I
nte
rna
t
io
na
l J
o
urna
l o
f
Rec
o
nfig
ura
ble a
nd
E
m
be
dd
e
d Sy
s
t
e
m
s
(
I
J
R
E
S)
Vo
l.
15
,
No
.
2
,
J
u
ly
20
26
,
p
p
.
514
~
5
2
3
I
SS
N:
2089
-
4864
,
DOI
:
1
0
.
1
1
5
9
1
/i
j
r
es
.
v
1
5
.
i
2
.
p
p
5
1
4
-
523
514
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
r
es.ia
esco
r
e.
co
m
M
eta
-
lea
rning
co
ntras
tive f
usio
n i
ntelligence
f
o
r do
m
a
in
-
a
da
ptive co
ntent
ca
tego
riz
a
tion
usi
ng
ada
ptive
DNN
J
a
na
ni Siv
a
priy
a
Venk
a
t
a
k
r
is
hn
a
n
1,
2
,
Sa
ra
v
a
na
M
o
o
r
t
hy
Ra
j
a
M
o
o
rt
hy
2
,
Ang
el
Sh
a
nm
ug
a
m
3
1
D
e
p
a
r
t
me
n
t
o
f
D
a
t
a
S
c
i
e
n
c
e
,
D
w
a
r
a
k
a
D
o
ss G
o
v
e
r
d
h
a
n
D
o
ss V
a
i
sh
n
a
v
C
o
l
l
e
g
e
,
C
h
e
n
n
a
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
K
o
n
g
u
n
a
d
u
A
r
t
s a
n
d
S
c
i
e
n
c
e
C
o
l
l
e
g
e
,
C
o
i
mb
a
t
o
r
e
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
y
b
e
r
se
c
u
r
i
t
y
w
i
t
h
C
o
mp
u
t
e
r
S
c
i
e
n
c
e
,
P
S
G
R
K
r
i
sh
n
a
mm
a
l
C
o
l
l
e
g
e
f
o
r
W
o
me
n
,
C
o
i
mb
a
t
o
r
e
,
I
n
d
i
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
2
,
2
0
2
5
R
ev
i
s
ed
J
u
n
3
,
2
0
2
6
A
cc
ep
ted
J
u
n
1
1
,
2
0
2
6
T
h
e
ra
p
id
g
ro
w
th
o
f
h
e
tero
g
e
n
e
o
u
s
d
ig
it
a
l
tex
t
so
u
rc
e
s,
i
n
c
lu
d
i
n
g
so
c
ial
m
e
d
ia,
n
e
w
s
stre
a
m
s,
a
n
d
sc
ien
ti
fic
li
tera
tu
re
,
p
o
se
s
sig
n
if
ica
n
t
c
h
a
ll
e
n
g
e
s
to
trad
it
io
n
a
l
c
o
n
ten
t
c
a
teg
o
riza
ti
o
n
m
o
d
e
ls
d
u
e
t
o
d
o
m
a
in
sh
if
t,
v
o
c
a
b
u
lary
d
rif
t,
a
n
d
se
m
a
n
ti
c
in
c
o
n
sist
e
n
c
ies
.
Ex
isti
n
g
d
e
e
p
lea
rn
in
g
a
p
p
ro
a
c
h
e
s
o
f
ten
re
l
y
o
n
d
o
m
a
in
-
sp
e
c
if
ic
p
a
tt
e
rn
s
a
n
d
sta
ti
c
re
p
re
se
n
tatio
n
s,
re
su
lt
in
g
in
d
e
g
ra
d
e
d
p
e
rf
o
rm
a
n
c
e
w
h
e
n
a
p
p
li
e
d
t
o
u
n
se
e
n
o
r
c
r
o
ss
-
d
o
m
a
in
d
a
ta.
M
o
re
o
v
e
r,
th
e
se
m
e
th
o
d
s
lac
k
sc
a
l
a
b
il
it
y
in
re
a
l
-
w
o
rld
sc
e
n
a
rio
s
c
h
a
ra
c
teriz
e
d
b
y
d
o
m
a
in
d
rif
t
a
n
d
li
m
it
e
d
lab
e
led
d
a
ta.
T
o
a
d
d
re
ss
th
e
se
c
h
a
ll
e
n
g
e
s,
th
is
stu
d
y
p
ro
p
o
se
s
a
m
e
ta
-
lea
rn
in
g
c
o
n
tras
ti
v
e
f
u
sio
n
in
telli
g
e
n
c
e
(M
CF
I)
f
ra
m
e
w
o
rk
f
o
r
d
o
m
a
in
-
a
d
a
p
ti
v
e
c
o
n
ten
t
c
a
teg
o
riza
ti
o
n
.
T
h
e
f
ra
m
e
w
o
rk
in
teg
ra
tes
d
o
m
a
in
c
o
n
tex
t
-
a
w
a
re
n
o
rm
a
li
z
a
ti
o
n
(D
CA
N)
f
o
r
ro
b
u
st
p
re
p
r
o
c
e
ss
in
g
,
b
in
a
ry
p
a
rti
c
le
s
w
a
r
m
o
p
ti
m
iza
ti
o
n
(BP
S
O)
f
o
r
se
lec
ti
n
g
d
o
m
a
in
-
in
v
a
rian
t
f
e
a
tu
re
s,
a
n
d
a
h
y
b
rid
a
rc
h
it
e
c
t
u
re
c
o
m
b
in
in
g
c
o
n
tras
ti
v
e
lea
rn
in
g
-
e
n
h
a
n
c
e
d
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
tati
o
n
s
f
ro
m
tran
sf
o
r
m
e
rs
(C
L
-
BERT
)
w
it
h
c
a
p
su
le n
e
tw
o
rk
-
e
n
h
a
n
c
e
d
tra
n
sf
o
rm
e
r
(Ca
p
s
-
T
ra
n
s).
A
m
e
ta
-
le
a
rn
in
g
stra
teg
y
is
e
m
p
lo
y
e
d
to
lea
rn
d
o
m
a
in
-
in
v
a
rian
t
re
p
re
se
n
tatio
n
s
a
n
d
e
n
a
b
le
ra
p
id
a
d
a
p
tatio
n
to
n
e
w
d
o
m
a
in
s,
w
h
il
e
c
o
n
tr
a
stiv
e
lea
rn
in
g
e
n
h
a
n
c
e
s
in
ter
-
d
o
m
a
in
se
p
a
ra
b
il
it
y
.
A
d
o
m
a
in
-
a
d
a
p
ti
v
e
d
e
c
isio
n
la
y
e
r
f
u
rth
e
r
re
f
in
e
s
f
e
a
tu
re
c
o
n
tr
ib
u
ti
o
n
s
d
y
n
a
m
i
c
a
ll
y
.
Ex
p
e
ri
m
e
n
tal
re
su
lt
s
o
n
m
u
lt
ip
le
b
e
n
c
h
m
a
rk
d
a
tas
e
t
s
d
e
m
o
n
stra
te
th
a
t
th
e
p
ro
p
o
s
e
d
M
CF
I
f
ra
m
e
w
o
rk
c
o
n
siste
n
tl
y
o
u
t
p
e
rf
o
rm
s
sta
te
-
of
-
th
e
-
a
rt
m
e
th
o
d
s
i
n
term
s
o
f
a
c
c
u
ra
c
y
,
F
1
-
sc
o
re
,
a
n
d
g
e
n
e
ra
li
z
a
ti
o
n
,
p
ro
v
i
d
in
g
a
sc
a
lab
le
a
n
d
e
ff
e
c
ti
v
e
so
lu
ti
o
n
f
o
r
c
ro
ss
-
d
o
m
a
in
tex
t
c
las
sif
ic
a
ti
o
n
.
K
ey
w
o
r
d
s
:
C
ateg
o
r
izatio
n
C
o
n
tr
ast
iv
e
lear
n
in
g
Do
m
ain
-
ad
ap
ti
v
e
co
n
te
n
t
Featu
r
e
en
g
i
n
ee
r
in
g
Me
ta
-
lear
n
in
g
S
w
ar
m
i
n
telli
g
e
n
ce
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
:
J
an
an
i Si
v
ap
r
i
y
a
Ve
n
k
a
tak
r
i
s
h
n
an
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
,
Ko
n
g
u
n
ad
u
A
r
t
s
an
d
Scie
n
ce
C
o
lle
g
e
C
o
i
m
b
ato
r
e,
I
n
d
ia
E
m
ail:
j
an
an
i.si
v
ap
r
i
y
ap
h
d
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
So
cial
m
ed
ia
h
as
tr
an
s
f
o
r
m
ed
m
o
d
er
n
co
m
m
u
n
icatio
n
b
y
e
n
a
b
lin
g
r
ea
l
-
ti
m
e,
d
y
n
a
m
ic,
an
d
l
ar
g
e
-
s
ca
le
s
h
ar
i
n
g
o
f
o
p
in
io
n
s
,
e
x
p
er
ien
c
es,
an
d
i
n
f
o
r
m
atio
n
ac
r
o
s
s
d
i
v
er
s
e
co
m
m
u
n
itie
s
.
W
it
h
t
h
e
r
ap
id
ad
v
an
ce
m
e
n
t
o
f
h
ig
h
-
s
p
ee
d
i
n
ter
n
et
a
n
d
co
m
m
u
n
ica
tio
n
tec
h
n
o
lo
g
ies
,
m
il
lio
n
s
o
f
u
s
er
s
ac
ti
v
el
y
p
ar
ticip
ate
in
s
o
cial
n
et
w
o
r
k
in
g
p
latf
o
r
m
s
,
g
e
n
er
ati
n
g
m
as
s
iv
e
v
o
lu
m
e
s
o
f
h
eter
o
g
e
n
eo
u
s
tex
tu
a
l
d
ata
[
1
]
,
[
2
]
.
T
h
ese
p
latf
o
r
m
s
co
n
t
in
u
o
u
s
l
y
p
r
o
d
u
ce
o
p
in
io
n
-
r
ich
co
n
ten
t,
m
ak
in
g
a
u
to
m
ated
te
x
t
clas
s
i
f
icatio
n
es
s
en
t
ial
f
o
r
ap
p
licatio
n
s
s
u
c
h
as
co
n
ten
t
m
o
d
er
atio
n
,
in
f
o
r
m
atio
n
r
etr
iev
al,
an
d
s
e
n
ti
m
en
t
an
al
y
s
is
[
3
]
.
Ho
w
e
v
er
,
th
e
u
n
s
tr
u
c
tu
r
ed
,
n
o
is
y
,
a
n
d
d
o
m
ain
-
v
ar
ian
t
n
at
u
r
e
o
f
s
u
ch
d
ata
i
n
tr
o
d
u
ce
s
s
i
g
n
i
f
ica
n
t c
h
alle
n
g
e
s
f
o
r
tr
ad
itio
n
al
tex
t c
la
s
s
i
f
icat
i
o
n
m
o
d
els.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
Meta
-
lea
r
n
in
g
co
n
tr
a
s
tive
fu
s
io
n
in
tellig
en
ce
fo
r
d
o
ma
in
-
a
d
a
p
tive
…
(
Ja
n
a
n
i
S
iva
p
r
iya
V
e
n
ka
ta
kris
h
n
a
n
)
515
Do
m
ain
ca
teg
o
r
izatio
n
,
w
h
ic
h
in
v
o
l
v
es
cla
s
s
i
f
y
i
n
g
te
x
t
b
ase
d
o
n
s
u
b
j
ec
t
o
r
co
n
tex
t,
r
eq
u
ir
es
m
o
d
els
ca
p
ab
le
o
f
g
en
er
alizi
n
g
ac
r
o
s
s
d
iv
er
s
e
lin
g
u
is
tic
p
atter
n
s
an
d
d
o
m
ai
n
s
[
4
]
,
[
5
]
.
I
n
r
ea
l
-
w
o
r
l
d
s
ce
n
ar
io
s
,
tex
t
u
al
d
ata
s
u
ch
as
n
e
w
s
ar
ticle
s
,
s
o
cial
m
ed
ia
p
o
s
ts
,
an
d
s
cien
ti
f
i
c
d
is
cu
s
s
io
n
s
e
x
h
ib
it
d
o
m
a
in
-
s
p
ec
if
ic
v
o
ca
b
u
lar
y
v
ar
iatio
n
s
an
d
s
e
m
a
n
tic
i
n
co
n
s
is
te
n
cies,
co
m
p
l
icatin
g
cla
s
s
i
f
icatio
n
tas
k
s
[
6
]
,
[
7
]
.
A
lth
o
u
g
h
d
ee
p
l
ea
r
n
in
g
m
o
d
els,
p
a
r
t
icu
l
ar
ly
tr
an
s
f
o
r
m
er
-
b
ase
d
a
r
ch
it
ec
tu
r
es
s
u
ch
as
b
i
d
i
r
e
cti
o
n
al
en
co
d
er
r
ep
r
esen
ta
ti
o
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
[
8
]
,
h
av
e
a
ch
iev
e
d
r
em
ar
k
ab
l
e
s
u
c
ce
s
s
,
th
ey
o
f
t
en
s
u
f
f
er
f
r
o
m
d
o
m
ain
s
h
if
t,
w
h
er
e
t
r
ain
in
g
an
d
test
in
g
d
at
a
d
is
t
r
i
b
u
ti
o
n
s
d
if
f
er
.
Do
m
ain
a
d
a
p
ta
ti
o
n
te
ch
n
iq
u
es
h
av
e
b
e
en
p
r
o
p
o
s
e
d
t
o
a
d
d
r
ess
th
is
is
s
u
e
[
9
]
,
[
1
0
]
;
h
o
w
ev
er
,
th
ei
r
ef
f
e
ctiv
en
ess
is
lim
ited
in
l
a
r
g
e
-
s
ca
l
e
h
ete
r
o
g
en
e
o
u
s
en
v
ir
o
n
m
en
ts
w
ith
s
ca
r
ce
la
b
e
led
d
at
a
[
1
1
]
.
