o
ac
h
th
r
o
u
g
h
Ap
ac
h
e
Kaf
k
a
-
Z
o
o
k
ee
p
e
r
is
d
ev
elo
p
ed
to
attain
an
o
n
y
m
o
u
s
au
th
en
ticatio
n
th
r
o
u
g
h
ac
ce
s
s
co
n
tr
o
l lis
t a
s
well
as a
s
y
n
ch
r
o
n
o
u
s
m
o
d
el
d
is
s
em
in
atio
n
s
.
iii)
A
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
ac
h
is
co
m
p
ar
ed
an
d
esti
m
ated
th
r
o
u
g
h
th
e
ex
is
tin
g
FL
a
p
p
r
o
ac
h
es
b
ased
o
n
ef
f
ec
tiv
e
n
ess
,
s
ec
u
r
ity
as we
ll a
n
d
ac
cu
r
ac
y
th
r
o
u
g
h
em
p
i
r
ical
s
tu
d
ies.
T
h
is
r
esear
ch
p
ap
e
r
is
o
r
g
a
n
is
ed
as
f
o
llo
ws:
s
ec
tio
n
2
d
em
o
n
s
tr
ates
a
p
r
o
p
o
s
ed
m
eth
d
o
lo
g
y
.
Sectio
n
3
o
u
tlin
es
th
e
FL
f
o
r
p
r
iv
ac
y
p
r
eser
v
i
n
g
f
r
am
e
wo
r
k
.
Sectio
n
4
d
em
o
n
s
tr
ates
th
e
r
esu
lts
an
d
d
is
cu
s
s
io
n
,
an
d
s
ec
tio
n
5
p
r
o
v
id
es a
co
n
clu
s
io
n
.
2.
P
RO
P
O
SE
D
M
E
T
H
O
DO
L
O
G
Y
Fig
u
r
e
1
d
e
m
o
n
s
tr
ates
a
s
y
s
tem
m
o
d
el
o
f
in
tr
o
d
u
ce
d
m
eth
o
d
.
I
n
itially
,
th
e
p
r
o
p
o
s
ed
m
e
th
o
d
o
lo
g
y
d
escr
ib
es
th
e
s
y
s
tem
m
o
d
el
a
n
d
th
e
n
d
e
f
in
ed
t
h
e
m
o
d
el
f
r
a
m
ewo
r
k
o
f
th
is
r
esear
ch
.
T
h
e
d
etailed
d
escr
ip
tio
n
an
d
wo
r
k
in
g
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
o
l
o
g
y
a
r
e
d
escr
ib
ed
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
Fig
u
r
e
1
.
Sy
s
tem
s
y
s
tem
m
o
d
el
o
f
in
tr
o
d
u
ce
d
m
eth
o
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Gra
d
ien
t d
escen
t o
p
timiz
a
tio
n
b
a
s
ed
w
eig
h
ted
fed
era
ted
lea
r
n
in
g
fo
r
p
r
iva
cy
-
…
(
Gu
r
u
r
a
j
P
r
a
ka
s
h
Mu
r
th
y)
881
2
.
1
.
Sy
s
t
e
m
mo
del
T
h
er
e
a
r
e
m
u
ltip
le
u
n
its
in
th
e
s
y
s
tem
m
o
d
el
s
u
ch
as
u
s
er
s
,
tr
ain
er
s
a
n
d
an
ag
g
r
e
g
atio
n
s
er
v
er
.
T
h
is
is
d
if
f
er
en
t
f
r
o
m
co
n
v
en
tio
n
al
b
in
ar
y
u
n
its
(
u
s
er
s
an
d
ag
g
r
e
g
atio
n
s
er
v
er
)
an
d
h
elp
s
u
s
er
s
r
em
ain
d
is
co
n
n
ec
ted
.
A
tr
ain
in
g
m
o
d
el
in
f
lu
en
ce
s
an
ap
p
r
o
ac
h
f
o
r
co
m
p
r
eh
e
n
d
in
g
tr
ain
in
g
as
well
as
f
o
r
ec
asti
n
g
.
E
v
en
tu
ally
,
an
FL
ap
p
r
o
ac
h
ac
q
u
ir
es
an
o
p
tim
izatio
n
tr
ain
in
g
p
ar
am
eter
a
n
d
f
o
r
ec
a
s
ts
o
u
tco
m
es.
E
ac
h
co
m
m
u
n
icatio
n
lin
e
is
ce
r
tain
th
r
o
u
g
h
s
ec
u
r
e
tu
n
n
ellin
g
p
r
o
t
o
co
l
(
STP)
.
i)
User
:
in
th
e
m
o
d
el
f
r
am
ewo
r
k
,
a
u
s
er
is
a
n
a
g
en
t
w
h
o
g
iv
es
cip
h
er
te
x
t
d
ata
f
o
r
t
r
ain
in
g
a
n
d
f
o
r
ec
asti
n
g
in
an
FL
o
p
er
atio
n
.
I
n
itially
,
ea
ch
u
s
er
p
r
o
d
u
ce
s
p
u
b
lic
an
d
p
r
iv
ate
k
ey
p
air
s
,
an
d
af
ter
tr
an
s
m
its
a
k
ey
to
a
s
er
v
er
f
o
r
t
h
e
m
o
d
el
lin
k
.
Af
ter
ac
q
u
i
r
in
g
a
lin
k
p
u
b
li
c
k
ey
,
a
u
s
er
en
c
o
d
es
an
d
en
cr
y
p
ts
th
e
d
ata
u
s
in
g
th
is
k
ey
,
an
d
th
en
tr
an
s
m
its
th
e
r
esu
ltin
g
cip
h
e
r
tex
t
t
o
th
e
tr
ain
e
r
f
o
r
p
r
o
ce
s
s
in
g
.
M
o
r
eo
v
e
r
,
a
u
s
er
h
as to
d
ec
r
y
p
t a
n
d
d
ec
o
d
e
an
i
n
f
o
r
m
atio
n
to
ac
q
u
ir
e
p
r
ed
icti
o
n
r
esu
lts
.
ii)
T
r
ain
er
:
as
an
ML
en
tity
,
a
tr
ain
er
in
v
o
lv
es
a
DL
ap
p
r
o
ac
h
f
o
r
t
r
ain
in
g
m
o
d
el
p
ar
am
et
er
s
an
d
h
as
t
o
g
ath
er
en
cr
y
p
ted
in
f
o
r
m
atio
n
f
r
o
m
u
s
er
s
in
a
lo
ca
l
ar
ea
.
C
e
r
tain
ly
,
th
e
tr
ain
er
s
wo
r
k
to
g
e
th
er
th
r
o
u
g
h
a
s
er
v
er
f
o
r
tr
ain
i
n
g
a
m
o
d
el
th
r
o
u
g
h
th
e
FL
ap
p
r
o
ac
h
.
B
y
d
if
f
er
en
t
iter
atio
n
s
,
th
e
tr
ain
er
s
o
b
tain
an
ef
f
icien
t tr
ain
ed
s
y
s
tem
u
s
ed
f
o
r
p
r
e
d
ictio
n
b
y
u
s
er
s
’
r
e
q
u
ests
.
iii)
Ser
v
er
:
as
a
co
n
tr
o
ller
an
d
a
g
g
r
eg
ato
r
,
a
s
er
v
e
r
h
as
d
is
tr
ib
u
ted
a
lin
k
ed
p
u
b
lic
k
e
y
an
d
esti
m
ated
a
f
ed
er
ated
a
v
er
ag
i
n
g
(
Fed
Av
g
)
th
r
o
u
g
h
ag
g
r
eg
ated
weig
h
ts
f
r
o
m
tr
ain
e
r
s
.
Fu
r
th
e
r
m
o
r
e
,
a
s
er
v
er
r
eq
u
ir
es
v
ar
io
u
s
iter
atio
n
s
an
d
co
llab
o
r
atio
n
to
ac
q
u
ir
e
an
o
p
tim
is
atio
n
m
o
d
el
p
ar
am
ete
r
in
tr
ain
in
g
p
r
o
ce
d
u
r
e.
I
n
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
,
e
v
e
n
if
s
er
v
e
r
co
n
s
p
ir
es
th
r
o
u
g
h
v
ar
io
u
s
e
n
titi
es,
it
d
o
es
n
o
t
ac
q
u
ir
e
a
n
y
in
f
o
r
m
atio
n
f
r
o
m
u
s
er
s
.
2
.
2
.
M
o
del f
ra
m
ewo
r
k
T
h
e
m
o
d
el
f
r
am
ewo
r
k
in
v
o
l
v
es
m
u
ltip
le
en
titi
es
lik
e
clien
ts
,
s
er
v
er
an
d
Ap
ac
h
e
Kaf
k
a
m
ess
ag
e
q
u
eu
e.
Her
e,
a
s
er
v
er
p
r
ep
a
r
es
a
f
r
am
ew
o
r
k
an
d
tr
an
s
m
its
an
u
n
ex
p
er
ien
ce
d
ap
p
r
o
ac
h
f
o
r
r
eq
u
esti
n
g
a
b
r
o
k
er
in
Kaf
k
a.
Sin
ce
Kaf
k
a
o
p
er
at
es
as
a
r
eliab
le
th
ir
d
-
p
ar
ty
s
y
s
tem
,
wh
en
clien
ts
r
e
q
u
est
a
n
etwo
r
k
f
r
o
m
Kaf
k
a
r
eq
u
est
b
r
o
k
er
,
th
ey
r
ec
eiv
e
t
h
is
u
n
tr
ain
ed
m
o
d
el
f
o
r
th
eir
u
s
e.
Af
ter
clien
ts
r
eq
u
est
an
ap
p
r
o
ac
h
f
r
o
m
t
h
e
Kaf
k
a
r
eq
u
est
b
r
o
k
er
,
it
d
ep
e
n
d
s
o
n
an
u
n
tr
ain
ed
m
o
d
el
f
o
r
th
e
clien
t.
T
h
e
s
u
g
g
ested
f
r
am
ewo
r
k
en
ab
le
d
a
ce
n
tr
al
b
r
o
k
e
r
to
f
ix
a
p
o
licy
th
at
allo
ws
d
ef
in
ed
clien
ts
o
n
ly
ac
co
r
d
in
g
to
a
1
-
out
-
of
-
n
esti
m
atio
n
p
lan
th
r
o
u
g
h
p
r
ed
e
f
in
ed
n
f
ea
tu
r
e
s
to
as
s
o
ciate
a
m
o
d
el.
T
h
is
r
esear
ch
co
n
s
id
er
s
th
at
th
er
e
is
a
r
ec
o
g
n
ized
p
r
o
tecte
d
ar
ea
b
etwe
en
ce
n
tr
a
l
s
er
v
er
an
d
clien
ts
.
A
s
im
p
le
en
cr
y
p
tio
n
f
o
r
v
er
if
icatio
n
is
ac
q
u
ir
ed
th
r
o
u
g
h
a
ce
n
tr
al
b
r
o
k
er
.
E
v
er
y
clien
t
tr
ain
s
a
m
o
d
el
ac
c
o
r
d
in
g
to
th
eir
lo
ca
l
d
ataset.
