I
n
d
on
e
s
i
a
n
Jo
u
r
n
al
of
E
l
e
c
t
r
i
c
al
E
n
g
i
n
e
e
r
i
n
g a
n
d
C
om
p
u
t
e
r
S
c
i
e
n
c
e
V
ol
.
40
, N
o.
2
,
N
ov
e
m
b
e
r
2025
, pp.
11
00
~
110
8
I
S
S
N
:
2502
-
4752, D
O
I
:
10.11591/
i
j
e
e
c
s
.
v
40
.i
2
.
pp
11
00
-
11
08
1100
Jou
r
n
al
h
om
e
page
:
ht
t
p:
/
/
i
j
e
e
c
s
.i
ae
s
c
or
e
.c
om
E
n
h
a
n
c
i
n
g
d
oc
u
m
e
n
t
t
e
x
t
c
l
ass
i
f
i
c
at
i
o
n
u
si
n
g
h
y
b
r
i
d
d
e
e
p
c
o
n
t
e
x
t
u
al
a
n
d
c
or
r
e
l
at
i
o
n
n
e
t
w
or
k
S
h
i
l
p
a,
S
h
r
i
d
e
vi
S
om
a
D
e
pa
r
t
m
e
n
t
of
C
o
m
put
e
r
S
c
i
e
n
c
e
a
n
d
E
ng
i
n
e
e
r
i
ng
,
P
D
A
C
ol
l
e
g
e
of
E
ngi
n
e
e
r
i
ng,
V
i
s
ve
s
va
r
a
ya
T
e
c
h
n
ol
o
gi
c
a
l
U
ni
ve
r
s
i
t
y,
B
e
l
a
g
a
vi
,
I
ndi
a
A
r
t
i
c
l
e
I
n
f
o
A
B
S
T
R
A
C
T
A
r
t
i
c
l
e
h
i
s
t
o
r
y
:
R
e
c
e
i
v
e
d
J
ul
27
,
2024
R
e
v
i
s
e
d
A
u
g
28
,
2025
A
c
c
e
pt
e
d
O
c
t
15
,
2025
Do
c
u
m
e
n
t
a
n
a
l
y
s
i
s
i
n
v
o
l
v
e
s
th
e
e
x
tr
a
c
t
i
o
n
a
n
d
p
r
o
c
e
s
s
i
n
g
o
f
i
n
f
o
r
m
a
ti
o
n
f
r
o
m
d
o
c
u
m
e
n
ts
,
a
tas
k
i
n
c
r
e
a
s
i
n
g
l
y
a
u
to
m
a
te
d
th
r
o
u
g
h
th
e
u
s
e
o
f
d
e
e
p
l
e
a
r
n
i
n
g
(
DL
)
tec
h
n
o
l
o
g
i
e
s
.
De
s
p
i
te
th
e
h
ig
h
p
r
e
d
i
c
t
iv
e
p
o
we
r
o
f
DL
m
o
d
e
l
s
,
th
e
i
r
b
l
a
c
k
-
b
o
x
n
a
tu
r
e
p
o
s
e
s
c
h
a
l
l
e
n
g
e
s
to
t
r
a
n
s
p
a
r
e
n
c
y
a
n
d
i
n
ter
p
r
e
tab
il
i
ty
,
h
i
n
d
e
r
i
n
g
th
e
ir
i
n
teg
r
a
t
i
o
n
i
n
to
th
e
i
n
d
u
s
tr
y
.
T
h
i
s
p
a
p
e
r
i
n
tr
o
d
u
c
e
s
th
e
h
y
b
r
i
d
d
e
e
p
c
o
n
tex
t
u
a
l
a
n
d
c
o
r
r
e
l
a
t
i
o
n
n
e
two
r
k
(
HD
C
C
Ne
t)
,
a
n
o
v
e
l
m
e
th
o
d
o
l
o
g
y
d
e
s
i
g
n
e
d
to
i
m
p
r
o
v
e
b
o
th
t
h
e
a
c
c
u
r
a
c
y
a
n
d
i
n
ter
p
r
e
tab
il
i
ty
o
f
m
u
l
ti
-
c
a
teg
o
r
y
c
l
a
s
s
i
f
i
c
a
ti
o
n
tas
k
s
.
HD
C
C
Ne
t
l
e
v
e
r
a
g
e
s
a
h
y
b
r
i
d
l
a
y
e
r
c
a
teg
o
r
y
c
o
r
r
e
l
a
t
i
o
n
m
o
d
u
l
e
to
d
e
e
p
e
n
c
a
teg
o
r
y
c
o
n
n
e
c
ti
o
n
s
,
th
e
r
e
b
y
e
n
h
a
n
c
i
n
g
th
e
u
n
d
e
r
s
tan
d
i
n
g
a
n
d
p
r
e
d
i
c
ti
o
n
o
f
c
a
te
g
o
r
y
i
n
ter
r
e
l
a
t
i
o
n
s
.
T
o
a
d
d
r
e
s
s
p
o
ten
ti
a
l
p
r
e
d
i
c
t
i
o
n
d
i
v
e
r
g
e
n
c
e
,
r
e
s
i
d
u
a
l
c
o
n
n
e
c
ti
o
n
s
a
r
e
i
n
c
o
r
p
o
r
a
ted
,
e
n
s
u
r
i
n
g
s
tab
l
e
a
n
d
r
e
li
a
b
l
e
p
e
r
f
o
r
m
a
n
c
e
.
F
u
r
th
e
r
m
o
r
e
,
HD
C
C
Ne
t
r
e
d
u
c
e
s
m
o
d
e
l
p
a
r
a
m
e
ter
s
,
a
c
c
e
ler
a
t
i
n
g
c
o
n
v
e
r
g
e
n
c
e
a
n
d
m
a
k
i
n
g
th
e
m
o
d
e
l
m
o
r
e
e
f
f
icie
n
t.
T
h
i
s
e
f
f
icie
n
c
y
i
s
p
a
r
ti
c
u
l
a
r
l
y
b
e
n
e
f
i
c
i
a
l
f
o
r
p
r
a
c
ti
c
a
l
a
p
p
li
c
a
t
i
o
n
s
,
a
ll
o
w
i
n
g
f
a
s
ter
d
e
p
l
o
y
m
e
n
t
a
n
d
s
c
a
l
a
b
i
li
ty
.
B
y
b
r
i
d
g
i
n
g
th
e
g
a
p
b
e
twe
e
n
DL
’
s
c
a
p
a
b
i
li
t
i
e
s
a
n
d
i
n
d
u
s
tr
y
n
e
e
d
s
f
o
r
tr
a
n
s
p
a
r
e
n
c
y
,
HD
C
C
Ne
t
p
r
o
v
i
d
e
s
a
r
o
b
u
s
t
s
o
l
u
ti
o
n
f
o
r
a
u
to
m
a
ted
d
o
c
u
m
e
n
t
p
r
o
c
e
s
s
i
n
g
,
p
a
v
i
n
g
th
e
wa
y
f
o
r
b
r
o
a
d
e
r
a
d
o
p
ti
o
n
o
f
DL
tec
h
n
o
l
o
g
i
e
s
i
n
b
u
s
i
n
e
s
s
e
n
v
i
r
o
n
m
e
n
t
s
.
K
e
y
w
or
d
s
:
D
e
e
p l
e
a
r
ni
n
g
H
D
C
C
N
e
t
N
a
t
ur
a
l
l
a
ng
ua
g
e
p
r
oc
e
s
s
i
ng
T
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on
T
e
xt
r
e
p
r
e
s
e
nt
a
t
i
on
T
h
i
s
i
s
a
n
o
p
e
n
a
c
c
e
s
s
a
r
ti
c
l
e
u
n
d
e
r
th
e
C
C
B
Y
-
SA
l
i
c
e
n
s
e
.
C
or
r
e
s
pon
di
n
g A
u
t
h
or
:
S
hi
l
pa
D
e
pa
r
t
m
e
nt
of
C
o
m
put
e
r
S
c
i
e
nc
e
a
nd E
ng
i
ne
e
r
i
ng
, P
D
A
C
ol
l
e
g
e
of
E
n
g
i
ne
e
r
i
n
g
V
i
s
v
e
s
v
a
r
a
ya
T
e
c
hnol
og
i
c
a
l
U
ni
v
e
r
s
i
t
y
B
e
l
a
g
a
v
i
,
I
ndi
a
E
m
a
i
l
:
s
hi
l
pa
_122023
@
r
e
di
f
f
m
a
i
l
.c
o
m
1.
I
N
T
R
O
D
U
C
T
I
O
N
D
e
e
p l
e
a
r
ni
ng
(
D
L
)
ha
s
m
a
de
m
a
j
or
a
d
v
a
nc
e
m
e
nt
s
i
n doc
u
m
e
nt
a
n
a
l
y
s
i
s
o
v
e
r
t
he
pa
s
t
f
e
w
y
e
a
r
s
[
1]
.
T
hi
s
m
e
t
hod
of
f
e
r
s
a
l
ot
of
pot
e
nt
i
a
l
f
or
a
ut
o
m
a
t
i
ng
a
d
m
i
ni
s
t
r
a
t
i
v
e
t
a
s
ks
r
e
l
a
t
e
d
t
o
pr
oc
e
s
s
i
ng
doc
u
m
e
nt
s
.
I
n
pr
e
di
c
t
i
on
t
a
s
ks
,
de
e
p
n
e
ur
a
l
ne
t
w
or
ks
h
a
v
e
pr
o
v
e
n
t
o
pe
r
f
or
m
w
e
l
l
.
H
o
w
e
v
e
r
,
one
s
i
gni
f
i
c
a
nt
ba
r
r
i
e
r
t
o
t
he
s
a
f
e
i
nt
e
g
r
a
t
i
on
of
t
he
s
e
t
e
c
hnol
og
i
e
s
i
nt
o
c
or
por
a
t
e
pr
oc
e
s
s
e
s
c
ont
i
nue
s
t
o
be
a
l
a
c
k
of
t
r
a
ns
pa
r
e
nc
y.
U
ns
t
r
uc
t
ur
e
d
da
t
a
w
a
s
f
r
e
que
nt
l
y
us
e
d
t
o
r
e
pr
e
s
e
nt
t
e
xt
da
t
a
i
n
t
he
e
a
r
l
y
da
y
s
of
d
i
g
i
t
a
l
c
o
m
put
i
ng.
T
e
c
hnol
og
i
c
a
l
a
d
v
a
n
c
e
s
i
n
t
e
xt
da
t
a
s
t
or
a
g
e
h
a
v
e
l
e
d
t
o
a
m
a
j
or
gr
o
w
t
h
i
n
t
he
f
i
e
l
d
of
i
nf
or
m
a
t
i
on
r
e
t
r
i
e
v
a
l
(
I
R
)
.
T
ha
nks
t
o
a
d
v
a
nc
e
s
i
n
t
e
c
hnol
og
y
,
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on
a
nd
t
e
xt
ua
l
da
t
a
pr
oc
e
s
s
i
ng
m
a
y
no
w
be
don
e
m
o
r
e
a
ut
o
m
a
t
i
c
a
l
l
y
[
2]
.
E
xpe
r
t
s
c
o
nc
l
ude
d
a
t
t
he
be
g
i
nni
ng
of
t
he
pr
oj
e
c
t
t
ha
t
m
a
t
he
m
a
t
i
c
a
l
i
nde
xi
ng
w
a
s
i
ns
uf
f
i
c
i
e
nt
t
o
a
t
t
a
i
n
hi
g
he
r
l
e
v
e
l
s
of
a
c
c
ur
a
c
y
.
A
s
t
he
be
ne
f
i
t
s
of
a
l
l
ow
i
ng
c
o
m
put
e
r
s
t
o
c
om
p
r
e
he
nd
a
nd
i
nt
e
r
pr
e
t
hu
m
a
n
l
a
n
g
ua
g
e
be
c
o
m
e
m
or
e
a
ppa
r
e
nt
,
na
t
ur
a
l
l
a
ng
u
a
g
e
pr
oc
e
s
s
i
ng
(
N
L
P
)
,
i
s
be
c
o
m
i
n
g
m
or
e
a
n
d
m
o
r
e
popul
a
r
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
E
nhanc
i
ng
d
oc
um
e
nt
t
e
x
t
c
l
as
s
i
f
i
c
at
i
on us
i
n
g hy
b
r
i
d
d
e
e
p c
ont
e
x
t
ual
an
d c
or
r
e
l
at
i
o
n ne
t
w
or
k
(
Shi
l
pa
)
1101
T
he
r
e
a
r
e
s
e
v
e
r
a
l
not
e
w
or
t
h
y
s
i
m
i
l
a
r
i
t
i
e
s
a
c
r
os
s
v
a
r
i
ous
doc
u
m
e
nt
c
ol
l
e
c
t
i
ons
.
W
he
n
t
he
d
e
s
t
i
na
t
i
on
dom
a
i
n
i
s
s
i
m
i
l
a
r
t
o
or
e
qua
l
t
o
t
he
s
our
c
e
dom
a
i
n,
t
r
a
ns
f
e
r
l
e
a
r
ni
n
g
i
s
t
he
pr
oc
e
s
s
of
m
o
vi
ng
know
l
e
d
g
e
or
i
nf
or
m
a
t
i
o
n f
r
o
m
one
do
m
a
i
n t
o a
not
he
r
[
3]
.
U
s
i
n
g
t
r
a
ns
f
e
r
l
e
a
r
ni
n
g
t
e
c
hni
que
s
oc
c
a
s
i
ona
l
l
y
r
e
qui
r
e
s
s
pe
c
i
f
i
c
f
e
a
t
ur
e
s
t
o
e
xi
s
t
i
n
bot
h
t
he
s
our
c
e
a
nd
de
s
t
i
na
t
i
on
dom
a
i
ns
.
I
t
m
i
g
ht
be
di
f
f
i
c
ul
t
t
o
t
r
a
i
n
t
r
a
ns
f
e
r
l
e
a
r
ni
n
g
a
l
g
or
i
t
h
m
s
f
or
t
he
t
a
r
g
e
t
do
m
a
i
n
c
l
a
s
s
i
f
i
e
r
,
e
s
pe
c
i
a
l
l
y
i
f
t
he
t
a
r
g
e
t
t
e
xt
s
l
a
c
k
c
a
t
e
g
or
i
e
s
.
O
ne
pr
obl
e
m
i
s
us
i
ng
t
e
r
m
i
nol
ogy
s
pe
c
i
f
i
c
t
o
t
he
de
s
t
i
na
t
i
on
dom
a
i
n
t
ha
t
doe
s
not
e
xi
s
t
i
n
t
he
s
our
c
e
dom
a
i
n.
D
oc
um
e
nt
s
a
r
e
f
r
e
que
nt
l
y
c
a
t
e
g
o
r
i
z
e
d
i
nt
o
one
or
m
or
e
g
r
oups
w
i
t
hi
n
a
hi
g
h
-
di
m
e
ns
i
ona
l
,
s
pa
r
s
e
s
pa
c
e
dur
i
ng
t
he
c
l
a
s
s
i
f
i
c
a
t
i
on
pr
oc
e
s
s
.
N
u
m
e
r
ous
s
t
a
t
i
s
t
i
c
a
l
a
nd
a
r
t
i
f
i
c
i
a
l
i
nt
e
l
l
i
g
e
nc
e
m
e
t
hods
ha
v
e
be
e
n
us
e
d
t
o
s
ol
v
e
t
he
pr
obl
e
m
of
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on [
4]
.
T
he
f
our
pr
i
m
a
r
y
s
t
a
g
e
s
of
doc
u
m
e
nt
c
a
t
e
g
o
r
i
z
a
t
i
on
a
r
e
pr
e
-
p
r
oc
e
s
s
i
ng
,
doc
u
m
e
nt
p
r
e
di
c
t
i
on,
c
l
a
s
s
i
f
i
e
r
t
r
a
i
ni
ng
,
a
nd
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on.
D
upl
i
c
a
t
e
da
t
a
a
r
e
r
e
m
o
v
e
d
a
t
t
he
pr
e
-
pr
oc
e
s
s
i
ng
s
t
a
g
e
of
da
t
a
a
na
l
y
s
i
s
.