Fu
r
th
e
r
m
o
r
e
,
r
e
ce
n
t
s
tu
d
i
es
h
ig
h
lig
h
t
th
at
m
u
l
ti
-
d
o
m
ain
ad
a
p
tiv
e
m
o
d
els s
tr
u
g
g
le
t
o
m
ain
tain
c
o
n
s
is
ten
t
p
e
r
f
o
r
m
an
ce
u
n
d
e
r
d
o
m
ain
s
k
e
w
an
d
s
em
an
tic
d
r
if
t
[
1
2
]
.
A
d
v
an
ce
d
d
o
m
a
in
ad
ap
tatio
n
ap
p
r
o
ac
h
es,
s
u
ch
as
ca
teg
o
r
y
atten
tio
n
n
et
w
o
r
k
s
[
1
3
]
,
aim
t
o
r
ed
u
ce
in
ter
-
d
o
m
ain
d
is
cr
ep
an
c
ies
b
u
t
o
f
ten
o
v
er
lo
o
k
f
ea
tu
r
e
r
ed
u
n
d
an
c
y
,
s
ca
lab
ilit
y
,
an
d
ad
ap
tiv
e
f
ea
tu
r
e
w
ei
g
h
t
in
g
.
Si
m
i
lar
l
y
,
r
ec
en
t
m
u
l
ti
m
o
d
al
an
d
co
n
ten
t
class
i
f
icat
io
n
f
r
a
m
e
w
o
r
k
s
d
e
m
o
n
s
tr
ate
i
m
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
th
r
o
u
g
h
d
ee
p
ar
ch
itect
u
r
es
[
1
4
]
,
[
1
5
]
,
y
et
r
e
m
a
in
h
ea
v
i
l
y
d
ep
en
d
en
t
o
n
d
o
m
ai
n
-
s
p
ec
if
ic
l
ab
eled
d
atasets
an
d
ex
h
ib
it l
i
m
ited
g
e
n
er
aliza
tio
n
ca
p
ab
ilit
ies
[
1
6
]
.
R
ec
en
t
d
ev
elo
p
m
e
n
ts
in
co
n
tr
asti
v
e
lear
n
i
n
g
an
d
d
o
m
a
i
n
g
en
er
aliza
t
io
n
,
i
n
clu
d
i
n
g
m
o
m
e
n
t
u
m
co
n
tr
ast
[
1
7
]
,
p
r
o
x
y
-
b
ased
co
n
tr
asti
v
e
lear
n
i
n
g
[
1
8
]
,
an
d
m
e
m
o
r
y
-
b
ased
s
u
p
er
v
i
s
ed
co
n
tr
asti
v
e
lear
n
in
g
[
1
9
]
,
h
av
e
s
h
o
w
n
p
r
o
m
is
e
i
n
i
m
p
r
o
v
in
g
r
ep
r
esen
tatio
n
r
o
b
u
s
tn
e
s
s
.
A
d
d
itio
n
all
y
,
m
eta
-
lear
n
in
g
ap
p
r
o
ac
h
es
en
ab
le
r
ap
id
ad
ap
tatio
n
to
u
n
s
ee
n
d
o
m
ai
n
s
[
2
0
]
,
w
h
ile
f
e
w
-
s
h
o
t
lea
r
n
in
g
m
et
h
o
d
s
en
h
an
ce
c
lass
if
i
ca
tio
n
u
n
d
er
li
m
ited
d
ata
co
n
d
itio
n
s
[
2
1
]
–
[
2
3
]
.
T
ec
h
n
iq
u
es
s
u
c
h
as
g
r
ad
ien
t
s
u
r
g
er
y
[
2
4
]
an
d
cr
o
s
s
-
d
o
m
ai
n
f
ea
tu
r
e
alig
n
m
en
t
[
2
5
]
f
u
r
t
h
er
co
n
tr
ib
u
te
to
i
m
p
r
o
v
i
n
g
g
e
n
er
aliza
tio
n
ac
r
o
s
s
d
o
m
ai
n
s
.
Ho
w
e
v
er
,
th
e
s
e
ap
p
r
o
ac
h
e
s
ar
e
o
f
ten
ap
p
lied
in
d
ep
en
d
en
tl
y
a
n
d
lack
i
n
teg
r
atio
n
w
it
h
f
ea
t
u
r
e
o
p
ti
m
izat
io
n
s
tr
ate
g
ies.
C
o
n
s
eq
u
en
tl
y
,
s
e
v
er
al
r
esear
ch
g
ap
s
r
e
m
a
in
.
Fir
s
t,
m
o
s
t
ex
is
ti
n
g
m
et
h
o
d
s
r
el
y
s
o
lel
y
o
n
d
ee
p
r
ep
r
esen
tatio
n
lear
n
in
g
w
i
th
o
u
t
i
n
co
r
p
o
r
atin
g
s
y
s
te
m
a
tic
f
ea
tu
r
e
en
g
i
n
ee
r
in
g
o
r
o
p
ti
m
iz
atio
n
-
b
ased
f
ea
t
u
r
e
s
elec
tio
n
.
Seco
n
d
,
li
m
ited
att
en
tio
n
h
a
s
b
ee
n
g
i
v
en
to
in
te
g
r
atin
g
m
eta
-
lear
n
in
g
w
i
th
d
o
m
ai
n
ad
ap
tatio
n
f
o
r
r
ap
id
cr
o
s
s
-
d
o
m
ai
n
g
e
n
er
aliza
t
io
n
.
T
h
ir
d
,
c
o
n
tr
asti
v
e
lear
n
in
g
tech
n
iq
u
es
ar
e
n
o
t
f
u
ll
y
lev
er
a
g
ed
i
n
co
n
j
u
n
ct
io
n
w
it
h
ad
ap
tiv
e
f
u
s
io
n
a
n
d
f
ea
t
u
r
e
r
ew
e
ig
h
ti
n
g
m
ec
h
an
is
m
s
.
F
in
all
y
,
s
ca
lab
ilit
y
u
n
d
er
ex
tr
e
m
e
d
o
m
ai
n
d
r
if
t a
n
d
v
o
ca
b
u
lar
y
v
ar
iatio
n
r
e
m
ain
s
a
cr
itical
ch
alle
n
g
e
i
n
r
ea
l
-
w
o
r
ld
ap
p
licatio
n
s
[
2
6
]
.
T
o
ad
d
r
ess
th
ese
li
m
itat
io
n
s
,
th
is
s
tu
d
y
p
r
o
p
o
s
es
a
m
eta
-
lear
n
in
g
co
n
tr
asti
v
e
f
u
s
io
n
in
telli
g
e
n
ce
(
MCF
I
)
f
r
a
m
e
w
o
r
k
th
at
in
t
eg
r
a
t
e
s
d
o
m
ain
-
aw
a
r
e
p
r
e
p
r
o
c
e
s
s
in
g
,
s
w
a
r
m
in
t
el
l
ig
en
c
e
-
b
as
ed
f
e
a
tu
r
e
s
e
le
c
t
i
o
n
,
c
o
n
t
r
as
t
iv
e
r
e
p
r
es
en
t
at
i
o
n
l
e
a
r
n
i
n
g
,
an
d
a
d
a
p
t
iv
e
d
e
e
p
n
eu
r
a
l
m
o
d
e
l
in
g
.
T
h
e
p
r
o
p
o
s
e
d
a
p
p
r
o
a
c
h
en
h
an
c
es
c
r
o
s
s
-
d
o
m
a
in
g
e
n
e
r
a
l
i
za
t
i
o
n
,
r
e
d
u
ce
s
f
e
atu
r
e
r
e
d
u
n
d
an
cy
,
a
n
d
d
y
n
am
i
ca
l
ly
a
d
a
p
ts
t
o
e
v
o
lv
in
g
t
e
x
t
u
a
l
d
i
s
t
r
i
b
u
t
i
o
n
s
t
h
r
o
u
g
h
o
p
t
im
iz
e
d
f
e
at
u
r
e
s
e
l
ec
t
i
o
n
an
d
c
o
n
t
r
as
ti
v
e
m
u
l
tim
o
d
a
l
f
u
s
i
o
n
.
B
y
l
ev
e
r
ag
in
g
m
et
a
-
le
a
r
n
e
d
t
r
an
s
f
o
r
m
e
r
-
b
a
s
e
d
e
n
c
o
d
e
r
s
a
n
d
d
o
m
a
in
-
in
v
a
r
i
an
t
r
e
p
r
es
en
t
a
ti
o
n
s
,
th
e
M
C
F
I
f
r
am
e
w
o
r
k
p
r
o
v
i
d
es
a
s
c
a
l
a
b
l
e
a
n
d
r
o
b
u
s
t
s
o
lu
t
i
o
n
f
o
r
c
o
n
t
en
t
c
at
eg
o
r
i
z
a
ti
o
n
in
h
e
te
r
o
g
en
e
o
u
s
b
ig
-
d
ata
en
v
ir
o
n
m
e
n
t
s
w
it
h
li
m
ited
lab
eled
d
ata
.
2.
RE
L
AT
E
D
WO
RK
T
h
e
r
ec
en
t
g
r
o
w
t
h
o
f
s
o
cial
m
ed
ia
an
d
cr
o
s
s
-
p
latf
o
r
m
c
o
m
m
u
n
icatio
n
h
as
s
ti
m
u
lated
ex
ten
s
i
v
e
r
esear
ch
in
r
ea
l
-
ti
m
e
te
x
t
clas
s
if
icatio
n
,
d
o
m
ain
ad
ap
tatio
n
,
m
u
lti
m
o
d
al
f
u
s
io
n
,
an
d
m
is
i
n
f
o
r
m
at
io
n
d
etec
tio
n
.
E
ar
ly
s
tu
d
ie
s
p
r
i
m
ar
il
y
f
o
cu
s
ed
o
n
o
r
g
an
izi
n
g
a
n
d
cla
s
s
if
y
i
n
g
u
s
er
-
g
en
er
ated
co
n
te
n
t
u
n
d
er
d
y
n
a
m
ic
co
n
d
itio
n
s
.
I
j
az
et
a
l.
[
1
]
c
o
n
d
u
ct
ed
a
co
m
p
r
eh
en
s
i
v
e
s
y
s
te
m
at
ic
r
ev
ie
w
o
n
r
ea
l
-
ti
m
e
te
x
t
clas
s
if
ica
tio
n
o
f
s
o
cia
l
m
ed
ia
s
tr
ea
m
s
,
h
ig
h
li
g
h
ti
n
g
k
e
y
c
h
alle
n
g
e
s
s
u
c
h
as
s
ca
lab
ilit
y
,
s
tr
ea
m
i
n
g
laten
c
y
,
a
n
d
d
o
m
ain
v
ar
iab
ilit
y
.
T
h
eir
f
i
n
d
in
g
s
in
d
icate
t
h
at
w
h
ile
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
ac
h
iev
e
h
ig
h
ac
cu
r
ac
y
,
th
e
y
o
f
ten
s
tr
u
g
g
le
to
g
en
er
aliz
e
ac
r
o
s
s
d
o
m
ai
n
s
an
d
ev
o
l
v
in
g
t
ex
tu
a
l d
is
tr
ib
u
tio
n
s
.
I
n
th
e
co
n
te
x
t
o
f
cr
o
s
s
-
p
lat
f
o
r
m
d
y
n
a
m
ics,
Xi
et
a
l.
[
2
]
p
r
o
p
o
s
ed
an
in
f
o
r
m
a
tio
n
d
if
f
u
s
io
n
m
o
d
el
to
an
al
y
ze
th
e
s
p
r
ea
d
o
f
to
p
ics
ac
r
o
s
s
s
o
cial
m
ed
ia
p
lat
f
o
r
m
s
.
T
h
eir
w
o
r
k
d
e
m
o
n
s
tr
ated
t
h
at
te
x
tu
a
l
co
n
te
n
t
p
r
o
p
ag
ates
ac
r
o
s
s
h
eter
o
g
e
n
e
o
u
s
d
o
m
ai
n
s
,
i
n
te
n
s
i
f
y
i
n
g
d
o
m
ai
n
-
s
h
if
t
ch
al
len
g
es
in
cla
s
s
if
icatio
n
tas
k
s
an
d
e
m
p
h
a
s
izi
n
g
t
h
e
n
ee
d
f
o
r
ad
ap
tiv
e
lear
n
in
g
m
o
d
els.
Fo
u
n
d
atio
n
al
w
o
r
k
o
n
tr
an
s
f
er
ab
le
r
ep
r
esen
tatio
n
lear
n
i
n
g
was
i
n
tr
o
d
u
ce
d
b
y
R
o
g
er
s
et
a
l.
[
3
]
,
w
h
o
p
r
o
p
o
s
ed
a
d
ee
p
lea
r
n
in
g
f
r
a
m
e
w
o
r
k
f
o
r
s
ca
lab
le
d
o
m
ain
ad
ap
tatio
n
th
r
o
u
g
h
tr
a
n
s
f
er
ab
le
f
ea
t
u
r
e
lear
n
i
n
g
.
Si
m
i
lar
l
y
,
Yi
n
et
a
l.
[
4
]
d
ev
elo
p
ed
a
w
ea
k
l
y
s
u
p
er
v
i
s
ed
d
o
m
a
in
ad
ap
tatio
n
ap
p
r
o
ac
h
f
o
r
asp
ec
t
ex
tr
ac
tio
n
u
s
i
n
g
m
u
ltil
e
v
el
in
ter
ac
tio
n
tr
an
s
f
er
.
