I
nte
rna
t
io
na
l J
o
urna
l o
f
I
nfo
rm
a
t
ics a
nd
Co
m
m
un
ica
t
io
n T
ec
hn
o
lo
g
y
(
I
J
-
I
CT
)
Vo
l.
15
,
No
.
4
,
Dec
em
b
er
20
26
,
p
p
.
1
4
5
3
~1
4
5
8
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct
.
v15
i
4
.
pp
1
4
5
3
-
1
4
5
8
1453
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
Federate
d
dee
p le
a
rning
f
o
r early
d
etec
tion
and pro
g
ress
io
n
mo
delling
of diab
etic
ne
phro
pa
thy
Ara
v
ind
K
um
a
r
K
y
ri
k
a
,
Su
m
it
K
um
a
r
B
a
nchh
o
r
A
mi
t
y
I
n
st
i
t
u
t
e
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
A
m
i
t
y
U
n
i
v
e
r
si
t
y
C
h
h
a
t
t
i
sg
a
r
h
,
R
a
i
p
u
r
,
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
Dec
1
,
2
0
2
5
R
ev
is
ed
Ma
y
2
6
,
2
0
2
6
Acc
ep
ted
J
u
l 5
,
2
0
2
6
On
e
o
f
th
e
m
a
in
c
a
u
se
s
o
f
e
n
d
-
s
tag
e
re
n
a
l
d
ise
a
se
is
d
ia
b
e
ti
c
n
e
p
h
r
o
p
a
th
y
(DN
).
P
a
ti
e
n
t
p
r
iv
a
c
y
-
se
n
siti
v
e
d
a
ta
is
a
re
strictio
n
i
n
c
e
n
tralize
d
m
a
c
h
in
e
-
lea
rn
in
g
a
p
p
r
o
a
c
h
e
s
fo
r
th
e
h
e
a
lt
h
c
a
re
d
o
m
a
in
.
T
h
is
re
se
a
rc
h
in
tro
d
u
c
e
s
a
p
riv
a
c
y
-
p
re
se
rv
in
g
fe
d
e
ra
ted
d
eep
-
lea
rn
in
g
(F
e
d
-
D
N)
m
e
th
o
d
fo
r
e
a
rly
d
iag
n
o
sis
o
f
DN
a
n
d
m
o
d
e
ll
i
n
g
d
ise
a
se
p
ro
g
re
ss
io
n
a
n
d
it
s
imp
a
c
t.
F
e
d
-
DN
c
o
m
b
i
n
e
s
sa
fe
a
g
g
re
g
a
ti
o
n
with
d
i
ffe
re
n
ti
a
l
p
riv
a
c
y
(
DP)
u
si
n
g
Op
a
c
u
s
a
n
d
m
u
l
ti
-
m
o
d
a
l
d
a
ta
so
u
rc
e
s
(e
lec
tro
n
ic
h
e
a
lt
h
re
c
o
r
d
s,
fu
n
d
u
s
ima
g
in
g
,
a
n
d
h
isto
p
a
th
o
lo
g
y
).
B
o
th
sim
u
latio
n
tes
ts
wit
h
m
u
lt
i
-
m
o
d
a
l
d
a
ta
se
ts
d
istri
b
u
te
d
a
c
ro
ss
h
o
s
p
it
a
ls
sh
o
w
a
n
e
fficie
n
t
p
riv
a
c
y
-
u
ti
li
t
y
trad
e
-
o
ff,
with
F
e
d
-
DN
a
c
h
iev
i
n
g
n
e
a
r
-
c
e
n
tralize
d
p
e
rfo
rm
a
n
c
e
(AU
C
≈
0
.
9
0
)
(Ac
c
u
ra
c
y
=
8
7
%
)
w
h
il
e
m
a
in
tai
n
in
g
a
p
ri
v
a
c
y
b
u
d
g
e
t
o
f
ε
≈
3
.
1
.
Th
is
is
a
g
o
o
d
e
x
a
m
p
le o
f
th
e
trad
e
-
o
f
f
a
c
h
iev
e
d
b
e
twe
e
n
m
o
d
e
l
a
c
c
u
ra
c
y
a
n
d
p
riv
a
c
y
p
ro
tec
ti
o
n
.
Th
e
tec
h
n
i
q
u
e
d
o
e
s
n
o
t
re
v
e
a
l
p
a
ti
e
n
t
i
n
fo
rm
a
ti
o
n
,
wh
i
c
h
e
n
a
b
le
s
c
o
ll
a
b
o
ra
ti
v
e
,
sc
a
lab
le
h
e
a
lt
h
c
a
re
a
n
a
ly
ti
c
s.
F
u
rt
h
e
rm
o
re
,
it
e
n
h
a
n
c
e
s
tru
st
am
o
n
g
p
a
ti
e
n
ts an
d
i
n
stit
u
ti
o
n
s.
K
ey
w
o
r
d
s
:
Diab
etic
n
ep
h
r
o
p
ath
y
Dif
f
er
en
tial p
r
iv
ac
y
Fed
er
ated
lear
n
in
g
Hea
lth
ca
r
e
AI
Mu
lti
-
m
o
d
al
d
ee
p
lear
n
in
g
Secu
r
e
ag
g
r
e
g
atio
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
:
Su
m
it Ku
m
ar
B
an
ch
h
o
r
Am
ity
I
n
s
titu
te
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
,
Am
ity
Un
iv
er
s
ity
C
h
h
attis
g
ar
h
R
aip
u
r
,
C
h
h
attis
g
ar
h
,
I
n
d
ia
E
m
ail:
s
k
b
an
ch
h
o
r
@
r
p
r
.
am
ity
.
ed
u
1.
I
NT
RO
D
UCT
I
O
N
Diab
etic
n
ep
h
r
o
p
ath
y
(
DN)
is
a
p
r
ev
alen
t
an
d
s
ev
er
e
d
iab
etes
m
ellitu
s
co
m
p
licatio
n
an
d
a
m
ajo
r
ca
u
s
e
o
f
ch
r
o
n
ic
k
id
n
ey
d
is
ea
s
e
(
C
KD)
an
d
en
d
-
s
tag
e
r
en
al
d
is
ea
s
e
(
E
S
R
D)
[
1
]
,
[
2
]
.
Ma
c
h
in
e
lear
n
in
g
(
ML
)
h
o
ld
s
o
u
t
th
e
p
o
ten
tial f
o
r
ea
r
lier
d
iag
n
o
s
is
an
d
m
o
r
e
p
r
ec
is
e
p
r
o
g
r
ess
io
n
m
o
d
elin
g
,
b
u
t c
en
tr
alize
d
tr
ain
in
g
is
u
s
u
ally
im
p
r
ac
tical
d
u
e
to
p
r
i
v
ac
y
laws
lik
e
th
e
h
ea
lth
in
s
u
r
an
ce
p
o
r
tab
ilit
y
an
d
ac
co
u
n
tab
ilit
y
ac
t
(
HI
PAA)
an
d
g
en
er
al
d
ata
p
r
o
tectio
n
r
e
g
u
latio
n
(
GDPR
)
[
3
]
,
[
4
]
.