Af
ter
a
s
er
v
er
d
esire
s
a
tr
ain
e
d
m
o
d
el,
clien
ts
will
r
ep
ly
to
a
r
esp
o
n
s
e
b
r
o
k
e
r
in
Kaf
k
a
th
r
o
u
g
h
a
tr
ain
ed
an
d
e
n
cr
y
p
ted
ap
p
r
o
ac
h
.
E
v
en
tu
ally
,
a
r
esp
o
n
s
e
b
r
o
k
er
will
tr
an
s
m
it
u
p
d
ated
ap
p
r
o
ac
h
es
to
a
s
er
v
er
,
an
d
th
ey
p
er
f
o
r
m
ef
f
ec
tiv
e
co
m
b
in
atio
n
m
eth
o
d
to
o
b
tain
o
v
er
all
o
u
t
co
m
e.
T
h
is
s
tr
u
ctu
r
e
is
p
er
f
e
ct
f
o
r
m
u
lti
-
o
r
g
an
izatio
n
FL
s
ce
n
ar
io
s
f
o
r
s
o
m
e
clien
ts
.
I
t
is
o
f
ten
ap
p
lied
in
an
ac
tu
al
m
an
u
f
ac
tu
r
i
n
g
en
v
ir
o
n
m
en
t
th
r
o
u
g
h
Kaf
k
a
clien
t
-
s
er
v
er
co
m
m
u
n
icatio
n
f
r
a
m
ewo
r
k
,
a
s
well
as
m
ess
ag
e
s
to
r
ag
e
as
well
as
tr
an
s
m
is
s
io
n
f
u
n
ctio
n
s
.
R
ec
en
tly
,
f
r
o
m
a
s
ec
u
r
ity
p
o
in
t
-
of
-
v
iew,
s
ec
u
r
e
co
m
b
in
atio
n
as
well
as
DP
m
ec
h
an
is
m
s
wer
e
in
teg
r
ated
i
n
to
th
e
GDO
-
W
FL
f
r
am
ewo
r
k
to
en
s
u
r
e
p
r
i
v
ac
y
p
r
eser
v
atio
n
a
n
d
ef
f
ec
tiv
ely
m
itig
ate
lar
g
e
-
s
ca
le
attac
k
s
,
a
s
s
u
m
in
g
th
e
s
er
v
er
an
d
clien
ts
ac
t
as
p
ass
iv
e
ad
v
er
s
ar
ies.
Fig
u
r
e
2
d
em
o
n
s
tr
ates
th
e
im
p
o
r
tan
t
s
tr
u
ctu
r
e
o
f
th
e
in
tr
o
d
u
ce
d
ap
p
r
o
ac
h
.
A
co
m
p
r
eh
e
n
s
iv
e
d
escr
ip
tio
n
o
f
th
is
s
tr
u
ctu
r
e
is
p
r
o
v
id
ed
in
a
s
u
b
s
eq
u
en
t sectio
n
.
Fig
u
r
e
2
.
I
m
p
o
r
tan
t stru
ctu
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
2
.
2
.
1
.
P
ha
s
e
1
:
s
et
up
a
nd
k
ey
g
ener
a
t
io
n
T
h
is
s
tep
in
v
o
lv
es
a
g
en
e
r
atio
n
o
f
k
e
y
am
o
n
g
clien
ts
.
Ass
u
m
e
a
g
r
o
u
p
o
f
clien
ts
as
=
1
,
2
,
…
,
.
Af
ter
th
at,
a
s
er
v
er
s
elec
ts
a
s
u
itab
le
g
r
o
u
p
G
,
wh
ile
a
co
m
p
u
tatio
n
al
Dif
f
ie
-
Hellm
an
is
s
u
e
is
co
m
p
lex
.
G
is
a
r
ec
u
r
r
in
g
q
ess
en
tial th
r
o
u
g
h
g
en
er
ato
r
g
.
T
h
e
b
asic c
o
n
s
tr
ai
n
ts
ar
e
p
r
o
d
u
ce
d
u
s
in
g
(
1
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
878
-
8
8
7
882
(
,
)
,
|
|
−
,
∈
:
=
<
>
(
1
)
Ass
u
m
e
≠
∈
b
ec
o
m
es
a
s
tatic
c
o
u
p
le
o
f
clien
ts
.
Af
ter
th
at
a
s
u
b
s
eq
u
en
t
p
r
o
ce
d
u
r
e
wh
ich
in
v
o
lv
es
th
e
s
ee
d
as
well
as
n
o
is
e
g
en
er
atio
n
h
as
to
b
e
im
p
l
em
en
ted
f
o
r
ea
ch
p
air
o
f
clien
ts
.
Af
ter
th
e
s
etu
p
an
d
k
e
y
g
en
e
r
atio
n
,
th
e
s
ee
d
m
u
s
t b
e
g
en
e
r
ated
.
i)
Gen
er
atio
n
o
f
s
ee
d
–
C
lien
t
i
s
elec
ts
a
p
r
o
p
o
n
e
n
t
∈
∗
an
d
tr
an
s
m
its
to
a
s
er
v
er
t
h
r
o
u
g
h
v
alu
e
o
f
.
Af
ter
t
h
at
s
er
v
er
tr
an
s
m
its
a
p
air
(
,
)
to
clien
t
j
.
–
C
lien
t
j
s
elec
ts
a
p
r
o
p
o
s
al
∈
∗
an
d
tr
an
s
m
its
to
a
s
er
v
er
th
r
o
u
g
h
th
e
v
alu
e
o
f
.
Af
ter
th
a
t
s
er
v
er
tr
an
s
m
its
a
p
air
(
,
)
to
clien
t
i
.
–
A
b
asic
k
ey
o
f
th
e
clien
t
i
an
d
j
is
,
=
.
At
ter
m
in
atio
n
o
f
th
is
s
tep
,
ea
ch
clien
t
∈
k
ee
p
s
a
g
r
o
u
p
o
f
b
asic k
e
y
s
th
r
o
u
g
h
d
if
f
e
r
en
t c
lien
ts
,
wh
ich
is
in
(
2
)
.
=
,
1
,
,
2
,
…
,
,
(
2
)
ii)
No
is
e
g
en
er
atio
n
T
o
ig
n
o
r
e
i
n
tr
u
s
io
n
attac
k
s
,
clien
ts
h
av
e
ex
ec
u
ted
DP
th
r
o
u
g
h
in
c
r
ea
s
in
g
n
o
is
es
to
weig
h
ts
af
ter
m
o
d
e
l
tr
ain
in
g
,
w
h
ich
is
r
elate
d
to
th
e
GDO.
I
n
th
at,
a
L
ap
lace
n
o
is
e
g
e
n
er
atio
n
s
tr
ateg
y
o
f
DP
is
u
s
ed
.
A
L
ap
lace
m
ec
h
an
is
m
co
n
s
er
v
es (
∈
,
0
)
DP
.
2
.
2
.
2
.
P
ha
s
e
2
:
m
o
del upd
a
t
e
As
an
o
u
tco
m
e
o
f
th
e
im
p
lem
en
tatio
n
o
f
p
r
im
ar
y
s
tag
e,
wh
er
e
a
s
ee
d
an
d
n
o
is
e
ar
e
p
r
o
d
i
ce
d
,
ea
ch
co
n
tr
ib
u
tin
g
clien
t
is
p
r
e
p
ar
e
d
to
m
ask
th
e
m
o
d
el
u
s
in
g
th
e
g
en
e
r
ated
n
o
is
e
an
d
k
ey
p
ai
r
s
.
Simu
ltan
eo
u
s
ly
,
clien
ts
h
av
e
u
p
d
ated
a
s
y
s
tem
b
y
r
ep
ly
in
g
to
a
r
e
q
u
est
d
ir
ec
ted
f
r
o
m
a
ce
n
tr
al
s
er
v
e
r
.
Sy
s
t
em
s
will
r
ea
ch
an
d
s
tay
ass
o
ciate
d
with
Kaf
k
a
b
r
o
k
er
s
ass
o
ciate
d
th
r
o
u
g
h
Z
o
o
k
ee
p
e
r
.
I
n
th
is
co
n
d
itio
n
,
ev
en
if
clien
ts
ar
e
in
v
o
lv
ed
in
v
ar
io
u
s
s
u
r
r
o
u
n
d
i
n
g
s
o
f
h
ar
d
war
e
s
ettin
g
s
a
n
d
n
etwo
r
k
b
an
d
wid
t
h
,
th
is
m
o
d
el
ev
en
tu
ally
attain
s
s
y
n
ch
r
o
n
ic
co
m
m
u
n
icatio
n
.
2
.
2
.
3
.
P
ha
s
e
3
:
s
ec
ure
a
g
g
re
g
a
t
io
n
I
n
th
is
s
tag
e,
a
s
er
v
er
ca
lcu
la
tes
an
ag
g
r
eg
ated
clien
t
d
ata
f
o
r
a
p
r
o
v
id
e
d
u
p
d
ate.
C
o
n
s
i
d
er
∈
b
ec
o
m
es
a
len
g
t
h
o
f
u
s
er
s
’
in
f
o
r
m
atio
n
o
f
an
u
n
k
n
o
wn
u
p
d
ate.
T
h
is
d
is
tan
ce
is
v
ar
ie
d
f
o
r
ea
ch
u
p
d
ate
h
o
wev
er
s
tatic
at
an
in
d
iv
id
u
al
u
p
d
ate.
C
o
n
s
id
er
∈
{
0
,
1
}
b
ec
o
m
es
th
e
clien
t’
s
d
ata;
b
ec
o
m
es
a
p
s
eu
d
o
r
an
d
o
m
g
en
er
at
o
r
o
f
o
u
tp
u
t size
l
.
3.
F
E
DE
R
AT
E
D
L
E
AR
NING
F
O
R
P
RIVACY
P
R
E
SE
RVI
NG
F
RAME
WO
RK
I
n
th
e
p
r
o
p
o
s
ed
FL
m
o
d
el
,
ea
ch
k
clien
t
tr
ain
s
a
l
o
ca
l
m
o
d
el
th
r
o
u
g
h
a
s
im
ilar
s
h
a
r
ed
g
l
o
b
al
ap
p
r
o
ac
h
;
h
o
we
v
er
,
it
is
tr
ai
n
e
d
o
n
d
if
f
e
r
en
t
lo
ca
l
d
atasets
in
s
tead
o
f
th
e
ce
n
tr
al
s
er
v
e
r
.
Pu
r
s
u
in
g
th
at,
clien
ts
s
ec
u
r
ely
tr
an
s
m
it
th
e
u
p
d
at
es
f
r
o
m
th
ei
r
lo
ca
l
tr
ai
n
in
g
to
th
e
a
g
g
r
e
g
atio
n
s
er
v
e
r
v
ia
s
ec
u
r
e
s
o
ck
ets
lay
er
/
tr
an
s
p
o
r
t
la
y
er
s
ec
u
r
it
y
(
SS
L
/TL
S
)
-
au
th
en
ticated
co
n
n
ec
tio
n
s
m
an
ag
e
d
b
y
a
co
m
m
u
n
icatio
n
ad
m
in
is
tr
ato
r
.