A
lt
h
o
u
g
h
ef
f
ec
ti
v
e,
th
eir
m
et
h
o
d
w
as
li
m
ited
to
asp
ec
t
-
lev
el
task
s
r
ath
er
th
a
n
g
en
er
al
co
n
ten
t c
ate
g
o
r
izatio
n
.
Z
h
an
g
et
a
l.
[
5
]
in
tr
o
d
u
ce
d
a
G
A
N
-
b
ased
cr
o
s
s
-
d
o
m
ai
n
s
en
t
i
m
en
t
tr
an
s
f
er
m
o
d
el
ca
p
ab
le
o
f
g
en
er
ati
n
g
tar
g
et
-
d
o
m
ain
co
n
te
n
t
u
s
i
n
g
s
o
u
r
ce
-
d
o
m
a
in
an
n
o
tatio
n
s
.
Desp
ite
its
ef
f
ec
ti
v
e
n
ess
,
t
h
e
m
o
d
el
s
u
f
f
er
s
f
r
o
m
in
s
tab
il
it
y
a
n
d
h
i
g
h
co
m
p
u
ta
tio
n
al
co
m
p
l
e
x
it
y
.
L
ia
n
g
et
a
l.
[
6
]
later
p
r
o
p
o
s
ed
a
h
y
b
r
id
B
iGR
U
-
b
ased
f
r
a
m
e
w
o
r
k
f
o
r
cr
o
s
s
-
d
o
m
ain
s
en
ti
m
en
t
cla
s
s
i
f
icatio
n
,
co
m
b
i
n
in
g
g
lo
b
al
s
e
m
a
n
tic
an
d
lo
ca
l
co
n
tex
t
u
al
f
ea
t
u
r
es.
Ho
w
e
v
er
,
th
e
ap
p
r
o
ac
h
lack
s
i
n
teg
r
at
io
n
o
f
co
n
tr
a
s
ti
v
e
an
d
m
eta
-
lear
n
i
n
g
s
tr
ate
g
ies.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
2
,
J
u
l
y
202
6
:
51
4
-
5
2
3
516
W
ith
t
h
e
r
is
e
o
f
m
is
i
n
f
o
r
m
at
io
n
d
etec
tio
n
,
m
u
lti
m
o
d
al
d
o
m
ai
n
-
ad
ap
ti
v
e
f
r
a
m
e
w
o
r
k
s
h
a
v
e
g
ain
ed
atten
tio
n
.
L
o
n
g
et
a
l.
[
7
]
p
r
o
p
o
s
ed
a
m
e
m
o
r
y
-
g
u
id
ed
m
u
lti
-
d
o
m
ai
n
f
a
k
e
n
e
w
s
d
et
ec
tio
n
m
o
d
el,
w
h
il
e
Kash
y
ap
et
a
l.
[
8
]
i
n
tr
o
d
u
ce
d
a
r
o
b
u
s
t
cr
o
s
s
-
m
o
d
al
f
ea
t
u
r
e
al
ig
n
m
e
n
t
f
r
a
m
e
w
o
r
k
f
o
r
m
is
i
n
f
o
r
m
at
io
n
d
etec
tio
n
.
Z
h
u
et
a
l.
[
9
]
f
u
r
th
er
p
r
o
p
o
s
ed
a
d
o
m
a
in
-
ad
ap
tiv
e
r
ep
r
esen
tatio
n
lear
n
i
n
g
m
o
d
el
f
o
r
m
u
lti
m
o
d
al
d
ee
p
f
ak
e
d
etec
tio
n
.
A
lt
h
o
u
g
h
th
e
s
e
ap
p
r
o
ac
h
es
i
m
p
r
o
v
e
cr
o
s
s
-
m
o
d
al
al
ig
n
m
e
n
t,
th
e
y
o
f
te
n
o
v
er
lo
o
k
f
ea
tu
r
e
o
p
ti
m
izatio
n
an
d
ad
ap
tiv
e
f
ea
t
u
r
e
s
elec
tio
n
m
ec
h
a
n
i
s
m
s
.
R
ec
en
t
ad
v
a
n
ce
m
e
n
t
s
i
n
tr
a
n
s
f
o
r
m
er
-
b
ased
ar
ch
itect
u
r
es
e
m
p
h
a
s
ize
e
f
f
icien
c
y
a
n
d
s
ca
lab
ilit
y
.
L
i
u
et
a
l.
[
1
0
]
p
r
o
p
o
s
ed
C
L
A
R
I
T
Y,
a
lig
h
t
w
ei
g
h
t
m
u
lti
m
o
d
al
tr
an
s
f
o
r
m
er
f
o
r
h
ar
m
f
u
l
co
n
te
n
t
d
etec
tio
n
,
w
h
il
e
L
ee
et
a
l.
[
1
1
]
in
tr
o
d
u
ce
d
a
n
atten
tio
n
-
b
ased
f
r
a
m
e
w
o
r
k
f
o
r
m
i
s
lead
in
g
co
n
te
n
t
d
etec
tio
n
.
Desp
ite
th
eir
r
o
b
u
s
tn
es
s
,
th
e
s
e
m
o
d
els e
x
h
i
b
it li
m
ited
ad
ap
tab
ilit
y
u
n
d
er
s
ev
er
e
d
o
m
a
in
d
r
if
t.
E
ar
lier
w
o
r
k
b
y
L
iu
[
1
2
]
ex
p
lo
r
ed
cr
o
s
s
-
d
o
m
a
in
m
ed
ia
s
tr
ea
m
lear
n
i
n
g
u
s
in
g
n
o
r
m
aliza
tio
n
an
d
g
r
ap
h
-
b
ased
r
an
k
i
n
g
tech
n
iq
u
e
s
.
Ho
w
e
v
er
,
th
eir
ap
p
r
o
ac
h
la
ck
ed
d
ee
p
s
e
m
a
n
tic
ali
g
n
m
en
t.
Mo
r
e
r
ec
en
tl
y
,
Sh
a
n
to
et
a
l.
[
1
3
]
p
r
o
p
o
s
ed
a
h
ier
ar
ch
ical
m
u
l
ti
m
o
d
al
class
i
f
icatio
n
f
r
a
m
e
w
o
r
k
f
o
r
B
en
g
ali
co
n
te
n
t,
d
em
o
n
s
tr
ati
n
g
i
m
p
r
o
v
ed
s
tr
u
ct
u
r
ed
class
i
f
icatio
n
,
th
o
u
g
h
w
it
h
li
m
ited
cr
o
s
s
-
d
o
m
ain
g
e
n
er
aliza
tio
n
.
A
lto
g
et
h
er
,
th
e
cu
r
r
en
t
liter
at
u
r
e
s
h
o
w
s
th
at
d
o
m
ai
n
ad
ap
tatio
n
,
m
u
lti
m
o
d
al
f
u
s
io
n
,
s
e
n
ti
m
e
n
t
tr
an
s
f
er
,
an
d
m
is
in
f
o
r
m
atio
n
d
etec
tio
n
h
av
e
b
ee
n
ad
v
a
n
ce
d
co
n
s
id
er
ab
ly
.
Ho
w
ev
er
,
th
er
e
ar
e
s
til
l
a
f
e
w
li
m
ita
tio
n
s
:
i)
lack
o
f
s
u
f
f
icie
n
t
f
ea
tu
r
e
s
e
l
ec
tio
n
b
ased
o
n
o
p
ti
m
izat
io
n
w
it
h
d
ee
p
d
o
m
ain
ad
ap
tatio
n
,
ii)
less
u
s
e
o
f
m
eta
-
lear
n
in
g
to
q
u
ick
l
y
g
en
er
alize
to
u
n
s
ee
n
d
o
m
ain
s
,
iii)
lack
o
f
co
n
tr
asti
v
e
ali
g
n
m
e
n
t
b
et
w
ee
n
h
eter
o
g
e
n
eo
u
s
d
o
m
ai
n
s
,
a
n
d
iv
)
th
e
is
s
u
e
o
f
s
ca
lab
ilit
y
w
h
en
ap
p
lied
to
lar
g
e
-
s
ca
le,
ch
a
n
g
i
n
g
s
t
r
ea
m
s
o
f
s
o
cial
m
ed
ia.
T
h
ese
d
ef
icien
c
ies
d
e
m
o
n
s
tr
ate
th
a
t
a
s
in
g
le
f
r
a
m
e
w
o
r
k
co
m
b
in
in
g
f
ea
t
u
r
e
en
g
in
ee
r
i
n
g
,
s
w
ar
m
-
i
n
telli
g
e
n
ce
o
p
tim
izatio
n
,
co
n
tr
ast
iv
e
lear
n
in
g
,
an
d
ad
ap
tiv
e
tr
an
s
f
o
r
m
er
a
r
ch
itect
u
r
es
is
n
ec
es
s
ar
y
to
o
f
f
er
r
o
b
u
s
t
an
d
s
ca
le
-
ab
le
d
o
m
ai
n
-
ad
ap
ti
v
e
co
n
te
n
t
ca
teg
o
r
izatio
n
.
3.
M
E
T
H
O
D
Fig
u
r
e
1
d
e
m
o
n
s
tr
ates
th
e
g
e
n
er
al
s
tr
u
ct
u
r
e
o
f
th
e
g
i
v
e
n
M
C
FI
f
r
a
m
e
w
o
r
k
o
f
d
o
m
ai
n
-
ad
ap
t
iv
e
co
n
ten
t
ca
teg
o
r
izatio
n
.
I
t
h
a
s
th
r
ee
s
tep
s
:
p
r
ep
r
o
ce
s
s
in
g
d
o
m
ai
n
co
n
tex
t
-
a
w
ar
e
n
o
r
m
aliza
tio
n
(
DC
A
N)
,
f
ea
t
u
r
e
s
elec
tio
n
b
ased
o
n
b
i
n
ar
y
p
ar
ticle
s
w
ar
m
o
p
ti
m
izat
io
n
(
B
P
SO)
,
an
d
co
n
te
n
t
ca
te
g
o
r
izatio
n
u
s
in
g
t
h
e
h
y
b
r
id
MCF
I
m
o
d
el
co
n
tr
a
s
ti
v
e
lear
n
in
g
-
e
n
h
an
ce
d
(
C
L
-
B
E
R
T
+Cap
s
-
T
r
an
s
)
.
C
o
n
te
x
t
u
al
to
k
e
n
r
ef
i
n
e
m
e
n
t,
ad
ap
tiv
e
s
to
p
-
w
o
r
d
r
e
m
o
v
al,
m
o
r
p
h
o
lo
g
ical
n
o
r
m
aliza
t
io
n
,
n
o
is
e
f
ilte
r
in
g
a
n
d
s
e
m
a
n
tic
a
u
g
m
e
n
tati
o
n
ar
e
p
er
f
o
r
m
ed
i
n
th
e
f
ir
s
t
s
tag
e,
o
n
th
e
r
a
w
m
u
lti
-
d
o
m
a
in
tex
t.
T
h
e
DC
A
N
m
ec
h
a
n
i
s
m
m
ai
n
tai
n
s
d
o
m
ai
n
in
v
ar
ia
n
t
lin
g
u
is
tic
clu
e
s
a
s
it
n
o
r
m
ali
s
es c
o
n
te
x
tu
al
v
ar
iatio
n
in
h
eter
o
g
e
n
eo
u
s
d
o
m
ai
n
s
p
r
o
d
u
ci
n
g
c
lean
,
s
e
m
an
ticall
y
co
n
s
is
ten
t
tex
t
u
al
r
ep
r
esen
tatio
n
s
.
Fig
u
r
e
1
.
W
o
r
k
f
lo
w
d
ia
g
r
a
m
f
o
r
d
o
m
ain
ad
ap
tiv
e
co
n
ten
t c
a
teg
o
r
izatio
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
Meta
-
lea
r
n
in
g
co
n
tr
a
s
tive
fu
s
io
n
in
tellig
en
ce
fo
r
d
o
ma
in
-
a
d
a
p
tive
…
(
Ja
n
a
n
i
S
iva
p
r
iya
V
e
n
ka
ta
kris
h
n
a
n
)
517
T
h
e
s
ec
o
n
d
s
tep
i
s
k
n
o
w
n
a
s
co
m
p
r
eh
e
n
s
iv
e
f
ea
t
u
r
e
e
n
g
in
ee
r
in
g
th
at
is
u
s
ed
to
o
b
tain
le
x
ical
,
s
y
n
tactic,
s
e
m
a
n
tic
an
d
co
n
te
x
tu
a
l
r
ep
r
esen
tatio
n
s
in
o
r
d
er
to
cr
ea
te
a
h
ig
h
-
d
i
m
en
s
io
n
al
f
ea
tu
r
e
p
o
o
l.
A
n
ef
f
icien
t
s
u
b
s
et
o
f
d
o
m
ain
-
r
o
b
u
s
t
f
ea
tu
r
e
s
is
t
h
e
n
f
o
u
n
d
u
s
in
g
a
s
w
ar
m
-
in
te
lli
g
en
ce
-
b
ase
d
B
P
SO
alg
o
r
ith
m
.
T
h
e
f
itn
es
s
cr
iter
io
n
m
a
x
i
m
iz
es
th
e
cr
o
s
s
-
d
o
m
ai
n
d
is
cr
i
m
in
ab
ilit
y
an
d
m
i
n
i
m
izes
t
h
e
r
ed
u
n
d
a
n
c
y
o
f
f
ea
t
u
r
es
,
th
er
ef
o
r
e,
m
a
x
i
m
izi
n
g
t
h
e
g
en
er
aliza
tio
n
ab
ilit
y
a
n
d
d
ec
r
ea
s
in
g
co
m
p
u
tatio
n
al
co
s
ts
.