Fed
er
ated
lear
n
in
g
(
FL)
allo
ws
c
o
llab
o
r
ativ
e
m
o
d
el
tr
ain
in
g
ac
r
o
s
s
h
o
s
p
itals
with
o
u
t
s
h
ar
in
g
r
aw
p
atien
t
r
ec
o
r
d
s
[
5
]
,
[
6
]
.
Ho
wev
er
,
r
ea
l
-
w
o
r
ld
d
ep
l
o
y
m
en
t
is
co
n
f
r
o
n
ted
with
ch
allen
g
es
lik
e
n
o
n
-
i
n
d
ep
e
n
d
en
t
an
d
id
en
t
ically
d
is
tr
ib
u
ted
(
I
I
D)
lo
ca
l
d
ata,
co
m
m
u
n
icatio
n
lim
itatio
n
s
,
an
d
p
r
iv
ac
y
leak
a
g
e
v
ia
m
o
d
el
u
p
d
ates
[
7
]
,
[
8
]
.
T
h
is
p
ap
e
r
in
tr
o
d
u
ce
s
f
ed
er
ated
d
ee
p
-
lear
n
in
g
(
Fed
-
D,
wh
ich
is
a
f
ed
e
r
ated
s
y
s
tem
th
at
m
ix
es
d
if
f
er
en
ti
al
p
r
iv
ac
y
(
DP)
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
DP
-
SGD)
with
s
ec
u
r
e
a
g
g
r
e
g
a
tio
n
[
9
]
-
[
1
1
]
to
lim
it
leak
ag
e,
an
d
p
er
s
o
n
aliza
tio
n
m
eth
o
d
s
to
ad
d
r
ess
h
eter
o
g
en
eity
,
u
s
ed
f
o
r
m
u
lti
-
m
o
d
al
DN
p
r
e
d
ictio
n
an
d
p
r
o
g
r
ess
io
n
m
o
d
elin
g
.
2.
T
H
E
P
RO
P
O
SE
D
WO
RK
Fed
Av
g
an
d
c
o
m
m
u
n
icatio
n
-
awa
r
e
FL
alg
o
r
ith
m
s
a
r
e
t
h
e
b
ac
k
b
o
n
e
o
f
c
r
o
s
s
-
in
s
titu
tio
n
al
ML
ap
p
licatio
n
s
[
4
]
-
[6
]
.
Fo
r
i
n
s
tan
ce
,
FL
h
as
s
u
cc
ess
f
u
lly
b
ee
n
ap
p
lied
to
im
ag
in
g
task
s
s
u
ch
as
b
r
ai
n
tu
m
o
r
s
eg
m
en
tatio
n
[
7
]
,
m
e
d
ical
d
iag
n
o
s
is
[
1
2
]
,
[
1
3
]
an
d
m
o
d
e
llin
g
o
f
el
ec
tr
o
n
ic
h
ea
lth
r
ec
o
r
d
s
(
E
HR
)
[
1
4
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
4
,
Dec
em
b
er
20
26
:
1
4
5
3
-
1
4
5
8
1454
Oth
er
p
r
iv
ac
y
-
p
r
eser
v
in
g
tech
n
i
q
u
es,
lik
e
s
ec
u
r
e
ag
g
r
eg
at
io
n
[
9
]
,
[
1
1
]
a
n
d
DP
[
1
0
]
,
[
1
5]
,
[
1
6
]
h
a
v
e
b
ee
n
ap
p
lied
to
FL
to
p
r
o
tect
ag
a
in
s
t
m
em
b
er
s
h
ip
in
f
er
en
ce
[
1
7
]
,
[
1
8
]
an
d
r
ec
o
n
s
tr
u
ctio
n
attac
k
s
.
W
h
ile
th
e
in
teg
r
atio
n
o
f
tab
u
lar
E
HR
with
im
ag
in
g
(
f
u
n
d
u
s
,
h
is
to
p
at
h
o
lo
g
y
)
in
cr
ea
s
es
d
iag
n
o
s
tic
ac
cu
r
ac
y
[
1
3
]
,
[
1
9
]
,
[
2
0
]
,
m
u
lti
-
m
o
d
al
FL
is
s
till
n
o
t f
u
lly
ex
p
lo
r
e
d
an
d
is
th
e
g
a
p
th
at
Fed
-
DN
will a
d
d
r
ess
.
Alg
o
r
ith
m
p
r
o
b
lem
f
o
r
m
u
latio
n
:
a)
I
n
p
u
t: C
lin
ical
tex
t d
ata
f
r
o
m
v
ar
io
u
s
h
o
s
p
itals
with
o
u
t c
o
m
p
r
o
m
is
in
g
p
atien
t d
ata.
b)
Pre
-
p
r
o
ce
s
s
in
g
: Cl
ea
n
clin
ical
tex
ts
an
d
to
k
en
ize
t
h
em
to
o
b
t
ain
r
elev
an
t c
lin
ical
f
ea
tu
r
es.
c)
L
o
ca
l
t
r
ain
in
g
: T
h
e
m
o
d
el
is
tr
ain
ed
lo
ca
lly
at
ea
ch
h
o
s
p
ital
with
DP
-
SGD,
g
r
ad
ien
t c
lip
p
i
n
g
,
an
d
Gau
s
s
ian
n
o
is
e
f
o
r
p
r
i
v
ac
y
.
d)
Fed
er
ated
ag
g
r
eg
atio
n
: Sec
u
r
e
ly
m
er
g
es th
e
n
o
is
y
m
o
d
el
u
p
d
ates f
r
o
m
th
e
p
ar
ticip
atin
g
h
o
s
p
itals
e)
Per
s
o
n
aliza
tio
n
:
A
p
r
o
x
im
al
t
er
m
tak
es
ca
r
e
o
f
n
o
n
-
I
I
D
d
a
ta
an
d
g
en
er
ates
a
p
er
s
o
n
alize
d
m
o
d
el
f
o
r
ev
er
y
h
o
s
p
ital,
wh
ile
r
etain
in
g
g
lo
b
al
k
n
o
wled
g
e
.
3.
M
E
T
H
O
D
S
L
et
h
o
s
p
itals
p
ar
ticip
ate,
ea
ch
with
a
p
r
iv
ate
d
ataset
=
{
(
(
)
,
(
)
)
}
=
1
T
h
e
g
lo
b
al
e
m
p
ir
ical
o
b
jectiv
e
is
(
)
=
∑
=
1
(
)
(
)
=
1
∑
ℓ
=
1
(
(
(
)
)
,
(
)
)
wh
er
e
d
en
o
tes m
o
d
el
p
ar
am
et
er
s
an
d
ℓ
th
e
p
er
-
e
x
am
p
le
lo
s
s
(
e.
g
.
,
cr
o
s
s
-
en
tr
o
p
y
)
.
