I
A
E
S
I
n
t
e
r
n
at
io
n
al
Jou
r
n
al
of
A
r
t
if
ic
ia
l
I
n
t
e
ll
ig
e
n
c
e
(
I
J
-
AI
)
V
ol
.
14
, N
o.
3
,
J
une
20
25
, pp.
2158
~
2168
I
S
S
N
:
2252
-
8938
,
D
O
I
:
10.11591/
ij
a
i.
v
14
.i
3
.pp
2158
-
2168
2158
Jou
r
n
al
h
om
e
page
:
ht
tp
:
//
ij
ai
.
ia
e
s
c
or
e
.c
om
Co
-
t
r
ai
n
i
n
g p
se
u
d
o
-
l
ab
e
l
i
n
g f
or
t
e
xt
c
l
ass
i
f
i
c
at
i
on
w
i
t
h
su
p
p
or
t
ve
c
t
o
r
m
ac
h
i
n
e
an
d
l
o
n
g sh
or
t
-
t
e
r
m
m
e
m
or
y
S
r
i
H
an
d
ayan
i
1,
2
,
R
iz
al
I
s
n
an
t
o
3
,
B
u
d
i
War
s
it
o
4
1
D
oc
t
or
a
l
P
r
ogr
a
m
of
I
nf
or
m
a
t
i
on
S
ys
t
e
m
s
, D
i
pone
gor
o U
ni
ve
r
s
i
t
y,
S
e
m
a
r
a
ng, I
ndone
s
i
a
2
D
e
pa
r
t
m
e
nt
of
I
nf
or
m
a
t
i
on a
nd C
om
m
uni
c
a
t
i
on T
e
c
hnol
ogy
, S
e
m
a
r
a
ng U
ni
ve
r
s
i
t
y, S
e
m
a
r
a
ng, I
ndone
s
i
a
3
D
e
pa
r
t
m
e
nt
of
C
om
put
e
r
E
ngi
ne
e
r
i
ng, D
i
pone
gor
o U
ni
ve
r
s
i
t
y, S
e
m
a
r
a
ng, I
nd
one
s
i
a
4
D
e
pa
r
t
m
e
nt
of
S
t
a
t
i
s
t
i
c
s
,
F
a
c
ul
t
y of
S
c
i
e
nc
e
a
nd M
a
t
he
m
a
t
i
c
s
,
D
i
pone
gor
o U
n
i
ve
r
s
i
t
y, S
e
m
a
r
a
ng, I
ndone
s
i
a
A
r
t
ic
le
I
n
f
o
A
B
S
T
R
A
C
T
A
r
ti
c
le
h
is
to
r
y
:
R
e
c
e
iv
e
d
M
a
y
4
,
2024
R
e
vi
s
e
d
F
e
b
3
,
2025
A
c
c
e
pt
e
d
M
a
r
15
,
2025
The
scarcity
of
l
abeled
data
may
hamper
training
text
-
processing
mod
els.
In
response
to
this
issue,
a
novel
and
intrigui
ng
strategy
that
combines
t
he
c
o
-
training
method
and
pseudo
-
labeling
design
is
applied
to
enhan
ce
the
model'
s
performance.
This
method,
a
component
of
an
efficient
semi
-
supervised
learning
paradigm
for
processing
and
comprehending
te
xt,
is
a
fresh
perspective
in
the
field.
The
model,
which
combines
a
support
vector
machine
(SVM)
for
classifica
tion
and
long
short
-
term
memory
(LSTM)
for
text
sequence
interpretation,
is
a
unique
approach.
By
introducing
s
amples
that
may
be
marginalized
in
the
labeled
data,
the
c
o
-
training
approac
h
could
help
solve
the
class
imbalance
problem
by
using
a
small
amount
of
l
abeled
data
and
the
rest
unlabeled.