W
or
ds
t
ha
t
a
r
e
unc
o
m
m
on
i
n
t
hi
s
s
pe
c
i
f
i
c
c
ont
e
xt
a
r
e
e
v
a
l
ua
t
e
d
a
s
pa
r
t
of
t
he
r
e
v
i
e
w
pr
oc
e
s
s
.
N
-
g
r
a
m
m
ode
l
s
a
r
e
f
r
e
qu
e
nt
l
y
us
e
d
i
n
a
w
i
de
r
a
n
ge
of
a
ppl
i
c
a
t
i
ons
f
or
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on.
O
ne
of
t
he
m
ode
l
s
us
e
d
m
os
t
f
r
e
que
nt
l
y
i
n
N
L
P
i
s
t
he
ba
g
-
of
-
w
or
ds
m
ode
l
.
W
he
n
us
e
r
-
g
e
ne
r
a
t
e
d
i
nf
or
m
a
t
i
on
i
s
e
nc
l
os
e
d
i
n
s
qua
r
e
br
a
c
ke
t
s
a
nd
f
ol
l
ow
e
d
by
a
nu
m
be
r
,
i
t
i
s
r
e
g
a
r
de
d
a
s
a
s
our
c
e
c
i
t
a
t
i
on.
T
hi
s
s
e
t
up
ha
s
be
e
n
i
ns
t
a
l
l
e
d.
T
e
r
m
s
t
ha
t
a
r
e
r
e
g
ul
a
r
l
y
us
e
d
ha
v
e
a
pr
ope
r
t
y
c
a
l
l
e
d
t
e
r
m
f
r
e
que
n
c
y
(
T
F
)
,
w
hi
c
h
i
s
t
he
ba
s
i
c
c
onc
e
pt
unde
r
l
y
i
ng
t
he
ba
g
-
of
-
w
or
ds
a
ppr
oa
c
h.
D
oc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
i
s
ba
s
e
d
on
t
he
i
de
a
s
of
t
e
r
m
f
r
e
que
n
c
y
a
n
d
i
nv
e
r
s
e
doc
u
m
e
nt
f
r
e
que
nc
y
(
T
F
-
I
D
F
)
,
w
hi
c
h
c
o
m
bi
ne
s
T
F
a
nd
I
D
F
w
i
t
h
I
D
F
[
5]
.
T
he
T
e
ns
or
F
l
ow
(
T
F
)
f
r
a
m
e
w
or
k
i
s
a
f
unda
m
e
nt
a
l
c
o
m
pone
nt
of
our
w
or
k.
A
t
e
xt
c
l
a
s
s
i
f
i
e
r
’
s
c
a
pa
c
i
t
y
t
o
r
a
pi
dl
y
i
nc
r
e
a
s
e
r
e
a
da
bi
l
i
t
y
t
hr
oug
h
w
or
d
f
r
e
que
nc
y
i
s
one
of
i
t
s
ke
y
a
d
v
a
nt
a
g
e
s
.
B
y
c
onc
e
nt
r
a
t
i
ng
on
t
he
m
os
t
c
r
u
c
i
a
l
s
e
c
t
i
ons
of
t
he
doc
um
e
nt
, t
he
c
l
a
s
s
i
f
i
e
r
’
s
s
t
udy
m
a
y
l
oc
a
t
e
t
he
ke
yw
or
ds
a
s
s
oc
i
a
t
e
d
w
i
t
h t
he
c
ont
e
nt
.
H
o
w
e
v
e
r
,
i
t
’
s
c
r
uc
i
a
l
t
o
r
e
m
e
m
be
r
t
ha
t
books
c
a
n
i
nc
l
ude
a
l
o
t
of
t
e
c
hni
c
a
l
j
a
r
g
on.
T
he
c
o
m
bi
na
t
i
o
n
of
t
he
ba
g
-
of
-
w
o
r
ds
a
r
c
hi
t
e
c
t
ur
e
a
nd
t
he
i
nt
e
g
r
a
t
e
d
f
r
e
que
nc
y
-
ba
s
e
d
f
e
a
t
ur
e
s
y
i
e
l
ds
a
s
pa
r
s
e
a
nd
v
a
s
t
f
e
a
t
ur
e
s
pa
c
e
.
A
l
g
or
i
t
h
m
i
c
f
e
a
t
ur
e
s
e
l
e
c
t
i
on
c
a
n
a
s
s
i
s
t
i
n
e
f
f
i
c
i
e
nt
l
y
a
nd
r
a
pi
dl
y
l
oc
a
t
i
ng
a
c
e
r
t
a
i
n
c
ol
l
e
c
t
i
on
of
pe
r
t
i
ne
nt
qua
l
i
t
i
e
s
[
6]
.
T
o
bui
l
d
m
ode
l
s
,
t
he
s
u
ppor
t
v
e
c
t
or
m
a
c
hi
n
e
s
(
S
V
M
)
m
a
c
hi
ne
l
e
a
r
ni
ng
t
e
c
hni
que
m
a
ke
s
us
e
of
c
e
r
t
a
i
n
a
t
t
r
i
but
e
s
.
T
he
g
e
ne
t
i
c
pr
o
g
r
a
m
m
i
n
g
(
G
P
)
a
ppr
oa
c
h
a
l
l
ow
s
t
he
c
om
put
e
r
t
o
a
ut
onom
ous
l
y
c
hoos
e
a
s
e
t
of
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
w
hi
l
e
bui
l
di
ng
c
l
a
s
s
i
f
i
e
r
s
,
i
n
c
ont
r
a
s
t
t
o
m
ode
l
s
t
ha
t
r
e
l
y
on
pr
e
s
e
l
e
c
t
e
d
f
e
a
t
ur
e
s
[
7]
.
M
or
e
o
v
e
r
,
c
on
v
e
r
t
i
ng
w
or
ds
i
nt
o
num
e
r
i
c
a
l
v
e
c
t
or
s
i
s
a
ke
y
s
t
one
of
DL
m
e
t
hodol
og
i
e
s
. C
l
a
s
s
i
f
i
e
r
s
s
houl
d be
t
r
a
i
ne
d us
i
ng
t
he
c
on
v
e
r
s
i
on pr
oc
e
dur
e
.
D
oc
u
m
e
nt
s
e
gm
e
nt
a
t
i
on
s
ol
ut
i
on
s
by
di
v
i
di
ng
l
e
n
g
t
h
y
t
e
xt
s
i
nt
o
s
m
a
l
l
e
r
,
m
o
r
e
di
g
e
s
t
i
bl
e
phr
a
s
e
s
o
r
c
hunks
by
us
i
n
g
s
l
i
de
r
w
i
ndow
s
or
s
e
que
nt
i
a
l
c
ut
t
i
ng
a
l
g
or
i
t
h
m
s
.
W
he
n
a
r
r
a
n
g
e
d
hi
e
r
a
r
c
hi
c
a
l
l
y
,
phr
a
s
e
s
or
s
e
c
t
i
ons
a
ppe
a
r
be
f
or
e
t
he
e
nt
i
r
e
r
e
pr
e
s
e
nt
a
t
i
on
of
t
he
t
e
xt
.
S
a
i
f
ul
l
a
h
e
t
a
l
.
[
8]
u
s
e
d
a
t
e
c
hn
i
que
i
n
t
he
i
r
s
t
udy
t
ha
t
i
nv
ol
v
e
d
br
e
a
ki
ng
up
t
he
r
a
w
da
t
a
i
nt
o
s
m
a
l
l
e
r
pi
e
c
e
s
.
T
he
uni
t
s
w
e
r
e
t
he
n
a
na
l
y
z
e
d
us
i
ng
t
he
B
E
R
T
ba
s
i
c
m
ode
l
.
F
or
c
l
a
s
s
i
f
i
c
a
t
i
on,
t
hi
s
m
ode
l
m
a
ke
s
us
e
of
on
e
r
e
c
u
r
r
e
nt
l
a
y
e
r
o
r
one
e
xt
r
a
t
r
a
ns
f
or
m
e
r
.
O
nc
e
t
hi
s
f
e
a
t
ur
e
i
s
put
i
nt
o
pr
a
c
t
i
c
e
,
a
l
l
of
t
he
out
pu
t
s
m
a
y
be
s
ha
r
e
d,
w
hi
c
h
m
a
ke
s
us
e
of
a
n
i
n
t
e
r
a
c
t
i
v
e
t
r
a
ns
f
or
m
e
r
m
ode
l
,
a
nd
i
m
p
r
o
v
e
d
s
e
nt
e
nc
e
m
ode
l
i
ng
e
v
e
n
m
or
e
.
S
e
nt
e
nc
e
r
e
p
r
e
s
e
nt
a
t
i
on
a
c
c
ur
a
c
y
i
s
i
m
pr
o
v
e
d
s
i
nc
e
t
hi
s
m
e
t
hod
c
ons
i
de
r
s
t
he
c
ont
e
xt
of
t
he
f
ul
l
doc
u
m
e
nt
.
T
he
e
xi
s
t
i
ng
a
ppr
oa
c
he
s
,
w
hi
c
h
a
s
s
um
e
hi
e
r
a
r
c
hi
c
a
l
r
e
p
r
e
s
e
nt
a
t
i
ons
f
or
l
ong
t
e
xt
s
,
of
t
e
n
f
a
i
l
t
o
c
a
pt
ur
e
t
he
i
nt
e
r
a
c
t
i
ons
a
c
r
os
s
s
e
v
e
r
a
l
f
e
a
t
ur
e
l
e
v
e
l
s
a
nd
s
t
r
u
gg
l
e
t
o
c
o
m
m
uni
c
a
t
e
i
nf
or
m
a
t
i
on
a
t
t
he
s
a
m
e
l
e
v
e
l
.
H
o
w
e
v
e
r
,
t
he
c
ur
r
e
nt
a
l
g
or
i
t
h
m
s
ha
v
e
not
f
ul
l
y
a
s
s
e
s
s
e
d t
he
s
t
r
uc
t
ur
a
l
i
nf
or
m
a
t
i
on i
nc
l
ude
d i
n l
e
ng
t
hy
t
e
xt
s
.
D
L
ha
s
r
e
v
ol
u
t
i
oni
z
e
d
doc
um
e
nt
a
na
l
y
s
i
s
,
of
f
e
r
i
ng
s
i
g
ni
f
i
c
a
nt
pot
e
nt
i
a
l
t
o
a
ut
om
a
t
e
doc
u
m
e
n
t
pr
oc
e
s
s
i
ng
i
n bus
i
ne
s
s
w
or
kf
l
o
w
s
. D
e
s
pi
t
e
t
he
i
r
pr
e
di
c
t
i
v
e
po
w
e
r
, t
he
bl
a
c
k
-
box na
t
ur
e
of
D
L
m
ode
l
s
hi
nde
r
s
s
a
f
e
i
nt
e
g
r
a
t
i
on
i
n
t
he
i
nd
us
t
r
y
due
t
o
a
l
a
c
k
of
t
r
a
ns
pa
r
e
nc
y
a
nd
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
.
B
us
i
ne
s
s
e
s
r
e
qui
r
e
not
onl
y
a
c
c
ur
a
t
e
m
ode
l
s
but
a
l
s
o
one
s
t
ha
t
c
a
n
be
unde
r
s
t
oo
d
a
nd
t
r
us
t
e
d
by
us
e
r
s
,
m
a
ki
ng
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
a
c
r
i
t
i
c
a
l
f
a
c
t
or
i
n
a
dopt
i
ng
D
L
s
ol
ut
i
ons
.
T
hi
s
pa
pe
r
a
ddr
e
s
s
e
s
t
he
c
ha
l
l
e
n
g
e
of
e
nha
nc
i
ng
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
w
hi
l
e
m
a
i
nt
a
i
ni
ng
hi
g
h
pe
r
f
or
m
a
nc
e
i
n
doc
um
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
t
a
s
ks
.
I
n
t
he
r
e
a
l
m
of
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on,
t
r
a
di
t
i
ona
l
m
e
t
hods
ha
v
e
r
e
l
i
e
d
he
a
v
i
l
y
on
m
a
nua
l
f
e
a
t
ur
e
e
ng
i
ne
e
r
i
ng
a
nd
s
i
m
pl
e
m
a
c
hi
ne
l
e
a
r
ni
n
g
a
l
g
or
i
t
h
m
s
,
w
hi
c
h,
a
l
t
houg
h
i
nt
e
r
pr
e
t
a
bl
e
,
of
t
e
n
l
a
c
k
t
he
pr
e
di
c
t
i
v
e
po
w
e
r
of
m
ode
r
n
D
L
a
ppr
oa
c
he
s
.
R
e
c
e
nt
a
d
v
a
n
c
e
m
e
nt
s
i
n
D
L
ha
v
e
de
m
ons
t
r
a
t
e
d
s
upe
r
i
or
pe
r
f
or
m
a
nc
e
by
a
ut
o
m
a
t
i
c
a
l
l
y
e
xt
r
a
c
t
i
ng
c
o
m
pl
e
x
f
e
a
t
ur
e
s
f
r
o
m
r
a
w
t
e
xt
da
t
a
.
H
ow
e
v
e
r
,
t
he
s
e
m
ode
l
s
’
de
c
i
s
i
ons
a
r
e
of
t
e
n
opa
que
,
m
a
ki
ng
i
t
di
f
f
i
c
ul
t
t
o
di
a
g
nos
e
e
r
r
or
s
or
unde
r
s
t
a
nd
t
he
m
ode
l
’
s
r
e
a
s
oni
ng
pr
oc
e
s
s
.
T
hi
s
l
a
c
k
of
t
r
a
ns
pa
r
e
nc
y
c
a
n
l
e
a
d
t
o
r
e
s
i
s
t
a
nc
e
t
o
a
dopt
i
ng
t
he
s
e
t
e
c
hnol
og
i
e
s
i
n s
e
ns
i
t
i
v
e
a
ppl
i
c
a
t
i
ons
w
he
r
e
unde
r
s
t
a
ndi
ng
t
he
de
c
i
s
i
on
-
m
a
ki
ng
pr
o
c
e
s
s
i
s
c
r
uc
i
a
l
.
E
nha
nc
e
d
c
a
t
e
g
or
y
c
or
r
e
l
a
t
i
on
l
e
a
r
ni
ng
:
our
m
e
t
hodol
og
y
i
nt
r
oduc
e
s
a
h
y
br
i
d
de
e
p
c
ont
e
xt
ua
l
a
nd
c
or
r
e
l
a
t
i
on
ne
t
w
or
k
(
H
D
C
C
N
E
T
)
,
w
hi
c
h
s
i
g
ni
f
i
c
a
nt
l
y
d
e
e
pe
ns
t
he
c
onne
c
t
i
ons
be
t
w
e
e
n
c
a
t
e
g
or
i
e
s
,
l
e
a
di
ng
t
o
i
m
pr
o
v
e
d
unde
r
s
t
a
ndi
ng
a
nd
pr
e
di
c
t
i
on
of
c
a
t
e
g
or
y
i
nt
e
r
r
e
l
a
t
i
ons
.
T
hi
s
e
nha
n
c
e
s
t
he
ov
e
r
a
l
l
a
c
c
ur
a
c
y
a
nd r
e
l
i
a
bi
l
i
t
y
of
m
ul
t
i
-
c
a
t
e
g
or
y
c
l
a
s
s
i
f
i
c
a
t
i
on t
a
s
ks
.
I
m
pr
o
v
e
d
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
:
b
y
i
nc
or
por
a
t
i
ng
r
e
s
i
dua
l
c
onne
c
t
i
ons
a
nd
f
oc
us
i
ng
on
e
s
s
e
nt
i
a
l
f
e
a
t
ur
e
s
,
ou
r
a
ppr
oa
c
h
not
onl
y
pr
e
v
e
nt
s
di
v
e
r
g
e
n
c
e
i
n
pr
e
di
c
t
i
ons
but
a
l
s
o
of
f
e
r
s
g
r
e
a
t
e
r
t
r
a
n
s
pa
r
e
nc
y
a
n
d
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
.