C
lien
ts
p
er
f
o
r
m
DP
-
SGD
with
g
r
ad
ien
t c
lip
p
in
g
n
o
r
m
an
d
Gau
s
s
ian
n
o
is
e:
Δ
̃
=
Δ
+
(
0
,
2
2
)
T
h
e
s
er
v
er
a
g
g
r
eg
ates n
o
is
y
u
p
d
ates f
r
o
m
p
ar
ticip
atin
g
clien
ts
at
r
o
u
n
d
:
+
1
=
+
1
|
|
∑
Δ
̃
∈
T
o
im
p
r
o
v
e
p
e
r
s
o
n
aliza
tio
n
u
n
d
er
h
ete
r
o
g
e
n
eo
u
s
d
is
tr
ib
u
tio
n
s
,
we
in
clu
d
e
a
p
r
o
x
i
m
al
p
er
s
o
n
aliza
tio
n
ter
m
:
min
(
)
+
∥
−
∥
2
2
3
.
1
.
Sy
s
t
e
m
o
v
er
v
iew
As
s
h
o
wn
in
Fig
u
r
e
1
,
Fed
-
D
N
h
as
th
r
ee
m
ajo
r
co
m
p
o
n
e
n
ts
:
First,
a
ce
n
tr
al
s
er
v
er
th
at
c
o
o
r
d
in
ates
co
m
m
u
n
icatio
n
r
o
u
n
d
s
,
s
ets
u
p
s
ec
u
r
e
ag
g
r
eg
atio
n
,
a
n
d
m
ai
n
tain
s
th
e
g
lo
b
al
m
o
d
el;
Seco
n
d
,
clien
ts
,
n
am
ely
,
th
e
h
o
s
p
itals
th
at
tr
ain
a
lo
ca
l
DP
-
SGD
m
o
d
el
o
n
th
eir
o
wn
m
u
lti
-
m
o
d
al
p
atien
t
d
ata
an
d
o
p
tio
n
ally
f
in
e
-
tu
n
e
a
p
er
s
o
n
alis
ed
m
o
d
el
h
ea
d
to
r
ef
lect
th
e
d
ata
h
eter
o
g
e
n
eity
o
f
ea
ch
h
o
s
p
ital;
T
h
ir
d
,
a
co
m
m
o
n
d
ata
ca
talo
g
u
e
th
at
p
r
o
v
i
d
es c
o
n
s
is
ten
t f
ea
tu
r
e
m
ap
p
in
g
s
,
d
ata
r
ep
r
esen
tatio
n
,
an
d
d
ata
p
r
e
p
r
o
ce
s
s
in
g
ac
r
o
s
s
all
clien
ts
.
3
.
2
.
M
ulti
-
m
o
da
l
m
o
del a
rc
hite
ct
ure
Fig
u
r
e
2
illu
s
tr
ates
a
m
u
ltimo
d
al
h
ea
lth
ca
r
e
s
y
s
tem
in
teg
r
atin
g
E
HR
d
ata
f
r
o
m
clin
ical
tex
t,
m
ed
ica
l
im
ag
in
g
,
an
d
p
atien
t
lab
in
f
o
r
m
atio
n
.
T
h
e
o
u
tp
u
ts
o
f
th
is
lay
er
in
clu
d
e
im
a
g
in
g
a
n
d
d
ia
g
n
o
s
is
p
r
o
ce
s
s
ed
f
r
o
m
th
ese
in
p
u
ts
an
d
s
en
t
to
a
f
u
s
io
n
lay
er
.
T
h
e
f
u
s
io
n
lay
e
r
co
m
b
in
es
all
o
f
th
e
in
f
o
r
m
atio
n
to
cr
ea
te
an
in
teg
r
ated
d
iag
n
o
s
is
an
d
s
u
m
m
ar
y
.
E
n
co
d
er
s
:
−
E
HR
en
co
d
er
:
Den
s
e
o
r
t
r
an
s
f
o
r
m
er
-
b
ased
en
co
d
e
r
f
o
r
tab
u
l
ar
/time
-
s
er
ies
f
ea
tu
r
es
o
f
clin
i
ca
l
tex
t
d
ata,
m
ed
ical
im
ag
in
g
,
an
d
p
atien
t l
ab
d
ata
[
1
4
]
.
−
Fu
n
d
u
s
en
co
d
er
:
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
(
C
NN)
b
ac
k
b
o
n
e
(
e.
g
.
,
R
esNet)
f
o
r
r
etin
al
f
u
n
d
u
s
im
ag
es [
1
9
]
.
−
His
to
p
ath
o
lo
g
y
e
n
co
d
e
r
: CNN o
r
p
atch
-
tr
an
s
f
o
r
m
er
f
o
r
tis
s
u
e
p
atch
es [
1
3
]
.
L
aten
t v
ec
to
r
s
e
hr
,
f
undus
,
his
to
ar
e
co
n
ca
ten
ated
an
d
f
u
s
ed
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
F
ed
era
ted
d
ee
p
lea
r
n
in
g
f
o
r
ea
r
ly
d
etec
tio
n
a
n
d
p
r
o
g
r
ess
io
n
m
o
d
ellin
g
o
f
… (
A
r
a
vin
d
K
u
ma
r
K
yrika
)
1455
f
us
e
=
R
e
L
U
(
[
e
hr
;
f
undus
;
his
to
]
+
)
Fin
al
p
r
ed
ictio
n
:
̂
=
(
f
us
e
+
)
Fig
u
r
e
1
.
FL
ar
c
h
itectu
r
e
Fig
u
r
e
2
.
T
h
e
co
m
p
lete
s
y
s
tem
ar
ch
itectu
r
e
d
iag
r
am
3
.
3
.
P
ri
v
a
cy
a
nd
s
ec
urit
y
mecha
nis
m
s
Fig
u
r
e
3
d
is
p
lay
s
th
e
p
r
iv
ac
y
th
r
ea
ts
an
d
m
itig
atio
n
s
tr
ateg
ies
in
ML
.
Mo
d
el
in
v
e
r
s
io
n
a
ttem
p
ts
to
ex
tr
ac
t
s
en
s
itiv
e
d
ata
f
r
o
m
a
tr
ain
ed
m
o
d
el.
Me
m
b
er
s
h
ip
in
f
er
en
ce
r
ef
er
s
to
wh
eth
er
a
s
p
ec
if
ic
d
ata
r
ec
o
r
d
is
u
s
ed
f
o
r
tr
ai
n
in
g
.
DP
h
el
p
s
s
af
eg
u
ar
d
in
d
iv
id
u
a
l
d
ata,
an
d
s
ec
u
r
e
a
g
g
r
e
g
atio
n
h
elp
s
e
n
s
u
r
e
th
at
d
ata
r
e
m
ain
s
p
r
iv
ate
d
u
r
in
g
m
o
d
el
tr
ain
i
n
g
.
−
DP: Per
-
ex
am
p
le
o
r
p
er
-
s
am
p
l
e
g
r
ad
ien
t c
lip
p
in
g
a
n
d
Gau
s
s
ian
n
o
is
e
ad
d
itio
n
(
DP
-
SGD)
with
Op
ac
u
s
f
o
r
p
r
iv
ac
y
ac
c
o
u
n
tin
g
[
5
]
,
[
2
1
]
,
[
22
].