This
study
assesses
the
model'
s
perfor
mance
using
a
student
dataset
from
higher
education
institutions
to
esta
blish
a
threshold
for
each
model'
s
degree
of
confidenc
e
and
ascerta
in
how
much
the
model
can
be
generalized
depending
on
the
threshold.
The
SVM
thr
esho
ld
was
calculated
as
>
=
0.88,
and
the
LSTM
threshold
was
calculated
as
>
=
0.5
using a mixture of confidence metrics.
K
e
y
w
o
r
d
s
:
Co
-
tr
a
in
in
g
L
ong s
hor
t
-
te
r
m
m
e
m
or
y
P
s
e
udo
-
la
be
li
ng
S
e
m
i
-
s
upe
r
vi
s
e
d l
e
a
r
ni
ng
S
uppor
t
ve
c
to
r
m
a
c
hi
ne
This is an
open
acce
ss artic
le unde
r the
CC BY
-
SA
license.
C
or
r
e
s
pon
di
n
g A
u
th
or
:
S
r
i
H
a
nda
ya
ni
D
e
pa
r
tm
e
nt
T
e
c
hnol
ogy of
I
nf
or
m
a
ti
on a
nd
C
om
m
uni
c
a
ti
on, S
e
m
a
r
a
ng U
ni
ve
r
s
it
y
S
oe
ka
r
no H
a
tt
a
, W
e
s
t
T
lo
gos
a
r
i,
S
e
m
a
r
a
ng C
it
y, C
e
nt
r
a
l
J
a
va
,
I
ndone
s
ia
E
m
a
il
:
s
r
i
@
us
m
.a
c
.i
d
1.
I
N
T
R
O
D
U
C
T
I
O
N
T
e
xt
c
lu
s
te
r
in
g
in
th
e
e
r
a
of
bi
g
da
ta
a
nd
m
a
c
hi
ne
le
a
r
ni
ng
is
one
of
th
e
m
os
t
c
r
it
ic
a
l
ta
s
ks
in
te
xt
m
in
in
g,
w
hi
c
h
a
im
s
to
gr
oup
te
xt
s
a
c
c
or
di
ng
to
th
e
ir
c
la
s
s
if
i
c
a
ti
on.
T
e
xt
c
lu
s
te
r
in
g
ha
s
be
e
n
be
n
e
f
ic
ia
l
in
m
a
ny
a
ppl
ic
a
ti
ons
,
s
uc
h
a
s
r
e
c
om
m
e
nda
ti
on
s
y
s
te
m
s
[
1]
,
[
2]
,
da
ta
m
in
in
g
m
a
na
ge
m
e
nt
[
3]
,
a
nd
pos
it
io
n
de
te
c
ti
on
[
4]
.
W
it
h
th
e
pr
e
s
e
nc
e
of
th
e
di
gi
ta
l
e
r
a
,
m
os
t
pe
opl
e
e
nj
oy
s
ha
r
in
g
a
nd
f
in
di
ng
va
r
io
us
c
ont
e
nt
on
s
oc
ia
l
m
e
di
a
a
nd
us
in
g
te
xt
i
s
ve
r
y
in
f
lu
e
nt
ia
l
in
c
a
r
r
yi
ng
in
f
or
m
a
ti
on.
T
e
xt
c
lu
s
te
r
in
g
c
a
n
id
e
nt
if
y
tr
e
ndi
ng
to
pi
c
s
,
de
te
c
t
s
e
nt
im
e
nt
pa
tt
e
r
n
s
,
or
r
e
c
om
m
e
nd
pe
r
s
on
a
li
z
e
d
c
ont
e
nt
,
of
f
e
r
in
g
a
w
e
a
lt
h
of
va
lu
a
bl
e
in
s
ig
ht
s
on
s
oc
ia
l
m
e
di
a
.
T
e
xt
c
lu
s
te
r
in
g
is
a
c
ha
ll
e
ngi
ng
ta
s
k.