An
ag
g
r
eg
atio
n
s
er
v
er
ass
o
ciate
s
th
em
as
well
as
g
en
er
ates
an
u
p
d
at
ed
g
lo
b
al
a
p
p
r
o
ac
h
th
r
o
u
g
h
i
d
ea
l
co
n
s
tr
ain
ts
.
A
n
o
tatio
n
w
d
en
o
tes
in
itial
weig
h
ts
an
d
r
d
em
o
n
s
tr
ates
a
c
o
u
n
t
o
f
FL
d
is
k
s
,
t
h
at
is
co
n
tin
u
ed
p
r
i
o
r
attain
in
g
a
c
o
n
v
er
g
e
n
ce
s
tag
e.
W
h
ile
ev
e
r
y
lo
ca
l
clien
t’
s
weig
h
t
is
g
iv
en
to
an
ag
g
r
eg
atio
n
s
er
v
er
at
th
e
co
m
m
u
n
icatio
n
r
o
u
n
d
t
,
a
s
u
b
s
eq
u
e
n
t
(
3
)
is
ac
q
u
ir
ed
f
r
o
m
Fed
Av
g
alg
o
r
ith
m
,
wh
ich
is
u
tili
ze
d
f
o
r
u
p
d
atin
g
m
o
d
el
weig
h
ts
.
+
1
=
∑
=
1
+
1
(
3
)
Her
e,
m
ea
n
s
a
to
tal
s
ize
o
f
ev
er
y
clien
t
d
ataset,
a
n
d
d
en
o
tes
a
s
ize
o
f
ev
er
y
clien
t
d
ataset.
+
1
d
em
o
n
s
tr
ates a
n
u
p
d
ated
g
lo
b
al
ap
p
r
o
ac
h
af
ter
a
n
iter
atio
n
.
A
s
er
v
er
p
r
im
ar
ily
s
elec
ts
clien
ts
wh
o
co
n
s
tr
ain
ass
o
ciatio
n
th
r
o
u
g
h
ac
tiv
e
d
e
v
ices,
af
t
er
th
at
th
e
v
ar
io
u
s
p
ar
ts
o
f
a
s
y
s
tem
in
ter
r
elate
s
u
b
s
eq
u
en
tly
to
c
o
m
p
let
e
an
en
tire
p
r
o
ce
d
u
r
e:
i)
On
=
0
,
a
s
er
v
er
p
r
o
d
u
ce
d
a
m
o
d
el
was
p
r
o
d
u
ce
d
f
r
o
m
th
e
g
l
o
b
al
d
ata
m
o
d
el.
I
n
th
is
ca
s
e,
th
e
n
u
m
b
er
o
f
p
ar
a
m
eter
s
is
esti
m
ated
.
ii)
E
ac
h
k
clien
t
(
[
1
,
.
.
.
,
]
)
is
r
eq
u
ir
ed
to
lev
er
ag
e
a
g
lo
b
al
ap
p
r
o
ac
h
to
tr
an
s
f
er
it,
n
ev
er
th
eless
o
f
if
th
ey
o
f
f
er
to
FL
p
r
o
ce
d
u
r
e
o
r
n
o
t.
T
h
r
o
u
g
h
in
d
i
v
id
u
al
p
r
iv
ate
d
ata,
e
v
er
y
k
clien
ts
r
e
-
ed
u
ca
te
a
g
lo
b
al
ap
p
r
o
ac
h
lo
ca
lly
in
p
ar
allel
an
d
d
esig
n
s
a
n
ew
g
r
o
u
p
o
f
lo
ca
l
weig
h
ts
+
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Gra
d
ien
t d
escen
t o
p
timiz
a
tio
n
b
a
s
ed
w
eig
h
ted
fed
era
ted
lea
r
n
in
g
fo
r
p
r
iva
cy
-
…
(
Gu
r
u
r
a
j
P
r
a
ka
s
h
Mu
r
th
y)
883
iii)
C
h
o
s
en
clien
ts
u
til
is
e
an
in
f
o
r
m
atio
n
g
ath
e
r
ed
f
r
o
m
I
o
T
d
ev
ices
u
n
d
er
th
eir
ac
ce
s
s
to
en
h
an
ce
an
ap
p
r
o
ac
h
u
n
d
er
esti
m
atio
n
wh
er
ea
s
k
ee
p
in
g
a
lo
ca
l d
ata’
s
p
r
iv
ac
y
.
iv
)
T
o
k
ee
p
clien
t
p
r
iv
ac
y
,
o
n
ly
th
e
p
ar
am
eter
s
o
f
an
u
p
d
ated
m
o
d
el
ar
e
tr
a
n
s
m
itted
to
a
ce
n
tr
al
s
er
v
er
.
v)
Af
ter
o
b
tain
in
g
th
e
m
o
d
if
icati
o
n
s
,
a
s
er
v
er
a
g
g
r
e
g
ated
weig
h
ts
f
r
o
m
d
i
f
f
er
en
t
n
o
d
e
m
o
d
e
ls
to
g
en
er
ate
an
u
n
k
n
o
wn
m
o
d
el
(
3
)
.
A
Fed
Av
g
ap
p
r
o
ac
h
is
u
tili
s
ed
f
o
r
ag
g
r
eg
atio
n
.
T
h
u
s
,
t
h
e
p
ar
am
eter
s
ar
e
esti
m
ated
ac
co
r
d
in
g
t
o
a
d
atab
ase
s
ize
at
ev
er
y
n
o
d
e
.
v
i)
Up
d
ate
m
o
d
el
p
ar
am
eter
s
ar
e
r
etu
r
n
ed
to
clien
ts
th
r
o
u
g
h
a
c
en
tr
al
s
er
v
er
.
v
i
i
)
E
v
e
r
y
c
l
ie
n
t
u
t
i
l
i
s
e
s
u
n
k
n
o
w
n
n
e
t
w
o
r
k
p
a
r
a
m
e
t
er
s
a
n
d
c
r
e
a
t
e
s
m
o
d
if
i
c
a
t
io
n
s
a
c
co
r
d
in
g
t
o
a
n
u
n
k
n
o
wn
d
a
t
a
.
v
iii)
Fo
r
en
d
u
r
in
g
m
o
d
el
u
n
d
er
s
tan
d
in
g
an
d
en
h
an
ce
m
en
t,
s
tep
s
4
to
7
ar
e
r
ep
etitiv
e.
3
.
1
.
G
ra
dient
des
ce
nt
o
ptim
iza
t
io
n ba
s
ed
weig
hte
d f
eder
a
t
ed
lea
rning
An
im
p
o
r
ta
n
t
lim
itatio
n
in
FL
is
a
n
etwo
r
k
b
an
d
wid
th
,
wh
ic
h
r
estricts
a
s
p
ee
d
at
lo
c
al
u
p
d
ates
f
r
o
m
m
u
ltip
le
estab
lis
h
m
en
ts
ar
e
ag
g
r
eg
ated
i
n
a
clo
u
d
.
T
o
tac
k
le
th
is
is
s
u
e,
Fed
Av
g
u
tili
ze
s
lo
ca
l
d
ata
f
o
r
GDO
p
r
io
r
p
e
r
f
o
r
m
in
g
th
e
weig
h
ted
av
er
ag
e
a
g
g
r
e
g
atio
n
ap
p
r
o
ac
h
u
p
lo
a
d
ed
th
r
o
u
g
h
ev
er
y
n
o
d
e.
An
a
p
p
r
o
ac
h
p
r
o
f
its
iter
ativ
ely
,
u
p
d
ate
a
g
l
o
b
al
m
o
d
el
in
e
v
er
y
tr
ain
in
g
r
o
u
n
d
ac
co
r
d
in
g
t
o
co
n
tr
ib
u
tio
n
s
f
r
o
m
co
n
tr
ib
u
tin
g
o
r
g
an
izatio
n
s
.
C
o
n
v
e
n
tio
n
al
c
en
tr
alize
d
lear
n
in
g
m
eth
o
d
s
in
teg
r
ate
d
ata
f
r
o
m
v
ar
io
u
s
o
r
g
an
is
atio
n
s
in
to
an
in
d
iv
id
u
al
d
atab
ase.
T
h
is
lea
d
s
to
s
u
b
s
tan
tial
m
ess
ag
e
c
o
s
ts
an
d
h
az
ar
d
s
f
o
r
d
ata
c
o
n
f
i
d
en
tiality
.
T
o
s
o
lv
e
th
ese
lim
itatio
n
s
,
th
e
p
r
iv
ac
y
-
p
r
eser
v
in
g
a
p
p
r
o
ac
h
p
r
ep
ar
e
d
th
r
o
u
g
h
f
o
r
ec
asti
n
g
ap
p
r
o
ac
h
is
p
r
o
p
o
s
ed
f
o
r
FL.
T
h
is
s
o
lu
tio
n
b
e
g
in
s
th
r
o
u
g
h
u
tilzin
g
Fed
Av
g
ap
p
r
o
ac
h
f
o
r
p
a
r
am
eter
a
g
g
r
e
g
atio
n
,
g
at
h
er
in
g
in
clin
e
d
ata
f
r
o
m
d
if
f
er
en
t
n
o
d
es.
Af
ter
th
at
d
ev
elo
p
ed
a
n
im
p
r
o
v
ed
ty
p
e
o
f
Fed
Av
g
to
r
e
d
u
ce
co
m
m
u
n
icatio
n
o
v
er
h
ea
d
as
well
as
em
p
lo
y
s
ig
n
if
ican
t
co
m
b
i
n
atio
n
.
T
h
is
is
p
r
ed
o
m
in
an
tly
ad
v
an
tag
eo
u
s
f
o
r
lar
g
e
-
s
ca
le
an
d
d
is
tr
ib
u
ted
f
o
r
ec
asti
n
g
f
o
llo
wi
n
g
:
i)
Own
s
u
p
p
o
r
ts
f
r
o
m
clien
ts
ar
e
weig
h
ted
ac
co
r
d
in
g
t
o
th
eir
d
ata
q
u
an
tity
as we
ll a
s
m
o
d
el
e
f
f
ec
tiv
en
ess
.
ii)
An
en
h
an
ce
d
FL
ap
p
r
o
ac
h
t
h
r
o
u
g
h
Fed
Av
g
is
u
tili
ze
d
f
o
r
v
ig
o
r
o
u
s
ag
g
r
eg
atio
n
,
m
e
ets
f
o
r
s
y
s
tem
d
y
n
am
ic
as we
ll a
s
an
d
d
awd
l
er
s
.
iii)
R
ath
er
th
an
ea
s
y
av
er
ag
in
g
,
weig
h
ted
av
er
ag
in
g
is
u
s
ed
,
wh
er
e
th
e
weig
h
ts
ar
e
id
en
tifie
d
in
ter
m
s
o
f
ev
er
y
clien
t
’
s
d
ata
d
is
tr
ib
u
tio
n
,
q
u
ality
o
r
esti
m
atio
n
i
n
d
ic
es.