T
hi
s
a
l
l
o
w
s
us
e
r
s
t
o
unde
r
s
t
a
nd
t
he
m
ode
l
’
s
de
c
i
s
i
on
-
m
a
ki
n
g
pr
oc
e
s
s
,
a
ddr
e
s
s
i
ng
a
c
r
i
t
i
c
a
l
ne
e
d f
or
t
r
us
t
w
or
t
h
y
D
L
m
ode
l
s
i
n t
he
i
ndus
t
r
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
,
V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
11
00
-
11
08
1102
E
f
f
i
c
i
e
nc
y
a
nd
s
c
a
l
a
bi
l
i
t
y
:
our
m
e
t
hod
r
e
duc
e
s
t
he
nu
m
b
e
r
of
m
ode
l
pa
r
a
m
e
t
e
r
s
c
o
m
p
a
r
e
d
t
o
t
r
a
di
t
i
ona
l
a
ppr
oa
c
he
s
, a
c
c
e
l
e
r
a
t
i
ng
c
on
v
e
r
g
e
nc
e
a
nd
m
a
ki
n
g
t
he
m
od
e
l
m
or
e
e
f
f
i
c
i
e
nt
.
T
hi
s
e
f
f
i
c
i
e
nc
y i
s
pa
r
t
i
c
ul
a
r
l
y
be
ne
f
i
c
i
a
l
f
or
pr
a
c
t
i
c
a
l
a
ppl
i
c
a
t
i
ons
,
e
na
bl
i
ng
f
a
s
t
e
r
de
pl
oym
e
nt
a
nd
s
c
a
l
a
bi
l
i
t
y
i
n
v
a
r
i
ous
bus
i
ne
s
s
w
or
kf
l
o
w
s
.
2.
R
E
L
A
T
E
D
WO
R
K
Th
e
f
i
e
l
d
of
doc
u
m
e
nt
a
na
l
y
s
i
s
ha
s
s
e
e
n
s
ubs
t
a
nt
i
a
l
a
dv
a
nc
e
m
e
nt
s
w
i
t
h
t
he
i
nt
e
g
r
a
t
i
on
of
DL
t
e
c
hni
que
s
,
e
nha
nc
i
ng
t
he
e
f
f
i
c
i
e
n
c
y
a
nd
a
c
c
u
r
a
c
y
of
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
t
a
s
ks
.
E
a
r
l
y
a
ppr
oa
c
he
s
r
e
l
i
e
d
he
a
v
i
l
y
on
m
a
nua
l
f
e
a
t
ur
e
e
n
g
i
ne
e
r
i
ng
a
nd
t
r
a
di
t
i
ona
l
m
a
c
h
i
ne
l
e
a
r
ni
ng
a
l
g
or
i
t
h
m
s
,
s
uc
h
a
s
l
og
i
s
t
i
c
r
e
g
r
e
s
s
i
on
(
L
R
)
a
nd
S
V
M
s
,
w
hi
c
h,
w
hi
l
e
i
nt
e
r
pr
e
t
a
bl
e
,
of
t
e
n
l
a
c
k
e
d
t
he
pr
e
di
c
t
i
v
e
pow
e
r
of
DL
m
od
e
l
s
.
M
ode
r
n
m
e
t
hodol
og
i
e
s
l
e
v
e
r
a
g
e
de
e
p
ne
ur
a
l
ne
t
w
or
ks
,
pa
r
t
i
c
ul
a
r
l
y
c
on
v
ol
ut
i
ona
l
ne
ur
a
l
ne
t
w
or
ks
(
C
N
N
s
)
a
nd
g
r
a
ph
c
on
v
ol
ut
i
ona
l
ne
t
w
or
ks
(
G
C
N
s
)
,
w
hi
c
h
ha
v
e
de
m
ons
t
r
a
t
e
d
s
upe
r
i
or
pe
r
f
or
m
a
nc
e
by
a
ut
o
m
a
t
i
c
a
l
l
y
e
xt
r
a
c
t
i
ng
c
o
m
pl
e
x
f
e
a
t
ur
e
s
f
r
o
m
r
a
w
t
e
xt
da
t
a
.
N
ot
a
bl
e
w
or
ks
i
nc
l
ude
t
he
us
e
of
hi
e
r
a
r
c
hi
c
a
l
t
r
a
ns
f
or
m
e
r
s
a
nd
B
E
R
T
-
ba
s
e
d
m
ode
l
s
,
w
hi
c
h
ha
v
e
s
how
n
s
i
g
ni
f
i
c
a
nt
i
m
pr
o
v
e
m
e
nt
s
i
n
ha
ndl
i
ng
l
on
g
doc
u
m
e
nt
s
a
nd
c
a
pt
ur
i
ng
i
nt
r
i
c
a
t
e
t
e
xt
r
e
p
r
e
s
e
nt
a
t
i
ons
. D
e
s
pi
t
e
t
he
s
e
a
d
v
a
nc
e
m
e
nt
s
, a
pe
r
s
i
s
t
e
nt
c
ha
l
l
e
nge
r
e
m
a
i
ns
t
he
bl
a
c
k
-
box
na
t
ur
e
of
DL
m
ode
l
s
,
w
hi
c
h
hi
nde
r
s
t
he
i
r
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
a
nd
t
hus
t
he
i
r
a
c
c
e
pt
a
nc
e
i
n
t
he
i
ndus
t
r
y
.
R
e
c
e
nt
s
t
udi
e
s
ha
v
e
f
oc
us
e
d
on
e
nha
nc
i
n
g
m
ode
l
t
r
a
ns
pa
r
e
n
c
y
t
hr
ou
g
h
t
e
c
hni
que
s
l
i
ke
a
t
t
e
nt
i
on
m
e
c
ha
ni
s
m
s
a
nd
e
xpl
a
i
na
bl
e
A
I
f
r
a
m
e
w
o
r
ks
,
a
i
m
i
ng
t
o
br
i
d
g
e
t
he
g
a
p
b
e
t
w
e
e
n
hi
g
h
p
e
r
f
or
m
a
nc
e
a
nd
t
he
ne
e
d
f
or
m
ode
l
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
.
S
a
i
f
ul
l
a
h
e
t
al
.
[
9]
r
e
por
t
e
d
a
pr
ope
r
t
y
c
ol
l
e
c
t
i
on
t
ha
t
w
a
s
done
by
ha
nd.
T
he
a
r
t
i
c
l
e
s
w
e
r
e
t
he
n
c
a
t
e
g
or
i
z
e
d
us
i
ng
t
he
pr
e
v
i
ous
l
y
m
e
nt
i
one
d
a
t
t
r
i
but
e
s
a
s
a
g
ui
de
.
T
he
r
a
t
i
os
of
t
e
x
t
ua
l
t
o
non
-
t
e
x
t
ua
l
por
t
i
ons
,
c
ol
um
n
l
a
y
out
s
,
c
ont
e
nt
de
ns
i
t
y
,
a
nd
f
ont
s
i
z
e
s
i
n
c
om
pa
r
i
s
on
w
e
r
e
a
m
on
g
t
he
m
a
ny
f
a
c
t
or
s
c
ons
i
de
r
e
d t
hr
oug
hout
t
he
s
e
l
e
c
t
i
on pr
oc
e
s
s
.
T
he
da
t
a
w
a
s
c
a
t
e
g
or
i
z
e
d
us
i
ng
a
de
c
i
s
i
on
t
r
e
e
t
ha
t
w
a
s
t
r
a
i
ne
d
us
i
ng
t
he
pr
ov
i
de
d
a
t
t
r
i
but
e
s
,
i
t
di
s
c
us
s
e
d
t
he
c
onc
e
pt
s
of
doc
um
e
nt
s
i
m
i
l
a
r
i
t
y
a
nd
a
que
r
y
i
ng
t
e
c
hni
que
de
s
i
g
ne
d
s
pe
c
i
f
i
c
a
l
l
y
f
or
doc
u
m
e
nt
i
m
a
g
e
da
t
a
ba
s
e
s
.
T
h
e
s
t
r
uc
t
ur
a
l
s
i
m
i
l
a
r
i
t
y
t
ha
t
i
s
g
e
o
m
e
t
r
i
c
a
l
l
y
i
n
v
a
r
i
a
nt
i
s
f
ound.
T
he
f
ol
l
ow
i
ng
s
t
a
t
e
m
e
nt
i
s
a
r
e
s
pons
e
t
o
pr
e
v
i
ous
r
e
s
e
a
r
c
h
c
onduc
t
e
d
by
r
e
l
e
v
a
nt
pa
r
t
i
e
s
.
P
uj
a
r
e
t
al
.
[
10]
e
m
pl
o
y
e
d
A
da
B
oos
t
i
n
c
o
m
bi
na
t
i
on
w
i
t
h
a
n
e
ns
e
m
bl
e
of
K
-
m
e
a
ns
c
l
us
t
e
r
i
ng
-
ba
s
e
d
c
l
a
s
s
i
f
i
e
r
s
t
o
f
i
nd
a
r
t
i
c
l
e
s
f
or
t
he
i
r
s
t
udy
qui
c
kl
y
.
T
he
i
de
nt
i
f
i
c
a
t
i
on
pr
oc
e
s
s
w
a
s
ba
s
e
d
on
t
he
l
ow
-
l
e
v
e
l
bi
na
r
y
i
m
a
g
e
pi
xe
l
de
ns
i
t
y
da
t
a
pr
o
c
e
s
s
i
ng
.
A
no
v
e
l
m
e
t
hod f
or
a
ut
o
m
a
t
i
c
a
l
l
y
r
e
c
og
ni
z
i
n
g
pi
c
t
ur
e
a
nc
hor
t
e
m
pl
a
t
e
s
f
r
o
m
doc
u
m
e
nt
i
m
a
g
e
s
.
T
he
t
e
m
pl
a
t
e
s
m
a
y
b
e
us
e
d
f
or
a
v
a
r
i
e
t
y
of
t
a
s
ks
,
s
uc
h
a
s
da
t
a
e
xt
r
a
c
t
i
on
a
nd
doc
um
e
nt
c
a
t
e
g
or
i
z
a
t
i
on.
I
n
t
he
i
r
i
nqui
r
y
[
11]
pr
opos
e
d
ut
i
l
i
z
i
ng
c
ode
books
a
s
a
m
e
a
ns
of
a
s
s
e
s
s
i
ng
doc
u
m
e
nt
pi
c
t
ur
e
s
i
m
i
l
a
r
i
t
y
.
T
he
doc
u
m
e
nt
i
s
r
e
c
ur
s
i
v
e
l
y
br
oke
n
i
nt
o
s
m
a
l
l
e
r
bi
t
s
us
i
ng
t
hi
s
t
e
c
hni
que
.
T
he
a
t
t
r
i
but
e
s
m
e
nt
i
one
d
be
f
or
e
w
e
r
e
ut
i
l
i
z
e
d
t
o
r
e
t
r
i
e
v
e
doc
u
m
e
nt
s
f
r
o
m
t
he
da
t
a
ba
s
e
t
ha
t
s
ha
r
e
d
t
he
s
a
m
e
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
.
T
he
a
ut
ho
r
s
e
nha
nc
e
d
t
he
i
r
pr
e
v
i
ous
f
i
ndi
ng
s
by
t
r
a
i
ni
ng
a
nd
r
e
t
r
i
e
v
i
n
g
doc
u
m
e
nt
phot
os
t
ha
t
be
l
o
ng
t
o
t
he
s
a
m
e
c
a
t
e
g
or
y
us
i
ng
a
n
uns
upe
r
v
i
s
e
d
r
a
ndo
m
f
or
e
s
t
c
l
a
s
s
i
f
i
e
r
.
T
he
a
c
hi
e
v
e
m
e
nt
w
a
s
a
t
t
a
i
ne
d
by
opt
i
m
i
z
i
n
g
t
he
c
a
l
c
ul
a
t
e
d
r
e
pr
e
s
e
nt
a
t
i
ons
.
I
n
t
he
s
ubs
e
que
nt
ye
a
r
[
12]
,
t
he
a
ut
hor
s
a
c
hi
e
v
e
d
s
t
a
t
e
-
of
-
t
he
-
a
r
t
pe
r
f
or
m
a
nc
e
on
t
he
t
a
x
f
or
m
a
n
d
t
a
bl
e
r
e
t
r
i
e
v
a
l
t
a
s
ks
us
i
ng
t
he
s
a
m
e
t
e
c
hni
que
s
.
T
he
pe
r
f
or
m
a
nc
e
e
v
a
l
u
a
t
i
on
’
s
unde
r
l
y
i
ng
a
s
s
um
pt
i
on w
a
s
t
ha
t
i
ns
uf
f
i
c
i
e
nt
t
r
a
i
ni
ng
da
t
a
pr
o
v
i
de
d
a
de
t
a
i
l
e
d
a
c
c
ount
of
t
he
f
i
r
s
t
a
ppl
i
c
a
t
i
on
of
de
e
p
C
N
N
i
n
t
he
f
i
e
l
d
of
doc
u
m
e
nt
pi
c
t
ur
e
c
a
t
e
g
or
i
z
a
t
i
on.
T
hi
s
a
ppr
oa
c
h
out
pe
r
f
or
m
e
d
pr
e
v
i
ous
ha
nd
-
c
od
e
d
f
e
a
t
ur
e
e
ng
i
ne
e
r
i
n
g
m
e
t
hodol
og
i
e
s
i
n
t
e
r
m
s
of
s
pe
e
d.
I
n
t
he
i
r
r
e
s
e
a
r
c
h
publ
i
c
a
t
i
on
i
nt
r
oduc
e
d
t
he
D
e
e
pD
oc
C
l
a
s
s
i
f
i
e
r
a
s
a
n
e
xa
m
pl
e
of
t
he
pot
e
nt
i
a
l
a
ppl
i
c
a
t
i
ons
of
t
r
a
ns
f
e
r
l
e
a
r
ni
n
g
[
13]
.
T
he
s
t
udy
m
a
k
e
s
us
e
of
a
d
e
e
p
C
N
N
w
i
t
h
t
he
A
l
e
xN
e
t
a
r
c
hi
t
e
c
t
ur
e
.
T
he
w
e
i
g
ht
s
of
t
he
ne
t
w
or
k
w
e
r
e
i
ni
t
i
a
l
i
z
e
d
us
i
ng
a
pr
e
-
t
r
a
i
ne
d
m
ode
l
t
ha
t
w
a
s
t
r
a
i
ne
d
on
t
he
l
a
r
g
e
I
m
a
ge
N
e
t
da
t
a
s
e
t
.
T
he
da
t
a
s
e
t
c
ons
i
s
t
s
of
1.28
m
i
l
l
i
on
t
r
a
i
ni
ng
pi
c
t
ur
e
s
t
ha
t
a
r
e
c
a
t
e
g
or
i
z
e
d
i
nt
o
1000
di
s
t
i
nc
t
g
r
oups
.
T
he
pe
r
f
or
m
a
nc
e
of
e
a
r
l
i
e
r
a
ppr
oa
c
he
s
w
a
s
s
i
g
ni
f
i
c
a
nt
l
y
i
m
pr
o
v
e
d
b
y
t
r
a
ns
f
or
m
i
ng
t
h
e
or
i
g
i
na
l
c
on
v
ol
ut
i
ona
l
l
a
y
e
r
s
i
nt
o
f
l
e
xi
bl
e
f
e
a
t
ur
e
e
xt
r
a
c
t
or
s
.
T
o
i
nc
r
e
a
s
e
t
he
a
c
c
ur
a
c
y
of
t
he
c
l
a
s
s
i
f
i
c
a
t
i
on
f
i
ndi
ng
s
,
t
he
t
e
xt
ua
l
da
t
a
f
r
o
m
a
c
o
m
m
e
r
c
i
a
l
opt
i
c
a
l
c
ha
r
a
c
t
e
r
r
e
c
o
g
ni
t
i
on
(
O
C
R
)
s
y
s
t
e
m
w
i
t
h
t
he
r
a
w
i
m
a
ge
da
t
a
[
14]
.
A
f
t
e
r
i
t
w
a
s
r
e
c
o
v
e
r
e
d,
an
N
L
P
m
ode
l
t
r
a
ns
l
a
t
e
d
t
he
t
e
xt
i
nt
o
t
he
f
e
a
t
ur
e
s
pa
c
e
.