C
onve
r
s
e
ly
,
th
e
di
ve
r
s
it
y
of
te
xt
c
a
te
gor
ie
s
of
te
n
pr
e
s
e
nt
s
a
s
ig
ni
f
ic
a
nt
c
ha
ll
e
nge
th
a
t
c
a
nnot
be
ig
nor
e
d
-
da
ta
d
a
ta
im
ba
la
nc
e
.
T
hi
s
s
it
ua
ti
on
a
r
is
e
s
w
h
e
n
th
e
c
la
s
s
e
s
i
n a
da
ta
s
e
t
a
r
e
not
e
ve
nl
y di
s
tr
ib
ut
e
d, l
e
a
di
ng t
o
one
c
l
a
s
s
ha
vi
ng s
ig
ni
f
ic
a
nt
ly
m
or
e
s
a
m
pl
e
s
t
ha
n t
he
ot
he
r
s
.
I
n
th
e
c
ont
e
xt
of
c
la
s
s
if
ic
a
ti
on,
th
is
im
ba
la
nc
e
c
a
n
pos
e
a
s
e
r
io
us
pr
obl
e
m
.
M
a
c
hi
ne
le
a
r
ni
ng
a
lg
or
it
hm
s
t
ypi
c
a
ll
y pr
io
r
it
iz
e
t
he
m
a
jo
r
it
y c
la
s
s
a
nd ma
y ove
r
lo
ok t
he
l
e
s
s
r
e
pr
e
s
e
nt
a
ti
ve
m
in
or
it
y c
la
s
s
.
O
ne
s
tu
dy
hi
ghl
ig
ht
s
th
e
c
ha
ll
e
ng
e
s
of
da
ta
m
e
a
s
ur
e
m
e
nt
i
n
te
xt
c
la
s
s
if
ic
a
ti
on.
I
t
pr
opos
e
s
a
pr
om
is
in
g
s
ol
ut
io
n
-
th
e
us
e
of
te
xt
ge
ne
r
a
ti
on
w
it
h
m
ode
G
P
T
-
2
a
nd
lo
ng
s
hor
t
-
te
r
m
m
e
m
or
y
(
L
S
T
M
)
to
ba
la
nc
e
th
e
d
a
ta
s
e
t.
T
hi
s
a
ppr
oa
c
h
s
ho
w
s
pot
e
nt
ia
l
in
im
pr
ovi
n
g
th
e
pe
r
f
or
m
a
nc
e
of
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
l
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Co
-
tr
ai
ni
ng ps
e
udo
-
la
be
li
ng f
or
t
e
x
t
c
la
s
s
if
ic
at
io
n w
it
h s
uppo
r
t
v
e
c
to
r
m
ac
hi
n
e
…
(
Sr
i
H
anday
ani
)
2159
[
5]
.
T
e
xt
c
lu
s
te
r
in
g
m
e
th
ods
ha
ve
a
ls
o
be
e
n
tr
ie
d
to
de
a
l
w
it
h
hi
ghl
y
im
ba
la
nc
e
d
da
ta
,
w
hi
c
h
te
nd
to
le
a
d
to
poor
s
ol
ut
io
ns
w
he
n
m
in
or
da
ta
c
lu
s
te
r
s
or
c
lu
s
te
r
s
w
it
h
s
m
a
ll
pr
opor
ti
ons
of
c
a
s
e
s
di
s
a
pp
e
a
r
[
6]
,
[
7]
.
T
he
y
m
a
y a
ls
o obta
in
e
a
s
y s
ol
ut
io
ns
i
n w
hi
c
h a
ll
t
e
xt
i
ns
ta
nc
e
s
a
r
e
i
n t
he
s
a
m
e
c
lu
s
te
r
[
8]
.
F
ur
th
e
r
m
o
r
e
, c
ont
r
a
s
ti
ve
la
ngua
ge
le
a
r
ni
ng
m
e
th
ods
c
om
pa
r
e
th
e
f
ir
s
t
a
nd
s
e
c
ond
w
o
r
ds
to
he
lp
c
or
r
e
c
t
e
r
r
or
s
a
nd
im
pr
ove
w
or
d
m
a
s
te
r
y.