T
h
is
ap
p
r
o
ac
h
g
iv
es
m
o
r
e
in
s
p
ir
atio
n
to
clien
ts
th
r
o
u
g
h
m
o
s
t a
p
p
r
o
p
r
ia
te
o
r
h
ig
h
er
-
ex
ce
llen
ce
d
ata.
iv
)
I
n
s
tead
o
f
s
im
p
le
ca
lcu
latio
n
,
r
ec
en
t
co
m
b
in
atio
n
ap
p
r
o
ac
h
es
h
av
e
b
ee
n
d
ev
elo
p
ed
th
a
t
in
co
r
p
o
r
ate
s
tatis
t
ical
ch
ar
ac
ter
is
tics
o
f
clien
t
u
p
d
ates
lik
e
v
ar
ian
ce
in
te
r
v
als,
to
en
a
b
le
a
m
o
r
e
d
etail
ed
an
d
r
o
b
u
s
t
g
lo
b
al
u
p
d
ate.
v)
A
p
u
r
p
o
s
e
o
f
u
tili
zin
g
weig
h
ted
av
er
ag
i
n
g
is
to
d
elib
er
ate
d
ata’
s
ir
r
eg
u
lar
d
is
p
er
s
io
n
o
f
d
ata
a
n
d
ex
ce
llen
ce
o
v
er
clien
ts
.
T
h
u
s
,
th
e
clien
ts
,
th
r
o
u
g
h
less
o
r
m
in
im
al
d
ata
f
r
o
m
co
n
tr
o
llin
g
a
g
lo
b
al
ap
p
r
o
ac
h
u
p
d
ate,
a
r
e
r
em
o
v
ed
.
R
ath
er
th
an
ev
en
ly
av
er
a
g
in
g
an
a
p
p
r
o
ac
h
u
p
d
ate
f
r
o
m
ev
er
y
clien
t,
t
h
e
weig
h
ts
ar
e
ap
p
lied
to
e
v
er
y
c
lien
t’
s
u
p
d
ate.
T
h
e
weig
h
ts
w
er
e
esti
m
ated
ac
co
r
d
i
n
g
to
ea
c
h
clien
t’
s
d
ata
d
is
p
er
s
al,
q
u
ality
as we
ll a
s
tr
ain
in
g
ef
f
ec
tiv
e
n
ess
.
v
i)
E
v
alu
atio
n
o
f
a
weig
h
t:
f
o
r
cl
ien
t
i
,
ass
u
m
e
d
en
o
tes
th
e
s
iz
e
o
f
a
d
ata;
d
en
o
tes
a
q
u
ality
s
co
r
e;
illu
s
tr
ates a
n
ef
f
ec
tiv
en
ess
o
f
tr
ain
in
g
.
T
h
e
weig
h
t
f
o
r
a
clien
t is ex
p
r
ess
ed
in
(
4
)
an
d
(
5
)
.
=
×
∑
=
1
+
×
+
(
1
−
−
)
×
(
4
)
=
×
∑
=
1
+
×
+
(
1
−
−
)
×
(
5
)
Her
e,
λ
an
d
μ
d
em
o
n
s
tr
ates
th
e
h
y
p
e
r
p
ar
am
eter
s
id
e
n
tify
i
n
g
th
e
im
p
o
r
tan
ce
o
f
d
ata
s
ize
as
well
as
q
u
ality
o
f
a
d
ata
in
d
iv
id
u
ally
.
T
h
ese
ass
u
r
an
ce
s
p
r
o
v
id
e
a
m
o
s
t
-
b
alan
ce
d
an
d
ac
cu
r
ate
d
e
p
ictio
n
o
f
d
ata
f
r
o
m
ea
ch
c
o
o
p
e
r
atin
g
clien
t.
At
ev
er
y
co
m
m
u
n
icatio
n
r
o
u
n
d
,
ev
er
y
d
ev
ice
esti
m
ates
lo
ca
l
u
p
d
ate
an
d
th
en
s
en
d
s
it
to
a
ce
n
tr
al
s
er
v
er
f
o
r
a
g
g
r
e
g
atio
n
.
T
h
is
iter
atio
n
co
n
tin
u
es
till
an
ap
p
r
o
a
ch
m
ee
ts
o
r
a
p
r
escr
ib
ed
co
u
n
t o
f
co
m
m
u
n
ic
atio
n
r
o
u
n
d
s
is
en
co
u
n
ter
ed
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
th
e
in
v
esti
g
atio
n
r
esu
l
ts
,
a
s
ig
n
if
ican
ce
o
f
an
in
tr
o
d
u
ce
d
m
eth
o
d
o
f
GDO
-
W
FL
p
r
iv
ac
y
-
p
r
eser
v
in
g
is
esti
m
ated
.
T
h
e
ex
p
er
im
en
ts
o
f
th
e
in
tr
o
d
u
ce
d
ap
p
r
o
ac
h
ar
e
r
ea
lized
o
n
Py
th
o
n
3
.
1
0
.
1
2
th
r
o
u
g
h
W
in
d
o
ws
1
0
OS,
1
6
GB
R
AM
an
d
I
n
tel
i5
p
r
o
c
ess
o
r
.
A
s
ig
n
if
ican
ce
o
f
in
tr
o
d
u
ce
d
a
p
p
r
o
ac
h
is
v
alid
ated
th
r
o
u
g
h
u
s
in
g
n
u
m
er
o
u
s
p
er
f
o
r
m
an
ce
m
etr
ics s
u
ch
as a
cc
u
r
ac
y
,
co
m
m
u
n
icatio
n
c
o
s
t,
co
m
p
u
tatio
n
al
co
s
t
,
an
d
r
u
n
n
in
g
tim
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
878
-
8
8
7
884
4
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
is
r
esear
ch
u
tili
s
es
b
en
ch
m
ar
k
MN
I
ST
an
d
C
I
FAR
-
1
0
d
atasets
f
o
r
im
p
lem
en
tin
g
th
e
s
ig
n
if
ican
ce
o
f
th
e
in
tr
o
d
u
ce
d
ap
p
r
o
ac
h
.
MN
I
ST
[
2
9
]
is
a
h
an
d
wr
itte
n
d
ig
ital
r
ep
ai
r
d
ataset
in
v
o
l
v
in
g
6
0
,
0
0
0
tr
ain
i
n
g
in
s
tan
ce
s
as
well
as
1
0
,
0
0
0
t
esti
n
g
s
am
p
les,
an
d
2
8
×2
8
s
ize
g
r
ey
-
lev
el
im
a
g
e.
MN
I
ST
is
p
ar
titi
o
n
ed
in
to
m
u
ltip
le
s
u
b
-
d
atasets
f
o
r
FL,
wh
ile
a
s
ize
o
f
e
v
er
y
s
u
b
-
d
ata
s
et
is
1
0
0
in
s
tan
ce
s
.
As
ass
o
ciate
d
with
MN
I
ST,
C
I
FAR
-
10
[
3
0
]
in
v
o
lv
es
th
e
m
o
s
t
co
m
p
lex
in
teg
r
atio
n
,
co
m
p
r
is
in
g
6
0
,
0
0
0
3
2
×
3
2
co
lo
r
in
s
tan
ce
s
in
1
0
ca
teg
o
r
ies,
with
6
,
0
0
0
in
s
t
an
ce
s
p
er
class
.
MN
I
ST
an
d
C
I
FAR
-
1
0
d
atasets
ar
e
b
r
o
ad
ly
u
tili
zin
g
in
ML
o
p
er
atio
n
s
lik
e
i
m
ag
e
class
if
icatio
n
.
W
h
er
ea
s
th
ese
d
atasets
lack
in
v
o
lv
em
e
n
t
o
f
p
er
s
o
n
ally
id
e
n
tifia
b
le
in
f
o
r
m
atio
n
(
PII
)
,
th
e
y
r
em
ain
v
u
ln
e
r
ab
le
to
s
p
ec
if
ic
p
r
iv
ac
y
attac
k
s
.
A
m
ajo
r
s
ec
u
r
ity
th
r
ea
t
is
in
v
o
lv
em
en
t
im
p
licatio
n
,
wh
er
e
an
attac
k
er
tr
ies
to
d
eter
m
in
e
a
s
p
ec
i
f
ic
d
ata
p
o
in
t
is
p
o
r
tio
n
o
f
an
ac
tu
al
d
atab
ase.
Simu
ltan
eo
u
s
ly
,
wh
ile
MN
I
ST
an
d
C
I
FAR
-
1
0
d
atasets
lack
in
v
o
lv
em
en
t
in
PII
,
i
t
is
co
n
ce
iv
a
b
le
to
re
-
class
if
y
en
titi
es
th
r
o
u
g
h
in
t
eg
r
atin
g
a
d
ataset
with
o
u
ts
id
e
d
ata.
T
h
er
e
is
a
g
r
ea
ter
is
s
u
e
th
at
an
attac
k
er
h
as
tr
ied
to
r
e
-
d
eter
m
in
e
th
o
s
e
in
d
iv
id
u
als ac
co
r
d
in
g
to
th
eir
f
ea
t
u
r
es.
4
.
2
.
P
er
f
o
r
m
a
nce
ev
a
lua
t
io
n
T
h
e
s
ig
n
if
ican
ce
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
is
esti
m
ated
with
th
e
p
r
ev
io
u
s
m
eth
o
d
s
b
y
u
s
i
n
g
v
a
r
io
u
s
p
er
f
o
r
m
an
ce
m
etr
ices.
T
h
e
ex
is
tin
g
m
eth
o
d
s
,
s
u
ch
as
FL,
Fed
Av
g
an
d
o
p
tim
ized
f
e
d
er
ate
lear
n
in
g
(
OFL)
ar
e
esti
m
ated
an
d
co
m
p
ar
e
d
with
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL.
T
h
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
is
esti
m
ated
b
y
u
s
in
g
v
a
r
io
u
s
p
er
f
o
r
m
an
ce
m
etr
ices
s
u
ch
as
n
u
m
b
er
o
f
r
o
u
n
d
s
,
n
u
m
b
er
o
f
cli
en
ts
an
d
n
u
m
b
er
o
f
g
r
ad
ien
ts
o
r
p
ac
k
ets p
er
u
s
er
.
T
ab
le
1
d
em
o
n
s
tr
ates
th
e
p
er
f
o
r
m
an
ce
ev
al
u
atio
n
o
f
ac
c
u
r
a
cy
b
ased
o
n
th
e
n
u
m
b
e
r
o
f
r
o
u
n
d
s
.
T
h
e
n
u
m
b
er
o
f
r
o
u
n
d
s
s
u
ch
as
5
,
1
0
,
1
5
,
2
0
,
an
d
2
5
,
is
co
n
s
id
er
ed
to
esti
m
ate
th
e
s
ig
n
if
ican
ce
o
f
th
e
in
tr
o
d
u
ce
d
ap
p
r
o
ac
h
.
I
n
MN
I
ST
d
ataset,
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
ac
h
attain
s
th
e
b
etter
ac
cu
r
ac
y
o
f
9
9
.
3
%
,
9
8
.
5
%
,
9
8
.
7
%
,
9
8
.
8
%
,
an
d
9
8
.