E
m
pl
oye
d
a
n
e
xt
r
e
m
e
l
e
a
r
ni
n
g
m
a
c
hi
ne
t
o
m
a
ni
pul
a
t
e
f
r
oz
e
n
c
on
v
ol
ut
i
ona
l
l
a
y
e
r
s
t
ha
t
w
e
r
e
t
r
a
i
ne
d
us
i
ng
a
n
A
l
e
xN
e
t
m
od
e
l
. T
he
gr
oup w
a
s
a
bl
e
t
o i
nc
r
e
a
s
e
out
put
w
i
t
hout
c
o
m
pr
o
m
i
s
i
ng
a
c
c
ur
a
c
y
r
e
qui
r
e
m
e
nt
s
.
P
a
ppa
g
a
r
i
e
t
al
.
[
1]
e
v
a
l
ua
t
e
d
t
he
pe
r
f
or
m
a
nc
e
of
v
i
s
ua
l
g
e
o
m
e
t
r
y
g
r
oup
(
V
G
G
)
,
R
e
s
N
e
t
,
a
nd
G
oo
gL
e
N
e
t
i
n
c
l
a
s
s
i
f
y
i
n
g
doc
u
m
e
nt
pi
c
t
ur
e
s
us
i
ng be
nc
h
m
a
r
k da
t
a
s
e
t
s
. T
h
e
s
t
udy
f
i
nds
t
ha
t
s
i
g
ni
f
i
c
a
nt
ga
i
ns
m
a
y
be
a
c
hi
e
v
e
d
w
i
t
h
pr
e
t
r
a
i
ne
d
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
on
ne
t
w
o
r
ks
,
w
hi
c
h
ha
v
e
b
e
e
n
t
r
a
i
ne
d
on
a
l
a
r
g
e
a
m
ount
of
c
a
t
e
g
or
i
z
e
d
da
t
a
.
T
he
y
pr
e
s
e
nt
e
d
a
t
w
o
-
s
t
r
e
a
m
n
e
t
w
or
k
t
ha
t
c
oul
d
g
e
ne
r
a
t
e
out
put
ba
s
e
d
on
bot
h
t
e
xt
ua
l
a
nd
v
i
s
ua
l
i
nput
s
.
I
n
c
ont
r
a
s
t
t
o
pr
e
v
i
ous
s
t
udi
e
s
,
a
n
a
l
g
or
i
t
hm
w
a
s
e
m
pl
oye
d
t
o
e
v
a
l
ua
t
e
t
he
t
e
xt
ua
l
s
t
r
e
a
m
’
s
f
e
a
t
ur
e
s
a
nd
pi
npoi
nt
t
he
m
os
t
s
i
g
ni
f
i
c
a
nt
one
s
.
T
he
y
c
o
m
bi
ne
d
O
C
R
pr
e
di
c
t
i
o
ns
w
i
t
h
i
m
a
g
e
d
a
t
a
us
i
ng
a
m
e
t
hodol
ogy
a
ki
n
t
o
pr
e
v
i
ous
w
or
k.
H
ow
e
v
e
r
,
t
he
pe
r
f
or
m
a
nc
e
w
a
s
e
nha
nc
e
d
by
e
m
pl
o
y
i
ng
bi
di
r
e
c
t
i
ona
l
e
nc
ode
r
r
e
pr
e
s
e
nt
a
t
i
ons
f
r
o
m
t
r
a
ns
f
or
m
e
r
s
(
B
E
R
T
)
a
s
t
he
N
L
P
f
r
a
m
e
w
or
k,
w
hi
c
h
l
ooke
d
a
t
t
he
v
i
a
bi
l
i
t
y
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
E
nhanc
i
ng
d
oc
um
e
nt
t
e
x
t
c
l
as
s
i
f
i
c
at
i
on us
i
n
g hy
b
r
i
d
d
e
e
p c
ont
e
x
t
ual
an
d c
or
r
e
l
at
i
o
n ne
t
w
or
k
(
Shi
l
pa
)
1103
a
ppl
y
i
ng
s
e
l
f
-
s
upe
r
v
i
s
e
d
r
e
pr
e
s
e
nt
a
t
i
on
l
e
a
r
ni
ng
f
or
doc
u
m
e
nt
i
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on.
H
o
w
e
v
e
r
,
t
he
s
t
ud
y
c
onduc
t
e
d
by
t
he
r
e
s
e
a
r
c
h
e
r
s
w
a
s
l
i
m
i
t
e
d
t
o
a
f
e
w
ol
d
-
f
a
s
hi
one
d
s
e
l
f
-
s
upe
r
v
i
s
e
d
a
c
t
i
v
i
t
i
e
s
,
s
uc
h
a
s
J
i
g
s
a
w
puz
z
l
e
s
.
T
he
s
t
udy
e
f
f
e
c
t
i
v
e
l
y
de
m
ons
t
r
a
t
e
d
t
he
l
i
m
i
t
a
t
i
ons
of
t
he
s
t
r
a
t
e
gy
i
n
t
e
r
m
s
of
g
e
ne
r
a
t
i
ng
m
e
a
ni
ngf
ul
r
e
pr
e
s
e
nt
a
t
i
ons
.
3.
P
R
O
P
O
S
E
D
M
E
T
H
O
D
T
hi
s
m
ode
l
pr
opos
e
s
a
H
D
C
C
N
e
t
m
ode
l
c
o
m
pos
e
d
of
t
w
o
pr
i
m
a
r
y
m
odul
e
s
:
t
he
a
nd
t
he
m
odul
e
.
T
he
m
odul
e
e
nha
nc
e
s
doc
u
m
e
nt
r
e
pr
e
s
e
nt
a
t
i
on
by
i
nc
or
por
a
t
i
ng
c
a
t
e
g
or
y
i
nf
or
m
a
t
i
on,
t
r
e
a
t
i
ng
c
a
t
e
g
or
y
i
nf
or
m
a
t
i
on
a
s
pos
i
t
i
v
e
s
a
m
pl
e
s
of
doc
u
m
e
nt
i
nf
or
m
a
t
i
on
t
hr
oug
h
c
ont
r
a
s
t
i
v
e
l
e
a
r
ni
ng
.
T
hi
s
pr
oc
e
s
s
a
l
l
ow
s
f
or
a
n
i
n
-
de
pt
h
e
xpl
or
a
t
i
on
of
t
he
r
e
l
a
t
i
ons
hi
p
s
be
t
w
e
e
n
doc
u
m
e
nt
s
a
nd
t
he
i
r
c
or
r
e
s
pondi
ng
c
a
t
e
g
or
i
e
s
.
B
E
R
T
i
s
e
m
pl
oye
d
a
s
a
f
e
a
t
ur
e
e
xt
r
a
c
t
or
t
o
obt
a
i
n
s
e
m
a
nt
i
c
f
e
a
t
ur
e
s
f
r
o
m
bot
h
t
he
doc
u
m
e
nt
s
a
nd
t
he
c
a
t
e
g
or
i
e
s
,
w
hi
c
h
a
r
e
t
he
n
pr
oc
e
s
s
e
d
by
t
he
m
odul
e
.
T
h
e
out
put
f
r
om
t
h
e
m
odu
l
e
s
e
r
v
e
s
a
s
t
he
i
nput
f
or
t
he
m
odul
e
.
I
n
,
t
he
or
i
g
i
na
l
c
a
t
e
g
or
y
pr
e
di
c
t
i
ons
a
r
e
f
ur
t
he
r
r
e
f
i
ne
d
us
i
ng
c
or
r
e
l
a
t
i
on
know
l
e
dge
by
t
r
a
i
ni
ng
m
ul
t
i
pl
e
w
e
i
g
ht
m
a
t
r
i
c
e
s
.
T
hi
s
m
odul
e
f
oc
us
e
s
on
i
de
nt
i
f
y
i
ng
r
e
l
e
v
a
nt
c
o
m
bi
na
t
i
ons
of
or
i
g
i
na
l
c
a
t
e
g
or
y pr
e
di
c
t
i
ons
t
o e
nha
nc
e
t
he
o
v
e
r
a
l
l
pr
e
di
c
t
i
on a
c
c
ur
a
c
y
.
B
y
i
nt
e
g
r
a
t
i
ng
c
a
t
e
g
or
y
s
e
m
a
nt
i
c
i
nf
or
m
a
t
i
on
a
nd
c
a
t
e
g
or
y
c
or
r
e
l
a
t
i
on,
H
D
C
C
N
E
T
a
c
hi
e
v
e
s
pr
e
c
i
s
e
m
ul
t
i
-
c
a
t
e
g
or
y
c
l
a
s
s
i
f
i
c
a
t
i
on r
e
s
ul
t
s
. F
i
g
ur
e
1 s
how
s
t
he
pr
opos
e
d H
D
C
C
N
a
r
c
hi
t
e
c
t
ur
e
.
F
i
g
ur
e
1
.
P
r
opos
e
d H
D
C
C
N
a
r
c
hi
t
e
c
t
ur
e
3.1.
I
n
i
t
i
al
a
n
al
ys
i
s
A
s
s
u
m
e
t
he
da
t
a
s
e
t
=
{
(
,
)
}
=
1
,
he
r
e
i
s
t
he
or
i
g
i
na
l
doc
um
e
nt
a
nd
t
he
−
ℎ
t
e
xt
a
s
,
=
{
1
,
2
,
…
,
}
,
r
e
pr
e
s
e
nt
s
t
he
l
e
n
g
t
h
of
t
he
i
nput
doc
u
m
e
nt
i
s
t
he
−
ℎ
w
or
d
of
t
he
doc
u
m
e
nt
.
∈
{
0
,
1
}
de
not
e
s
t
he
a
dj
a
c
e
nt
c
a
t
e
g
or
y
s
e
t
f
or
,
i
s
t
he
c
a
t
e
g
or
y
s
e
t
of
t
he
da
t
a
s
e
t
a
nd
P
i
s
t
he
t
ot
a
l
num
b
e
r
of
e
xa
m
pl
e
s
i
n
t
he
da
t
a
s
e
t
,
t
he
c
l
a
s
s
i
f
i
e
r
he
r
e
c
om
put
e
s
t
he
pr
oba
bi
l
i
t
y
f
or
e
a
c
h
c
a
t
e
g
o
r
y
b
e
i
ng
t
r
ue
w
he
r
e
=
{
1
,
2
,
…
,
}
,
t
he
l
o
s
s
be
t
w
e
e
n
a
nd
i
s
g
i
v
e
n
by
(
1)
.
B
e
f
or
e
be
i
ng
g
i
v
e
n
a
s
i
nput
t
o
t
he
B
E
R
T
m
od
e
l
, t
he
doc
u
m
e
nt
i
s
c
ons
i
de
r
e
d, c
ons
i
de
r
i
ng
a
m
ode
l
w
i
t
h
l
a
y
e
r
s
t
he
hi
dde
n r
e
pr
e
s
e
nt
a
t
i
on i
s
de
not
e
d
by
t
he
−
ℎ
l
a
y
e
r
a
s
g
i
v
e
n i
n (
2)
.
(
,
)
=
−
1
∑
∈
[
+
(
1
−
)
(
1
−
)
]
(
1)
(
)
(
)
=
{
(
)
,
1
(
)
,
…
…
.
,
(
)
}
(
2)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
,
V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
11
00
-
11
08
1104
3.2.
D
e
e
p
c
on
t
e
xt
u
al
r
e
p
r
e
s
e
n
t
at
i
on
i
n
t
e
gr
at
or
(
D
C
R
I
)
N
e
t
A
s
t
he
s
e
m
a
nt
i
c
c
ha
r
a
c
t
e
r
i
s
t
i
c
of
t
he
doc
u
m
e
nt
a
nd
c
a
t
e
g
o
r
y
,
w
e
m
a
ke
us
e
of
t
he
e
m
be
dd
e
d
l
a
y
e
r
i
n
e
nha
nc
e
d
B
E
R
T
.
N
e
v
e
r
t
he
l
e
s
s
,
i
n
v
e
s
t
i
g
a
t
i
ng
t
he
c
onne
c
t
i
ons
be
t
w
e
e
n
t
he
s
e
m
a
nt
i
c
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
of
t
he
t
e
xt
a
nd t
he
c
a
t
e
g
or
y r
e
qui
r
e
s
m
or
e
t
ha
n j
us
t
us
i
ng
t
he
l
a
y
e
r
of
t
he
f
i
na
l
l
a
y
e
r
.
I
n t
he
t
r
a
i
ni
ng
ph
a
s
e
, t
he
doc
u
m
e
nt
f
e
a
t
ur
e
s
a
nd
c
a
t
e
g
o
r
y
f
e
a
t
ur
e
s
a
r
e
s
y
nc
hr
onous
l
y
e
xt
r
a
c
t
e
d,
a
nd
t
he
t
e
xt
f
e
a
t
ur
e
i
s
de
not
e
d
a
s
g
i
v
e
n
i
n
(
3)
.
T
o a
ddr
e
s
s
t
he
a
f
or
e
m
e
nt
i
one
d i
s
s
ue
, w
e
t
hus
s
ugg
e
s
t
a
doc
u
m
e
nt
-
c
a
t
e
g
or
y
c
ont
r
a
s
t
i
v
e
l
e
a
r
ni
ng
l
os
s
f
unc
t
i
on.
t
he
e
nc
odi
ng
of
doc
um
e
nt
f
e
a
t
ur
e
s
ut
i
l
i
z
i
ng
t
he
f
e
a
t
ur
e
s
of
a
s
m
ul
t
i
-
c
a
t
e
g
or
y
s
e
t
s
r
e
l
a
t
e
d
t
o
t
he
doc
u
m
e
nt
he
nc
e
t
he
num
be
r
of
s
a
m
pl
e
s
i
s
2M
,
K
=
{
1,
…,
2M
}
.
T
he
i
nde
x
of
t
he
−
ℎ
do
c
u
m
e
nt
a
s
c
a
t
e
g
or
y
(
)
a
nd t
he
ne
g
a
t
i
v
e
s
a
m
pl
e
r
e
m
a
i
ns
a
s
i
n K
. T
h
e
l
os
s
f
unc
t
i
on o
f
t
he
e
xpr
e
s
s
i
on i
s
s
h
ow
n i
n (
4)
.
=
(
(
−
1
)
,
(
−
2
)
,
(
−
3
)
,
(
−
4
)
,
(
−
5
)
)
(
3)
=
∑
2
=
1
−
(
.
(
)
)
∑
∈
/
(
.
)
(
4)
H
e
r
e
i
s
t
he
t
e
m
p
e
r
a
t
ur
e
c
o
-
e
f
f
i
c
i
e
nt
t
ha
t
c
a
n
di
s
t
i
ng
ui
s
h be
t
w
e
e
n pos
i
t
i
v
e
a
nd n
e
ga
t
i
v
e
s
a
m
pl
e
s
.
i
s
t
he
c
onc
a
t
e
na
t
e
d
v
e
c
t
or
of
t
he
l
a
y
e
r
s
obt
a
i
ne
d
f
r
o
m
t
o
.
T
w
o
m
a
i
n
f
a
c
t
or
s
c
ont
r
i
but
e
t
o
C
a
t
e
g
or
y
s
uc
c
e
s
s
.
U
s
i
ng
t
he
s
e
m
a
nt
i
c
i
nf
or
m
a
t
i
on
of
t
he
c
a
t
e
g
o
r
y
i
s
t
he
f
i
r
s
t
s
t
e
p
i
n
c
r
e
a
t
i
ng
a
d
oc
u
m
e
nt
r
e
pr
e
s
e
nt
a
t
i
on
w
i
t
h
c
a
t
e
g
or
y
i
nf
or
m
a
t
i
on
g
ui
da
nc
e
.
U
s
i
ng
t
hi
s
da
t
a
,
a
l
a
n
g
ua
g
e
m
ode
l
c
a
l
l
e
d
B
E
R
T
i
s
i
ns
t
r
uc
t
e
d
t
o
e
xt
r
a
c
t
c
ha
r
a
c
t
e
r
i
s
t
i
c
s
f
r
o
m
t
he
doc
u
m
e
nt
i
nf
or
m
a
t
i
on
t
ha
t
a
r
e
pe
r
t
i
ne
nt
t
o
t
h
e
c
a
t
e
g
or
y
.