H
ow
e
ve
r
,
th
e
s
e
m
e
th
ods
s
ti
ll
ne
e
d
be
tt
e
r
s
ol
ut
io
ns
,
e
s
pe
c
ia
ll
y
in
hi
ghl
y
im
ba
la
nc
e
d
te
xt
c
ol
le
c
ti
ons
[
7]
.
I
t
'
s
im
por
ta
nt
to
not
e
th
a
t
im
ba
la
nc
e
d
da
ta
,
if
p
r
oc
e
s
s
e
d
di
r
e
c
tl
y,
c
a
n
le
a
d
to
da
ta
pr
opor
ti
ons
th
a
t
r
e
duc
e
th
e
pe
r
f
or
m
a
nc
e
of
t
he
c
la
s
s
if
ic
a
ti
on a
lg
or
it
hm
[
9]
.
T
he
a
ut
hor
s
pr
opos
e
a
s
ol
ut
io
n
to
r
e
duc
in
g
da
ta
a
nd
im
pr
ovi
ng
m
ode
l
pe
r
f
o
r
m
a
nc
e
on
unl
a
be
le
d
da
ta
us
in
g
ps
e
udo
-
la
be
li
ng
.
T
hi
s
te
c
hni
que
,
w
hi
c
h
in
vol
ve
s
tr
a
in
in
g
a
m
ode
l
on
la
be
l
e
d
da
ta
to
pr
oduc
e
ps
e
udo
la
be
ls
f
or
unl
a
be
le
d
da
ta
,
of
f
e
r
s
s
e
ve
r
a
l
be
ne
f
it
s
.
P
s
e
udo
-
la
be
li
ng
in
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
ha
s
be
e
n
de
ve
lo
pe
d
to
c
la
s
s
if
y
da
ta
s
e
ts
w
it
h
pa
r
ti
a
l
la
b
e
ls
[
10]
.
I
t
ha
s
a
ls
o
be
e
n
d
e
ve
lo
pe
d
in
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
f
or
pa
tt
e
r
n
r
e
c
ogni
ti
on
[
11]
,
be
c
a
us
e
it
c
a
n
s
ig
ni
f
ic
a
nt
ly
in
c
r
e
a
s
e
le
a
r
ni
ng
e
f
f
e
c
ti
ve
ne
s
s
th
r
ough
th
e
di
s
tr
ib
ut
io
n
of
in
f
or
m
a
ti
on
f
r
o
m
unl
a
be
le
d
da
ta
a
nd
in
f
o
r
m
a
ti
o
n
f
r
om
la
be
le
d
da
ta
[
12]
.
P
s
e
udo
-
la
be
li
ng
w
il
l
a
ls
o
be
us
e
d
f
or
th
e
c
la
s
s
if
ic
a
ti
on
pr
oc
e
s
s
[
13]
,
[
14]
,
f
r
om
u
nl
a
be
le
d
te
s
t
da
ta
s
o
th
a
t
th
e
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
a
lg
or
it
hm
c
a
n
be
u
s
e
d
pr
ope
r
ly
. T
hus
,
th
e
c
la
s
s
if
ic
a
ti
o
n
pr
oc
e
s
s
th
a
t
w
il
l
be
c
a
r
r
ie
d
out
c
a
n
f
a
c
il
it
a
te
th
e
s
e
a
r
c
h
f
or
in
f
or
m
a
ti
on
ba
s
e
d
on
c
e
r
ta
in
c
a
te
gor
ie
s
ne
e
de
d
.
S
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
a
ls
o
f
unc
ti
ons
by
tr
a
in
in
g
th
e
m
ode
l
on
la
b
e
le
d
a
nd
unl
a
be
le
d
da
ta
.