5
%
as
well
as
in
C
I
FA
R
d
ataset,
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
a
ch
attain
s
th
e
b
etter
ac
cu
r
ac
y
o
f
9
1
.
5
%
,
8
8
.
7
%
,
8
9
.
1
%
,
8
9
.
5
%
,
an
d
8
9
.
7
%
b
ase
d
o
n
t
h
e
d
if
f
er
en
t
n
u
m
b
e
r
o
f
r
o
u
n
d
s
o
f
5
,
1
0
,
1
5
,
2
0
,
an
d
2
5
in
d
iv
i
d
u
ally
.
T
ab
le
2
d
e
m
o
n
s
tr
ates
th
e
p
e
r
f
o
r
m
an
ce
e
v
alu
atio
n
o
f
ac
c
u
r
a
cy
b
ased
o
n
th
e
n
u
m
b
er
o
f
cl
ien
ts
.
T
h
e
n
u
m
b
er
o
f
r
o
u
n
d
s
s
u
ch
as
2
,
4
,
6
,
8
,
an
d
1
0
is
co
n
s
id
er
e
d
to
esti
m
ate
th
e
s
ig
n
if
ican
c
e
o
f
th
e
in
tr
o
d
u
ce
d
ap
p
r
o
ac
h
.
I
n
MN
I
ST
d
ataset,
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
ac
h
attain
s
th
e
b
etter
ac
cu
r
ac
y
o
f
9
4
.
5
%
,
9
4
.
7
%
,
9
5
.
0
%
,
9
5
.
3
%
,
an
d
9
5
.
8
%
as
well
as
in
C
I
FA
R
d
ataset,
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
a
ch
attain
s
th
e
b
etter
ac
cu
r
ac
y
o
f
8
7
.
3
%
,
8
7
.
5
%
,
8
7
.
9
%
,
8
8
.
1
%
,
an
d
8
8
.
3
%
b
ased
o
n
th
e
d
if
f
er
e
n
t
n
u
m
b
er
o
f
r
o
u
n
d
s
o
f
2
,
4
,
6
,
8
,
an
d
1
0
in
d
iv
i
d
u
ally
.
T
ab
le
1
.
Per
f
o
r
m
an
ce
ev
alu
ati
o
n
o
f
ac
cu
r
ac
y
r
esu
lts
(
%)
b
ased
o
n
n
u
m
b
er
o
f
r
o
u
n
d
s
D
a
t
a
s
e
t
M
e
t
h
o
d
s
N
u
mb
e
r
o
f
r
o
u
n
d
s
5
10
15
20
25
M
N
I
S
T
FL
9
3
.
2
9
3
.
5
9
3
.
8
9
4
.
0
9
4
.
1
F
e
d
A
V
g
9
5
.
8
9
5
.
9
9
6
.
3
9
6
.
7
9
6
.
8
O
F
L
9
6
.
4
9
6
.
5
9
7
.
1
9
7
.
4
9
7
.
5
GDO
-
WFL
9
9
.
3
9
8
.
5
9
8
.
7
9
8
.
8
9
8
.
8
C
I
F
A
R
-
10
FL
8
3
.
3
8
3
.
4
8
3
.
8
8
4
.
2
8
4
.
3
F
e
d
A
V
g
8
4
.
3
8
4
.
5
8
4
.
7
8
4
.
9
8
5
.
0
O
F
L
8
6
.
1
8
6
.
4
8
6
.
5
8
6
.
9
8
7
.
2
GDO
-
WFL
9
1
.
5
8
8
.
7
8
9
.
1
8
9
.
5
8
9
.
7
T
ab
le
2
.
Per
f
o
r
m
an
ce
ev
alu
ati
o
n
o
f
ac
cu
r
ac
y
r
esu
lts
(
%)
b
ased
o
n
n
u
m
b
er
o
f
clien
ts
D
a
t
a
s
e
t
M
e
t
h
o
d
s
N
u
mb
e
r
o
f
c
l
i
e
n
t
s
2
4
6
8
10
M
N
I
S
T
FL
9
0
.
6
9
0
.
8
9
1
.
2
9
1
.
5
9
1
.
6
F
e
d
A
V
g
9
2
.
5
9
2
.
8
9
2
.
9
9
3
.
1
9
3
.
3
O
F
L
9
3
.
5
9
3
.
6
9
3
.
9
9
4
.
2
9
4
.
4
GDO
-
WFL
9
4
.
5
9
4
.
7
9
5
.
0
9
5
.
3
9
5
.
8
C
I
F
A
R
-
10
FL
8
0
.
3
8
1
.
6
8
1
.
9
8
2
.
1
8
2
.
5
F
e
d
A
V
g
8
3
.
3
8
3
.
7
8
3
.
8
8
4
.
2
8
4
.
5
O
F
L
8
5
.
3
8
5
.
7
8
5
.
9
8
6
.
0
8
6
.
3
GDO
-
WFL
8
7
.
3
8
7
.
5
8
7
.
9
8
8
.
1
8
8
.
3
T
ab
le
3
d
em
o
n
s
tr
ates
th
e
p
er
f
o
r
m
an
ce
ev
alu
atio
n
o
f
c
o
m
m
u
n
icatio
n
co
s
t
b
ased
o
n
g
r
ad
ien
ts
/p
ac
k
ets
p
er
u
s
er
.
T
h
e
n
u
m
b
er
o
f
g
r
ad
ien
ts
/p
ac
k
ets
p
er
u
s
er
s
u
c
h
as
1
,
0
0
0
,
2
,
0
0
0
,
3
,
0
0
0
,
4
,
000
,
a
n
d
5
,
0
0
0
ar
e
co
n
s
id
er
ed
to
esti
m
ate
a
s
ig
n
if
ican
ce
o
f
in
tr
o
d
u
c
ed
ap
p
r
o
a
ch
.
I
n
MN
I
ST
d
ataset,
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Gra
d
ien
t d
escen
t o
p
timiz
a
tio
n
b
a
s
ed
w
eig
h
ted
fed
era
ted
lea
r
n
in
g
fo
r
p
r
iva
cy
-
…
(
Gu
r
u
r
a
j
P
r
a
ka
s
h
Mu
r
th
y)
885
ap
p
r
o
ac
h
attain
s
th
e
b
etter
c
o
m
m
u
n
icatio
n
c
o
s
t
o
f
4
9
0
KB
,
5
1
0
KB
,
5
3
0
KB
,
6
1
0
KB
,
an
d
6
5
0
KB
as
well
as
in
C
I
FAR
d
ataset,
th
e
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
ac
h
att
ain
s
th
e
b
etter
co
m
m
u
n
icatio
n
co
s
t
o
f
4
5
0
KB
,
490
KB
,
5
2
0
KB
,
6
4
0
KB
,
an
d
6
0
KB
b
ased
o
n
t
h
e
d
if
f
e
r
e
n
t
n
u
m
b
er
o
f
g
r
ad
ien
ts
/p
ac
k
et
s
p
er
u
s
er
o
f
1
,
0
0
0
,
2
,
0
0
0
,
3
,
0
0
0
,
4
,
0
0
0
,
a
n
d
5
,
0
0
0
in
d
iv
id
u
ally
.
T
ab
le
3
.
Per
f
o
r
m
an
ce
ev
alu
ati
o
n
o
f
co
m
m
u
n
icatio
n
c
o
s
t (
KB
)
b
ased
o
n
g
r
ad
ien
ts
/p
ac
k
ets
p
er
u
s
er
D
a
t
a
s
e
t
M
e
t
h
o
d
s
N
u
mb
e
r
o
f
g
r
a
d
i
e
n
t
s
/
p
a
c
k
e
t
s
p
e
r
u
ser
1
,
0
0
0
2
,
0
0
0
3
,
0
0
0
4
,
0
0
0
5
,
0
0
0
M
N
I
S
T
FL
5
7
0
6
0
0
6
1
0
6
6
0
7
0
0
F
e
d
A
V
g
5
4
0
5
8
0
5
8
0
6
4
0
6
9
0
O
F
L
5
2
0
5
4
0
5
6
0
6
3
0
6
7
0
GDO
-
WFL
4
9
0
5
1
0
5
3
0
6
1
0
6
5
0
C
I
F
A
R
-
10
FL
5
5
0
5
9
0
6
4
0
7
1
0
7
4
0
F
e
d
A
V
g
5
2
0
5
5
0
6
1
0
6
9
0
7
2
0
O
F
L
4
9
0
5
3
0
5
7
0
6
6
0
7
0
0
GDO
-
WFL
4
5
0
4
9
0
5
2
0
6
4
0
6
7
0
4
.
3
.
Co
m
pa
ra
t
iv
e
a
na
ly
s
is
T
h
e
co
m
p
a
r
a
t
iv
e
an
a
ly
s
i
s
o
f
t
h
e
p
r
o
p
o
s
ed
m
e
th
o
d
b
a
s
e
d
o
n
th
e
ex
i
s
t
i
n
g
m
e
th
o
d
i
s
d
es
c
r
i
b
ed
i
n
t
h
i
s
s
e
c
t
i
o
n
.
T
a
b
le
4
d
e
m
o
n
s
t
r
a
t
e
s
t
h
e
c
o
m
p
ar
a
t
iv
e
a
n
a
ly
s
i
s
o
f
t
h
e
e
x
i
s
t
i
n
g
m
e
th
o
d
w
i
t
h
t
h
e
p
r
o
p
o
s
ed
m
e
t
h
o
d
.
T
h
e
ef
f
e
c
t
iv
en
e
s
s
o
f
t
h
e
p
r
o
p
o
s
e
d
m
e
t
h
o
d
i
s
v
a
l
i
d
a
t
e
d
b
a
s
ed
o
n
t
w
o
d
i
f
f
e
r
en
t
d
a
t
a
s
e
t
s
l
i
k
e
M
N
I
S
T
a
n
d
C
I
F
A
R
-
1
0
.
T
ab
le
4
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
o
f
ac
cu
r
ac
y
r
esu
lts
(
%)
o
f
th
e
p
r
o
p
o
s
ed
m
eth
o
d
with
th
e
e
x
i
s
tin
g
m
eth
o
d
M
e
t
h
o
d
D
a
t
a
s
e
t
s
M
N
I
S
T
C
I
F
A
R
-
10
C
e
n
t
r
a
l
i
z
e
d
l
e
a
r
n
i
n
g
[
2
1
]
9
8
.
2
8
9
.
3
P
r
o
p
o
se
d
G
D
O
-
W
F
L
[
2
2
]
9
9
.
3
9
1
.
5
4
.
4
.
Dis
cus
s
io
n
T
h
e
p
r
o
p
o
s
ed
GDO
-
W
FL
ap
p
r
o
ac
h
o
u
t
p
er
f
o
r
m
s
ex
is
tin
g
FL
m
o
d
els
b
y
en
s
u
r
in
g
p
r
i
v
ac
y
a
n
d
m
o
d
el
p
er
f
o
r
m
an
ce
e
v
en
u
n
d
er
h
eter
o
g
en
eo
u
s
clien
t
s
ettin
g
s
.