T
h
e
s
elec
ted
f
ea
tu
r
es
ar
e
th
en
in
p
u
tted
in
t
h
e
MCF
I
-
b
ased
co
n
te
n
t
ca
te
g
o
r
izatio
n
m
o
d
u
le
in
th
e
las
t
p
h
ase.
An
en
co
d
er
b
ased
o
n
m
eta
-
lear
n
i
n
g
tr
ai
n
ed
C
L
-
B
E
R
T
ac
q
u
ir
es
d
o
m
ain
-
i
n
v
ar
ian
t
r
ep
r
esen
tatio
n
s
b
y
ep
is
o
d
ic
tr
ain
i
n
g
an
d
co
n
tr
asti
v
e
lear
n
in
g
,
w
h
er
e
s
e
m
a
n
tical
l
y
s
i
m
ilar
e
x
a
m
p
le
s
ac
r
o
s
s
d
o
m
ai
n
s
ar
e
a
lig
n
ed
.
On
a
s
i
m
i
lar
n
o
te,
th
e
C
ap
s
-
T
r
an
s
p
er
f
o
r
m
s
h
ier
ar
ch
ical
an
d
m
u
l
ti
-
s
ca
le
co
n
te
x
tu
a
l
d
ep
en
d
en
cie
s
th
r
o
u
g
h
tr
an
s
f
o
r
m
er
la
y
er
s
,
ca
p
s
u
le
la
y
er
s
,
an
d
d
y
n
a
m
ic
r
o
u
t
in
g
.
T
h
e
m
er
g
ed
r
ep
r
esen
tat
io
n
s
ar
e
t
h
e
n
f
ed
to
a
d
o
m
ai
n
-
ad
ap
tiv
e
class
i
f
ier
in
o
r
d
er
to
p
r
o
d
u
ce
th
e
en
d
co
n
t
en
t
ca
te
g
o
r
ies.
T
h
is
co
m
b
in
ed
ar
ch
i
tect
u
r
e
g
u
ar
a
n
tees
h
i
g
h
cr
o
s
s
d
o
m
ai
n
ad
ap
tatio
n
,
en
h
a
n
ce
d
s
e
m
a
n
tic
ali
g
n
m
en
t a
n
d
h
ig
h
er
clas
s
i
f
icatio
n
p
er
f
o
r
m
a
n
ce
.
3
.
1
.
Do
m
a
in co
nte
x
t
-
a
w
a
re
no
r
m
a
liza
t
io
n
T
h
e
s
u
g
g
ested
DC
AN
f
r
a
m
e
wo
r
k
is
th
e
s
tr
u
c
tu
r
ed
an
d
d
o
m
a
in
-
s
e
n
s
i
t
iv
e
p
r
ep
r
o
ce
s
s
in
g
w
h
i
ch
en
ab
le
s
cr
ea
tin
g
s
tab
le
s
e
m
an
ticall
y
co
n
s
i
s
te
n
t
tex
t
u
al
r
ep
r
esen
tatio
n
s
in
cr
o
s
s
-
d
o
m
ai
n
lear
n
in
g
.
An
d
ass
u
m
e
t
h
at
in
p
u
t
co
r
p
u
s
is
f
o
r
m
ed
b
y
h
eter
o
g
en
eo
u
s
d
o
cu
m
e
n
ts
r
ep
r
esen
tin
g
m
o
r
e
t
h
an
o
n
e
d
o
m
ai
n
w
i
th
d
is
tr
ib
u
tio
n
al
d
if
f
er
e
n
ce
s
,
lex
ica
l
ch
an
g
e
an
d
s
em
a
n
tic
ch
a
n
g
e.
DC
A
N
o
p
ti
m
ize
s
tex
t
u
al
d
ata
in
s
y
s
te
m
a
tic
r
ef
in
e
m
e
n
ts
th
a
t
co
n
s
is
t
o
f
f
i
v
e
clo
s
el
y
in
ter
a
ctin
g
s
tep
s
n
a
m
el
y
co
n
te
x
t
u
a
l
to
k
en
r
ef
i
n
e
m
e
n
t,
d
o
m
a
in
-
a
d
ap
tiv
e
s
to
p
-
w
o
r
d
r
e
m
o
v
al,
m
o
r
p
h
o
lo
g
ical
n
o
r
m
aliza
tio
n
,
e
n
tr
o
p
y
-
b
ased
n
o
is
e
r
e
m
o
v
al,
a
n
d
e
m
b
ed
d
in
g
-
co
n
s
tr
ai
n
ed
s
e
m
a
n
tic
au
g
m
e
n
tatio
n
.
T
h
r
o
u
g
h
j
o
in
t
o
p
ti
m
izatio
n
o
f
s
e
m
a
n
tic
ali
g
n
m
e
n
t,
p
r
o
b
ab
ilis
tic
r
elev
an
ce
an
d
cr
o
s
s
-
d
o
m
ai
n
s
tab
ilit
y
,
DC
A
N
g
e
n
er
ates
a
n
o
r
m
al
ize
d
co
r
p
u
s
th
at
e
v
o
k
e
s
t
h
e
lea
s
t
in
tr
a
-
d
o
m
ain
v
ar
ia
n
ce
a
n
d
th
e
m
o
s
t
in
t
er
-
d
o
m
ai
n
d
is
cr
i
m
i
n
ab
ilit
y
.
A
s
s
u
m
i
n
g
t
h
e
m
u
l
ti
-
d
o
m
ai
n
co
r
p
u
s
b
e
co
m
p
u
ted
as
:
=
{
(
,
,
)
}
=
1
w
h
er
e,
d
en
o
tes
t
h
e
ℎ
d
o
cu
m
en
t
,
d
en
o
tes
its
cla
s
s
lab
el,
∈
{
1
,
2
,
…
,
}
d
en
o
tes
th
e
d
o
m
ain
in
d
e
x
,
i
s
th
e
to
tal
n
u
m
b
er
o
f
d
o
cu
m
e
n
ts
,
an
d
is
th
e
n
u
m
b
er
o
f
d
o
m
ain
s
.
E
ac
h
d
o
cu
m
e
n
t is r
ep
r
ese
n
ted
as a
to
k
en
s
eq
u
en
ce
:
=
{
1
,
2
,
…
,
}
w
h
er
e
is
t
h
e
len
g
t
h
o
f
d
o
cu
m
en
t
.
E
ac
h
to
k
e
n
is
m
ap
p
ed
to
a
co
n
tex
tu
al
e
m
b
ed
d
in
g
d
ef
i
n
ed
as
:
e
=
(
,
)
h
er
ein
,
(
⋅
)
is
a
p
ar
am
eter
ized
co
n
tex
tu
al
e
m
b
ed
d
i
n
g
f
u
n
ctio
n
,
r
ep
r
esen
ts
th
e
lo
ca
l
co
n
tex
t
w
i
n
d
o
w
o
f
to
k
en
,
e
∈
ℝ
is
th
e
e
m
b
ed
d
in
g
v
ec
t
o
r
.
Af
ter
to
k
e
n
r
ef
i
n
e
m
en
t is p
er
f
o
r
m
ed
v
ia
s
i
m
ilar
it
y
-
b
a
s
ed
p
r
o
j
ec
tio
n
is
r
ep
r
esen
ted
in
:
(
)
=
a
r
g
ma
x
∈
c
os
(
e
,
e
)
w
h
er
e,
is
th
e
v
o
ca
b
u
lar
y
s
p
ac
e,
c
os
(
⋅
)
d
en
o
tes co
s
in
e
s
i
m
ilar
it
y
,
a
n
d
(
)
is
th
e
r
e
f
in
ed
to
k
e
n
.
T
h
e
DC
A
N
f
r
a
m
e
w
o
r
k
th
u
s
p
r
o
v
id
es
a
m
at
h
e
m
atica
ll
y
b
ased
,
d
o
m
ai
n
co
n
s
cio
u
s
,
n
o
r
m
aliza
tio
n
m
ec
h
a
n
i
s
m
w
h
ic
h
i
m
p
r
o
v
e
s
s
e
m
an
tic
s
tab
ilit
y
o
r
d
o
m
ai
n
in
v
ar
ian
ce
a
n
d
th
e
p
r
ed
is
p
o
s
itio
n
o
f
d
is
cr
i
m
in
a
tiv
e
f
ea
t
u
r
es to
f
o
llo
w
o
n
ad
ap
tiv
e
d
ee
p
n
eu
r
al
m
o
d
elin
g
.
T
h
e
d
o
m
ain
-
co
n
d
it
io
n
al
p
r
o
b
ab
ilit
y
o
f
to
k
en
is
ca
lcu
lated
as
:
(
∣
∣
)
=
,
∑
′
,
′
let
u
s
as
s
u
m
e
t
h
at,
,
d
en
o
tes f
r
e
q
u
en
c
y
o
f
to
k
e
n
in
d
o
m
a
in
.
Af
ter
th
a
t c
o
m
p
u
te
s
t
h
e
g
lo
b
al
p
r
o
b
a
b
ilit
y
in
:
(
)
=
∑
=
1
(
∣
∣
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
2
,
J
u
l
y
202
6
:
51
4
-
5
2
3
518
w
h
er
e
is
t
h
e
p
r
io
r
p
r
o
b
ab
ilit
y
o
f
d
o
m
a
in
.
I
d
en
tify
th
e
d
o
m
a
in
r
ele
v
an
ce
s
co
r
e
is
d
en
o
ted
as:
(
)
=
1
∑
∣
(
∣
)
−
(
)
∣
=
1
T
o
k
en
eli
m
i
n
atio
n
cr
iter
io
n
a
s
d
ef
in
ed
as
:
is
r
em
o
v
ed
if
(
)
<
h
er
ein
,
is
th
e
d
o
m
ai
n
-
d
is
cr
i
m
i
n
ab
ilit
y
th
r
es
h
o
ld
.
E
ac
h
r
ef
i
n
ed
to
k
en
u
n
d
er
g
o
es
ca
n
o
n
ical
tr
a
n
s
f
o
r
m
atio
n
:
(
)
=
(
(
)
)
ass
u
m
in
g
t
h
at,
(
⋅
)
is
th
e
m
o
r
p
h
o
l
o
g
ical
n
o
r
m
a
lizatio
n
o
p
er
ato
r
.
C
r
o
s
s
-
d
o
m
ai
n
en
tr
o
p
y
f
o
r
to
k
en
is
co
m
p
u
ted
in
:
(
)
=
−
∑
(
∣
=
1
)
l
og
(
∣
)
No
r
m
a
lized
en
tr
o
p
y
d
ef
i
n
ed
:
̂
(
)
=
(
)
l
o
g
Stab
ilit
y
cr
iter
io
n
:
is
r
etain
ed
if
̂
(
)
≤
w
h
er
e
is
t
h
e
en
tr
o
p
y
s
tab
ili
t
y
th
r
es
h
o
ld
.
Fo
r
ea
ch
n
o
r
m
alize
d
to
k
e
n
(
)
,
a
n
au
g
m
en
ted
to
k
e
n
(
)
is
s
elec
ted
s
u
c
h
th
at
:
c
os
(
e
(
)
,
e
(
)
)
≥
ass
u
m
in
g
th
a
t,
is
th
e
s
i
m
ilar
it
y
th
r
esh
o
ld
,
an
d
e
(
)
an
d
e
(
)
d
en
o
te
e
m
b
ed
d
i
n
g
v
ec
to
r
s
.
T
h
e
au
g
m
e
n
ted
to
k
e
n
s
et
b
ec
o
m
es:
̃
=
{
(
)
,
(
)
}
T
h
e
DC
AN
-
p
r
o
ce
s
s
ed
d
o
cu
m
en
t is d
ef
in
ed
as:
=
{
̃
1
,
̃
2
,
…
,
̃
′
}
h
er
e
,
′
≤
af
ter
r
ef
in
e
m
e
n
t
an
d
f
ilt
er
in
g
.
D
C
A
N
ai
m
s
to
m
i
n
i
m
iz
e
in
tr
a
-
d
o
m
ai
n
v
ar
ian
ce
w
h
ile
m
ax
i
m
izin
g
in
ter
-
d
o
m
ain
s
ep
ar
ab
ilit
y
:
min
(
∑
Var
=
1
(
)
)
−
(
Dis
c
)
w
h
er
e,
d
en
o
tes
n
o
r
m
a
lizatio
n
o
p
er
atio
n
s
,
Va
r
(
)
is
in
tr
a
-
d
o
m
ai
n
v
ar
ian
ce
,
Dis
c
r
ep
r
esen
ts
in
ter
-
d
o
m
ai
n
d
is
cr
i
m
i
n
ab
ilit
y
,
a
n
d
is
a
b
alan
ci
n
g
co
ef
f
icie
n
t.
3
.
2
.
B
ina
ry
pa
rt
icle
s
w
a
r
m
i
nte
llig
ence
o
pti
m
iza
t
io
n
T
h
e
s
u
g
g
ested
f
r
a
m
e
w
o
r
k
u
s
e
s
a
B
P
SO
m
ec
h
a
n
is
m
to
ca
r
r
y
o
u
t
d
o
m
ai
n
r
o
b
u
s
t
f
ea
t
u
r
e
s
el
ec
tio
n
o
n
h
ig
h
-
d
i
m
en
s
io
n
al
te
x
t
u
al
s
p
ac
es.
A
llo
w
t
h
e
DC
A
N
-
p
r
o
ce
s
s
e
d
co
r
p
u
s
to
p
r
o
d
u
ce
an
ex
ten
s
i
v
e
p
o
o
l
o
f
f
ea
tu
r
es
co
n
s
is
tin
g
o
f
le
x
ical,
s
y
n
tact
i
c,
s
e
m
a
n
tic,
co
n
te
x
tu
a
l
a
n
d
s
t
atis
tical
f
ea
t
u
r
es.