L
a
be
le
d
da
t
a
pr
ovi
de
s
in
f
or
m
a
ti
on
a
bout
th
e
r
e
la
ti
ons
hi
p
be
twe
e
n i
nput
a
nd output
va
r
ia
bl
e
s
, w
hi
le
unl
a
b
e
le
d da
ta
c
a
pt
ur
e
s
t
he
unde
r
ly
in
g da
ta
s
tr
uc
tu
r
e
[
15]
.
S
e
ve
r
a
l
s
tu
di
e
s
in
hi
ghe
r
e
duc
a
ti
on
ha
ve
be
e
n
c
onduc
te
d
w
it
h
a
c
ol
la
bor
a
ti
ve
s
pi
r
it
,
a
im
in
g
to
id
e
nt
if
y
a
nd
im
pr
ove
s
tu
de
nt
a
c
c
e
pt
a
nc
e
r
a
te
s
.
O
ne
s
uc
h
c
o
ll
a
bor
a
ti
ve
e
f
f
or
t
in
vol
ve
s
th
e
a
ppl
ic
a
ti
on
of
pr
e
di
c
ti
ve
m
ode
ls
w
it
h
th
e
a
id
of
m
a
c
hi
ne
le
a
r
ni
ng.
T
hi
s
a
ppr
oa
c
h
c
a
n
he
lp
U
ni
ve
r
s
it
ie
s
id
e
nt
if
y
s
tu
de
nt
s
w
ho
ne
e
d
he
lp
e
a
r
ly
,
th
e
r
e
by
e
na
bl
in
g
in
te
r
ve
nt
io
ns
th
a
t
c
a
n
i
m
pr
ove
s
tu
de
nt
gr
a
dua
ti
on
r
a
te
s
[
16]
.
A
not
he
r
a
r
e
a
of
in
te
r
e
s
t
is
a
hybr
id
m
a
c
hi
ne
le
a
r
ni
ng
a
ppr
oa
c
h
th
a
t
e
f
f
e
c
ti
ve
ly
c
om
bi
ne
s
uns
up
e
r
vi
s
e
d
a
nd s
upe
r
vi
s
e
d
le
a
r
ni
ng
m
e
th
ods
,
le
a
di
ng
to
im
pr
ove
a
c
c
ur
a
c
y
in
pr
e
di
c
ti
ng
s
tu
de
nt
a
c
a
d
e
m
ic
pe
r
f
or
m
a
nc
e
[
17]
.
A
ddi
ti
ona
ll
y,
r
e
s
e
a
r
c
h
on
e
n
s
e
m
bl
e
m
ode
ls
,
w
hi
c
h
in
vol
ve
t
he
c
om
bi
na
ti
on
of
th
r
e
e
di
s
ti
nc
t
a
lg
or
it
hm
s
(
de
c
is
io
n
tr
e
e
,
lo
gi
s
ti
c
r
e
gr
e
s
s
io
n,
a
nd
ne
ur
a
l
ne
twor
ks
)
,
ha
s
y
ie
ld
e
d
a
n
e
f
f
e
c
ti
ve
to
ol
f
or
im
pr
ovi
ng
s
tu
de
nt
r
e
te
nt
io
n
r
a
te
s
.
T
he
s
e
to
ol
s
,
de
ve
lo
p
e
d
th
r
ough
c
ol
la
bor
a
ti
on,
ha
ve
th
e
pot
e
nt
ia
l
to
he
lp
hi
ghe
r
e
duc
a
ti
on
in
s
ti
tu
ti
ons
r
e
duc
e
th
e
r
is
k
of
dr
opout
a
nd
th
e
a
s
s
oc
i
a
te
d
f
in
a
nc
ia
l
lo
s
s
e
s
[
18]
. T
he
r
e
i
s
a
l
s
o
th
e
a
ppl
ic
a
ti
on
of
th
e
F
e
a
tu
r
e
s
e
le
c
ti
on
a
nd
c
ons
tr
uc
ti
on
f
or
r
a
di
a
l
ba
s
is
f
unc
ti
on
(
F
S
C
4R
B
F
)
us
in
g
a
lg
or
it
hm
.