I
t
a
l
s
o
i
nc
or
por
a
t
e
s
a
n or
g
a
ni
z
e
d
f
r
a
m
e
w
o
r
k f
or
c
onne
c
t
i
ng
t
he
s
e
m
a
nt
i
c
da
t
a
.
I
n t
hi
s
t
e
c
hni
que
, t
he
c
a
t
e
g
or
y
f
e
a
t
ur
e
e
nc
odi
ng
i
s
us
e
d a
s
a
pos
i
t
i
v
e
s
a
m
pl
e
f
or
t
he
c
a
t
e
g
or
y
a
nd t
he
doc
u
m
e
nt
c
ont
e
nt
s
.
3.3.
D
e
e
p
c
or
r
e
l
at
i
on
N
e
t
(
D
C
N
e
t
)
T
he
t
w
o
p
a
r
t
s
of
a
r
e
t
he
r
a
w
c
a
t
e
g
or
y
pr
e
di
c
t
i
ons
a
nd
t
he
c
o
m
put
a
t
i
ona
l
uni
t
,
w
hi
c
h
c
on
v
e
r
t
s
t
he
r
a
w
c
a
t
e
g
or
y
pr
e
di
c
t
i
ons
i
nt
o i
m
pr
o
v
e
d
c
a
t
e
g
or
y
pr
e
di
c
t
i
ons
de
pe
ndi
ng
on c
a
t
e
g
or
y
c
or
r
e
l
a
t
i
ons
.
T
he
m
ode
l
i
s
de
f
i
ne
d
b
y
t
he
g
i
v
e
n
(
5
)
.
H
e
r
e
A
a
nd
A
a
r
e
t
he
out
put
a
nd
i
nput
of
t
he
m
o
de
l
,
he
r
e
i
s
t
he
r
a
w
c
a
t
e
g
or
y
p
r
e
di
c
t
i
on
be
f
or
e
m
ode
l
a
nd
A
i
s
t
he
c
a
t
e
g
or
y
pr
e
di
c
t
i
on
m
ode
l
w
i
t
h
c
or
r
e
l
a
t
i
on r
e
pr
e
s
e
nt
e
d b
y
t
he
.
=
+
(
)
(
5)
T
he
s
i
m
pl
e
s
t
de
s
i
g
n
f
or
a
c
a
t
e
g
or
y
c
o
r
r
e
l
a
t
i
on
m
odul
e
a
dds
a
l
i
ne
a
r
l
a
ye
r
a
f
t
e
r
y
,
s
i
m
i
l
a
r
t
o
C
or
ne
t
,
f
or
c
or
r
e
l
a
t
e
d
c
a
t
e
g
or
y
pr
e
di
c
t
i
on.
H
o
w
e
v
e
r
,
t
hi
s
a
ppr
oa
c
h
y
i
e
l
ds
s
ha
l
l
ow
c
a
t
e
g
or
y
c
or
r
e
l
a
t
i
ons
.
T
o
de
e
pe
n
t
he
c
onne
c
t
i
ons
,
w
e
pr
opos
e
a
m
ul
t
i
-
l
a
y
e
r
c
a
t
e
g
o
r
y
c
or
r
e
l
a
t
i
on
m
odul
e
w
h
e
r
e
e
a
c
h
l
a
y
e
r
’
s
out
pu
t
s
e
r
v
e
s
a
s
t
he
ne
xt
l
a
y
e
r
’
s
i
nput
,
i
m
pr
o
v
i
n
g
c
a
t
e
g
or
y
c
o
r
r
e
l
a
t
i
ons
.
T
o
pr
e
v
e
nt
di
v
e
r
g
e
n
c
e
i
n
pr
e
di
c
t
i
ons
,
w
e
i
nc
or
por
a
t
e
r
e
s
i
dua
l
c
onne
c
t
i
ons
us
i
ng
t
he
or
i
g
i
na
l
c
a
t
e
g
o
r
y
pr
e
di
c
t
i
ons
.
T
hi
s
de
s
i
g
n,
w
hi
c
h
ha
s
f
e
w
e
r
pa
r
a
m
e
t
e
r
s
t
ha
n
t
he
t
ot
a
l
num
be
r
of
c
a
t
e
g
or
i
e
s
,
f
oc
us
e
s
on
l
e
a
r
ni
ng
d
e
e
p
c
or
r
e
l
a
t
i
ons
a
nd
a
c
c
e
l
e
r
a
t
e
s
m
ode
l
c
on
v
e
r
ge
nc
e
.
H
e
r
e
i
s
t
he
i
nput
f
or
t
he
c
or
r
e
l
a
t
i
on l
a
y
e
r
a
nd
i
s
t
he
a
c
t
i
v
a
t
i
on f
unc
t
i
on. C
a
t
e
g
or
y pr
e
di
c
t
i
on a
s
t
he
out
put
of
a
c
hi
e
v
e
s
hi
g
h
a
c
c
ur
a
c
y
i
n
g
u
i
di
ng
t
he
c
a
t
e
g
o
r
y
c
or
r
e
l
a
t
i
on
pr
e
di
c
t
i
on
a
s
i
nput
of
.
T
he
c
a
t
e
g
or
y
pr
e
di
c
t
i
on i
s
t
he
i
nput
f
or
. T
he
e
f
f
e
c
t
i
s
a
m
pl
i
f
i
e
d t
hr
oug
h r
e
s
i
dua
l
c
onne
c
t
i
ons
.
(
)
=
(
+
)
(
6)
3.4.
H
D
C
C
N
E
T
m
od
e
l
t
r
ai
n
i
n
g
T
he
e
nd
-
to
-
e
nd
ba
s
e
d
m
ul
t
i
-
c
a
t
e
g
or
y
c
l
a
s
s
i
f
i
c
a
t
i
on
m
ode
l
i
s
m
a
de
up
of
a
nd
m
od
e
l
,
t
he
g
oa
l
i
s
t
o
r
e
duc
e
t
he
t
a
r
g
e
t
l
os
s
a
s
t
ha
t
c
ons
i
s
t
s
of
,
(
,
)
a
nd
(
,
)
,
t
hi
s
i
s
de
f
i
ne
d
a
s
g
i
v
e
n
i
n
(
7)
.
H
e
r
e
i
s
t
he
c
oe
f
f
i
c
i
e
nt
a
s
s
oc
i
a
t
e
d
w
i
t
h
t
he
c
ont
r
a
s
t
l
e
a
r
ni
ng
of
t
h
e
c
a
t
e
g
or
i
z
e
d
doc
u
m
e
nt
be
t
w
e
e
n
t
he
l
os
s
e
s
,
i
s
t
he
l
a
s
t
pr
oba
bi
l
i
t
y
of
t
he
−
ℎ
doc
um
e
nt
s
e
m
a
nt
i
c
i
nf
o
a
s
i
nput
.
i
s
t
he
l
a
s
t
pr
oba
bi
l
i
t
y
of
s
e
m
a
nt
i
c
i
nf
o
of
t
he
c
a
t
e
g
or
y
s
e
t
r
e
l
e
v
a
nt
w
i
t
h
t
he
−
ℎ
doc
um
e
n
t
a
s
i
nput
.
=
+
(
,
)
+
(
,
)
(
7)
4.
P
E
R
F
O
R
M
A
N
C
E
E
V
A
L
U
A
T
I
O
N
T
he
pe
r
f
or
m
a
nc
e
e
v
a
l
ua
t
i
on
de
m
ons
t
r
a
t
e
s
t
ha
t
t
he
pr
opos
e
d
s
y
s
t
e
m
(
P
S
)
m
e
t
hod
s
i
g
ni
f
i
c
a
nt
l
y
out
pe
r
f
or
m
s
ot
he
r
s
t
a
t
e
-
of
-
t
he
-
a
r
t
t
e
c
hni
que
s
i
n
doc
um
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
on
t
he
e
xA
A
P
D
2
a
nd
e
x
P
F
D
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
E
nhanc
i
ng
d
oc
um
e
nt
t
e
x
t
c
l
as
s
i
f
i
c
at
i
on us
i
n
g hy
b
r
i
d
d
e
e
p c
ont
e
x
t
ual
an
d c
or
r
e
l
at
i
o
n ne
t
w
or
k
(
Shi
l
pa
)
1105
da
t
a
s
e
t
s
,
r
e
s
pe
c
t
i
v
e
l
y
.
T
hi
s
not
a
bl
e
pe
r
f
or
m
a
nc
e
ga
p
hi
g
hl
i
g
ht
s
t
he
P
S
m
e
t
hod
’
s
a
d
v
a
n
c
e
d
t
e
c
hni
que
s
a
nd
opt
i
m
i
z
a
t
i
ons
,
m
a
ki
ng
i
t
a
r
obu
s
t
a
nd
r
e
l
i
a
bl
e
c
hoi
c
e
f
o
r
hi
g
h
-
a
c
c
ur
a
c
y
c
l
a
s
s
i
f
i
c
a
t
i
on
t
a
s
ks
.
T
he
c
l
us
t
e
r
i
n
g
of
ot
he
r
m
e
t
hods
’
s
c
or
e
s
w
i
t
hi
n
na
r
r
o
w
e
r
r
a
n
g
e
s
s
ugge
s
t
s
t
he
i
r
c
o
m
pa
r
a
bl
e
e
f
f
e
c
t
i
v
e
ne
s
s
,
y
e
t
P
S
’
s
s
upe
r
i
or
r
e
s
ul
t
s
unde
r
s
c
or
e
i
t
s
pot
e
nt
i
a
l
a
dv
a
nt
a
g
e
s
i
n pr
a
c
t
i
c
a
l
a
ppl
i
c
a
t
i
ons
.
4.1.
D
at
as
e
t
e
xA
A
P
D
2:
t
he
e
xt
e
nde
d
a
nnot
a
t
e
d
a
ut
om
a
t
i
c
phone
m
e
da
t
a
(
e
x
A
A
P
D
)
[1
5
]
da
t
a
s
e
t
i
s
a
c
o
m
pr
e
he
ns
i
ve
r
e
s
our
c
e
d
e
s
i
g
ne
d
f
or
a
d
v
a
nc
e
d s
pe
e
c
h pr
o
c
e
s
s
i
ng
r
e
s
e
a
r
c
h.
I
t
f
e
a
t
ur
e
s
hi
g
h
-
qu
a
l
i
t
y
a
udi
o r
e
c
or
di
ng
s
w
i
t
h
de
t
a
i
l
e
d, t
i
m
e
-
a
l
i
g
ne
d phone
m
e
a
nnot
a
t
i
ons
, s
uppor
t
i
ng
a
w
i
de
a
r
r
a
y
o
f
a
ppl
i
c
a
t
i
ons
i
nc
l
udi
ng
s
pe
e
c
h a
n
d
phone
m
e
r
e
c
o
g
ni
t
i
on,
s
pe
a
ke
r
i
de
nt
i
f
i
c
a
t
i
on,
a
nd
e
m
ot
i
on
a
na
l
y
s
i
s
.
T
he
da
t
a
s
e
t
e
nc
o
m
p
a
s
s
e
s
m
ul
t
i
pl
e
l
a
ng
ua
g
e
s
,
pr
o
v
i
di
n
g
a
v
a
l
ua
bl
e
t
ool
f
or
c
r
os
s
-
l
i
ng
ui
s
t
i
c
s
t
ud
i
e
s
a
nd
c
om
pa
r
i
s
ons
.
E
a
c
h
r
e
c
or
di
ng
c
o
m
e
s
w
i
t
h
e
xt
e
ns
i
v
e
m
e
t
a
da
t
a
c
o
v
e
r
i
n
g
s
pe
a
ke
r
d
e
m
o
g
r
a
phi
c
s
,
r
e
c
or
di
ng
c
ondi
t
i
ons
,
a
nd
l
i
n
g
ui
s
t
i
c
de
t
a
i
l
s
,
e
nha
nc
i
ng
i
t
s
r
e
s
e
a
r
c
h
ut
i
l
i
t
y
.
T
he
di
v
e
r
s
i
t
y
of
s
pe
a
ke
r
s
,
a
c
c
e
nt
s
,
a
nd
e
n
v
i
r
on
m
e
nt
s
r
e
pr
e
s
e
nt
e
d
i
n
t
he
da
t
a
s
e
t
e
ns
ur
e
s
i
t
s
a
ppl
i
c
a
bi
l
i
t
y
a
c
r
os
s
v
a
r
i
ous
c
ont
e
xt
s
a
nd r
e
s
e
a
r
c
h
g
oa
l
s
.
e
xP
F
D
:
t
he
e
xt
e
nde
d
phone
t
i
c
f
r
a
m
e
da
t
a
(
e
xP
F
D
)
da
t
a
s
e
t
i
s
a
s
ophi
s
t
i
c
a
t
e
d
r
e
s
our
c
e
de
s
i
g
ne
d
f
o
r
phone
t
i
c
a
nd
s
pe
e
c
h
pr
oc
e
s
s
i
ng
r
e
s
e
a
r
c
h.
I
t
i
nc
l
ude
s
hi
g
h
-
f
i
de
l
i
t
y
a
udi
o
r
e
c
o
r
di
ng
s
w
i
t
h
de
t
a
i
l
e
d,
t
i
m
e
-
a
l
i
g
ne
d
phone
t
i
c
f
r
a
m
e
a
nnot
a
t
i
ons
,
pr
o
v
i
di
ng
pr
e
c
i
s
e
c
a
t
e
g
or
y
f
or
phone
t
i
c
e
v
e
nt
s
.
T
he
d
a
t
a
s
e
t
s
uppor
t
s
m
ul
t
i
pl
e
l
a
ng
ua
g
e
s
a
nd
di
a
l
e
c
t
s
,
m
a
ki
ng
i
t
a
n
i
n
v
a
l
ua
bl
e
t
ool
f
or
c
om
pa
r
a
t
i
v
e
phone
t
i
c
s
t
ud
i
e
s
.
I
t
f
e
a
t
ur
e
s
e
xt
e
nde
d
a
nnot
a
t
i
ons
t
ha
t
e
nc
om
pa
s
s
phone
t
i
c
c
ont
e
xt
,
s
p
e
a
ke
r
a
t
t
r
i
but
e
s
,
a
nd
e
nv
i
r
on
m
e
n
t
a
l
c
ondi
t
i
ons
,
a
l
ong
s
i
de
c
o
m
pr
e
h
e
ns
i
v
e
phone
t
i
c
t
r
a
ns
c
r
i
pt
i
ons
.
R
i
c
h
m
e
t
a
da
t
a
a
c
c
o
m
pa
ni
e
s
e
a
c
h
r
e
c
or
di
ng
,
de
t
a
i
l
i
ng
s
pe
a
ke
r
de
m
o
g
r
a
phi
c
s
,
r
e
c
or
di
ng
s
e
t
ups
,
a
nd
l
i
ng
ui
s
t
i
c
ba
c
kg
r
ounds
.
T
hi
s
e
xt
e
ns
i
v
e
d
a
t
a
s
e
t
i
s
i
de
a
l
f
or
a
pp
l
i
c
a
t
i
ons
i
n
phone
t
i
c
r
e
s
e
a
r
c
h,
s
pe
e
c
h
pr
oc
e
s
s
i
ng
,
a
nd
l
a
n
g
ua
g
e
s
t
udi
e
s
,
e
na
bl
i
ng
i
n
-
de
pt
h
a
na
l
y
s
e
s
o
f
phone
t
i
c
pr
ope
r
t
i
e
s
a
nd
v
a
r
i
a
t
i
ons
a
c
r
os
s
di
v
e
r
s
e
l
i
ng
ui
s
t
i
c
c
ont
e
xt
s
.
R
e
s
e
a
r
c
he
r
s
m
us
t
a
dhe
r
e
t
o
e
t
hi
c
a
l
g
ui
de
l
i
ne
s
t
o e
ns
ur
e
t
he
r
e
s
pons
i
bl
e
us
e
a
nd i
n
t
e
g
r
i
t
y
of
t
he
da
t
a
.
4.2.