L
ater
al
s
ele
ctio
n
v
ar
iab
ilit
y
an
d
r
ed
u
n
d
an
c
y
o
f
f
ea
tu
r
es
m
ea
n
s
th
at
a
n
ad
ap
tiv
e
s
elec
tio
n
m
ec
h
an
i
s
m
i
s
n
ee
d
ed
to
d
eter
m
in
e
th
e
d
i
s
cr
i
m
i
n
ati
v
e
an
d
d
o
m
ai
n
-
i
n
v
ar
ia
n
t a
ttrib
u
te
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
Meta
-
lea
r
n
in
g
co
n
tr
a
s
tive
fu
s
io
n
in
tellig
en
ce
fo
r
d
o
ma
in
-
a
d
a
p
tive
…
(
Ja
n
a
n
i
S
iva
p
r
iya
V
e
n
ka
ta
kris
h
n
a
n
)
519
E
ac
h
p
ar
ticle
is
r
ep
r
esen
ted
as
a
b
in
ar
y
v
ec
to
r
as d
en
o
ted
in
:
p
=
[
1
,
2
,
…
,
]
w
h
er
e:
=
{
1
if
f
ea
tu
r
e
is
s
elec
ted
0
o
th
er
wis
e
I
n
B
P
SO,
v
elo
cit
y
co
n
tr
o
ls
t
h
e
p
r
o
b
a
b
ilit
y
o
f
s
elec
t
in
g
a
f
ea
t
u
r
e.
T
h
e
v
elo
cit
y
o
f
p
ar
ticle
at
iter
atio
n
is
:
(
+
1
)
=
(
)
+
1
1
(
−
(
)
)
+
2
2
(
−
(
)
)
ass
u
m
in
g
t
h
at,
r
ef
er
s
in
er
tia
w
ei
g
h
t,
1
lies
co
g
n
iti
v
e
co
ef
f
icien
t,
2
p
r
esen
ts
s
o
cial
co
ef
f
i
cien
t,
1
,
2
∼
(
0
,
1
)
,
lies
p
er
s
o
n
al
b
est p
o
s
itio
n
a
n
d
d
ep
icts
g
lo
b
al
b
est p
o
s
itio
n
.
Velo
cit
y
is
tr
a
n
s
f
o
r
m
ed
i
n
to
s
elec
tio
n
p
r
o
b
ab
ilit
y
u
s
i
n
g
s
i
g
m
o
id
f
u
n
ct
io
n
:
(
)
=
1
1
+
−
B
in
ar
y
u
p
d
ate
r
u
le
i
s
co
m
p
u
te
d
:
(
+
1
)
=
{
1
if
(
)
<
(
(
+
1
)
)
0
o
th
er
wis
e
T
h
e
f
itn
e
s
s
f
u
n
ctio
n
b
alan
ce
s
d
is
cr
i
m
i
n
ab
ilit
y
an
d
r
ed
u
n
d
an
c
y
:
(
p
)
=
⋅
(
p
)
−
⋅
(
p
)
w
h
er
e
,
r
ef
er
s
b
alan
cin
g
co
ef
f
icie
n
ts
,
(
p
)
s
tates
cr
o
s
s
-
d
o
m
ai
n
d
is
cr
im
in
ab
ili
t
y
an
d
(
p
)
m
e
n
ti
o
n
s
r
ed
u
n
d
an
c
y
m
ea
s
u
r
e
.
C
r
o
s
s
-
d
o
m
ai
n
d
is
cr
i
m
in
ab
ili
t
y
is
co
m
p
u
ted
i
n
:
(
)
=
1
∑
Tr
=
1
(
(
)
(
)
−
1
)
h
er
ein
,
r
ep
r
esen
ts
n
u
m
b
er
o
f
d
o
m
a
in
s
,
(
)
s
tates
b
et
w
ee
n
-
cla
s
s
s
ca
tter
m
atr
ix
i
n
d
o
m
ai
n
,
an
d
(
)
p
r
esen
t
s
w
it
h
i
n
-
clas
s
s
ca
tter
m
atr
ix
.
R
ed
u
n
d
an
c
y
m
ea
s
u
r
e
m
e
n
t
is
c
alcu
lated
in
:
(
p
)
=
1
∣
∣
∑
∣
co
r
r
(
,
∈
,
)
∣
h
er
ein
,
d
en
o
tes
th
e
s
elec
ted
f
ea
tu
r
e
s
u
b
s
et
an
d
c
or
r
(
⋅
)
s
tates
co
r
r
elatio
n
co
ef
f
icien
t.
Op
ti
m
izatio
n
o
b
j
ec
tiv
e
is
esti
m
ated
as
:
ma
x
[
(
p
)
−
(
p
)
]
s
u
b
j
ec
t to
:
∣
∣
p
∣
∣
0
≤
ass
u
m
in
g
t
h
at,
∣
∣
⋅
∣
∣
0
d
en
o
tes n
u
m
b
er
o
f
s
elec
ted
f
ea
t
u
r
es.
C
o
n
v
er
g
e
n
ce
is
ac
h
iev
ed
w
h
e
n
:
∣
(
(
+
1
)
)
−
(
(
)
)
∣
<
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
2
,
J
u
l
y
202
6
:
51
4
-
5
2
3
520
T
h
e
s
u
g
g
e
s
ted
B
P
SO
-
b
ased
b
eh
av
io
r
al
f
ea
t
u
r
e
s
elec
tio
n
m
e
ch
an
i
s
m
,
t
h
u
s
,
allo
w
s
ad
ap
tiv
e
s
elec
tio
n
o
f
h
i
g
h
-
d
o
m
ai
n
i
n
s
e
n
s
it
iv
e,
l
o
w
-
r
ed
u
n
d
a
n
c
y
f
ea
t
u
r
es
b
y
c
o
o
p
er
ativ
e
s
w
ar
m
s
ea
r
ch
a
n
d
p
r
o
b
ab
ilis
tic
b
in
ar
y
m
ax
i
m
izatio
n
,
w
h
ic
h
g
u
ar
an
t
ee
h
ig
h
d
is
cr
i
m
i
n
ati
v
e
ca
p
ac
it
y
in
cr
o
s
s
-
d
o
m
ain
co
n
te
n
t
cl
ass
i
f
icatio
n
s
tas
k
s
.
T
ab
le
1
d
escr
ib
es th
e
B
P
SO a
p
p
r
o
ac
h
p
ar
am
eter
s
co
n
f
i
g
u
r
at
io
n
.
T
ab
le
1
.
B
P
SO a
p
p
r
o
ac
h
p
ar
a
m
eter
s
co
n
f
i
g
u
r
atio
n
P
a
r
a
me
t
e
r
S
y
mb
o
l
D
e
scri
p
t
i
o
n
V
a
l
u
e
S
w
a
r
m si
z
e
N
u
mb
e
r
o
f
p
a
r
t
i
c
l
e
s
30
–
50
M
a
x
i
t
e
r
a
t
i
o
n
s
O
p
t
i
mi
z
a
t
i
o
n
st
e
p
s
1
0
0
I
n
e
r
t
i
a
w
e
i
g
h
t
Ex
p
l
o
r
a
t
i
o
n
–
e
x
p
l
o
i
t
a
t
i
o
n
b
a
l
a
n
c
e
0
.
7
C
o
g
n
i
t
i
v
e
c
o
e
f
f
i
c
i
e
n
t
1
P
e
r
so
n
a
l
l
e
a
r
n
i
n
g
w
e
i
g
h
t
1
.
5
S
o
c
i
a
l
c
o
e
f
f
i
c
i
e
n
t
2
G
l
o
b
a
l
l
e
a
r
n
i
n
g
w
e
i
g
h
t
1
.
5
D
i
scri
m
i
n
a
b
i
l
i
t
y
w
e
i
g
h
t
F
i
t
n
e
ss w
e
i
g
h
t
0
.
7
R
e
d
u
n
d
a
n
c
y
w
e
i
g
h
t
F
i
t
n
e
ss p
e
n
a
l
t
y
0
.
3
C
o
n
v
e
r
g
e
n
c
e
t
h
r
e
sh
o
l
d
S
t
o
p
p
i
n
g
c
r
i
t
e
r
i
o
n
(
10
{
−
4
}
)
3
.
3
.
M
et
a
-
le
a
rning
co
ntr
a
s
t
i
v
e
f
us
io
n inte
llig
ence
T
h
e
p
r
o
p
o
s
ed
MCF
I
is
a
d
o
m
a
in
-
ad
ap
tiv
e
m
o
d
el
th
at
i
s
d
ev
elo
p
ed
to
attain
s
tr
o
n
g
co
n
te
n
t
class
i
f
icatio
n
w
h
en
t
h
er
e
ar
e
cr
o
s
s
-
d
o
m
ai
n
d
is
tr
ib
u
tio
n
s
h
if
t
s
.
MCF
I
co
n
s
is
t
s
o
f
t
w
o
co
m
p
l
e
m
en
tar
y
m
o
d
u
le
s
o
f
r
ep
r
esen
tatio
n
lear
n
i
n
g
:
i)
a
co
n
tr
asti
v
e
lear
n
i
n
g
-
e
n
h
an
ce
d
B
E
R
T
(
C
L
-
B
E
R
T
)
s
e
m
an
t
ic
alig
n
er
an
d
d
o
m
ain
-
in
v
ar
ia
n
t
e
m
b
ed
d
in
g
g
en
er
at
o
r
an
d
ii)
a
ca
p
s
u
le
n
et
w
o
r
k
-
en
h
a
n
ce
d
tr
an
s
f
o
r
m
er
(
C
ap
s
-
T
r
an
s
)
h
ier
ar
ch
ical
s
e
m
a
n
tic
a
g
g
r
eg
atio
n
a
n
d
s
tr
u
ctu
r
al
r
elatio
n
s
h
ip
r
ep
r
esen
tati
o
n
.
T
h
e
C
L
-
B
E
R
T
is
a
m
o
d
el
th
at
cr
o
s
s
d
o
m
ai
n
al
ig
n
s
s
e
m
a
n
tical
l
y
s
i
m
ilar
s
a
m
p
les
t
h
r
o
u
g
h
co
n
tr
asti
v
e
g
o
als,
an
d
t
h
u
s
,
i
m
p
o
s
e
s
i
n
ter
-
d
o
m
a
in
in
v
ar
ian
ce
i
n
e
m
b
ed
d
in
g
s
p
ac
e.
B
u
t
w
h
er
ea
s
co
n
tr
asti
v
e
ali
g
n
m
e
n
t
f
ac
i
litates
th
e
s
ep
ar
atio
n
o
f
g
lo
b
al
r
ep
r
esen
tatio
n
s
,
h
ier
ar
ch
ical
r
ela
tio
n
s
a
m
o
n
g
s
e
m
an
tic
ele
m
e
n
ts
in
d
o
cu
m
e
n
t
s
d
e
m
a
n
d
i
n
s
tit
u
tio
n
o
f
s
tr
u
ctu
r
al
ag
g
r
e
g
atio
n
.
T
h
e
C
ap
s
-
T
r
an
s
m
o
d
u
le
tak
e
s
th
i
s
in
to
co
n
s
id
er
atio
n
an
d
id
en
ti
f
i
e
s
p
ar
t
-
w
h
o
le
s
e
m
an
tic
in
ter
a
ctio
n
s
b
y
m
o
d
eli
n
g
d
y
n
a
m
ic
r
o
u
t
in
g
to
al
lo
w
r
o
b
u
s
t
s
tr
u
ctu
r
al
f
ea
t
u
r
e
e
n
co
d
in
g
.
T
h
e
f
in
al
o
u
tp
u
t
o
f
MC
FI
is
a
f
u
s
ed
d
o
m
ai
n
-
ad
ap
tiv
e
r
ep
r
esen
tatio
n
v
ec
to
r
:
z
=
ℱ
(
z
,
z
)
L
et
u
s
ass
u
m
e
th
a
t,
z
d
en
o
tes
c
o
n
tr
as
ti
v
el
y
ali
g
n
ed
s
e
m
an
tic
e
m
b
ed
d
in
g
s
,
z
d
en
o
tes
ca
p
s
u
le
-
s
tr
u
ct
u
r
ed
h
ier
ar
ch
ical
e
m
b
ed
d
in
g
s
,
an
d
ℱ
(
⋅
)
d
en
o
tes ad
ap
tiv
e
f
u
s
io
n
.
T
h
e
r
esu
ltin
g
r
ep
r
esen
tatio
n
is
cr
o
s
s
-
d
o
m
ai
n
,
i
n
cr
ea
s
i
n
g
in
cr
o
s
s
-
d
o
m
ain
d
is
cr
i
m
i
n
a
b
ilit
y
an
d
s
tr
u
ct
u
r
al
co
n
s
i
s
te
n
c
y
a
n
d
p
r
o
v
id
in
g
g
r
ea
ter
s
tab
ilit
y
i
n
ca
te
g
o
r
izatio
n
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
I
n
o
r
d
er
to
test
th
e
ef
f
icie
n
c
y
o
f
th
e
s
u
g
g
ested
MCF
I
s
tr
u
ct
u
r
e,
th
e
ex
p
er
im
e
n
t
s
w
er
e
ca
r
r
ied
o
u
t w
ith
th
e
h
elp
o
f
o
n
e
o
f
t
h
e
p
u
b
licl
y
av
ailab
le
te
x
t
ca
te
g
o
r
izin
g
d
at
a
s
ets g
at
h
er
ed
at
t
h
e
Ka
g
g
le
r
ep
o
s
ito
r
y
.