−
Secu
r
e
Ag
g
r
eg
ati
o
n
: Cry
p
to
g
r
ap
h
ic
p
r
o
to
co
ls
to
ag
g
r
eg
ate
e
n
cr
y
p
ted
clien
t u
p
d
ates so
th
e
s
er
v
er
ca
n
n
o
t
o
b
s
er
v
e
in
d
i
v
id
u
al
u
p
d
ates [
9
]
,
[1
1
].
C
o
m
b
in
ed
,
DP
an
d
s
ec
u
r
e
ag
g
r
eg
atio
n
m
itig
ate
g
r
a
d
ien
t
-
in
v
er
s
io
n
an
d
m
em
b
e
r
s
h
ip
-
in
f
er
en
ce
attac
k
s
[
1
7
]
,
[
18
].
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
4
,
Dec
em
b
er
20
26
:
1
4
5
3
-
1
4
5
8
1456
Fig
u
r
e
3
.
Priv
ac
y
attac
k
an
d
d
ef
en
s
e
d
iag
r
am
3
.
4
.
H
i
g
h
-
lev
el
a
lg
o
rit
hm
T
h
is
alg
o
r
ith
m
is
ca
lled
Fed
-
DN
an
d
is
co
m
p
o
s
ed
o
f
a
ce
n
tr
al
s
er
v
er
a
n
d
p
ar
ticip
atin
g
h
o
s
p
itals
.
T
h
e
s
er
v
er
in
itializes
th
e
g
lo
b
al
m
o
d
el,
p
ick
s
a
s
et
o
f
clien
ts
f
o
r
ea
ch
co
m
m
u
n
icati
o
n
r
o
u
n
d
,
s
en
d
s
o
u
t
th
e
g
lo
b
al
m
o
d
el,
g
ets
b
ac
k
n
o
is
y
u
p
d
ates
th
at
h
av
e
b
e
en
en
cr
y
p
ted
,
s
af
ely
co
m
b
in
es
th
e
m
,
an
d
u
p
d
ates
th
e
g
lo
b
al
m
o
d
el.
L
o
ca
l
DP
-
SGD
is
im
p
lem
en
ted
o
n
ea
ch
h
o
s
p
ital's
p
r
iv
ate
p
atien
t
d
ata.
T
h
e
n
o
r
m
C
is
u
s
ed
to
clip
g
r
ad
ien
ts
,
an
d
Gau
s
s
ian
n
o
is
e
with
th
e
m
u
ltip
lier
σ
i
s
ad
d
ed
f
o
r
p
r
iv
ac
y
.
I
t
is
p
o
s
s
ib
le
to
f
in
e
-
tu
n
e
o
p
tio
n
al
p
e
r
s
o
n
aliza
tio
n
to
lo
ca
l
d
ata
with
th
e
m
o
d
el.
L
astl
y
,
th
e
n
o
is
y
u
p
d
ate
is
s
en
t
to
th
e
s
er
v
er
,
b
u
t
it
is
en
cr
y
p
ted
f
o
r
s
ec
u
r
e
ag
g
r
eg
ati
o
n
.
Ser
v
er
:
in
itialize
0
.
Fo
r
=
0
,
…
,
−
1
:
s
a
m
p
le
s
u
b
s
et
,
s
en
d
,
r
ec
eiv
e
en
cr
y
p
ted
n
o
is
y
u
p
d
ates
Δ
̃
,
ag
g
r
eg
ate
a
n
d
u
p
d
at
e
+
1
.
C
lien
t
:
r
ec
eiv
e
,
p
er
f
o
r
m
D
P
-
SGD
with
clip
p
in
g
n
o
r
m
an
d
n
o
is
e
m
u
ltip
lier
,
o
p
tio
n
al
p
er
s
o
n
aliza
tio
n
f
in
e
-
tu
n
in
g
,
s
e
n
d
en
cr
y
p
ted
Δ
̃
.
4.
RE
SU
L
T
S
T
h
e
r
esu
lts
s
h
o
wn
in
T
ab
le
1
d
em
o
n
s
tr
ate
th
at
Fed
-
DN
o
u
tp
er
f
o
r
m
s
th
e
l
o
ca
l
-
o
n
l
y
a
n
d
Fed
Av
g
m
o
d
els,
an
d
p
er
f
o
r
m
s
clo
s
e
to
th
e
ce
n
tr
alize
d
C
NN.
Fed
-
DN
ac
h
iev
es
n
ea
r
-
ce
n
tr
alize
d
p
er
f
o
r
m
a
n
ce
wh
il
e
p
r
o
v
id
i
n
g
f
o
r
m
al
DP
g
u
ar
an
te
es
an
d
im
p
r
o
v
e
d
ad
a
p
tatio
n
t
o
n
o
n
-
I
I
D
clien
t
d
is
tr
ib
u
tio
n
s
v
ia
p
er
s
o
n
aliza
tio
n
[
2
3
]
,
[
2
4
]
.
T
h
e
s
m
all
ac
cu
r
ac
y
g
ap
v
s
ce
n
tr
alize
d
m
o
d
els ar
is
es f
r
o
m
DP n
o
is
e
an
d
s
im
u
lat
ed
d
ata
lim
itatio
n
s
.
Alg
o
r
ith
m
1
d
em
o
n
s
tr
ate
C
o
m
b
in
ed
DP
+
Per
s
o
n
aliza
tio
n
is
o
b
s
er
v
ed
to
p
r
o
d
u
ce
n
ea
r
-
ce
n
tr
alize
d
p
er
f
o
r
m
an
ce
(
AUC
≈
0
.
9
0
)
(
Acc
u
r
ac
y
=
8
7
%)
wh
ile
m
ain
t
ain
in
g
a
p
r
iv
ac
y
b
u
d
g
et
o
f
ε
≈
3
.
1
(A
.
I
n
g
e
n
er
al,
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
m
o
d
els
ca
n
b
e
e
n
h
an
ce
d
b
y
co
m
b
i
n
in
g
m
o
d
alities
.
T
ab
le
1
.
Su
m
m
a
r
y
o
f
ex
p
er
im
en
t
al
p
er
f
o
r
m
an
ce
(
s
im
u
lated
d
ata)
M
e
t
h
o
d
DP
P
e
r
so
n
a
l
i
z
a
t
i
o
n
P
r
i
v
a
c
y
B
u
d
g
e
t
(
\
e
p
si
l
o
n
)
AUC
A
c
c
u
r
a
c
y
C
e
n
t
r
a
l
i
z
e
d
C
N
N
No
No
–
0
.
9
2
8
9
%
C
e
n
t
r
a
l
i
z
e
d
C
N
N
+
D
P
Y
e
s
No
3
.
8
0
.
8
9
8
6
%
Lo
c
a
l
-
o
n
l
y
No
Y
e
s
–
0
.
7
8
7
5
%
Lo
c
a
l
-
o
n
l
y
+
D
P
Y
e
s
Y
e
s
3
.
5
0
.