S
t
at
e
-
of
-
ar
t
t
e
c
h
n
i
q
u
e
s
L
R
,
K
i
m
C
N
N
[
1
6
]
,
F
a
s
t
T
e
xt
[
1
7
]
,
X
M
L
-
C
N
N
[
1
8
]
,
H
A
N
[
19
]
,
a
nd
r
e
g
ul
a
r
i
z
e
d
L
S
T
M
[
2
0
]
a
r
e
e
xa
m
pl
e
s
of
c
on
v
e
nt
i
ona
l
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
t
e
c
hni
que
s
.
H
e
d
w
i
g
5
i
s
a
DL
t
ool
ki
t
t
ha
t
of
f
e
r
s
pr
e
-
i
m
pl
e
m
e
nt
e
d
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
m
ode
l
s
. I
t
pr
o
v
i
de
s
v
a
r
i
ous
a
ppr
oa
c
he
s
f
or
do
c
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on.
T
he
L
D
C
ha
s
de
v
e
l
ope
d
t
he
hi
e
r
a
r
c
hi
c
a
l
g
r
a
ph
c
on
v
ol
ut
i
ona
l
n
e
t
w
or
k
(
H
G
C
N
)
,
a
m
o
de
l
c
a
pa
bl
e
of
pr
oc
e
s
s
i
ng
bot
h t
he
m
i
c
r
o l
e
v
e
l
o
f
a
w
or
d
g
r
a
ph a
nd t
he
m
a
c
r
o
l
e
v
e
l
of
a
s
e
c
t
i
on gr
a
ph. S
e
ve
r
a
l
doc
u
m
e
nt
c
a
t
e
g
or
i
z
a
t
i
on
t
e
c
hni
que
s
ba
s
e
d
on
G
C
N
ha
v
e
b
e
e
n
de
v
e
l
op
e
d,
i
nc
l
udi
ng
T
e
xt
G
C
N
[
2
1
]
a
nd
B
e
r
t
G
C
N
[2
2
]
. T
he
f
oc
us
of
t
he
s
e
t
e
c
hni
que
s
i
s
t
o c
l
a
s
s
i
f
y
node
s
on
g
r
a
phs
a
t
t
he
c
o
r
pus
l
e
v
e
l
.
I
n
a
dd
i
t
i
on, t
he
r
e
a
r
e
ot
he
r
t
e
c
hni
que
s
a
v
a
i
l
a
bl
e
f
or
c
a
t
e
g
or
i
z
i
ng
node
s
on
doc
um
e
nt
-
l
e
v
e
l
ne
t
w
or
ks
,
i
nc
l
udi
ng
T
L
G
N
N
[
2
3
]
a
nd H
y
pe
r
G
A
T
[
2
4
].
T
he
t
r
a
ns
f
or
m
e
r
-
ba
s
e
d
L
D
C
t
e
c
hni
que
s
c
o
m
e
i
n
s
e
v
e
r
a
l
f
or
m
s
.
A
m
on
g
t
he
s
e
i
s
B
E
R
T
+
T
e
xt
R
a
nk
15
]
,
a
t
r
unc
a
t
i
on
ba
s
e
l
i
ne
e
xt
e
ns
i
on
of
B
E
R
T
.
I
t
e
nt
a
i
l
s
us
i
ng
T
e
xt
R
a
nk
t
o
g
e
t
a
n
a
ddi
t
i
ona
l
s
e
t
of
512
t
oke
ns
.
A
not
he
r
m
e
t
hod
t
ha
t
c
a
n
be
us
e
d
i
ns
t
e
a
d
of
t
he
B
E
R
T
+
T
e
xt
R
a
nk
m
ode
l
i
s
B
E
R
T
+
R
a
ndom
[
2
5
]
.
I
t
c
hoos
e
s
up t
o 512 t
oke
ns
a
t
r
a
ndom
f
r
o
m
t
e
xt
s
. F
ur
t
he
r
m
or
e
, R
oB
E
R
T
[
1
]
i
s
a
n
e
xt
e
ns
i
on of
B
E
R
T
t
ha
t
pe
r
m
i
t
s
i
t
s
us
e
i
n
L
D
C
.
O
n
t
op
of
t
he
s
e
gm
e
nt
r
e
pr
e
s
e
nt
a
t
i
ons
,
R
oB
E
R
T
a
ppl
i
e
s
a
r
e
c
ur
r
e
nc
e
l
a
y
e
r
a
f
t
e
r
s
e
gm
e
nt
a
t
i
on.
T
o
B
E
R
T
[
1
]
,
A
t
r
a
ns
f
or
m
e
r
l
a
ye
r
i
s
s
ug
ge
s
t
e
d
a
s
a
s
ubs
t
i
t
u
t
e
f
or
t
he
r
e
c
ur
r
e
nt
l
a
y
e
r
i
n
t
he
R
oB
E
R
T
m
ode
l
.
T
o
m
a
na
ge
l
e
ng
t
h
y
t
e
xt
s
,
a
T
r
a
ns
f
or
m
e
r
m
ode
l
v
a
r
i
a
t
i
on
c
a
l
l
e
d
t
he
L
ong
f
or
m
e
r
[
2
6
]
c
o
m
bi
ne
s
l
oc
a
l
a
nd
g
l
oba
l
a
t
t
e
nt
i
on
pr
oc
e
s
s
e
s
.
T
he
B
i
g
B
i
r
d
[
2
7
]
i
s
a
n
a
ddi
t
i
ona
l
L
ong
f
or
m
e
r
e
xt
e
ns
i
on
t
ha
t
a
dds
r
a
ndo
m
a
t
t
e
nt
i
on.
I
n
a
ddi
t
i
on,
a
hi
e
r
a
r
c
hi
c
a
l
m
ode
l
c
a
l
l
e
d
t
he
H
i
-
T
r
a
ns
f
or
m
e
r
[
3
]
c
onc
e
nt
r
a
t
e
s
on obt
a
i
ni
ng
a
doc
um
e
nt
’
s
w
hol
e
c
ont
e
xt
t
o m
ode
l
s
e
nt
e
nc
e
s
[
28
].
4.3.
R
e
s
u
l
t
s
T
he
F
i
g
ur
e
2 c
ha
r
t
di
s
pl
a
y
s
t
he
T
e
s
t
.F
1 s
c
or
e
s
f
or
v
a
r
i
ous
c
l
a
s
s
i
f
i
c
a
t
i
on
m
e
t
hods
, r
e
v
e
a
l
i
ng
t
ha
t
m
os
t
m
e
t
hods
pe
r
f
or
m
s
i
m
i
l
a
r
l
y
,
w
i
t
h
s
c
or
e
s
r
a
n
g
i
ng
f
r
o
m
a
ppr
oxi
m
a
t
e
l
y
66
t
o
68.
M
e
t
ho
ds
s
uc
h
a
s
K
i
m
C
N
N
[1
6
]
,
T
L
-
G
N
N
[
2
2
]
,
X
M
L
-
C
N
N
[
1
8
]
,
F
a
s
t
T
e
xt
[
1
7
]
,
T
e
xt
G
C
N
[
2
1
]
,
L
S
T
M
(
R
e
g)
[
2
0
]
,
H
A
N
[
19
]
,
L
R
,
B
e
r
t
G
C
N
[
2
4
]
,
B
E
R
T
+
R
a
ndom
[
15
]
,
H
G
C
N
+
B
E
R
T
,
H
i
-
T
r
a
ns
f
or
m
e
r
[
1
5
]
,
B
E
R
T
+
T
e
xt
r
a
n
k
[
2
5
]
,
R
o
B
E
R
T
a
[
1
]
,
L
on
g
f
o
r
m
e
r
[
2
7
]
,
H
y
p
e
r
G
A
T
[
2
3
]
,
B
i
g
B
i
r
d
[3
]
,
H
G
C
N
-
B
i
g
B
i
r
d
[
1
5
]
,
T
oB
E
R
T
[
1
]
,
H
M
G
C
N
+
B
E
R
T
[
2
8
]
,
a
nd
H
M
G
C
N
+
B
i
g
B
i
r
d
[
2
8
]
a
l
l
f
a
l
l
w
i
t
hi
n
t
hi
s
r
a
ng
e
,
i
ndi
c
a
t
i
ng
c
o
m
pa
r
a
bl
e
e
f
f
e
c
t
i
v
e
ne
s
s
i
n
t
he
t
a
s
k
a
t
ha
nd.
H
o
w
e
v
e
r
,
t
he
P
S
m
e
t
hod
s
t
a
nds
out
s
i
g
ni
f
i
c
a
nt
l
y
,
a
c
hi
e
v
i
n
g
a
T
e
s
t
.F
1
s
c
or
e
of
a
ppr
oxi
m
a
t
e
l
y
79,
w
hi
c
h
i
s
not
i
c
e
a
bl
y
hi
g
he
r
t
ha
n
t
he
ot
he
r
s
.
T
hi
s
s
ug
g
e
s
t
s
t
ha
t
t
he
P
S
m
e
t
hod
i
nc
or
por
a
t
e
s
m
or
e
a
d
v
a
nc
e
d
t
e
c
hni
que
s
or
opt
i
m
i
z
a
t
i
ons
,
l
e
a
di
ng
t
o
i
t
s
s
upe
r
i
or
pe
r
f
or
m
a
n
c
e
.
T
he
c
l
e
a
r
g
a
p
be
t
w
e
e
n
P
S
a
nd
t
he
o
t
he
r
m
e
t
hod
s
hi
g
hl
i
g
ht
s
i
t
s
r
obus
t
ne
s
s
a
nd
pot
e
nt
i
a
l
a
d
v
a
nt
a
g
e
s
,
m
a
ki
n
g
i
t
a
pr
o
m
i
s
i
ng
c
hoi
c
e
f
or
t
a
s
ks
r
e
qui
r
i
n
g
hi
g
h a
c
c
ur
a
c
y
a
nd r
e
l
i
a
bi
l
i
t
y
i
n c
l
a
s
s
i
f
i
c
a
t
i
on.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
,
V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
11
00
-
11
08
1106
F
i
g
ur
e
2
.
T
e
s
t
.F
1 r
e
s
ul
t
s
f
or
e
xA
A
P
D
2 da
t
a
s
e
t
i
n c
o
m
pa
r
i
s
on w
i
t
h t
he
e
xi
s
t
i
ng
s
t
a
t
e
-
of
-
a
r
t
t
e
c
hni
que
s
w
i
t
h P
S
F
i
g
ur
e
3
de
pi
c
t
s
t
he
T
e
s
t
.F
1
s
c
or
e
s
f
or
v
a
r
i
ous
c
l
a
s
s
i
f
i
c
a
t
i
on
m
e
t
hods
f
or
e
xP
F
D
da
t
a
s
e
t
,
hi
g
hl
i
g
ht
i
ng
t
he
i
r
pe
r
f
or
m
a
n
c
e
di
f
f
e
r
e
nc
e
s
.
T
he
m
e
t
hods
r
a
n
g
e
f
r
o
m
T
L
-
G
N
N
,
w
hi
c
h
s
c
or
e
s
a
r
ound
76,
t
o
P
S
, w
hi
c
h a
c
hi
e
v
e
s
t
he
hi
g
he
s
t
s
c
or
e
a
t
a
pp
r
oxi
m
a
t
e
l
y
88. M
os
t
m
e
t
hods
, i
nc
l
udi
ng
B
e
r
t
G
C
N
[
2
4
]
, T
e
xt
G
C
N
[2
1
]
,
L
R
,
H
y
pe
r
G
A
T
[
2
3
]
,
K
i
m
C
N
N
[
1
6
]
,
F
a
s
t
T
e
xt
[
1
7
]
,
X
M
L
-
C
N
N
[
1
8
]
,
H
i
-
T
r
a
ns
f
or
m
e
r
,
H
A
N
,
L
S
T
M
(
R
e
g
)
,
L
on
g
f
or
m
e
r
,
B
i
g
B
i
r
d,
B
E
R
T
+
R
a
ndom
,
R
oB
E
R
T
a
,
H
G
C
N
-
B
E
R
T
,
H
G
A
T
-
B
i
g
B
i
r
d,
T
oB
E
R
T
,
H
G
C
N
+
B
i
g
B
i
r
d,
B
E
R
T
+
T
e
xt
r
a
nk,
H
M
G
C
N
+
B
E
R
T
,
a
nd
H
M
G
C
N
+
B
i
g
B
i
r
d,
f
a
l
l
w
i
t
hi
n
a
na
r
r
o
w
e
r
r
a
nge
of
78 t
o 84. T
hi
s
i
ndi
c
a
t
e
s
a
c
l
os
e
c
l
us
t
e
r
i
ng
of
pe
r
f
or
m
a
n
c
e
a
m
on
g
t
he
s
e
m
e
t
hods
, s
ugg
e
s
t
i
ng t
he
y
of
f
e
r
s
i
m
i
l
a
r
e
f
f
e
c
t
i
v
e
ne
s
s
f
or
t
he
g
i
v
e
n
t
a
s
k.
H
o
w
e
v
e
r
,
t
he
P
S
m
e
t
hod
di
s
t
i
nc
t
l
y
out
pe
r
f
or
m
s
a
l
l
ot
he
r
s
w
i
t
h
i
t
s
T
e
s
t
.F
1
s
c
or
e
of
88,
i
ndi
c
a
t
i
ng
a
s
i
g
ni
f
i
c
a
nt
pe
r
f
or
m
a
nc
e
a
d
v
a
nt
a
g
e
.
T
hi
s
s
ug
ge
s
t
s
t
ha
t
t
he
P
S
m
e
t
hod
ut
i
l
i
z
e
s
m
or
e
a
d
v
a
n
c
e
d
or
opt
i
m
i
z
e
d
t
e
c
hni
que
s
,
s
e
t
t
i
ng
i
t
a
pa
r
t
f
r
o
m
t
he
ot
he
r
m
e
t
hods
a
nd
m
a
ki
ng
i
t
a
pa
r
t
i
c
ul
a
r
l
y
r
obus
t
c
hoi
c
e
f
or
a
c
hi
e
v
i
n
g
hi
g
h a
c
c
ur
a
c
y
i
n c
l
a
s
s
i
f
i
c
a
t
i
on t
a
s
ks
.
F
i
g
ur
e
3
.
T
e
s
t
.F
1 r
e
s
ul
t
f
or
e
xP
F
D
da
t
a
s
e
t
i
n c
om
pa
r
i
s
on
w
i
t
h t
he
e
xi
s
t
i
ng
s
t
a
t
e
-
of
-
a
r
t
t
e
c
hni
que
s
w
i
t
h P
S
5.
C
O
N
C
L
U
S
I
O
N
I
n
c
onc
l
us
i
on,
t
he
H
D
C
C
N
E
T
pr
e
s
e
nt
s
a
s
i
g
ni
f
i
c
a
nt
a
d
v
a
nc
e
m
e
nt
i
n
t
he
f
i
e
l
d
of
doc
um
e
n
t
a
na
l
y
s
i
s
by
a
ddr
e
s
s
i
ng
t
he
dua
l
c
ha
l
l
e
n
g
e
s
of
a
c
c
ur
a
c
y
a
nd
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
i
n
m
ul
t
i
-
c
a
t
e
g
or
y
c
l
a
s
s
i
f
i
c
a
t
i
on
t
a
s
ks
.
B
y
i
nt
r
oduc
i
ng
a
m
ul
t
i
-
l
a
ye
r
c
a
t
e
g
or
y
c
or
r
e
l
a
t
i
on
m
o
dul
e
,
H
D
C
C
N
E
T
e
nha
nc
e
s
t
he
de
pt
h
of
c
a
t
e
g
or
y
0
10
20
30
40
50
60
70
80
90
va
l
ue
M
e
t
h
od
T
es
t
.
F1
72
74
76
78
80
82
84
86
88
90
va
l
ue
M
e
t
h
od
T
es
t
.