T
h
e
d
ata
is
m
ad
e
u
p
o
f
f
iv
e
co
n
te
n
t
ca
t
eg
o
r
ies
n
a
m
el
y
b
u
s
i
n
ess
,
tec
h
n
o
lo
g
y
,
s
p
o
r
ts
,
en
ter
tai
n
m
en
t,
an
d
ed
u
ca
tio
n
.
All
th
e
s
tep
s
o
f
p
r
ep
r
o
ce
s
s
in
g
,
f
e
atu
r
e
s
elec
tio
n
,
m
o
d
el
tr
ain
i
n
g
,
an
d
ev
al
u
atio
n
w
er
e
i
m
p
le
m
en
ted
in
P
y
th
o
n
en
v
ir
o
n
m
e
n
t
w
i
th
g
en
er
al
-
p
u
r
p
o
s
e
d
ee
p
lear
n
in
g
lib
r
ar
ies.
T
h
e
d
ata
w
a
s
d
iv
id
ed
in
to
7
0
%
an
d
3
0
%
tr
ain
in
g
an
d
test
in
g
r
esp
ec
tiv
el
y
to
h
a
v
e
u
n
b
iased
co
n
s
id
er
atio
n
.
E
p
is
o
d
ic
o
p
ti
m
i
za
tio
n
o
f
m
eta
-
lear
n
i
n
g
an
d
co
n
tr
asti
v
e
e
m
b
ed
d
in
g
ali
g
n
m
en
t
w
a
s
o
p
ti
m
ized
i
n
th
e
tr
ain
i
n
g
s
et,
an
d
p
er
f
o
r
m
a
n
ce
v
alid
atio
n
w
as
o
n
l
y
d
o
n
e
in
th
e
te
s
ti
n
g
s
e
t.
Fig
u
r
e
2
s
h
o
w
s
a
co
m
p
ar
ati
v
e
an
al
y
s
is
o
f
th
e
clas
s
if
icatio
n
p
er
f
o
r
m
an
ce
b
ased
o
n
ac
c
u
r
ac
y
an
d
F1
-
s
co
r
e
o
f
G
A
N,
B
iG
R
U,
D
A
R
L
,
a
n
d
t
h
e
p
r
o
p
o
s
ed
MCF
I
m
o
d
el.
I
t
is
n
o
ted
th
at
G
A
N
h
as
7
5
.
4
%
ac
c
u
r
ac
y
a
n
d
7
4
.
2
8
%
F1
-
s
co
r
e,
w
h
ic
h
m
ea
n
s
t
h
at
it
h
as
r
elativ
e
l
y
lo
w
er
class
i
f
ic
atio
n
p
o
ten
tial.
B
iGR
U
u
p
g
r
ad
es
th
e
ac
cu
r
ac
y
an
d
F1
-
s
c
o
r
e
to
8
4
.
2
1
%
an
d
8
3
.
1
9
%,
r
esp
ec
tiv
el
y
,
b
y
b
ein
g
ab
le
to
ca
p
tu
r
e
b
id
ir
ec
tio
n
a
l
co
n
tex
t
u
al
r
elatio
n
s
h
ip
s
.
DAR
L
also
p
r
o
v
id
es
f
u
r
t
h
er
r
esu
l
ts
o
f
8
9
.
4
5
%
ac
cu
r
ac
y
a
n
d
8
8
.
0
4
%
F1
-
s
co
r
e
b
ec
au
s
e
o
f
ad
ap
tiv
e
r
ep
r
esen
tatio
n
lear
n
i
n
g
s
tr
ateg
y
.
Nev
er
th
e
less
,
th
e
p
r
o
p
o
s
ed
MCF
I
m
o
d
el
is
m
o
r
e
p
r
ec
is
e
th
a
n
a
n
y
o
f
th
e
ex
i
s
ti
n
g
o
n
e
s
w
it
h
t
h
e
b
est
ac
cu
r
ac
y
o
f
9
2
.
1
7
an
d
F1
-
s
co
r
e
o
f
9
1
.
0
5
.
T
h
e
f
ac
t
th
at
b
o
t
h
m
etr
ic
s
ar
e
s
tead
il
y
i
n
cr
ea
s
i
n
g
p
r
o
v
es
th
e
s
tr
en
g
t
h
an
d
b
alan
ce
d
class
i
f
icatio
n
ca
p
ab
ilit
y
o
f
t
h
e
s
u
g
g
e
s
ted
f
r
a
m
e
w
o
r
k
.
F
ig
u
r
e
3
s
h
o
w
s
a
c
o
m
p
a
r
at
iv
e
a
n
a
ly
s
is
o
f
p
r
e
ci
s
i
o
n
an
d
r
e
c
al
l
o
f
GA
N
,
B
i
GR
U
,
DA
R
L
,
a
n
d
p
r
o
p
o
s
e
d
M
C
F
I
m
o
d
e
l
.
B
a
s
e
d
o
n
t
h
e
o
u
t
c
o
m
e
,
GA
N
h
a
s
a
p
r
e
c
is
i
o
n
o
f
7
4
.
2
%
,
w
h
i
ch
m
e
an
s
th
a
t
it
h
as
r
e
l
a
t
iv
ely
l
o
w
e
r
c
l
a
s
s
if
ic
a
t
i
o
n
th
an
a
h
ig
h
e
r
c
h
an
c
e
o
f
m
i
s
c
l
as
s
i
f
i
c
a
ti
o
n
an
d
m
is
s
i
n
g
th
e
r
e
le
v
a
n
t
c
as
e
s
.
B
i
GR
U
h
as
a
m
u
ch
h
ig
h
e
r
p
e
r
f
o
r
m
an
c
e
w
h
i
ch
is
at
8
5
.
1
4
%
p
r
e
c
is
i
o
n
an
d
8
4
.
1
9
%
r
e
c
a
ll,
w
h
i
ch
s
h
o
w
s
b
e
t
t
e
r
c
o
n
t
ex
tu
al
u
n
d
e
r
s
t
an
d
in
g
an
d
l
e
s
s
f
a
ls
e
p
r
e
d
i
c
ti
o
n
s
.
DA
R
L
a
ls
o
im
p
r
o
v
es
th
e
s
c
o
r
e
an
d
th
e
p
r
e
c
i
s
i
o
n
an
d
r
e
c
al
l
a
r
e
8
8
.
0
5
a
n
d
8
8
.
3
,
r
e
s
p
e
c
t
iv
e
ly
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
Meta
-
lea
r
n
in
g
co
n
tr
a
s
tive
fu
s
io
n
in
tellig
en
ce
fo
r
d
o
ma
in
-
a
d
a
p
tive
…
(
Ja
n
a
n
i
S
iva
p
r
iya
V
e
n
ka
ta
kris
h
n
a
n
)
521
Fig
u
r
e
4
co
m
p
ar
es t
h
e
er
r
o
r
r
a
te
o
f
GAN,
B
iGR
U,
D
AR
L
,
a
n
d
th
e
p
r
o
p
o
s
ed
MCF
I
m
o
d
el.
A
s
ca
n
b
e
s
ee
n
,
G
AN
h
as
th
e
lar
g
e
s
t
er
r
o
r
r
ate
o
f
2
6
.
5
%
,
w
it
h
m
o
r
e
f
al
s
e
cla
s
s
i
f
icatio
n
s
b
et
w
ee
n
co
n
ten
t
ca
te
g
o
r
ies.
B
iGR
U
g
r
ea
tl
y
m
in
i
m
ize
s
t
h
e
er
r
o
r
r
ate
at
1
5
.
3
s
h
o
w
i
n
g
b
etter
co
n
tex
t
lear
n
i
n
g
o
n
f
ea
t
u
r
es.
DAR
L
al
s
o
r
ed
u
ce
s
th
e
er
r
o
r
r
ate
to
1
2
.
1
7
,
w
h
ic
h
s
h
o
w
s
an
i
m
p
r
o
v
ed
r
ep
r
esen
tat
io
n
an
d
ad
ap
tiv
e
lear
n
i
n
g
ab
ili
t
y
.
Fig
u
r
e
2
.
C
o
m
p
ar
is
o
n
o
f
ac
c
u
r
ac
y
an
d
F1
-
s
co
r
e
p
er
f
o
r
m
an
ce
Fig
u
r
e
3
.
C
o
m
p
ar
is
o
n
o
f
p
r
ec
i
s
io
n
an
d
r
ec
all
p
er
f
o
r
m
a
n
ce
Fig
u
r
e
4
.
C
o
m
p
ar
is
o
n
o
f
er
r
o
r
r
ate
5.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
in
tr
o
d
u
ce
d
a
n
o
v
el
d
o
m
ai
n
-
ad
ap
tiv
e
co
n
te
n
t
ca
teg
o
r
izatio
n
f
r
a
m
e
w
o
r
k
in
teg
r
ati
n
g
DC
A
N,
B
P
SO
-
b
ased
f
ea
t
u
r
e
s
elec
tio
n
,
an
d
t
h
e
p
r
o
p
o
s
ed
MCF
I
m
o
d
el,
w
h
ic
h
co
m
b
i
n
es
C
L
-
B
E
R
T
w
it
h
a
ca
p
s
u
le
-
tr
an
s
f
o
r
m
er
ar
ch
itec
tu
r
e
.
T
h
e
f
r
a
m
e
w
o
r
k
w
as
i
m
p
le
m
e
n
ted
in
P
y
t
h
o
n
a
n
d
ev
al
u
ated
o
n
a
p
u
b
licl
y
a
v
ailab
l
e
Kag
g
le
d
ataset
co
m
p
r
i
s
in
g
f
i
v
e
ca
teg
o
r
ies:
b
u
s
i
n
ess
,
tec
h
n
o
lo
g
y
,
s
p
o
r
ts
,
en
ter
tain
m
e
n
t,
an
d
ed
u
ca
tio
n
.
A
7
0
:3
0
tr
ain
–
test
s
p
lit w
as
ad
o
p
ted
to
en
s
u
r
e
r
o
b
u
s
t
p
er
f
o
r
m
a
n
ce
ev
a
lu
atio
n
.
E
x
p
er
i
m
en
tal
r
esu
lt
s
clea
r
l
y
d
e
m
o
n
s
tr
ate
th
e
s
u
p
er
io
r
it
y
o
f
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
o
v
er
ex
is
ti
n
g
m
et
h
o
d
s
s
u
c
h
a
s
G
A
N,
B
iG
R
U,
a
n
d
DAR
L
.
T
h
e
MC
FI
m
o
d
el
ac
h
ie
v
ed
a
clas
s
i
f
icati
o
n
ac
cu
r
ac
y
o
f
9
2
.
1
7
%,
F1
-
s
co
r
e
o
f
9
1
.
0
5
%,
p
r
ec
is
io
n
o
f
9
2
.
2
7
%,
r
ec
all
o
f
9
1
.
1
5
%,
an
d
a
lo
w
er
r
o
r
r
ate
o
f
8
.
4
5
%.
No
tab
ly
,
it
o
u
tp
er
f
o
r
m
ed
DAR
L
—
t
h
e
s
tr
o
n
g
e
s
t
b
aselin
e
—
i
n
d
icatin
g
en
h
a
n
ce
d
d
is
cr
i
m
i
n
ati
v
e
ca
p
ab
ilit
y
a
n
d
b
alan
ce
d
p
r
ec
is
io
n
–
r
ec
all
p
er
f
o
r
m
an
ce
.
T
h
e
co
n
s
i
s
ten
t
i
m
p
r
o
v
e
m
e
n
t
ac
r
o
s
s
all
m
etr
ics
co
n
f
ir
m
s
t
h
e
m
o
d
el’
s
ab
ilit
y
t
o
r
ed
u
ce
b
o
th
f
alse
p
o
s
itiv
e
s
an
d
f
al
s
e
n
eg
a
tiv
e
s
w
h
i
le
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
9
-
4864
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
,
Vo
l.
15
,
No
.
2
,
J
u
l
y
202
6
:
51
4
-
5
2
3
522
p
r
eser
v
in
g
co
n
tex
t
u
al
a
n
d
h
ie
r
ar
ch
ical
s
e
m
a
n
tic
r
elatio
n
s
h
i
p
s
.
T
h
e
p
er
f
o
r
m
an
ce
g
ai
n
s
ar
e
attr
ib
u
ted
to
t
h
r
ee
k
e
y
co
m
p
o
n
e
n
ts
:
i
)
D
C
A
N
-
b
a
s
ed
p
r
ep
r
o
ce
s
s
in
g
f
o
r
n
o
is
e
r
e
d
u
ctio
n
a
n
d
d
o
m
ai
n
-
i
n
v
ar
ia
n
t
f
ea
t
u
r
e
ex
tr
ac
tio
n
;
ii
)
B
P
SO
-
d
r
iv
en
f
ea
tu
r
e
o
p
ti
m
izatio
n
to
eli
m
i
n
ate
r
ed
u
n
d
an
c
y
a
n
d
m
i
tig
a
te
o
v
er
f
itti
n
g
;
a
n
d
ii
i
)
th
e
h
y
b
r
id
C
L
-
B
E
R
T
w
i
th
ca
p
s
u
le
-
tr
an
s
f
o
r
m
er
ar
ch
itect
u
r
e
f
o
r
ca
p
tu
r
in
g
b
o
th
g
lo
b
al
s
e
m
a
n
tic
s
an
d
h
ier
ar
ch
ical
d
ep
en
d
en
cies.
F
UNDIN
G
I
NF
O
RM
AT
I
O
N
Au
t
h
o
r
s
s
tate
n
o
f
u
n
d
i
n
g
i
n
v
o
l
v
ed
.