7
5
7
2
%
F
e
d
A
v
g
No
No
–
0
.
8
8
8
6
%
F
e
d
A
v
g
+
D
P
Y
e
s
No
3
.
2
0
.
8
5
8
3
%
F
e
d
-
DN
No
Y
e
s
–
0
.
9
2
8
9
%
F
e
d
-
D
N
(
D
P
+
P
e
r
so
n
a
l
i
z
a
t
i
o
n
)
Y
e
s
Y
e
s
3
.
1
0
.
9
0
8
7
%
Alg
o
r
ith
m
1
.
Op
ac
u
s
DP
-
SGD
co
d
e
s
k
etch
for batch in dataloader:
optimizer.zero.grad()
loss = criterion(model(batch x), batch y)
loss.backward()
torch.nn.utils.clip_grad_norm_(model.parameters(), 1.0) # clip
add_noise(model.parameters(), sigma=1.0) # Opacus handles per
-
sample no
ise
optimizer.step()
eps = privacy_engine.accountant.get_epsilon(delta=1e
-
5)
# Privacy parameters: batch_size=32, epochs=2, sigma=1.0
-
> epsilon ≈ 3.1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
F
ed
era
ted
d
ee
p
lea
r
n
in
g
f
o
r
ea
r
ly
d
etec
tio
n
a
n
d
p
r
o
g
r
ess
io
n
m
o
d
ellin
g
o
f
… (
A
r
a
vin
d
K
u
ma
r
K
yrika
)
1457
Mu
lti
-
m
o
d
al
DN
p
r
ed
ictio
n
c
an
b
e
ac
h
iev
e
d
with
a
wo
r
k
a
b
le
p
r
iv
ac
y
-
u
tili
ty
t
r
ad
e
-
o
f
f
f
o
r
Fed
-
DN.
T
h
e
u
s
e
o
f
s
im
u
lated
d
ata
in
s
tead
o
f
ac
tu
al
m
u
lti
-
in
s
titu
tio
n
al
co
h
o
r
ts
,
th
e
en
g
in
ee
r
in
g
a
n
d
c
o
m
m
u
n
icatio
n
o
v
er
h
ea
d
f
o
r
ac
tu
al
d
ep
lo
y
m
e
n
ts
[
4
]
,
an
d
th
e
ac
cu
r
ac
y
d
e
g
r
ad
atio
n
b
r
o
u
g
h
t
o
n
b
y
DP
n
o
is
e
ar
e
s
o
m
e
o
f
th
e
lim
itatio
n
s
.
L
o
n
g
itu
d
in
al
m
o
d
ellin
g
an
d
tim
e
-
to
-
ev
e
n
t
an
al
y
s
is
,
in
ter
p
r
etab
ilit
y
f
o
r
clin
ic
al
u
p
tak
e
[
2
0
]
an
d
p
ilo
t m
u
lti
-
ce
n
ter
d
ep
lo
y
m
en
t
[
3
]
,
an
d
f
air
n
ess
an
d
b
ias ass
ess
m
en
t [
2
5
]
ar
e
f
u
tu
r
e
d
ir
ec
tio
n
s
.
5.
CO
NCLU
SI
O
N
Fin
ally
,
th
e
p
r
o
p
o
s
ed
Fed
-
DN
f
r
am
ewo
r
k
s
h
o
ws
th
e
p
o
ten
ti
al
o
f
a
p
r
i
v
ac
y
-
p
r
eser
v
in
g
f
ed
er
ated
d
ee
p
lear
n
in
g
m
o
d
el
f
o
r
t
h
e
ea
r
ly
d
iag
n
o
s
is
an
d
m
o
d
ellin
g
o
f
th
e
p
r
o
g
r
ess
io
n
o
f
d
ia
b
etic
n
ep
h
r
o
p
ath
y
.
T
h
e
f
r
a
m
ewo
r
k
in
co
r
p
o
r
ates
f
ed
er
ated
lear
n
in
g
,
DP
b
ased
o
n
O
p
ac
u
s
,
s
ec
u
r
e
ag
g
r
eg
atio
n
,
an
d
m
u
lti
-
m
o
d
al
h
ea
lth
ca
r
e
d
ata
to
f
ac
ilit
ate
h
o
s
p
itals
in
jo
in
tly
b
u
ild
in
g
a
m
o
d
el
with
o
u
t
ex
p
o
s
in
g
p
atie
n
ts
'
s
en
s
i
tiv
e
d
ata.
T
h
e
s
im
u
latio
n
r
esu
lts
s
h
o
w
th
at
Fed
-
DN
p
er
f
o
r
m
s
n
ea
r
ly
c
en
tr
ally
,
with
an
AUC o
f
ab
o
u
t 0
.
9
0
an
d
a
p
r
i
v
ac
y
b
u
d
g
et
o
f
ε
≈
3
.
1
,
b
alan
ci
n
g
p
r
ed
ictiv
e
p
er
f
o
r
m
an
ce
a
n
d
p
r
iv
ac
y
p
r
o
tectio
n
.
As
s
u
ch
,
Fed
-
DN
o
f
f
er
s
a
s
ca
lab
le
an
d
s
ec
u
r
e
s
o
lu
tio
n
f
o
r
c
o
llab
o
r
ativ
e
h
ea
lth
ca
r
e
an
aly
tics
,
tack
lin
g
t
h
e
cr
itical
is
s
u
es
o
f
d
ata
p
r
iv
ac
y
an
d
s
ec
u
r
ity
.
Fu
tu
r
e
r
esear
c
h
m
ay
in
v
o
l
v
e
th
e
ex
te
n
s
io
n
o
f
th
e
v
alid
atio
n
to
r
ea
l
-
wo
r
l
d
clin
ical
s
ettin
g
s
,
d
ea
lin
g
with
n
o
n
-
I
I
D
an
d
im
b
alan
ce
d
d
ata,
p
er
s
o
n
alis
atio
n
b
etwe
en
h
o
s
p
itals
,
an
d
o
p
tim
i
zin
g
co
m
m
u
n
icatio
n
an
d
p
r
i
v
ac
y
co
s
ts
.
ACK
NO
WL
E
DG
E
M
E
NT
T
h
e
au
th
o
r
s
ar
e
th
an
k
f
u
l
to
th
e
Dep
ar
tm
en
t
o
f
I
n
f
o
r
m
atio
n
T
ec
h
n
o
lo
g
y
f
o
r
p
r
o
v
id
in
g
th
e
to
o
ls
an
d
f
ac
ilit
ies
to
s
u
cc
ess
f
u
lly
co
n
d
u
ct
th
is
r
esear
ch
.
W
e
th
a
n
k
clin
ical
n
u
tr
itio
n
a
n
d
n
ep
h
r
o
lo
g
y
ex
p
e
r
ts
f
o
r
d
o
m
ain
g
u
id
an
ce
a
n
d
th
e
FL
co
m
m
u
n
i
ty
f
o
r
f
o
u
n
d
atio
n
al
m
et
h
o
d
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
s
tate
th
at
n
o
f
u
n
d
i
n
g
was
in
v
o
lv
e
d
.