Un
li
k
e
p
r
e
v
io
u
s
m
o
d
els,
it
d
y
n
am
ically
ac
co
u
n
ts
f
o
r
d
ata
q
u
ality
a
n
d
tr
ain
i
n
g
ef
f
e
ctiv
en
ess
th
r
o
u
g
h
weig
h
ted
a
v
er
ag
in
g
,
m
in
im
izin
g
th
e
e
f
f
e
ct
o
f
s
tr
ag
g
ler
s
a
n
d
lo
w
-
q
u
ality
co
n
t
r
ib
u
to
r
s
.
T
h
e
in
teg
r
atio
n
o
f
Kaf
k
a
-
Z
o
o
k
e
ep
er
en
ab
les
asy
n
ch
r
o
n
o
u
s
c
o
m
m
u
n
icatio
n
an
d
clien
t
an
o
n
y
m
ity
,
wh
ich
is
o
f
t
en
lack
in
g
in
tr
ad
itio
n
al
FL
s
ch
em
es.
Secu
r
e
ag
g
r
e
g
atio
n
co
m
b
in
ed
with
n
o
is
e
in
jectio
n
(
DP)
s
ig
n
if
ican
tly
r
ed
u
ce
s
p
r
iv
ac
y
leak
ag
e
r
is
k
.
T
h
e
ex
p
er
im
e
n
tal
r
esu
lts
ac
r
o
s
s
MN
I
ST
an
d
C
I
FAR
-
1
0
v
alid
ate
th
e
s
u
p
er
io
r
ac
cu
r
ac
y
an
d
r
ed
u
ce
d
co
m
m
u
n
icatio
n
o
v
e
r
h
ea
d
o
f
GDO
-
W
FL.
Ov
er
all,
th
e
m
o
d
el
d
em
o
n
s
tr
ates a
n
ef
f
icie
n
t,
s
ec
u
r
e,
an
d
s
ca
lab
le
s
o
lu
tio
n
f
o
r
FL
e
n
v
ir
o
n
m
en
ts
.
5.
CO
NCLU
SI
O
N
F
L
i
s
a
p
o
p
u
l
a
r
co
l
l
a
b
o
r
a
t
iv
e
l
e
a
r
n
i
n
g
wh
i
ch
co
m
b
i
n
ed
l
y
t
o
u
p
d
a
t
e
t
h
e
w
e
ig
h
t
s
o
r
g
r
ad
i
e
n
t
s
f
o
r
a
c
q
u
ir
i
n
g
a
g
lo
b
a
l
a
p
p
r
o
a
c
h
.
D
u
e
to
th
e
w
e
ig
h
t
s
o
r
g
r
ad
ie
n
t
s
a
r
e
s
e
n
s
i
t
i
v
e
d
a
t
a
,
i
t
h
a
s
b
e
en
i
n
v
e
s
t
i
g
a
t
e
d
u
n
d
e
r
v
a
r
io
u
s
p
r
iv
a
c
y
-
p
r
e
s
er
v
i
n
g
ap
p
r
o
ac
h
e
s
.
T
h
i
s
r
e
s
ea
r
c
h
p
r
o
p
o
s
e
s
a
n
o
v
e
l
p
r
i
v
a
c
y
-
p
r
e
s
er
v
in
g
F
L
a
p
p
r
o
a
ch
n
a
m
ed
G
D
O
-
W
F
L
,
w
h
i
c
h
s
o
l
v
e
s
th
e
p
r
o
b
l
e
m
o
f
ex
t
r
em
e
co
n
t
r
ib
u
t
io
n
o
f
lo
w
-
q
u
a
l
i
ty
d
a
ta
i
n
f
e
d
er
a
t
ed
t
r
a
in
i
n
g
.
T
h
r
o
u
g
h
d
e
v
e
l
o
p
i
n
g
a
c
o
m
p
l
ex
e
s
t
i
m
at
io
n
v
a
l
u
e
f
o
r
a
d
a
t
a
,
a
n
o
n
-
p
o
s
i
t
i
v
e
i
n
f
l
u
en
c
e
o
f
l
o
w
-
q
u
a
l
i
ty
d
a
t
a
o
n
f
ed
e
r
a
ted
t
r
a
in
i
n
g
i
s
m
i
n
im
i
z
e
d
,
w
h
il
e
m
a
k
e
s
u
r
e
a
p
r
iv
a
cy
a
s
w
e
l
l
a
s
s
e
c
u
r
i
ty
o
f
c
o
n
tr
i
b
u
t
o
r
d
a
t
a
b
y
s
e
c
u
r
e
m
o
d
e
l
.
T
h
e
e
x
p
e
r
im
e
n
ta
l
r
es
u
l
t
s
i
l
l
u
s
t
r
a
t
e
s
t
h
a
t
t
h
e
p
r
o
p
o
s
e
d
G
D
O
-
W
F
L
a
p
p
r
o
a
ch
a
t
t
a
i
n
s
th
e
o
v
e
r
a
l
l
ac
c
u
r
ac
y
o
f
9
9
.
3
%
an
d
9
1
.
5
%
o
n
M
N
I
S
T
a
n
d
C
I
F
A
R
-
1
0
d
at
a
s
e
t
s
a
s
c
o
m
p
ar
e
d
t
o
t
h
e
e
x
i
s
t
i
n
g
m
e
th
o
d
o
f
F
ed
l
a
b
X
m
e
t
h
o
d
.
I
n
f
u
t
u
r
e
r
e
s
e
a
r
c
h
,
i
t
’
s
i
m
p
o
r
t
an
t
t
o
f
o
cu
s
o
n
t
h
e
i
m
p
a
c
t
o
f
m
a
l
e
v
o
l
e
n
t
b
e
h
av
i
o
u
r
s
o
n
b
o
t
h
th
e
c
l
i
e
n
t
an
d
s
e
r
v
e
r
s
i
d
e
s
.
F
o
r
i
n
s
t
a
n
c
e
,
a
m
a
l
ic
i
o
u
s
c
l
i
e
n
t
m
ig
h
t
m
a
n
ip
u
l
a
t
e
i
t
s
g
r
a
d
i
en
t
to
i
n
f
l
u
en
c
e
th
e
a
cc
u
r
a
c
y
o
f
t
h
e
g
l
o
b
a
l
m
o
d
e
l,
w
h
i
le
a
m
a
l
i
ci
o
u
s
s
e
r
v
e
r
c
o
u
l
d
p
r
o
v
i
d
e
u
s
er
s
w
i
t
h
f
a
l
s
i
f
ie
d
ag
g
r
e
g
a
t
ed
r
e
s
u
l
t
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
878
-
8
8
7
886
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Gu
r
u
r
aj
Pra
k
ash
Mu
r
th
y
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
h
an
d
r
ash
ek
h
ar
Po
m
u
C
h
av
an
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
–
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
o
p
en
l
y
av
ailab
le
in
[
2
0
1
2
I
E
E
E
C
o
n
f
er
e
n
ce
o
n
C
o
m
p
u
ter
Vis
io
n
an
d
Patter
n
R
ec
o
g
n
itio
n
]
at
h
ttp
://d
o
i.o
r
g
/
1
0
.
1
1
0
9
/C
VPR
.
2
0
1
2
.
6
2
4
8
1
1
0
,
r
ef
er
e
n
ce
n
u
m
b
er
[
2
9
]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
–
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
t
h
e
f
i
n
d
i
n
g
s
o
f
th
is
s
tu
d
y
ar
e
o
p
en
l
y
a
v
ailab
le
in
[
T
ec
h
n
ical
R
ep
o
r
t,
Un
iv
er
s
ity
o
f
T
o
r
o
n
t
o
]
at
h
ttp
s
://www.
cs.t
o
r
o
n
to
.
ed
u
/~k
r
iz/lear
n
in
g
-
f
ea
t
u
r
es
-
2
0
0
9
-
T
R
.
p
d
f
,
r
ef
er
en
ce
n
u
m
b
er
[
3
0
]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
J.
Z
h
a
n
g
,
Y
.
L
i
u
,
D
.
W
u
,
S
.
L
o
u
,
B
.
C
h
e
n
,
a
n
d
S
.
Y
u
,
“
V
P
F
L:
a
v
e
r
i
f
i
a
b
l
e
p
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
s
c
h
e
me
f
o
r
e
d
g
e
c
o
m
p
u
t
i
n
g
s
y
s
t
e
ms
,
”
D
i
g
i
t
a
l
C
o
m
m
u
n
i
c
a
t
i
o
n
s
a
n
d
N
e
t
w
o
r
k
s
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
9
8
1
–
9
8
9
,
A
u
g
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
c
a
n
.
2
0
2
2
.
0
5
.
0
1
0
.
[
2
]
Z.
L
i
,
H
.
B
a
o
,
H
.
P
a
n
,
M
.
G
u
a
n
,
C
.
H
u
a
n
g
,
a
n
d
H
.
-
N
.
D
a
i
,
“
U
EFL
:
u
n
i
v
e
r
s
a
l
a
n
d
e
f
f
i
c
i
e
n
t
p
r
i
v
a
c
y
-
p
r
e
serv
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
I
EEE
I
n
t
e
r
n
e
t
o
f
T
h
i
n
g
s
J
o
u
r
n
a
l
,
v
o
l
.
1
2
,
n
o
.
1
0
,
p
p
.
1
4
3
3
3
–
1
4
3
4
7
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
JI
O
T.
2
0
2
5
.
3
5
2
5
7
3
1
.
[
3
]
L.
Z
h
o
n
g
,
L
.
Z
h
a
n
g
,
L
.
X
u
,
a
n
d
L.
W
a
n
g
,
“
M
P
C
-
b
a
se
d
p
r
i
v
a
c
y
-
p
r
e
serv
i
n
g
ser
v
e
r
l
e
ss
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
2
3
r
d
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
B
i
g
D
a
t
a
,
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
I
n
t
e
rn
e
t
o
f
T
h
i
n
g
s
En
g
i
n
e
e
r
i
n
g
(
I
C
BAIE)
,
J
u
l
.
2
0
2
2
,
p
p
.
4
9
3
–
4
9
7
.
d
o
i
:
1
0
.
1
1
0
9
/
I
C
B
A
I
E5
6
4
3
5
.
2
0
2
2
.
9
9
8
5
9
3
3
.
[
4
]
J.
L
i
u
,
X
.
Li
,
X
.
L
i
u
,
H
.
Z
h
a
n
g
,
Y
.
M
i
a
o
,
a
n
d
R
.
H
.
D
e
n
g
,
“
D
e
f
e
n
d
F
L
:
a
p
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
sc
h
e
m
e
a
g
a
i
n
s
t
p
o
i
s
o
n
i
n
g
a
t
t
a
c
k
s,”
I
EE
E
T
r
a
n
sa
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
r
k
s
a
n
d
L
e
a
r
n
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
6
,
n
o
.
5
,
p
p
.
9
0
9
8
–
9
1
1
1
,
M
a
y
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
LS.
2
0
2
4
.
3
4
2
3
3
9
7
.
[
5
]
F
.
W
a
n
g
,
Y
.
H
e
,
Y
.
G
u
o
,
P
.