F1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
E
nhanc
i
ng
d
oc
um
e
nt
t
e
x
t
c
l
as
s
i
f
i
c
at
i
on us
i
n
g hy
b
r
i
d
d
e
e
p c
ont
e
x
t
ual
an
d c
or
r
e
l
at
i
o
n ne
t
w
or
k
(
Shi
l
pa
)
1107
c
onne
c
t
i
ons
,
l
e
a
di
ng
t
o
i
m
pr
o
v
e
d
pr
e
di
c
t
i
on
a
c
c
ur
a
c
y
a
nd
a
m
or
e
nua
n
c
e
d
unde
r
s
t
a
n
di
ng
of
c
a
t
e
g
or
y
i
nt
e
r
r
e
l
a
t
i
ons
.
T
he
i
nc
or
por
a
t
i
on
of
r
e
s
i
dua
l
c
onne
c
t
i
ons
e
ns
ur
e
s
s
t
a
bi
l
i
t
y
a
nd
pr
e
v
e
nt
s
di
v
e
r
g
e
nc
e
i
n
pr
e
di
c
t
i
on
s
,
c
ont
r
i
but
i
ng
t
o
t
he
m
ode
l
’
s
r
e
l
i
a
bi
l
i
t
y
.
F
ur
t
he
r
m
or
e
,
H
D
C
C
N
E
T
’
s
de
s
i
g
n,
w
hi
c
h
r
e
duc
e
s
t
he
num
b
e
r
of
pa
r
a
m
e
t
e
r
s
,
not
onl
y
a
c
c
e
l
e
r
a
t
e
s
m
od
e
l
c
on
v
e
r
g
e
n
c
e
but
a
l
s
o
m
a
k
e
s
i
t
m
or
e
e
f
f
i
c
i
e
nt
a
nd
s
c
a
l
a
bl
e
f
or
pr
a
c
t
i
c
a
l
a
ppl
i
c
a
t
i
ons
.
T
hi
s
e
f
f
i
c
i
e
nc
y
i
s
pa
r
t
i
c
ul
a
r
l
y
v
a
l
ua
bl
e
i
n
bus
i
ne
s
s
e
nv
i
r
onm
e
nt
s
w
he
r
e
c
o
m
put
a
t
i
ona
l
r
e
s
our
c
e
s
a
nd
t
i
m
e
a
r
e
of
t
e
n
l
i
m
i
t
e
d.
B
y
br
i
dg
i
ng
t
he
g
a
p
be
t
w
e
e
n
t
he
pow
e
r
f
ul
c
a
pa
bi
l
i
t
i
e
s
of
DL
a
nd
t
he
pr
a
c
t
i
c
a
l
ne
e
d
f
or
t
r
a
ns
pa
r
e
nc
y
a
nd
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
,
H
D
C
C
N
E
T
pr
o
v
i
de
s
a
r
obus
t
a
nd
s
c
a
l
a
bl
e
s
ol
ut
i
on
f
or
a
ut
om
a
t
e
d
doc
u
m
e
nt
pr
oc
e
s
s
i
ng
.
T
he
c
ont
r
i
but
i
ons
of
H
D
C
C
N
E
T
pa
v
e
t
he
w
a
y
f
or
t
he
br
oa
de
r
a
dopt
i
on
of
DL
t
e
c
hnol
og
i
e
s
i
n
v
a
r
i
ous
bus
i
ne
s
s
w
or
kf
l
o
w
s
,
of
f
e
r
i
ng
a
r
e
l
i
a
bl
e
,
t
r
a
ns
pa
r
e
nt
,
a
nd
e
f
f
i
c
i
e
n
t
a
ppr
oa
c
h t
o doc
u
m
e
nt
a
n
a
l
y
s
i
s
. F
ut
ur
e
w
or
k
w
i
l
l
e
xpl
or
e
t
he
a
ppl
i
c
a
t
i
on of
H
D
C
C
N
E
T
t
o
ot
he
r
do
m
a
i
ns
a
n
d
f
ur
t
he
r
e
nha
nc
e
m
e
nt
s
t
o i
m
pr
o
v
e
i
t
s
r
obus
t
ne
s
s
a
nd a
ppl
i
c
a
bi
l
i
t
y
i
n di
v
e
r
s
e
r
e
a
l
-
w
or
l
d s
c
e
n
a
r
i
os
.
A
C
K
N
O
WL
E
D
G
E
M
E
N
T
I
w
oul
d
l
i
ke
t
o
e
xpr
e
s
s
our
s
i
nc
e
r
e
g
r
a
t
i
t
ude
t
o
a
l
l
t
hos
e
w
ho
ha
v
e
s
uppor
t
e
d
a
nd
c
ont
r
i
but
e
d
t
o
t
hi
s
r
e
s
e
a
r
c
h pr
oj
e
c
t
. P
r
i
m
a
r
i
l
y
,
I
e
xt
e
nd our
he
a
r
t
f
e
l
t
t
ha
nks
t
o our
g
ui
de
f
or
he
r
unw
a
v
e
r
i
n
g
g
u
i
da
nc
e
, i
n
v
a
l
ua
bl
e
i
ns
i
g
ht
s
, a
nd e
nc
our
a
g
e
m
e
nt
t
hr
oug
hout
t
he
r
e
s
e
a
r
c
h pr
oc
e
s
s
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
T
he
a
ut
hor
s
s
t
a
t
e
no f
undi
ng
i
n
v
ol
v
e
d.
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
A
ut
hor
s
s
t
a
t
e
no c
onf
l
i
c
t
of
i
nt
e
r
e
s
t
.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
D
a
t
a
a
v
a
i
l
a
bi
l
i
t
y
i
s
not
a
ppl
i
c
a
bl
e
t
o
t
hi
s
pa
pe
r
a
s
no
ne
w
da
t
a
w
e
r
e
c
r
e
a
t
e
d
o
r
a
na
l
y
z
e
d
i
n
t
hi
s
s
t
u
dy
.
R
E
F
E
R
E
N
C
E
S
[
1]
R
.
P
a
ppa
g
a
r
i
,
P
.
Z
e
l
a
s
ko,
J
.
V
i
l
l
a
l
ba
,
Y
.
C
a
r
m
i
e
l
,
a
n
d
N
.
D
e
h
a
k,
“
H
i
e
r
a
r
c
h
i
c
a
l
t
r
a
n
s
f
o
r
m
e
r
s
f
or
l
o
ng
doc
u
m
e
n
t
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
i
n
2019
I
E
E
E
A
ut
om
a
t
i
c
Spe
e
c
h
R
e
c
ogni
t
i
o
n
and
U
nde
r
s
t
andi
ng
W
or
k
s
hop
(
A
SR
U
)
,
D
e
c
.
2019,
pp.
838
–
844,
doi
:
10.
1109/
A
S
R
U
46091.
2019
.
900395
8.
[
2]
X
.
Z
h
a
ng,
F
.
W
e
i
,
a
n
d
M
.
Z
h
ou
,
“
H
I
B
E
R
T
:
doc
u
m
e
n
t
l
e
ve
l
pr
e
-
t
r
a
i
n
i
ng
of
h
i
e
r
a
r
c
h
i
c
a
l
bi
di
r
e
c
t
i
o
n
a
l
t
r
a
n
s
f
or
m
e
r
s
f
or
doc
um
e
nt
s
um
m
a
r
i
z
a
t
i
o
n,
”
i
n
P
r
o
c
e
e
di
ngs
of
t
he
57t
h
A
nnual
M
e
e
t
i
n
g
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
,
2019,
pp.
5059
–
5069,
doi
:
10.
18653/
v1
/
P
19
-
1499.
[
3]
C
.
W
u,
F
.
W
u
,
T
.
Q
i
,
a
n
d
Y
.
H
ua
ng
,
“
H
i
-
t
r
a
n
s
f
o
r
m
e
r
:
h
i
e
r
a
r
c
h
i
c
a
l
i
n
t
e
r
a
c
t
i
ve
t
r
a
n
s
f
or
m
e
r
f
or
e
f
f
i
c
i
e
n
t
a
n
d
e
f
f
e
c
t
i
ve
l
on
g
doc
um
e
nt
m
ode
l
i
ng,
”
i
n
P
r
o
c
e
e
di
ngs
of
t
he
59t
h
A
nnual
M
e
e
t
i
ng
of
t
he
A
s
s
oc
i
a
t
i
on
f
or
C
om
put
a
t
i
onal
L
i
ngui
s
t
i
c
s
and
t
he
11t
h
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
on
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng
(
V
ol
um
e
2:
Shor
t
P
ape
r
s
)
,
2021
,
vol
.
2,
pp
.
848
–
853,
doi
:
10.
18653/
v1/
2021.
a
c
l
-
s
h
or
t
.
107.
[
4]
S
a
i
f
ul
l
a
h,
S
.
A
g
n
e
,
A
.
D
e
n
g
e
l
,
a
n
d
S
.
A
hm
e
d,
“
T
h
e
r
e
a
l
i
t
y
of
h
i
gh
pe
r
f
o
r
m
i
ng
de
e
p
l
e
a
r
n
i
ng
m
ode
l
s
:
a
c
a
s
e
s
t
u
dy
on
doc
um
e
nt
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n,
”
I
E
E
E
A
c
c
e
s
s
,
vol
.
12,
pp.
103537
–
103564,
2024
,
doi
:
10.
1109/
A
C
C
E
S
S
.
2024.
3425910.
[
5]
S
.
S
a
i
f
ul
l
a
h,
S
.
A
gn
e
,
A
.
D
e
n
g
e
l
,
a
n
d
S
.
A
hm
e
d,
“
D
oc
X
c
l
a
s
s
i
f
i
e
r
:
t
ow
a
r
ds
a
r
obus
t
a
n
d
i
n
t
e
r
p
r
e
t
a
bl
e
de
e
p
n
e
u
r
a
l
n
e
t
w
or
k
f
o
r
doc
um
e
nt
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
I
nt
e
r
nat
i
onal
J
our
nal
on
D
o
c
um
e
nt
A
nal
y
s
i
s
and
R
e
c
ogni
t
i
on
,
vo
l
.
27,
n
o.
3,
pp
.
447
–
473,
S
e
p.
2024,
doi
:
10.
1007/
s
10032
-
024
-
0
0483
-
w.
[
6]
R
.
P
o
w
a
l
s
ki
,
Ł
.
B
o
r
c
hm
a
nn,
D
.
J
ur
ki
e
w
i
c
z
,
T
.
D
w
o
j
a
k,
M
.
P
i
e
t
r
us
z
ka
,
a
n
d
G
.
P
a
ł
ka
,
“
G
oi
ng
f
ul
l
-
T
I
L
T
boo
gi
e
on
doc
u
m
e
nt
un
de
r
s
t
a
n
d
i
ng
w
i
t
h
t
e
x
t
-
i
m
a
g
e
-
l
a
you
t
t
r
a
n
s
f
o
r
m
e
r
,
”
i
n
L
e
c
t
ur
e
N
ot
e
s
i
n
C
om
put
e
r
Sc
i
e
nc
e
(
i
n
c
l
udi
ng
s
ubs
e
r
i
e
s
L
e
c
t
ur
e
N
ot
e
s
i
n
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
n
c
e
and
L
e
c
t
ur
e
N
ot
e
s
i
n
B
i
oi
nf
or
m
at
i
c
s
)
,
vol
.
12822
L
N
C
S
,
2021
,
pp.
732
–
747.
[
7]
J
.
F
e
r
r
a
n
do
e
t
al
.
,
“
I
m
pr
ovi
ng
a
c
c
u
r
a
c
y
a
n
d
s
pe
e
di
ng
up
doc
u
m
e
n
t
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
h
r
ou
gh
pa
r
a
l
l
e
l
s
ys
t
e
m
s
,
”
i
n
L
e
c
t
ur
e
N
ot
e
s
i
n
C
om
put
e
r
Sc
i
e
nc
e
(
i
n
c
l
udi
ng
s
ubs
e
r
i
e
s
L
e
c
t
ur
e
N
ot
e
s
i
n
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
n
c
e
and
L
e
c
t
u
r
e
N
ot
e
s
i
n
B
i
oi
nf
or
m
at
i
c
s
)
,
vol
.
12138
L
N
C
S
,
2020
,
pp.
387
–
400.
[
8]
S
a
i
f
ul
l
a
h,
S
.
A
.
S
i
ddi
qui
,
S
.
A
gn
e
,
A
.
D
e
ng
e
l
,
a
n
d
S
.
A
hm
e
d
,
“
A
r
e
de
e
p
m
ode
l
s
r
ob
us
t
a
g
a
i
n
s
t
r
e
a
l
di
s
t
or
t
i
o
n
s
?
A
c
a
s
e
s
t
udy
o
n
doc
um
e
nt
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n,
”
i
n
2022
2
6t
h
I
n
t
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
P
a
t
t
e
r
n
R
e
c
ogni
t
i
on
(
I
C
P
R
)
,
A
u
g
.
2
022,
vol
.
2022
-
A
ug
us
,
pp.
1628
–
1635,
doi
:
10.
1109/
I
C
P
R
56361.
2022.
9956167.
[
9]
S
.
S
a
i
f
ul
l
a
h
,
S
.
A
gn
e
,
A
.
D
e
ng
e
l
,
a
n
d
S
.
A
hm
e
d
,
“
A
n
a
l
y
z
i
ng
t
h
e
pot
e
n
t
i
a
l
of
a
c
t
i
ve
l
e
a
r
ni
ng
f
or
doc
u
m
e
nt
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
on
,
”
I
nt
e
r
nat
i
onal
J
our
nal
o
n
D
o
c
um
e
nt
A
nal
y
s
i
s
and
R
e
c
ogni
t
i
on
(
I
J
D
A
R
)
,
vol
.
26,
no.
3,
pp.
187
–
209
,
S
e
p.
2023,
doi
:
10.
1007/
s
10032
-
023
-
00429
-
8.
[
10]
P
.
P
uj
a
r
,
A
.
K
u
m
a
r
,
a
n
d
V
.
K
u
m
a
r
,
“
E
f
f
i
c
i
e
n
t
pl
a
n
t
l
e
a
f
de
t
e
c
t
i
on
t
h
r
ou
gh
m
a
c
h
i
n
e
l
e
a
r
n
i
ng
a
ppr
oa
c
h
ba
s
e
d
on
c
or
n
l
e
a
f
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
I
A
E
S
I
nt
e
r
nat
i
onal
J
our
nal
of
A
r
t
i
f
i
c
i
a
l
I
nt
e
l
l
i
ge
n
c
e
(
I
J
-
A
I
)
,
vol
.
13,
n
o.
1,
p.
1139,
M
a
r
.
2024,
doi
:
10.
11591/
i
j
a
i
.
v13.
i
1.
pp1139
-
11
48.
[
11]
S
.
H
.
S
r
e
e
d
h
a
r
a
,
V
.
K
u
m
a
r
,
a
n
d
S
.
S
a
l
m
a
,
“
E
f
f
i
c
i
e
n
t
bi
g
da
t
a
c
l
us
t
e
r
i
ng
us
i
ng
a
d
h
oc
f
u
z
z
y
C
m
e
a
n
s
a
n
d
a
ut
o
-
e
nc
ode
r
C
N
N
,
”
i
n
L
e
c
t
ur
e
N
ot
e
s
i
n
N
e
t
w
or
k
s
and
Sy
s
t
e
m
s
,
vol
.
563,
2023,
pp.
353
–
368.
[
12]
R
.
K
h
a
s
a
w
n
e
h
a
n
d
R
.
K
o
r
n
r
e
i
c
h,
“
E
x
p
l
a
i
n
i
ng
da
t
a
-
dr
i
ve
n
doc
u
m
e
n
t
c
l
a
s
s
i
f
i
c
a
t
i
on
s
,
”
M
I
S
Q
uar
t
e
r
l
y
:
M
anage
m
e
nt
I
nf
or
m
at
i
on
Sy
s
t
e
m
s
,
vol
.
3,
n
o.
4,
pp.
781
–
791,
2014.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
,
V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
11
00
-
11
08
1108
[
13]
T
.
V
e
r
m
e
i
r
e
,
D
.
B
r
u
ghm
a
n
s
,
S
.
G
oe
t
h
a
l
s
,
R
.
M
.
B
.
de
O
l
i
ve
i
r
a
,
a
n
d
D
.