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
Ar
av
in
d
Ku
m
a
r
Ky
r
ik
a
✓
✓
✓
✓
✓
✓
Su
m
it Ku
m
ar
B
an
ch
h
o
r
✓
✓
✓
✓
✓
✓
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
T
h
e
au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
ter
est.
DATA AV
AI
L
AB
I
L
I
T
Y
Data
av
ailab
ilit
y
d
o
es n
o
t a
p
p
l
y
to
th
is
p
ap
e
r
as n
o
n
ew
d
ata
wer
e
cr
ea
ted
o
r
an
aly
ze
d
in
th
is
s
tu
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
H
.
B
.
M
c
M
a
h
a
n
,
E.
M
o
o
r
e
,
D
.
R
a
mag
e
,
S
.
H
a
mp
s
o
n
,
a
n
d
B
.
A
g
ü
e
r
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
,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
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
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
s
t
i
c
s
(
AI
S
T
ATS
)
,
v
o
l
.
5
4
,
p
p
.
1
2
7
3
–
1
2
8
2
,
2
0
1
7
.
[
2
]
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
re
n
d
s
®
i
n
m
a
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
.
[
3
]
K
.
B
o
n
a
w
i
t
z
e
t
a
l
.
,
“
P
r
a
c
t
i
c
a
l
se
c
u
r
e
a
g
g
r
e
g
a
t
i
o
n
f
o
r
p
r
i
v
a
c
y
-
p
r
e
ser
v
i
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
7
A
C
M
S
I
G
S
A
C
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
e
c
u
ri
t
y
,
O
c
t
.
2
0
1
7
,
p
p
.
1
1
7
5
–
1
1
9
1
,
d
o
i
:
1
0
.
1
1
4
5
/
3
1
3
3
9
5
6
.
3
1
3
3
9
8
2
.
[
4
]
C
.
D
w
o
r
k
a
n
d
A
.
R
o
t
h
,
“
T
h
e
a
l
g
o
r
i
t
h
mi
c
f
o
u
n
d
a
t
i
o
n
s o
f
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
F
o
u
n
d
a
t
i
o
n
s
a
n
d
T
re
n
d
s®
i
n
T
h
e
o
r
e
t
i
c
a
l
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
9
,
n
o
.
3
–
4
,
p
p
.
2
1
1
–
4
8
7
,
A
u
g
.
2
0
1
4
,
d
o
i
:
1
0
.
1
5
6
1
/
0
4
0
0
0
0
0
0
4
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
15
,
No
.
4
,
Dec
em
b
er
20
26
:
1
4
5
3
-
1
4
5
8
1458
[
5
]
M
.
A
b
a
d
i
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
n
g
w
i
t
h
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
1
6
A
C
M
S
I
G
S
AC
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
e
c
u
ri
t
y
,
O
c
t
.
2
0
1
6
,
p
p
.
3
0
8
–
3
1
8
,
d
o
i
:
1
0
.
1
1
4
5
/
2
9
7
6
7
4
9
.
2
9
7
8
3
1
8
.
[
6
]
I
.
M
i
r
o
n
o
v
,
“
R
é
n
y
i
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
,
”
i
n
2
0
1
7
I
E
EE
3
0
t
h
C
o
m
p
u
t
e
r
S
e
c
u
ri
t
y
Fo
u
n
d
a
t
i
o
n
s
S
y
m
p
o
si
u
m
(
C
S
F)
,
A
u
g
.
2
0
1
7
,
p
p
.
2
6
3
–
2
7
5
,
d
o
i
:
1
0
.
1
1
0
9
/
c
sf
.
2
0
1
7
.
1
1
.
[
7
]
M
.
J
.
S
h
e
l
l
e
r
,
G
.
A
.
R
e
i
n
a
,
B
.
E
d
w
a
r
d
s,
J
.
M
a
r
t
i
n
,
a
n
d
S
.
B
a
k
a
s,
“
M
u
l
t
i
-
i
n
s
t
i
t
u
t
i
o
n
a
l
d
e
e
p
l
e
a
r
n
i
n
g
m
o
d
e
l
i
n
g
w
i
t
h
o
u
t
sh
a
r
i
n
g
p
a
t
i
e
n
t
d
a
t
a
:
a
f
e
a
s
i
b
i
l
i
t
y
s
t
u
d
y
o
n
b
r
a
i
n
t
u
m
o
r
se
g
me
n
t
a
t
i
o
n
,
”
i
n
Br
a
i
n
l
e
s
i
o
n
:
G
l
i
o
m
a
,
M
u
l
t
i
p
l
e
S
c
l
e
ro
si
s,
S
t
ro
k
e
a
n
d
T
ra
u
m
a
t
i
c
Bra
i
n
I
n
j
u
r
i
e
s
,
S
p
r
i
n
g
e
r
I
n
t
e
r
n
a
t
i
o
n
a
l
P
u
b
l
i
s
h
i
n
g
,
2
0
1
9
,
p
p
.
9
2
–
1
0
4
.
[
8
]
N
.
R
i
e
k
e
e
t
a
l
.
,
“
T
h
e
f
u
t
u
r
e
o
f
d
i
g
i
t
a
l
h
e
a
l
t
h
w
i
t
h
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,
”
n
p
j
D
i
g
i
t
a
l
Me
d
i
c
i
n
e
,
v
o
l
.
3
,
n
o
.
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
7
4
6
-
020
-
0
0
3
2
3
-
1.
[
9
]
T.
S
.
B
r
i
si
m
i
,
R
.
C
h
e
n
,
T
.
M
e
l
a
,
A
.
O
l
s
h
e
v
sk
y
,
I
.
C
.
P
a
sc
h
a
l
i
d
i
s,
a
n
d
W
.
S
h
i
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
o
f
p
r
e
d
i
c
t
i
v
e
m
o
d
e
l
s
f
r
o
m
f
e
d
e
r
a
t
e
d
El
e
c
t
r
o
n
i
c
H
e
a
l
t
h
R
e
c
o
r
d
s,”
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Me
d
i
c
a
l
I
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
1
1
2
,
p
p
.
5
9
–
6
7
,
A
p
r
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
j
m
e
d
i
n
f
.
2
0
1
8
.
0
1
.
0
0
7
.
[
1
0
]
R
.
S
h
o
k
r
i
a
n
d
V
.
S
h
ma
t
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
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
2
n
d
A
C
M
S
I
G
S
A
C
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
e
c
u
r
i
t
y
,
O
c
t
.
2
0
1
5
,
p
p
.
1
3
1
0
–
1
3
2
1
,
d
o
i
:
1
0
.
1
1
4
5
/
2
8
1
0
1
0
3
.
2
8
1
3
6
8
7
.
[
1
1
]
M
.
F
r
e
d
r
i
k
so
n
,
S
.
Jh
a
,
a
n
d
T.