T
hi
s
a
lg
or
it
hm
us
e
s
th
e
gr
a
m
m
a
ti
c
a
l
e
vol
ut
io
n
m
e
th
od
f
or
f
e
a
tu
r
e
s
e
le
c
ti
on
a
nd
f
e
a
tu
r
e
c
ons
tr
uc
ti
on
in
th
e
R
B
F
ne
twor
k
,
w
hi
c
h
a
im
s
to
im
pr
ove
th
e
g
e
ne
r
a
li
z
a
ti
on
a
bi
li
ty
in
pr
e
di
c
ti
ng
s
tu
de
nt
a
c
a
de
m
ic
pe
r
f
or
m
a
nc
e
,
e
s
pe
c
ia
ll
y
in
pr
e
di
c
ti
ng
th
e
dur
a
ti
on
o
f
s
tu
dy
a
nd
f
in
a
l
gr
a
de
s
ba
s
e
d
on
pr
e
vi
ous
le
a
r
ni
ng
da
ta
,
to
e
f
f
e
c
ti
ve
ly
im
p
r
ove
th
e
pr
e
di
c
ti
on
a
bi
li
ty
in
th
e
c
ol
le
ge
e
nvi
r
onm
e
nt
[
19]
.
H
ow
e
ve
r
,
n
o
r
e
s
e
a
r
c
h
ha
s
a
ppl
ie
d
th
e
c
o
-
tr
a
in
in
g
m
e
th
od
in
c
ol
le
ge
m
a
na
ge
m
e
nt
. T
hr
ough
th
is
a
r
ti
c
le
,
th
e
a
ut
hor
s
tr
y
to
a
ppl
y
th
e
ps
e
udo
-
la
be
li
ng
te
c
hni
que
u
s
in
g
th
e
c
o
-
tr
a
in
in
g
m
e
th
od,
in
w
hi
c
h
th
e
s
tu
de
nt
da
ta
obt
a
in
e
d
is
te
xt
da
ta
f
r
om
a
que
s
ti
onna
ir
e
di
s
tr
ib
ut
e
d
a
t
a
c
ol
le
ge
to
de
te
r
m
in
e
th
e
pot
e
nt
ia
l
of
s
tu
de
nt
s
a
t
th
e
c
ol
le
ge
.
W
hi
le
pr
oc
e
s
s
in
g
th
e
que
s
ti
onna
ir
e
r
e
s
ul
t
s
,
th
e
da
ta
w
il
l
be
la
be
le
d
ba
s
e
d
on
pot
e
nt
ia
l
a
nd
non
-
pot
e
nt
ia
l
c
r
it
e
r
ia
.
A
s
m
a
ll
por
ti
on
of
th
e
s
tu
de
nt
da
ta
w
il
l
be
us
e
d
a
s
tr
a
in
in
g
da
ta
,
w
hi
c
h
th
e
n
he
lp
s
f
or
m
a
w
r
a
ppe
d
da
ta
s
e
t
to
c
la
s
s
if
y
la
be
ls
on
unl
a
be
le
d
da
ta
(
te
s
ti
ng
da
ta
)
us
in
g t
he
c
o
-
tr
a
in
in
g s
e
m
i
-
s
upe
r
vi
s
e
d l
e
a
r
ni
ng c
on
c
e
pt
.
P
s
e
udo
la
be
li
ng
in
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
c
a
n
d
e
gr
a
de
th
e
la
be
li
ng
qua
li
ty
be
c
a
us
e
it
doe
s
not
r
e
pr
e
s
e
nt
th
e
c
la
s
s
e
s
in
th
e
la
be
le
d
da
t
a
s
e
t.