A
U
T
H
O
R
C
O
N
T
R
I
B
U
T
I
O
N
S
S
T
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
e
s
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT
)
t
o
r
ec
o
g
n
ize
in
d
i
v
id
u
al
au
t
h
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
t
h
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
lla
b
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
J
an
an
i Si
v
ap
r
i
y
a
Ven
k
atak
r
i
s
h
n
a
n
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Sar
av
an
a
Mo
o
r
th
y
R
aj
a
Mo
o
r
th
y
✓
✓
✓
✓
✓
✓
✓
✓
✓
An
g
el
S
h
an
m
u
g
a
m
✓
✓
✓
✓
✓
✓
✓
✓
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
si
s
I
:
I
n
v
e
st
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
si
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
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
Au
t
h
o
r
s
s
tate
n
o
co
n
f
lic
t o
f
i
n
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
Data
av
ailab
ilit
y
i
s
n
o
t a
p
p
licab
le
to
th
is
p
ap
er
as n
o
n
e
w
d
ata
w
er
e
cr
ea
ted
o
r
an
al
y
ze
d
in
th
is
s
t
u
d
y
.
RE
F
E
R
E
NC
E
S
[
1
]
M
.
I
j
a
z
,
N
.
A
n
w
a
r
,
M
.
S
a
f
r
a
n
,
S
.
A
l
f
a
r
h
o
o
d
,
T
.
S
a
d
a
d
,
a
n
d
I
mr
a
n
,
“
D
o
mai
n
a
d
a
p
t
i
v
e
l
e
a
r
n
i
n
g
f
o
r
m
u
l
t
i
r
e
a
l
m
se
n
t
i
me
n
t
c
l
a
ssi
f
i
c
a
t
i
o
n
o
n
b
i
g
d
a
t
a
,
”
PLO
S
O
N
E
,
v
o
l
.
1
9
,
n
o
.
4
,
A
p
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
3
7
1
/
j
o
u
r
n
a
l
.
p
o
n
e
.
0
2
9
7
0
2
8
.
[
2
]
D
.
X
i
,
F
.
Z
h
u
a
n
g
,
G
.
Z
h
o
u
,
X
.
C
h
e
n
g
,
F
.
L
i
n
,
a
n
d
Q
.
H
e
,
“
D
o
mai
n
a
d
a
p
t
a
t
i
o
n
w
i
t
h
c
a
t
e
g
o
r
y
a
t
t
e
n
t
i
o
n
n
e
t
w
o
r
k
f
o
r
d
e
e
p
se
n
t
i
me
n
t
a
n
a
l
y
si
s,”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
T
h
e
We
b
C
o
n
f
e
re
n
c
e
2
0
2
0
,
A
p
r
.
2
0
2
0
,
p
p
.
3
1
3
3
–
3
1
3
9
,
d
o
i
:
1
0
.
1
1
4
5
/
3
3
6
6
4
2
3
.
3
3
8
0
0
8
8
.
[
3
]
D
.
R
o
g
e
r
s,
A
.
P
r
e
e
c
e
,
M
.
I
n
n
e
s,
a
n
d
I
.
S
p
a
si
c
,
“
R
e
al
-
t
i
me
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
u
se
r
-
g
e
n
e
r
a
t
e
d
c
o
n
t
e
n
t
o
n
so
c
i
a
l
me
d
i
a
:
S
y
st
e
mat
i
c
r
e
v
i
e
w
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
S
o
c
i
a
l
S
y
s
t
e
m
s
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
1
1
5
4
–
1
1
6
6
,
A
u
g
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
C
S
S
.
2
0
2
1
.
3
1
2
0
1
3
8
.
[
4
]
J.
Y
i
n
e
t
a
l
.
,
“
B
l
o
w
i
n
g
se
e
d
s
a
c
r
o
ss
g
a
r
d
e
n
s:
V
i
s
u
a
l
i
z
i
n
g
i
mp
l
i
c
i
t
p
r
o
p
a
g
a
t
i
o
n
o
f
c
r
o
ss
-
p
l
a
t
f
o
r
m
so
c
i
a
l
me
d
i
a
p
o
st
s,
”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
V
i
su
a
l
i
z
a
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r
G
r
a
p
h
i
c
s
,
v
o
l
.
3
1
,
n
o
.
1
,
p
p
.
1
8
5
–
1
9
5
,
J
a
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
T
V
C
G
.
2
0
2
4
.
3
4
5
6
1
8
1
.
[
5
]
R
.
Z
h
a
n
g
,
Z
.
W
a
n
g
,
K
.
Y
i
n
,
a
n
d
Z
.
H
u
a
n
g
,
“
Emo
t
i
o
n
a
l
t
e
x
t
g
e
n
e
r
a
t
i
o
n
b
a
s
e
d
o
n
c
r
o
ss
-
d
o
m
a
i
n
se
n
t
i
me
n
t
t
r
a
n
sf
e
r
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
7
,
p
p
.
1
0
0
0
8
1
–
1
0
0
0
8
9
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
9
.
2
9
3
1
0
3
6
.
[
6
]
T
.
L
i
a
n
g
,
W
.
W
a
n
g
,
a
n
d
F
.
L
v
,
“
W
e
a
k
l
y
su
p
e
r
v
i
se
d
d
o
ma
i
n
a
d
a
p
t
a
t
i
o
n
f
o
r
a
s
p
e
c
t
e
x
t
r
a
c
t
i
o
n
v
i
a
m
u
l
t
i
l
e
v
e
l
i
n
t
e
r
a
c
t
i
o
n
t
r
a
n
sf
e
r
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
N
e
u
r
a
l
N
e
t
w
o
rks
a
n
d
L
e
a
rn
i
n
g
S
y
s
t
e
m
s
,
v
o
l
.
3
3
,
n
o
.
1
0
,
p
p
.
5
8
1
8
–
5
8
2
9
,
O
c
t
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
N
N
L
S
.
2
0
2
1
.
3
0
7
1
4
7
4
.
[
7
]
M
.
L
o
n
g
,
J.
W
a
n
g
,
Y
.
C
a
o
,
J.
S
u
n
,
a
n
d
P
.
S
.
Y
u
,
“
D
e
e
p
l
e
a
r
n
i
n
g
o
f
t
r
a
n
sf
e
r
a
b
l
e
r
e
p
r
e
s
e
n
t
a
t
i
o
n
f
o
r
sc
a
l
a
b
l
e
d
o
mai
n
a
d
a
p
t
a
t
i
o
n
,
”
I
EEE
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
r
i
n
g
,
v
o
l
.
2
8
,
n
o
.
8
,
p
p
.
2
0
2
7
–
2
0
4
0
,
A
u
g
.
2
0
1
6
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
1
6
.
2
5
5
4
5
4
9
.
[
8
]
G
.
S
.
K
a
sh
y
a
p
,
N
.
J
a
i
n
,
E.
S
h
a
b
b
i
r
,
H
.
Jo
sh
i
,
U
.
N
a
se
e
m,
a
n
d
J
.
G
a
o
,
“
C
L
A
R
I
TY
:
A
l
i
g
h
t
w
e
i
g
h
t
m
u
l
t
i
mo
d
a
l
t
r
a
n
sf
o
r
me
r
f
o
r
h
a
r
mf
u
l
c
o
n
t
e
n
t
d
e
t
e
c
t
i
o
n
,
”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
7
,
n
o
.
2
,
p
p
.
1
0
7
3
–
1
0
8
3
,
F
e
b
.
2
0
2
6
,
d
o
i
:
1
0
.
1
1
0
9
/
T
A
I
.
2
0
2
5
.
3
5
9
1
5
8
5
.
[
9
]
Y
.
Z
h
u
e
t
a
l
.
,
“
M
e
mo
r
y
-
g
u
i
d
e
d
m
u
l
t
i
-
v
i
e
w
mu
l
t
i
-
d
o
m
a
i
n
f
a
k
e
n
e
w
s
d
e
t
e
c
t
i
o
n
,
”
I
EEE
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
En
g
i
n
e
e
ri
n
g
,
p
p
.
1
–
1
4
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
T
K
D
E.
2
0
2
2
.
3
1
8
5
1
5
1
.
[
1
0
]
H
.
L
i
u
,
W
.
W
a
n
g
,
H
.
S
u
n
,
A
.
R
o
c
h
a
,
a
n
d
H
.
L
i
,
“
R
o
b
u
st
d
o
mai
n
mi
si
n
f
o
r
mat
i
o
n
d
e
t
e
c
t
i
o
n
v
i
a
m
u
l
t
i
-
mo
d
a
l
f
e
a
t
u
r
e
a
l
i
g
n
me
n
t
,
”
I
EEE
T
r
a
n
s
a
c
t
i
o
n
s
o
n
I
n
f
o
rm
a
t
i
o
n
Fo
ren
s
i
c
s
a
n
d
S
e
c
u
ri
t
y
,
v
o
l
.
1
9
,
p
p
.
7
9
3
–
8
0
6
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
F
S
.
2
0
2
3
.
3
3
2
6
3
6
8
.
[
1
1
]
W.
-
Y
.
L
e
e
,
W
.
H
.
H
su
,
a
n
d
S
.
S
a
t
o
h
,
“
L
e
a
r
n
i
n
g
f
r
o
m
c
r
o
ss
-
d
o
mai
n
me
d
i
a
st
r
e
a
ms
f
o
r
e
v
e
n
t
-
of
-
i
n
t
e
r
e
st
d
i
s
c
o
v
e
r
y
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
M
u
l
t
i
m
e
d
i
a
,
v
o
l
.
2
0
,
n
o
.
1
,
p
p
.
1
4
2
–
1
5
4
,
Ja
n
.
2
0
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
T
M
M
.
2
0
1
7
.
2
7
2
6
1
8
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
R
ec
o
n
f
i
g
u
r
ab
le
&
E
m
b
ed
d
ed
Sy
s
t
I
SS
N:
2089
-
4864
Meta
-
lea
r
n
in
g
co
n
tr
a
s
tive
fu
s
io
n
in
tellig
en
ce
fo
r
d
o
ma
in
-
a
d
a
p
tive
…
(
Ja
n
a
n
i
S
iva
p
r
iya
V
e
n
ka
ta
kris
h
n
a
n
)
523
[
1
2
]
B
.
L
i
u
,
“
R
e
se
a
r
c
h
o
n
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
mi
sl
e
a
d
i
n
g
c
o
n
t
e
n
t
d
e
t
e
c
t
i
o
n
me
t
h
o
d
b
a
se
d
o
n
a
t
t
e
n
t
i
o
n
me
c
h
a
n
i
sm
,
”
I
E
EE
Ac
c
e
ss
,
v
o
l
.
1
3
,
p
p
.
1
3
2
4
1
0
–
1
3
2
4
2
5
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
5
.
3
5
9
1
4
5
5
.
[
1
3
]
A
.
M
.
S
h
a
n
t
o
,
F
.
S
.
T
a
m
i
m,
M
.
S
.
J.
P
r
i
y
a
,
M
.
M
.
H
o
q
u
e
,
a
n
d
E.
H
.
P
r
i
n
c
e
,
“
T
S
F
C
:
A
h
i
e
r
a
r
c
h
i
c
a
l
c
l
a
ssi
f
i
c
a
t
i
o
n
f
r
a
me
w
o
r
k
f
o
r
mu
l
t
i
mo
d
a
l
c
o
mm
e
r
c
i
a
l
c
o
n
t
e
n
t
s
i
n
B
e
n
g
a
l
i
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
4
,
p
p
.
2
8
4
1
1
–
2
8
4
3
4
,
2
0
2
6
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
E
S
S
.
2
0
2
6
.
3
6
5
9
7
1
2
.
[
1
4
]
J.
D
e
v
l
i
n
,
M
.
-
W
.
C
h
a
n
g
,
K
.
L
e
e
,
a
n
d
K
.
T
o
u
t
a
n
o
v
a
,
“
B
E
R
T
:
P
r
e
-
t
r
a
i
n
i
n
g
o
f
d
e
e
p
b
i
d
i
r
e
c
t
i
o
n
a
l
t
r
a
n
sf
o
r
me
r
s
f
o
r
l
a
n
g
u
a
g
e
u
n
d
e
r
st
a
n
d
i
n
g
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
N
A
AC
L
-
H
L
T
2
0
1
9
,
M
a
y
2
0
1
9
,
p
p
.
4
1
7
1
–
4
1
8
6
.
[
1
5
]
K
.
H
e
,
H
.
F
a
n
,
Y
.
W
u
,
S
.
X
i
e
,
a
n
d
R
.
G
i
r
sh
i
c
k
,
“
M
o
me
n
t
u
m
c
o
n
t
r
a
st
f
o
r
u
n
su
p
e
r
v
i
se
d
v
i
s
u
a
l
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
,
”
i
n
2
0
2
0
I
EEE/
C
VF
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
e
r
V
i
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
(
C
VPR)
,
Ju
n
.
2
0
2
0
,
p
p
.
9
7
2
6
–
9
7
3
5
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
4
2
6
0
0
.
2
0
2
0
.
0
0
9
7
5
.
[
1
6
]
X
.
C
h
e
n
,
H
.
F
a
n
,
R
.
G
i
r
sh
i
c
k
,
a
n
d
K
.
H
e
,
“
I
mp
r
o
v
e
d
b
a
s
e
l
i
n
e
s
w
i
t
h
mo
me
n
t
u
m
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
,
”
a
rXi
v
Pr
e
p
ri
n
t
,
M
a
r
.
2
0
2
0
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
0
0
3
.
0
4
2
9
7
.
[
1
7
]
S
.
G
i
d
a
r
i
s,
P
.
S
i
n
g
h
,
a
n
d
N
.