R
i
s
t
e
n
p
a
r
t
,
“
M
o
d
e
l
i
n
v
e
r
si
o
n
a
t
t
a
c
k
s
t
h
a
t
e
x
p
l
o
i
t
c
o
n
f
i
d
e
n
c
e
i
n
f
o
r
m
a
t
i
o
n
a
n
d
b
a
s
i
c
c
o
u
n
t
e
r
m
e
a
s
u
r
e
s,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
AC
M
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
a
n
d
C
o
m
m
u
n
i
c
a
t
i
o
n
s
S
e
c
u
ri
t
y
,
v
o
l
.
2
0
1
5
-
O
c
t
o
b
,
p
p
.
1
3
2
2
–
1
3
3
3
,
2
0
1
5
,
d
o
i
:
1
0
.
1
1
4
5
/
2
8
1
0
1
0
3
.
2
8
1
3
6
7
7
.
[
1
2
]
K
.
B
o
n
a
w
i
t
z
e
t
a
l
.
,
“
T
o
w
a
r
d
s
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
a
t
sca
l
e
:
S
y
st
e
m
d
e
si
g
n
,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
a
n
d
S
y
s
t
e
m
s
(
MLS
y
s)
,
v
o
l
.
1
,
p
p
.
3
7
4
–
3
8
8
,
2
0
1
9
.
[
1
3
]
A
.
Y
o
u
sef
p
o
u
r
e
t
a
l
.
,
“
O
p
a
c
u
s:
U
ser
-
f
r
i
e
n
d
l
y
d
i
f
f
e
r
e
n
t
i
a
l
p
r
i
v
a
c
y
i
n
P
y
T
o
r
c
h
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
r
Xi
v
:
2
1
0
9
.
1
2
2
9
8
,
2
0
2
1
.
[
1
4
]
V
.
G
u
l
s
h
a
n
e
t
a
l
.
,
“
D
e
v
e
l
o
p
m
e
n
t
a
n
d
v
a
l
i
d
a
t
i
o
n
o
f
a
d
e
e
p
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
f
o
r
d
e
t
e
c
t
i
o
n
o
f
d
i
a
b
e
t
i
c
r
e
t
i
n
o
p
a
t
h
y
i
n
r
e
t
i
n
a
l
f
u
n
d
u
s
p
h
o
t
o
g
r
a
p
h
s
,
”
J
A
MA
,
v
o
l
.
3
1
6
,
n
o
.
2
2
,
p
.
2
4
0
2
,
D
e
c
.
2
0
1
6
,
d
o
i
:
1
0
.
1
0
0
1
/
j
a
m
a
.
2
0
1
6
.
1
7
2
1
6
.
[
1
5
]
G
.
Li
t
j
e
n
s
e
t
a
l
.
,
“
A
s
u
r
v
e
y
o
n
d
e
e
p
l
e
a
r
n
i
n
g
i
n
me
d
i
c
a
l
i
m
a
g
e
a
n
a
l
y
si
s
,
”
Me
d
i
c
a
l
I
m
a
g
e
An
a
l
y
si
s
,
v
o
l
.
4
2
,
p
p
.
6
0
–
8
8
,
D
e
c
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
m
e
d
i
a
.
2
0
1
7
.
0
7
.
0
0
5
.
[
1
6
]
A
.
Es
t
e
v
a
e
t
a
l
.
,
“
A
g
u
i
d
e
t
o
d
e
e
p
l
e
a
r
n
i
n
g
i
n
h
e
a
l
t
h
c
a
r
e
,
”
N
a
t
u
r
e
M
e
d
i
c
i
n
e
,
v
o
l
.
2
5
,
n
o
.
1
,
p
p
.
2
4
–
2
9
,
J
a
n
.
2
0
1
9
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
1
-
018
-
0
3
1
6
-
z.
[
1
7
]
Q
.
Y
a
n
g
,
Y
.
Li
u
,
T.
C
h
e
n
,
a
n
d
Y
.
To
n
g
,
“
F
e
d
e
r
a
t
e
d
mac
h
i
n
e
l
e
a
r
n
i
n
g
:
C
o
n
c
e
p
t
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s,
”
A
C
M
T
ra
n
sa
c
t
i
o
n
s
o
n
I
n
t
e
l
l
i
g
e
n
t
S
y
s
t
e
m
s
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
1
0
,
n
o
.
2
,
p
p
.
1
–
1
9
,
J
a
n
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
4
5
/
3
2
9
8
9
8
1
.
[
1
8
]
M
.
J.
S
h
e
l
l
e
r
e
t
a
l
.
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
i
n
m
e
d
i
c
i
n
e
:
f
a
c
i
l
i
t
a
t
i
n
g
mu
l
t
i
-
i
n
st
i
t
u
t
i
o
n
a
l
c
o
l
l
a
b
o
r
a
t
i
o
n
s
w
i
t
h
o
u
t
s
h
a
r
i
n
g
p
a
t
i
e
n
t
d
a
t
a
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
r
t
s
,
v
o
l
.
1
0
,
n
o
.
1
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
8
-
0
2
0
-
6
9
2
5
0
-
1.
[
1
9
]
T.
Li
,
A
.
K
.
S
a
h
u
,
A
.
Ta
l
w
a
l
k
a
r
,
a
n
d
V
.
S
mi
t
h
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
:
c
h
a
l
l
e
n
g
e
s,
met
h
o
d
s
,
a
n
d
f
u
t
u
r
e
d
i
r
e
c
t
i
o
n
s,”
I
E
EE
S
i
g
n
a
l
Pro
c
e
ssi
n
g
Ma
g
a
zi
n
e
,
v
o
l
.
3
7
,
n
o
.
3
,
p
p
.
5
0
–
6
0
,
M
a
y
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
M
S
P
.
2
0
2
0
.
2
9
7
5
7
4
9
.
[
2
0
]
N
.
P
a
p
e
r
n
o
t
,
M
.
A
b
a
d
i
,
Ú
.
Er
l
i
n
g
ss
o
n
,
I
.
G
o
o
d
f
e
l
l
o
w
,
a
n
d
K
.
Ta
l
w
a
r
,
“
S
e
mi
-
s
u
p
e
r
v
i
se
d
k
n
o
w
l
e
d
g
e
t
r
a
n
sf
e
r
f
o
r
d
e
e
p
l
e
a
r
n
i
n
g
f
r
o
m p
r
i
v
a
t
e
t
r
a
i
n
i
n
g
d
a
t
a
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
e
a
r
n
i
n
g
Re
p
re
se
n
t
a
t
i
o
n
s
(
I
C
L
R)
.
,
2
0
1
7
.
[
2
1]
A
.
Z.
T
a
n
,
H
.
Y
u
,
L
.
C
u
i
,
a
n
d
Q
.
Y
a
n
g
,
“
T
o
w
a
r
d
s
p
e
r
s
o
n
a
l
i
z
e
d
f
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
,”
I
E
EE
T
r
a
n
sa
c
t
i
o
n
s
o
n
N
e
u
ra
l
N
e
t
w
o
rks
a
n
d
L
e
a
r
n
i
n
g
S
y
st
e
m
s
,
v
o
l
.
3
4
,
n
o
.
1
2
,
p
p
.