M
a
r
t
e
n
s
,
“
E
x
pl
a
i
n
a
bl
e
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
w
i
t
h
e
vi
de
n
c
e
c
oun
t
e
r
f
a
c
t
ua
l
,
”
P
at
t
e
r
n
A
nal
y
s
i
s
and
A
ppl
i
c
at
i
ons
,
vol
.
25,
n
o.
2,
pp.
315
–
335,
M
a
y
2022,
doi
:
10.
1007/
s
10044
-
0
21
-
01055
-
y.
[
14]
O
.
L
a
ng
e
t
al
.
,
“
E
x
pl
a
i
ni
ng
i
n
s
t
yl
e
:
t
r
a
i
n
i
ng
a
G
A
N
t
o
e
x
p
l
a
i
n
a
c
l
a
s
s
i
f
i
e
r
i
n
S
t
yl
e
S
pa
c
e
,
”
i
n
2
021
I
E
E
E
/
C
V
F
I
nt
e
r
nat
i
o
na
l
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
(
I
C
C
V
)
,
O
c
t
.
2021,
pp.
673
–
682,
doi
:
10.
1109/
I
C
C
V
48922.
2021.
00073.
[1
5
]
T
.
L
i
u
,
Y
.
H
u,
B
.
W
a
ng,
Y
.
S
un
,
J
.
G
a
o,
a
n
d
B
.
Y
i
n
,
“
H
i
e
r
a
r
c
h
i
c
a
l
gr
a
p
h
c
on
vol
ut
i
o
n
a
l
n
e
t
w
or
ks
f
or
s
t
r
uc
t
ur
e
d
l
ong
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
I
E
E
E
T
r
ans
ac
t
i
ons
on
N
e
ur
al
N
e
t
w
or
k
s
and
L
e
ar
ni
ng
Sy
s
t
e
m
s
,
vol
.
34,
n
o.
10
,
pp.
8071
–
8085
,
O
c
t
.
2023,
doi
:
10.
1109/
T
N
N
L
S
.
2022.
3185295.
[1
6
]
Y
.
K
i
m
,
“
C
o
n
vo
l
ut
i
o
n
a
l
n
e
ur
a
l
n
e
t
w
or
ks
f
or
s
e
n
t
e
n
c
e
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
i
n
P
r
oc
e
e
di
ngs
of
t
he
2014
C
on
f
e
r
e
nc
e
on
E
m
pi
r
i
c
al
M
e
t
hods
i
n
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng
(
E
M
N
L
P
)
,
2014,
pp.
1746
–
1751,
doi
:
10.
3115/
v1/
D
14
-
1181.
[1
7
]
A
.
J
oul
i
n
,
E
.
G
r
a
ve
,
P
.
B
oj
a
n
o
w
s
ki
,
a
n
d
T
.
M
i
kol
ov,
“
B
a
g
of
t
r
i
c
ks
f
or
e
f
f
i
c
i
e
n
t
t
e
x
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n,
”
i
n
P
r
oc
e
e
d
i
ngs
of
t
he
15t
h
C
onf
e
r
e
nc
e
o
f
t
he
E
ur
ope
an
C
hapt
e
r
of
t
he
A
s
s
oc
i
a
t
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
:
V
ol
um
e
2,
Shor
t
P
ape
r
s
,
2017,
vol
.
2,
pp.
427
–
431,
doi
:
10
.
18653/
v1/
E
17
-
206
8.
[1
8
]
J
.
L
i
u
,
W
.
-
C
.
C
h
a
ng,
Y
.
W
u,
a
n
d
Y
.
Y
a
n
g
,
“
D
e
e
p
l
e
a
r
n
i
ng
f
or
e
x
t
r
e
m
e
m
ul
t
i
-
l
a
be
l
t
e
x
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
i
n
P
r
oc
e
e
d
i
ngs
o
f
t
he
40t
h
I
nt
e
r
nat
i
onal
A
C
M
S
I
G
I
R
C
onf
e
r
e
nc
e
on
R
e
s
e
ar
c
h
and
D
e
v
e
l
opm
e
nt
i
n
I
nf
or
m
at
i
on
R
e
t
r
i
e
v
al
,
A
u
g
.
2017
,
pp
.
115
–
124,
doi
:
10.
1145/
3077136.
3080834.
[
19
]
Z
.
Y
a
ng,
D
.
Y
a
n
g,
C
.
D
ye
r
,
X
.
H
e
,
A
.
S
m
ol
a
,
a
n
d
E
.
H
ovy
,
“
H
i
e
r
a
r
c
h
i
c
a
l
a
t
t
e
nt
i
o
n
n
e
t
w
o
r
ks
f
or
doc
u
m
e
n
t
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
i
n
P
r
oc
e
e
di
ngs
of
t
he
2016
C
onf
e
r
e
nc
e
of
t
he
N
o
r
t
h
A
m
e
r
i
c
an
C
hapt
e
r
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
n
gui
s
t
i
c
s
:
H
um
a
n
L
anguage
T
e
c
hnol
ogi
e
s
,
2016
,
pp.
1480
–
1489,
doi
:
10.
18653/
v1/
N
16
-
1174.
[2
0
]
A
.
A
dh
i
ka
r
i
,
A
.
R
a
m
,
R
.
T
a
ng
,
a
n
d
J
.
L
i
n,
“
R
e
t
h
i
nki
ng
c
o
m
pl
e
x
n
e
u
r
a
l
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
ur
e
s
f
or
doc
u
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
i
n
P
r
oc
e
e
di
ngs
of
t
he
201
9
C
on
f
e
r
e
nc
e
o
f
t
he
N
or
t
h
,
2019,
vol
.
1,
pp.
4046
–
4051,
doi
:
10.
18653/
v1/
N
1
9
-
1408.
[2
1
]
L
.
Y
a
o,
C
.
M
a
o,
a
n
d
Y
.
L
uo
,
“
G
r
a
p
h
c
onvol
ut
i
o
n
a
l
n
e
t
w
or
ks
f
or
t
e
x
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n,
”
33r
d
A
A
A
I
C
onf
e
r
e
nc
e
on
A
r
t
i
f
i
c
i
a
l
I
nt
e
l
l
i
ge
n
c
e
,
A
A
A
I
2019,
31s
t
I
nnov
at
i
v
e
A
ppl
i
c
at
i
ons
of
A
r
t
i
f
i
c
i
al
I
n
t
e
l
l
i
ge
n
c
e
C
onf
e
r
e
nc
e
,
I
A
A
I
2019
and
t
he
9t
h
A
A
A
I
Sy
m
pos
i
um
on
E
duc
at
i
onal
A
dv
an
c
e
s
i
n
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
n
c
e
,
E
A
A
I
2019
,
pp.
7370
–
7377,
2019,
doi
:
10.
48550/
a
r
X
i
v.
1809.
05679.
[2
2
]
L
.
H
ua
ng
,
D
.
M
a
,
S
.
L
i
,
X
.
Z
h
a
ng,
a
n
d
H
.
W
a
ng
,
“
T
e
xt
l
e
ve
l
g
r
a
p
h
n
e
ur
a
l
n
e
t
w
o
r
k
f
or
t
e
x
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
,
”
i
n
P
r
oc
e
e
di
ngs
of
t
he
2019
C
onf
e
r
e
nc
e
on
E
m
pi
r
i
c
al
M
e
t
hods
i
n
N
at
ur
al
L
anguage
P
r
oc
e
s
s
i
ng
and
t
he
9t
h
I
nt
e
r
nat
i
onal
J
oi
nt
C
onf
e
r
e
nc
e
on
N
at
ur
al
L
anguage
P
r
o
c
e
s
s
i
ng
(
E
M
N
L
P
-
I
J
C
N
L
P
)
,
2019
,
pp.
3442
–
3448,
doi
:
10.
18653/
v1
/
D
19
-
1
345.
[2
3
]
K
.
D
i
ng
,
J
.
W
a
ng,
J
.
L
i
,
D
.
L
i
,
a
n
d
H
.
L
i
u,
“
B
e
m
o
r
e
w
i
t
h
l
e
s
s
:
h
ype
r
g
r
a
p
h
a
t
t
e
n
t
i
o
n
n
e
t
w
or
ks
f
o
r
i
n
duc
t
i
ve
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
i
n
P
r
oc
e
e
di
ngs
of
t
he
2020
C
onf
e
r
e
nc
e
on
E
m
pi
r
i
c
al
M
e
t
h
ods
i
n
N
at
ur
al
L
an
guage
P
r
o
c
e
s
s
i
ng
(
E
M
N
L
P
)
,
2020
,
pp.
4927
–
4936
,
doi
:
10.
18653/
v1/
2020.
e
m
n
l
p
-
m
a
i
n
.
399.
[2
4
]
Y
.
L
i
n
e
t
al
.
,
“
B
e
r
t
G
C
N
:
t
r
a
n
s
duc
t
i
ve
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on
by
c
om
bi
n
i
ng
G
N
N
a
n
d
B
E
R
T
,
”
i
n
F
i
ndi
ngs
of
t
he
A
s
s
o
c
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
:
A
C
L
-
I
J
C
N
L
P
2021
,
2021,
pp.
1456
–
1462
,
doi
:
10.
18653/
v1/
2021.
f
i
n
di
ng
s
-
a
c
l
.
126.
[2
5
]
H
.
P
a
r
k,
Y
.
V
ya
s
,
a
n
d
K
.
S
h
a
h,
“
E
f
f
i
c
i
e
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
of
l
ong
doc
u
m
e
n
t
s
us
i
ng
t
r
a
n
s
f
or
m
e
r
s
,
”
i
n
P
r
oc
e
e
di
ngs
of
t
he
60t
h
A
nnual
M
e
e
t
i
ng
of
t
he
A
s
s
oc
i
at
i
on
f
or
C
om
put
at
i
onal
L
i
ngui
s
t
i
c
s
(
V
ol
um
e
2:
Shor
t
P
ape
r
s
)
,
20
22,
vol
.
2,
pp.
702
–
709,
doi
:
10.
18653/
v1/
2022.
a
c
l
-
s
h
or
t
.
79.
[2
6
]
I
.
B
e
l
t
a
g
y
,
M
.
E
.
P
e
t
e
r
s
,
a
n
d
A
.
C
oh
a
n
,
“
L
o
ngf
or
m
e
r
:
t
h
e
l
ong
-
doc
u
m
e
n
t
t
r
a
n
s
f
o
r
m
e
r
,
”
ar
X
i
v
,
2020,
[
O
n
l
i
n
e
]
.
A
va
i
l
a
bl
e
:
h
t
t
p:
/
/
a
r
xi
v.
or
g
/
a
bs
/
2004.
05150.
[
2
7
]
M
.
Z
a
h
e
e
r
e
t
al
.
,
“
B
i
g
bi
r
d:
t
r
a
n
s
f
or
m
e
r
s
f
o
r
l
o
ng
e
r
s
e
que
n
c
e
s
,
”
A
dv
an
c
e
s
i
n
N
e
ur
al
I
nf
or
m
at
i
on
P
r
oc
e
s
s
i
ng
Sy
s
t
e
m
s
,
vol
.
2020
-
D
e
c
e
m
,
pp.
17283
–
1729
7,
J
a
n
.
2021,
[
O
nl
i
n
e
]
.
A
va
i
l
a
bl
e
:
h
t
t
p:
/
/
a
r
x
i
v.
or
g
/
a
bs
/
2
007.
14062.
[
28
]
T
.
L
i
u
,
Y
.
H
u,
J
.
G
a
o,
Y
.
S
un
,
a
n
d
B
.
Y
i
n,
“
H
i
e
r
a
r
c
h
i
c
a
l
m
ul
t
i
-
gr
a
n
ul
a
r
i
t
y
i
n
t
e
r
a
c
t
i
o
n
g
r
a
p
h
c
o
n
vol
ut
i
o
n
a
l
ne
t
w
o
r
k
f
or
l
on
g
doc
um
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on,
”
I
E
E
E
/
A
C
M
T
r
ans
ac
t
i
ons
on
A
udi
o,
Spe
e
c
h,
and
L
anguage
P
r
oc
e
s
s
i
ng
,
vol
.
32,
pp.
1762
–
1775,
2024
,
doi
:
10.
1109/
T
A
S
L
P
.
2024.
3369
530.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
M
r
s
.
S
h
il
p
a
r
e
c
e
i
v
e
d
h
e
r
b
a
c
h
e
l
o
r
s
d
e
g
r
e
e
i
n
C
o
m
p
u
ter
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
f
r
o
m
th
e
Vi
s
v
e
s
v
a
r
a
y
a
T
e
c
h
n
o
l
o
g
i
c
a
l
U
n
iv
e
r
s
i
ty
,
B
E
L
GA
U
M
-
I
n
d
i
a
i
n
2
0
1
0
a
n
d
M
a
s
ter
De
g
r
e
e
i
n
C
o
m
p
u
ter
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
f
r
o
m
s
a
m
e
U
n
iv
e
r
s
i
ty
i
n
2
0
1
2
.
S
h
e
i
s
c
u
r
r
e
n
tl
y
p
u
r
s
u
i
n
g
h
e
r
P
h
.
D.
d
e
g
r
e
e
f
r
o
m
th
e
s
a
m
e
u
n
iv
e
r
s
i
ty
.
S
h
e
i
s
p
r
e
s
e
n
t
l
y
wo
r
k
i
n
g
a
s
a
s
s
i
s
tan
t
p
r
o
f
e
s
s
o
r
i
n
C
o
m
p
u
ter
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
De
p
t.
S
h
a
r
n
b
a
s
v
a
U
n
iv
e
r
s
i
ty
Ka
l
a
b
u
r
a
g
i
,
Ka
r
n
a
tak
a
,
–
I
n
d
i
a
.
He
r
p
r
i
m
a
r
y
a
r
e
a
o
f
i
n
ter
e
s
t
i
s
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
,
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
,
a
n
d
pa
tt
e
r
n
r
e
c
o
g
n
i
ti
o
n
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
i
l:
s
h
il
p
a
_
1
2
2
0
2
3
@r
e
d
i
f
f
m
a
il
.
c
o
m
.
Dr
.
S
h
r
id
e
v
i
S
o
m
a
wo
r
k
i
n
g
p
r
e
s
e
n
tl
y
a
s
p
r
o
f
e
s
s
o
r
i
n
De
p
a
r
tm
e
n
t
o
f
C
o
m
p
u
ter
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
P
o
o
jy
a
Do
d
d
a
p
p
a
Ap
p
a
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
Ka
l
a
b
u
r
a
g
i
.
S
h
e
h
a
s
1
8
y
e
a
r
s
o
f
T
e
a
c
h
i
n
g
a
n
d
1
0
y
e
a
r
s
o
f
R
e
s
e
a
r
c
h
E
x
p
e
r
ien
c
e
,
a
n
d
c
o
m
p
l
e
ted
h
e
r
B
.
E
.
,
M
.
T
e
c
h
.
,
a
n
d
P
h
.
D
.
i
n
C
o
m
p
u
ter
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
.
He
r
r
e
s
e
a
r
c
h
a
r
e
a
i
n
c
l
u
d
e
s
d
i
g
i
ta
l
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
a
n
d
p
a
tt
e
r
n
r
e
c
o
g
n
i
ti
o
n
,
c
l
o
u
d
c
o
m
p
u
ti
n
g
,
i
n
ter
n
e
t
o
f
th
i
n
g
s
,
b
i
g
d
a
ta
a
n
a
l
y
ti
c
s
.
S
h
e
p
u
b
l
i
s
h
e
d
m
o
r
e
th
a
n
3
0
R
e
s
e
a
r
c
h
p
a
p
e
r
s
i
n
a
b
o
v
e
m
e
n
ti
o
n
e
d
a
r
e
a
s
,
a
l
s
o
Gu
i
d
i
n
g
R
e
s
e
a
r
c
h
S
tu
d
e
n
ts
.
S
h
e
h
a
s
a
l
s
o
r
e
c
e
i
v
e
d
g
r
a
n
t
f
o
r
e
s
tab
li
s
h
m
e
n
t
o
f
“
C
l
o
u
d
C
o
m
p
u
ti
n
g
L
a
b
”
f
r
o
m
VG
S
T
.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
i
l:
s
h
r
i
d
e
v
i
s
o
m
a
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.