K
o
mo
d
a
k
i
s
,
“
U
n
su
p
e
r
v
i
se
d
r
e
p
r
e
se
n
t
a
t
i
o
n
l
e
a
r
n
i
n
g
b
y
p
r
e
d
i
c
t
i
n
g
i
mag
e
r
o
t
a
t
i
o
n
s,
”
i
n
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
rn
i
n
g
Re
p
res
e
n
t
a
t
i
o
n
s
(
I
C
L
R)
,
2
0
1
8
.
[
1
8
]
Y
.
H
u
,
C
.
Z
h
a
n
g
,
M
.
Y
a
n
g
,
X
.
L
i
a
n
g
,
C
.
L
i
,
a
n
d
X
.
H
u
,
“
L
e
a
r
n
i
n
g
t
o
g
e
n
e
r
a
l
i
z
e
u
n
se
e
n
d
o
mai
n
s
v
i
a
m
u
l
t
i
-
so
u
r
c
e
me
t
a
l
e
a
r
n
i
n
g
f
o
r
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
P
a
t
t
e
r
n
Re
c
o
g
n
i
t
i
o
n
,
2
0
2
5
,
p
p
.
4
1
2
–
428
,
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
1
-
7
8
4
9
5
-
8
_
2
6
.
[
1
9
]
A
.
Ze
n
g
e
t
a
l
.
,
“
G
L
M
-
1
3
0
B
:
A
n
o
p
e
n
b
i
l
i
n
g
u
a
l
p
r
e
-
t
r
a
i
n
e
d
mo
d
e
l
,
”
a
rXi
v
P
rep
r
i
n
t
,
O
c
t
.
2
0
2
3
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
X
i
v
.
2
2
1
0
.
0
2
4
1
4
.
[
2
0
]
L
.
M
a
n
si
l
l
a
,
R
.
E
c
h
e
v
e
st
e
,
D
.
H
.
M
i
l
o
n
e
,
a
n
d
E
.
F
e
r
r
a
n
t
e
,
“
D
o
mai
n
g
e
n
e
r
a
l
i
z
a
t
i
o
n
v
i
a
g
r
a
d
i
e
n
t
s
u
r
g
e
r
y
,
”
i
n
2
0
2
1
I
EEE
/
C
VF
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
C
o
m
p
u
t
e
r V
i
s
i
o
n
(
I
C
C
V)
,
2
0
2
1
,
p
p
.
6
6
1
0
-
6
6
1
8
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
V
4
8
9
2
2
.
2
0
2
1
.
0
0
6
5
6
.
[
2
1
]
X
.
Y
a
o
e
t
a
l
.
,
“
P
C
L
:
P
r
o
x
y
-
b
a
se
d
c
o
n
t
r
a
s
t
i
v
e
l
e
a
r
n
i
n
g
f
o
r
d
o
m
a
i
n
g
e
n
e
r
a
l
i
z
a
t
i
o
n
,
”
i
n
2
0
2
2
I
EEE
/
C
V
F
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
e
r
Vi
si
o
n
a
n
d
Pa
t
t
e
rn
Re
c
o
g
n
i
t
i
o
n
(
C
V
P
R)
,
Ju
n
.
2
0
2
2
,
p
p
.
7
0
8
7
–
7
0
9
7
,
d
o
i
:
1
0
.
1
1
0
9
/
C
V
P
R
5
2
6
8
8
.
2
0
2
2
.
0
0
6
9
6
.
[
2
2
]
Q
.
T
a
n
,
R
.
H
e
,
L
.
B
i
n
g
,
a
n
d
H
.
T
.
N
g
,
“
D
o
mai
n
g
e
n
e
r
a
l
i
z
a
t
i
o
n
f
o
r
t
e
x
t
c
l
a
ssi
f
i
c
a
t
i
o
n
w
i
t
h
me
mo
r
y
-
b
a
se
d
su
p
e
r
v
i
se
d
c
o
n
t
r
a
s
t
i
v
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
9
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
2
0
2
2
,
p
p
.
6
9
1
6
–
6
9
2
6
.
[
2
3
]
T
.
D
o
p
i
e
r
r
e
,
C
.
G
r
a
v
i
e
r
,
a
n
d
W
.
L
o
g
e
r
a
i
s,
“
A
n
e
u
r
a
l
f
e
w
-
sh
o
t
t
e
x
t
c
l
a
s
si
f
i
c
a
t
i
o
n
r
e
a
l
i
t
y
c
h
e
c
k
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
1
6
t
h
C
o
n
f
e
re
n
c
e
o
f
t
h
e
Eu
r
o
p
e
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Asso
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
(
EA
C
L
)
,
2
0
2
1
,
p
p
.
9
3
5
–
9
4
3
.
[
2
4
]
Z.
-
X
.
Y
e
a
n
d
Z
.
-
H
.
L
i
n
g
,
“
M
u
l
t
i
-
l
e
v
e
l
mat
c
h
i
n
g
a
n
d
a
g
g
r
e
g
a
t
i
o
n
n
e
t
w
o
r
k
f
o
r
f
e
w
-
sh
o
t
r
e
l
a
t
i
o
n
c
l
a
ssi
f
i
c
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
5
7
t
h
A
n
n
u
a
l
Me
e
t
i
n
g
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s
,
2
0
1
9
,
p
p
.
2
8
7
2
–
2
8
8
1
.
[
2
5
]
H
.
Z
h
e
n
g
,
R
.
W
a
n
g
,
Y
.
Y
a
n
g
,
Y
.
L
i
,
a
n
d
M
.
X
u
,
“
I
n
t
e
l
l
i
g
e
n
t
f
a
u
l
t
i
d
e
n
t
i
f
i
c
a
t
i
o
n
b
a
se
d
o
n
mu
l
t
i
so
u
r
c
e
d
o
ma
i
n
g
e
n
e
r
a
l
i
z
a
t
i
o
n
t
o
w
a
r
d
s
a
c
t
u
a
l
d
i
a
g
n
o
si
s
s
c
e
n
a
r
i
o
,
”
I
EE
E
T
ra
n
s
a
c
t
i
o
n
s
o
n
I
n
d
u
s
t
ri
a
l
El
e
c
t
ro
n
i
c
s
,
v
o
l
.
6
7
,
n
o
.
2
,
p
p
.
1
2
9
3
–
1
3
0
4
,
F
e
b
.
2
0
2
0
,
d
o
i
:
1
0
.
1
1
0
9
/
TI
E.
2
0
1
9
.
2
8
9
8
6
1
9
.
[
2
6
]
H
.
Z
h
a
n
g
e
t
a
l
.
,
“
Ef
f
e
c
t
i
v
e
n
e
ss o
f
p
r
e
-
t
r
a
i
n
i
n
g
f
o
r
f
e
w
-
sh
o
t
i
n
t
e
n
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
Fi
n
d
i
n
g
s
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
EM
N
L
P
2
0
2
1
,
2
0
2
1
,
p
p
.
1
1
1
4
–
1
1
2
0
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
J
a
n
a
n
i
S
iv
a
p
r
iy
a
V
e
n
k
a
t
a
k
r
is
h
n
a
n
w
o
rk
in
g
a
s
a
n
A
s
sista
n
t
P
r
o
f
e
ss
o
r
in
th
e
De
p
a
rtme
n
t
o
f
Da
ta
S
c
ien
c
e
a
t
Dw
a
r
a
k
a
Do
ss
G
o
v
e
rd
h
a
n
Do
ss
V
a
ish
n
a
v
Co
ll
e
g
e
(A
u
to
n
o
m
o
u
s),
A
ff
il
iate
d
to
Un
iv
e
rsit
y
o
f
M
a
d
ra
s,
C
h
e
n
n
a
i,
T
a
m
il
Na
d
u
.
S
h
e
h
a
s
se
c
u
re
d
h
e
r
M.Sc
.
d
e
g
re
e
in
S
o
f
tw
a
re
S
c
ien
c
e
f
ro
m
JK
KN
Co
ll
e
g
e
o
f
A
rts
a
n
d
S
c
ien
c
e
,
Aff
il
iate
d
to
P
e
riy
a
r
Un
iv
e
rsit
y
,
S
a
lem
a
n
d
h
e
r
M
.
P
h
il
.
in
Co
m
p
u
ter
S
c
ien
c
e
f
ro
m
Dh
a
n
ra
j
Ba
id
Ja
in
Co
l
leg
e
(A
u
to
n
o
m
o
u
s),
A
ff
il
iate
d
to
Un
iv
e
rsit
y
o
f
M
a
d
ra
s,
Ch
e
n
n
a
i.
F
r
o
m
De
c
e
m
b
e
r
2
0
2
3
,
s
h
e
h
a
s
b
e
e
n
a
P
h
.
D.
sc
h
o
lar
in
C
o
m
p
u
ter
S
c
i
e
n
c
e
a
t
Ko
n
g
u
n
a
d
u
A
rts
a
n
d
S
c
i
e
n
c
e
Co
ll
e
g
e
(
A
u
to
n
o
m
o
u
s),
Aff
il
iate
d
to
Bh
a
ra
t
h
iar
Un
iv
e
rsit
y
,
Co
i
m
b
a
to
re
.
He
r
m
a
in
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
a
ta
sc
ien
c
e
a
n
d
m
a
c
h
in
e
lea
rn
in
g
,
tex
t
m
in
in
g
,
a
n
d
n
a
tu
ra
l
lan
g
u
a
g
e
p
ro
c
e
ss
i
n
g
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
jan
a
n
i.
siv
a
p
riy
a
phd
@
g
m
a
i
l.
c
o
m
.
S
a
r
a
v
a
n
a
M
o
o
r
t
h
y
Ra
j
a
M
o
o
r
th
y
w
o
rk
in
g
a
s
A
ss
o
c
iate
P
ro
f
e
ss
o
r
a
n
d
He
a
d
,
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
(S
F
),
K
o
n
g
u
n
a
d
u
A
rts
a
n
d
S
c
ien
c
e
Co
ll
e
g
e
,
Co
im
b
a
to
re
,
T
a
m
il
n
a
d
u
.
He
se
c
u
re
d
h
is
Do
c
o
ra
te
in
Co
m
p
u
ter
E
n
g
in
e
e
rin
g
a
n
d
p
o
st
g
ra
d
u
a
te
d
e
g
re
e
i
n
S
o
f
tw
a
r
e
En
g
in
e
e
rin
g
.
He
is
in
te
a
c
h
in
g
p
ro
f
e
ss
io
n
f
o
r
m
o
re
th
a
n
1
8
y
e
a
rs.
He
h
a
s
p
re
se
n
ted
2
0
p
a
p
e
rs
in
n
a
ti
o
n
a
l
a
n
d
i
n
tern
a
ti
o
n
a
l
jo
u
rn
a
l
,
c
o
n
f
re
n
c
e
a
n
d
se
m
in
a
rs.
He
is
re
g
istere
d
a
s
a
n
a
u
th
o
r
in
G
o
o
g
le
S
c
h
o
lar,
Re
se
a
rc
h
-
Ga
te
a
n
d
Orc
id
.
His
m
a
in
a
re
a
o
f
i
n
tere
st
in
c
lu
d
e
s
A
I
a
n
d
m
a
c
h
in
e
lea
rn
in
g
.
He
h
a
d
se
tu
p
i
n
A
S
IAN
-
Bo
o
k
o
f
Re
c
o
rd
f
o
r
c
o
n
d
u
c
ti
n
g
n
o
n
st
o
p
4
8
-
h
o
u
r
c
o
n
c
lav
e
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
d
rs
a
ra
v
a
n
a
m
o
o
rth
y
@
g
m
a
il
.
c
o
m
.
An
g
e
l
S
h
a
n
m
u
g
a
m
w
o
rk
in
g
a
s
A
ss
ist
a
n
t
P
ro
f
e
ss
o
r,
De
p
a
rt
m
e
n
t
o
f
Co
m
p
u
ter
S
c
ien
c
e
w
it
h
Cy
b
e
r
S
e
c
u
rit
y
,
P
S
G
R
Krish
n
a
m
m
a
l
Co
ll
e
g
e
f
o
r
W
o
m
e
n
,
Co
im
a
b
a
to
re
.
S
h
e
S
e
c
u
re
d
h
e
r
D
o
c
to
ra
te
i
n
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
p
o
st
g
ra
d
u
a
te
d
e
g
re
e
in
M
.
E
.
C
o
m
p
u
ter
En
g
in
e
e
rin
g
.
S
h
e
h
a
s
se
c
u
re
d
h
e
r
M
.
Sc
.
d
e
g
re
e
in
S
o
f
twa
re
S
c
ien
c
e
f
ro
m
JK
KN
Co
ll
e
g
e
o
f
A
rts
a
n
d
S
c
ien
c
e
,
A
ff
il
iate
d
to
P
e
riy
a
r
Un
iv
e
rsit
y
,
S
a
le
m
.
S
h
e
is
in
tea
c
h
in
g
p
ro
f
e
ss
io
n
f
o
r
m
o
re
th
a
n
1
4
y
e
a
rs.
S
h
e
h
a
s
p
re
se
n
ted
2
0
p
a
p
e
rs
in
n
a
ti
o
n
a
l
a
n
d
in
ter
n
a
ti
o
n
a
l
jo
u
rn
a
l
,
c
o
n
f
e
re
n
c
e
a
n
d
se
m
in
a
rs.
S
h
e
is r
e
g
istere
d
a
s a
n
a
u
th
o
r
i
n
G
o
o
g
le
S
c
h
o
lar
a
n
d
Re
se
a
rc
h
G
a
te.
He
r
m
a
in
a
re
a
o
f
in
tere
st
in
c
lu
d
e
s
A
I,
m
a
c
h
in
e
lea
rn
in
g
,
a
n
d
c
y
b
e
r
se
c
u
rit
y
.
Sh
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
n
g
e
lsh
a
n
m
u
g
a
m
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.