9
5
8
7
–
9
6
0
3
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
TN
N
LS.2
0
2
2
.
3
1
6
0
6
9
9
.
[2
2
]
J.
X
u
,
B
.
S
.
G
l
i
c
k
sb
e
r
g
,
C
.
S
u
,
P
.
W
a
l
k
e
r
,
J.
B
i
a
n
,
a
n
d
F
.
W
a
n
g
,
“
F
e
d
e
r
a
t
e
d
l
e
a
r
n
i
n
g
f
o
r
h
e
a
l
t
h
c
a
r
e
i
n
f
o
r
mat
i
c
s
,
”
J
o
u
rn
a
l
o
f
H
e
a
l
t
h
c
a
re
I
n
f
o
rm
a
t
i
c
s
Re
sea
r
c
h
,
v
o
l
.
5
,
n
o
.
1
,
p
p
.
1
–
1
9
,
M
a
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
1
6
6
6
-
0
2
0
-
0
0
0
8
2
-
4.
[
2
3]
M
.
K
i
t
a
d
a
,
Y
.
O
g
u
r
a
,
I
.
M
o
n
n
o
,
a
n
d
D
.
K
o
y
a
,
“
A
l
o
w
-
p
r
o
t
e
i
n
d
i
e
t
f
o
r
d
i
a
b
e
t
i
c
k
i
d
n
e
y
d
i
s
e
a
s
e
:
I
t
s
e
f
f
e
c
t
a
n
d
m
o
l
e
c
u
l
a
r
mec
h
a
n
i
sm
,
a
n
a
p
p
r
o
a
c
h
f
r
o
m
a
n
i
ma
l
st
u
d
i
e
s,
”
N
u
t
ri
e
n
t
s,
v
o
l
.
1
0
,
n
o
.
5
,
p
.
5
4
4
,
A
p
r
.
2
0
1
8
,
d
o
i
:
1
0
.
3
3
9
0
/
n
u
1
0
0
5
0
5
4
4
.
[
2
4]
X
.
F
.
L
i
e
t
a
l
.
,
“
Ef
f
i
c
a
c
y
o
f
l
o
w
-
p
r
o
t
e
i
n
d
i
e
t
i
n
d
i
a
b
e
t
i
c
n
e
p
h
r
o
p
a
t
h
y
:
a
m
e
t
a
-
a
n
a
l
y
si
s
o
f
r
a
n
d
o
mi
z
e
d
c
o
n
t
r
o
l
l
e
d
t
r
i
a
l
s,”
L
i
p
i
d
s
i
n
H
e
a
l
t
h
a
n
d
D
i
se
a
se
,
v
o
l
.
1
8
,
n
o
.
1
,
p
.
8
2
,
D
e
c
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
8
6
/
s1
2
9
4
4
-
0
1
9
-
1
0
0
7
-
6.
[
2
5]
R
.
S
h
o
k
r
i
,
M
.
S
t
r
o
n
a
t
i
,
C
.
S
o
n
g
,
a
n
d
V
.
S
h
m
a
t
i
k
o
v
,
“
M
e
mb
e
r
sh
i
p
i
n
f
e
r
e
n
c
e
a
t
t
a
c
k
s
a
g
a
i
n
st
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
s
,
”
i
n
2
0
1
7
I
EEE
s
y
m
p
o
s
i
u
m
o
n
s
e
c
u
ri
t
y
a
n
d
p
r
i
v
a
c
y
(
S
P)
,
I
EE
E
,
M
a
y
2
0
1
7
,
p
p
.
3
–
1
8
.
d
o
i
:
1
0
.
1
1
0
9
/
S
P
.
2
0
1
7
.
4
1
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Ar
a
v
in
d
K
u
m
a
r
K
y
r
ik
a
is
w
o
rk
i
n
g
a
s
a
n
As
sista
n
t
P
ro
fe
s
so
r
in
Co
m
p
u
ter
S
c
ien
c
e
a
n
d
h
a
s
m
o
re
th
a
n
1
8
y
e
a
rs
o
f
tea
c
h
in
g
e
x
p
e
rien
c
e
a
n
d
2
y
e
a
rs
o
f
I
n
d
u
str
y
Ex
p
e
rien
c
e
.
P
h
.
D
.
p
u
rsu
in
g
a
t
Am
it
y
Un
i
v
e
rsity
Ra
ip
u
r
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
ra
v
in
d
k
u
m
a
rk
y
r
ik
a
@g
m
a
il
.
c
o
m
.
S
u
m
it
K
u
m
a
r
Ba
n
c
h
h
o
r
re
c
e
iv
e
d
h
is
P
h
.
D.
fr
o
m
th
e
Na
ti
o
n
a
l
In
st
it
u
te
o
f
Tec
h
n
o
l
o
g
y
(NIT
),
Ra
i
p
u
r
,
I
n
d
ia,
in
2
0
1
8
.
He
is
c
u
rre
n
t
ly
se
rv
in
g
a
s
a
n
As
sista
n
t
P
r
o
fe
ss
o
r
i
n
Am
it
y
In
stit
u
te
o
f
In
f
o
rm
a
ti
o
n
Tec
h
n
o
l
o
g
y
a
t
Am
it
y
U
n
iv
e
rsit
y
Ch
h
a
tt
isg
a
rh
,
In
d
ia.
Wi
t
h
m
o
re
th
a
n
n
in
e
y
e
a
rs
o
f
tea
c
h
in
g
a
n
d
re
se
a
rc
h
e
x
p
e
rien
c
e
,
h
is
a
c
a
d
e
m
ic
a
n
d
re
se
a
rc
h
in
tere
sts
sp
a
n
b
io
m
e
d
ica
l
e
n
g
i
n
e
e
rin
g
a
n
d
e
lec
tro
n
ics
,
wit
h
a
f
o
c
u
s
o
n
t
h
e
I
n
tern
e
t
o
f
Th
i
n
g
s
(Io
T)
,
so
f
t
c
o
m
p
u
ti
n
g
a
n
d
m
a
c
h
in
e
lea
rn
i
n
g
a
p
p
li
c
a
ti
o
n
s,
ima
g
e
p
r
o
c
e
ss
in
g
,
a
n
d
sp
e
e
c
h
p
r
o
c
e
ss
in
g
.
He
h
a
s p
u
b
li
sh
e
d
m
o
re
th
a
n
3
0
re
se
a
rc
h
a
rti
c
les
in
re
p
u
te
d
in
ter
n
a
ti
o
n
a
l
jo
u
r
n
a
ls.
In
a
d
d
it
i
o
n
,
h
e
is
th
e
a
u
th
o
r
o
f
a
b
o
o
k
a
n
d
h
o
l
d
s
n
in
e
in
tellec
t
u
a
l
p
r
o
p
e
rt
y
ri
g
h
ts
(I
P
Rs),
in
c
l
u
d
i
n
g
p
a
ten
ts
a
n
d
c
o
p
y
ri
g
h
ts.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
sk
b
a
n
c
h
h
o
r@r
p
r.
a
m
it
y
.
e
d
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.