L
i
,
a
n
d
X
.
W
e
i
,
“
P
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
r
o
b
u
st
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
w
i
t
h
d
i
st
r
i
b
u
t
e
d
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
i
n
2
0
2
2
I
E
EE
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
T
r
u
st
,
S
e
c
u
ri
t
y
a
n
d
Pri
v
a
c
y
i
n
C
o
m
p
u
t
i
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
(
T
ru
st
C
o
m
)
,
D
e
c
.
2
0
2
2
,
p
p
.
5
9
8
–
6
0
5
.
d
o
i
:
1
0
.
1
1
0
9
/
Tr
u
s
t
C
o
m5
6
3
9
6
.
2
0
2
2
.
0
0
0
8
7
.
[
6
]
J.
W
a
n
g
,
R
.
W
a
n
g
,
L.
X
i
o
n
g
,
N
.
X
i
o
n
g
,
a
n
d
Z
.
Li
u
,
“
S
A
EV
:
sec
u
r
e
a
g
g
r
e
g
a
t
i
o
n
a
n
d
e
f
f
i
c
i
e
n
t
v
e
r
i
f
i
c
a
t
i
o
n
f
o
r
p
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
I
EE
E
I
n
t
e
r
n
e
t
o
f
T
h
i
n
g
s
J
o
u
r
n
a
l
,
v
o
l
.
1
1
,
n
o
.
2
4
,
p
p
.
3
9
6
8
1
–
3
9
6
9
6
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
JI
O
T.
2
0
2
4
.
3
4
4
5
9
6
4
.
[
7
]
Y
.
Q
i
a
o
,
A
.
A
d
h
i
k
a
r
y
,
K
.
T.
K
i
m,
C
.
Zh
a
n
g
,
a
n
d
C
.
S
.
H
o
n
g
,
“
K
n
o
w
l
e
d
g
e
d
i
s
t
i
l
l
a
t
i
o
n
a
ssi
st
e
d
r
o
b
u
st
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
:
t
o
w
a
r
d
s
e
d
g
e
i
n
t
e
l
l
i
g
e
n
c
e
,
”
i
n
I
C
C
2
0
2
4
-
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
J
u
n
.
2
0
2
4
,
p
p
.
8
4
3
–
8
4
8
.
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
5
1
1
6
6
.
2
0
2
4
.
1
0
6
2
2
9
5
6
.
[
8
]
Z.
Li
u
,
J.
G
u
o
,
W
.
Y
a
n
g
,
J.
F
a
n
,
K
.
-
Y
.
L
a
m,
a
n
d
J
.
Zh
a
o
,
“
D
y
n
a
mi
c
u
ser
c
l
u
st
e
r
i
n
g
f
o
r
e
f
f
i
c
i
e
n
t
a
n
d
p
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
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
D
e
p
e
n
d
a
b
l
e
a
n
d
S
e
c
u
re
C
o
m
p
u
t
i
n
g
,
p
p
.
1
–
1
2
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TD
S
C
.
2
0
2
4
.
3
3
5
5
4
5
8
.
[
9
]
Z.
Zh
a
n
g
a
n
d
R
.
H
u
,
“
B
y
z
a
n
t
i
n
e
-
r
o
b
u
s
t
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
w
i
t
h
v
a
r
i
a
n
c
e
r
e
d
u
c
t
i
o
n
a
n
d
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
i
n
2
0
2
3
I
EEE
C
o
n
f
e
re
n
c
e
o
n
C
o
m
m
u
n
i
c
a
t
i
o
n
s
a
n
d
N
e
t
w
o
rk
S
e
c
u
ri
t
y
(
C
N
S
)
,
O
c
t
.
2
0
2
3
,
p
p
.
1
–
9
.
d
o
i
:
1
0
.
1
1
0
9
/
C
N
S
5
9
7
0
7
.
2
0
2
3
.
1
0
2
8
8
9
3
8
.
[
1
0
]
Z.
L
u
,
S
.
L
u
,
X
.
T
a
n
g
,
a
n
d
J.
W
u
,
“
R
o
b
u
s
t
a
n
d
v
e
r
i
f
i
a
b
l
e
p
r
i
v
a
c
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
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
5
,
n
o
.
4
,
p
p
.
1
8
9
5
–
1
9
0
8
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TA
I
.
2
0
2
3
.
3
3
0
9
2
7
3
.
[
1
1
]
X
.
W
a
n
g
,
S
.
W
a
n
g
,
Y
.
L
i
,
F
.
F
a
n
,
S
.
Li
,
a
n
d
X
.
Li
n
,
“
D
i
f
f
e
r
e
n
t
i
a
l
l
y
p
r
i
v
a
t
e
a
n
d
h
e
t
e
r
o
g
e
n
e
i
t
y
-
r
o
b
u
st
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
w
i
t
h
t
h
e
o
r
e
t
i
c
a
l
g
u
a
r
a
n
t
e
e
,
”
I
EEE
T
r
a
n
sa
c
t
i
o
n
s
o
n
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
5
,
n
o
.
1
2
,
p
p
.
6
3
6
9
–
6
3
8
4
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TA
I
.
2
0
2
4
.
3
4
4
6
7
5
9
.
[
1
2
]
S
.
S
.
N
a
g
e
s
h
,
N
.
F
e
r
n
a
n
d
o
,
S
.
W
.
L
o
k
e
,
A
.
G
.
N
e
i
a
t
,
a
n
d
P
.
N
.
P
a
t
h
i
r
a
n
a
,
“
H
o
n
e
y
b
e
e
-
R
S
:
e
n
h
a
n
c
i
n
g
t
r
u
s
t
t
h
r
o
u
g
h
l
i
g
h
t
w
e
i
g
h
t
r
e
su
l
t
v
a
l
i
d
a
t
i
o
n
i
n
mo
b
i
l
e
c
r
o
w
d
c
o
mp
u
t
i
n
g
,
”
i
n
2
0
2
4
I
E
EE
2
3
rd
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
T
r
u
st
,
S
e
c
u
r
i
t
y
a
n
d
Pr
i
v
a
c
y
i
n
C
o
m
p
u
t
i
n
g
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
(
T
ru
st
C
o
m
)
,
D
e
c
.
2
0
2
4
,
p
p
.
2
5
5
3
–
2
5
5
8
.
d
o
i
:
1
0
.
1
1
0
9
/
Tr
u
st
C
o
m
6
3
1
3
9
.
2
0
2
4
.
0
0
3
5
6
.
[
1
3
]
H
.
Ze
n
g
e
t
a
l
.
,
“
B
S
R
-
F
L:
a
n
e
f
f
i
c
i
e
n
t
B
y
z
a
n
t
i
n
e
-
r
o
b
u
st
p
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
,
”
I
E
EE
T
ra
n
sa
c
t
i
o
n
s
o
n
C
o
m
p
u
t
e
rs
,
v
o
l
.
7
3
,
n
o
.
8
,
p
p
.
2
0
9
6
–
2
1
1
0
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TC
.
2
0
2
4
.
3
4
0
4
1
0
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Gra
d
ien
t d
escen
t o
p
timiz
a
tio
n
b
a
s
ed
w
eig
h
ted
fed
era
ted
lea
r
n
in
g
fo
r
p
r
iva
cy
-
…
(
Gu
r
u
r
a
j
P
r
a
ka
s
h
Mu
r
th
y)
887
[
1
4
]
S
.
N
a
z
i
r
a
n
d
M
.
K
a
l
e
e
m,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
m
e
d
i
c
a
l
i
ma
g
e
a
n
a
l
y
si
s
w
i
t
h
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
D
i
a
g
n
o
st
i
c
s
,
v
o
l
.
1
3
,
n
o
.
9
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
d
i
a
g
n
o
s
t
i
c
s1
3
0
9
1
5
3
2
.
[
1
5
]
L.
Zh
a
n
g
,
T.
Z
h
u
,
P
.
X
i
o
n
g
,
W
.
Z
h
o
u
,
a
n
d
P
.
S
.
Y
u
,
“
A
r
o
b
u
st
g
a
m
e
-
t
h
e
o
r
e
t
i
c
a
l
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
r
a
m
e
w
o
r
k
w
i
t
h
j
o
i
n
t
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
I
E
EE
T
ra
n
s
a
c
t
i
o
n
s
o
n
K
n
o
w
l
e
d
g
e
a
n
d
D
a
t
a
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
3
5
,
n
o
.
4
,
p
p
.
3
3
3
3
–
3
3
4
6
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
2
1
.
3
1
4
0
1
3
1
.
[
1
6
]
A.
-
T.
Tr
a
n
a
n
d
X
.
-
S
.
P
h
a
m,
“
A
n
o
v
e
l
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
s
c
h
e
me
f
o
r
t
h
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
C
O
V
I
D
-
1
9
i
n
c
h
e
st
X
-
r
a
y
i
ma
g
e
s,
”
i
n
2
0
2
3
1
5
t
h
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
K
n
o
w
l
e
d
g
e
a
n
d
S
y
st
e
m
s
E
n
g
i
n
e
e
ri
n
g
(
K
S
E)
,
O
c
t
.
2
0
2
3
,
p
p
.
1
–
6
.
d
o
i
:
1
0
.
1
1
0
9
/
K
S
E
5
9
1
2
8
.
2
0
2
3
.
1
0
2
9
9
4
3
3
.
[
1
7
]
M
.
S
h
e
n
e
t
a
l
.
,
“
S
e
c
u
r
e
d
e
c
e
n
t
r
a
l
i
z
e
d
a
g
g
r
e
g
a
t
i
o
n
t
o
p
r
e
v
e
n
t
m
e
m
b
e
r
s
h
i
p
p
r
i
v
a
c
y
l
e
a
k
a
g
e
i
n
e
d
g
e
-
b
a
se
d
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
t
w
o
r
k
S
c
i
e
n
c
e
a
n
d
En
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
1
,
n
o
.
3
,
p
p
.
3
1
0
5
–
3
1
1
9
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
S
E.
2
0
2
4
.
3
3
6
0
3
1
1
.
[
1
8
]
G
.
Z
h
e
n
g
,
L.
K
o
n
g
,
a
n
d
A
.
B
r
i
n
t
r
u
p
,
“
F
e
d
e
r
a
t
e
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
f
o
r
p
r
i
v
a
c
y
p
r
e
serv
i
n
g
,
c
o
l
l
e
c
t
i
v
e
su
p
p
l
y
c
h
a
i
n
r
i
s
k
p
r
e
d
i
c
t
i
o
n
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Pr
o
d
u
c
t
i
o
n
R
e
se
a
rc
h
,
v
o
l
.
6
1
,
n
o
.
2
3
,
p
p
.
8
1
1
5
–
8
1
3
2
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
8
0
/
0
0
2
0
7
5
4
3
.
2
0
2
2
.
2
1
6
4
6
2
8
.
[
1
9
]
K
.
O
.
-
A
g
y
e
men
g
,
Z.
Q
i
n
,
H
.
X
i
o
n
g
,
Y
.
L
i
u
,
T
.
Zh
u
a
n
g
,
a
n
d
Z
.
Q
i
n
,
“
M
S
D
P
:
mu
l
t
i
-
s
c
h
e
me
p
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
d
e
e
p
l
e
a
r
n
i
n
g
v
i
a
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
P
e
r
so
n
a
l
a
n
d
U
b
i
q
u
i
t
o
u
s
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
7
,
n
o
.
2
,
p
p
.
2
2
1
–
2
3
3
,
A
p
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s0
0
7
7
9
-
021
-
0
1
5
4
5
-
0.
[
2
0
]
T.
H
.
R
a
f
i
,
F
.
A
.
N
o
o
r
,
T.
H
u
ss
a
i
n
,
a
n
d
D
.
-
K
.
C
h
a
e
,
“
F
a
i
r
n
e
ss
a
n
d
p
r
i
v
a
c
y
p
r
e
ser
v
i
n
g
i
n
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
:
a
s
u
r
v
e
y
,
”
I
n
f
o
rm
a
t
i
o
n
F
u
si
o
n
,
v
o
l
.
1
0
5
,
M
a
y
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
n
f
f
u
s.
2
0
2
3
.
1
0
2
1
9
8
.
[
2
1
]
Y
.
Y
a
n
e
t
a
l
.
,
“
F
e
d
l
a
b
x
:
a
p
r
a
c
t
i
c
a
l
a
n
d
p
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
f
r
a
mew
o
r
k
f
o
r
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
C
o
m
p
l
e
x
&
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
,
v
o
l
.
1
0
,
n
o
.
1
,
p
p
.
6
7
7
–
6
9
0
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
0
7
4
7
-
0
2
3
-
0
1
1
8
4
-
3.
[
2
2
]
Y
.
C
h
e
n
,
B
.
W
a
n
g
,
H
.
Ji
a
n
g
,
P
.
D
u
a
n
,
Y
.
P
i
n
g
,
a
n
d
Z.
H
o
n
g
,
“
P
EPF
L:
a
f
r
a
mew
o
r
k
f
o
r
a
p
r
a
c
t
i
c
a
l
a
n
d
e
f
f
i
c
i
e
n
t
p
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
D
i
g
i
t
a
l
C
o
m
m
u
n
i
c
a
t
i
o
n
s
a
n
d
N
e
t
w
o
rks
,
v
o
l
.
1
0
,
n
o
.
2
,
p
p
.
3
5
5
–
3
6
8
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
d
c
a
n
.
2
0
2
2
.
0
5
.
0
1
9
.
[
2
3
]
S
u
mi
t
r
a
,
J
.
S
h
a
r
ma,
a
n
d
M
.
V
.
S
h
e
n
o
y
,
“
H
A
F
e
d
L
:
a
H
e
ss
i
a
n
-
a
w
a
r
e
a
d
a
p
t
i
v
e
p
r
i
v
a
c
y
p
r
e
s
e
r
v
i
n
g
h
o
r
i
z
o
n
t
a
l
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
sch
e
me
f
o
r
I
o
T
a
p
p
l
i
c
a
t
i
o
n
s,
”
I
EE
E
A
c
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
1
2
6
7
3
8
–
1
2
6
7
5
3
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
4
5
4
0
7
4
.
[
2
4
]
H
.
W
a
n
g
,
Q
.
W
a
n
g
,
Y
.
D
i
n
g
,
S
.
Ta
n
g
,
a
n
d
Y
.
W
a
n
g
,
“
P
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
b
a
s
e
d
o
n
p
a
r
t
i
a
l
l
o
w
-
q
u
a
l
i
t
y
d
a
t
a
,
”
J
o
u
rn
a
l
o
f
C
l
o
u
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
3
,
n
o
.
1
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
8
6
/
s1
3
6
7
7
-
0
2
4
-
0
0
6
1
8
-
8.
[
2
5
]
L.
Z
h
o
n
g
e
t
a
l
.
,
“
D
u
a
l
-
ser
v
e
r
-
b
a
se
d
l
i
g
h
t
w
e
i
g
h
t
p
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
N
e
t
w
o
r
k
a
n
d
S
e
r
v
i
c
e
Ma
n
a
g
e
m
e
n
t
,
v
o
l
.
2
1
,
n
o
.
4
,
p
p
.
4
7
8
7
–
4
8
0
0
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
S
M
.
2
0
2
4
.
3
3
9
9
5
3
4
.
[
2
6
]
L.
W
a
n
g
,
X
.
Z
h
a
o
,
Z.
L
u
,
L.
W
a
n
g
,
a
n
d
S
.
Z
h
a
n
g
,
“
E
n
h
a
n
c
i
n
g
p
r
i
v
a
c
y
p
r
e
serv
a
t
i
o
n
a
n
d
t
r
u
s
t
w
o
r
t
h
i
n
e
ss
f
o
r
d
e
c
e
n
t
r
a
l
i
z
e
d
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
I
n
f
o
rm
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
6
2
8
,
p
p
.
4
4
9
–
4
6
8
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
n
s.
2
0
2
3
.
0
1
.
1
3
0
.
[
2
7
]
X
.
S
u
n
,
Z.
Y
u
a
n
,
X
.
K
o
n
g
,
L.
X
u
e
,
L.
H
e
,
a
n
d
Y
.
L
i
n
,
“
C
o
mm
u
n
i
c
a
t
i
o
n
-
e
f
f
i
c
i
e
n
t
a
n
d
p
r
i
v
a
c
y
-
p
r
e
s
e
r
v
i
n
g
a
g
g
r
e
g
a
t
i
o
n
i
n
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
w
i
t
h
a
d
a
p
t
a
b
i
l
i
t
y
,
”
I
E
EE
I
n
t
e
r
n
e
t
o
f
T
h
i
n
g
s
J
o
u
r
n
a
l
,
v
o
l
.
1
1
,
n
o
.
1
5
,
p
p
.
2
6
4
3
0
–
2
6
4
4
3
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
JI
O
T.
2
0
2
4
.
3
3
9
6
2
1
7
.
[
2
8
]
F
.
S
h
a
n
,
Y
.
Lu
,
S
.
Li
,
S
.
M
a
o
,
Y
.
Li
,
a
n
d
X
.
W
a
n
g
,
“
Ef
f
i
c
i
e
n
t
a
d
a
p
t
i
v
e
d
e
f
e
n
se
sc
h
e
m
e
f
o
r
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
i
n
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
S
e
c
u
ri
t
y
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
8
9
,
M
a
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
i
sa.
2
0
2
5
.
1
0
3
9
9
2
.
[
2
9
]
D
.
C
i
r
e
sa
n
,
U
.
M
e
i
e
r
,
a
n
d
J
.
S
c
h
m
i
d
h
u
b
e
r
,
“
M
u
l
t
i
-
c
o
l
u
m
n
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
2
0
1
2
I
EE
E
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
s
i
o
n
a
n
d
Pa
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
J
u
n
.
2
0
1
2
,
p
p
.
3
6
4
2
–
3
6
4
9
.
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
.
2
0
1
2
.
6
2
4
8
1
1
0
.
[
3
0
]
A
.
K
r
i
z
h
e
v
sk
y
,
“
Le
a
r
n
i
n
g
m
u
l
t
i
p
l
e
l
a
y
e
r
s
o
f
f
e
a
t
u
r
e
s
f
r
o
m
t
i
n
y
i
ma
g
e
s.
”
T
e
c
h
n
i
c
a
l
Re
p
o
rt
,
U
n
i
v
e
r
si
t
y
o
f
T
o
r
o
n
t
o
,
To
r
o
n
t
o
,
O
n
t
a
r
i
o
,
2
0
0
9
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
c
s.
t
o
r
o
n
t
o
.
e
d
u
/
~
k
r
i
z
/
l
e
a
r
n
i
n
g
-
f
e
a
t
u
r
e
s
-
2
0
0
9
-
TR
.
p
d
f
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
G
u
r
u
r
a
j
Pra
k
a
sh
Mu
r
th
y
is
re
se
a
rc
h
sc
h
o
lar
a
t
P
ES
Un
i
v
e
rsity
.
C
o
m
p
lete
d
b
a
c
h
e
lo
r
o
f
En
g
in
e
e
ri
n
g
a
n
d
m
a
ste
r
o
f
En
g
i
n
e
e
rin
g
i
n
Visv
e
sw
a
ra
y
a
i
n
stit
u
te
o
f
tec
h
n
o
l
o
g
y
.
Wo
rk
e
d
a
s
so
ftwa
re
e
n
g
in
e
e
r
1
0
y
e
a
rs
in
IT
i
n
d
u
stry
a
n
d
p
re
se
n
tl
y
w
o
rk
i
n
g
a
s
a
n
a
ss
istan
t
p
ro
fe
ss
o
r
fr
o
m
p
a
st t
h
re
e
y
e
a
rs.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
g
u
r
u
r
a
jp
@p
e
s.e
d
u
.
Cha
n
d
r
a
shek
h
a
r
P
o
m
u
Cha
v
a
n
re
c
e
iv
e
d
h
is
B.
E.
d
e
g
re
e
in
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
fro
m
G
u
ru
Na
n
a
k
De
v
En
g
in
e
e
rin
g
Co
ll
e
g
e
,
Bi
d
a
r,
Ka
rn
a
tak
a
,
In
d
ia,
a
n
d
h
is
M
.
Tec
h
d
e
g
re
e
in
Ne
two
rk
a
n
d
In
tern
e
t
En
g
in
e
e
rin
g
fr
o
m
S
ri
Ja
y
a
c
h
a
m
a
ra
jen
d
ra
Co
ll
e
g
e
o
f
En
g
i
n
e
e
rin
g
,
M
y
s
o
re
,
Ka
rn
a
tak
a
,
In
d
ia,
wh
e
re
h
e
se
c
u
re
d
th
e
3
r
d
ra
n
k
in
t
h
e
u
n
iv
e
rsity
.
He
late
r
e
a
rn
e
d
h
is
P
h
.
D.
i
n
th
e
field
o
f
Wi
re
les
s
Ne
two
rk
in
g
fr
o
m
th
e
In
d
ian
In
st
it
u
te
o
f
S
c
ien
c
e
(IIS
c
),
Ba
n
g
a
l
o
re
,
In
d
ia.
His
c
o
re
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
wire
les
s
n
e
two
rk
s,
m
o
b
il
e
a
d
-
h
o
c
n
e
two
rk
s
(M
AN
ET
s),
Io
T,
a
rt
i
ficia
l
in
tell
ig
e
n
c
e
a
n
d
m
a
c
h
in
e
lea
rn
in
g
(AIM
L),
c
lo
u
d
c
o
m
p
u
ti
n
g
,
u
b
i
q
u
i
to
u
s
n
e
tw
o
rk
s,
n
e
two
r
k
se
c
u
rit
y
,
p
e
r
v
a
siv
e
c
o
m
p
u
ti
n
g
,
c
o
n
tex
t
-
a
wa
re
sy
ste
m
s,
a
n
d
p
o
st
-
q
u
a
n
t
u
m
c
ry
p
t
o
g
ra
p
h
y
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
c
p
c
h
a
v
a
n
@p
e
s.e
d
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.