T
he
c
o
-
tr
a
in
in
g
m
e
th
od
c
a
n
ut
il
iz
e
s
e
m
a
nt
ic
la
be
l
s
to
ove
r
c
om
e
th
is
pr
obl
e
m
[
20]
.
T
he
la
be
li
ng
f
r
a
m
e
w
or
k
w
it
h
th
e
c
o
-
t
r
a
in
in
g
m
e
th
od
c
a
n
s
ol
ve
th
e
m
ul
ti
-
vi
e
w
w
e
a
k
la
be
l
le
a
r
ni
ng
pr
obl
e
m
us
in
g
ps
e
udo
la
be
l
ve
c
to
r
s
[
21
]
.
C
om
pr
e
he
ns
iv
e
e
xpe
r
im
e
nt
a
l
s
tu
di
e
s
ha
ve
be
e
n
c
onduc
te
d
to
de
m
ons
tr
a
te
th
e
e
f
f
e
c
ti
ve
ne
s
s
of
th
e
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
a
nd
c
o
-
tr
a
in
in
g
m
e
th
od
f
or
m
ul
ti
-
la
be
l
[
22]
.
O
ve
r
th
e
pa
s
t
de
c
a
d
e
,
th
e
c
o
-
tr
a
in
in
g
m
e
th
od
ha
s
ga
in
e
d
popul
a
r
it
y
in
th
e
r
e
s
e
a
r
c
h
a
nd
in
dus
tr
y
c
om
m
uni
ti
e
s
a
nd
ha
s
be
e
n
s
uc
c
e
s
s
f
ul
ly
a
ppl
ie
d
in
s
e
ve
r
a
l
r
e
a
l
-
w
or
ld
a
ppl
i
c
a
ti
ons
[
23]
.
T
he
c
o
-
tr
a
in
in
g
m
e
th
od
ha
s
a
ls
o
be
e
n
us
e
d
f
or
tr
a
ns
f
e
r
le
a
r
ni
ng
(
T
L
)
[
24]
.
C
o
-
tr
a
in
in
g
is
a
n
e
m
e
r
gi
ng
to
pi
c
in
m
a
c
hi
ne
le
a
r
ni
ng
.
I
t
a
im
s
to
e
xt
r
a
c
t
knowle
dge
ga
in
e
d
in
a
s
our
c
e
ta
s
k
or
dom
a
in
a
nd
us
e
it
to
f
a
c
il
it
a
te
ta
r
ge
t
p
r
e
di
c
ti
ve
le
a
r
ni
ng
f
unc
ti
ons
in
a
di
f
f
e
r
e
nt
ta
s
k
or
dom
a
in
.
C
o
-
tr
a
in
in
g
c
a
n
im
pr
ove
th
e
ge
ne
r
a
li
z
a
ti
on
a
bi
li
ty
of
a
m
ode
l
by
ut
il
iz
in
g
in
f
or
m
a
ti
on
f
r
o
m
unl
a
be
le
d
da
ta
.
C
o
-
tr
a
in
in
g
e
na
bl
e
s
m
a
c
hi
ne
s
to
th
in
k
f
r
om
m
ul
ti
pl
e
pe
r
s
pe
c
ti
ve
s
li
ke
huma
ns
by divi
di
ng
da
ta
i
nt
o m
u
lt
ip
le
vi
e
w
s
, de
s
ig
ni
ng l
e
a
r
ne
r
s
s
c
ie
nt
if
ic
a
ll
y, a
nd e
s
ti
m
a
ti
ng c
onf
id
e
nc
e
la
be
ls
a
c
c
ur
a
te
ly
.
T
hi
s
m
e
th
od
c
a
n
a
ls
o
im
pr
ove
c
la
s
s
if
ic
a
ti
on a
c
c
ur
a
c
y
a
nd
m
ode
l
c
onve
r
ge
nc
e
.
I
n
a
ddi
ti
on,
co
-
tr
a
in
in
g'
s
ve
r
s
a
ti
li
ty
in
va
r
io
us
r
e
s
e
a
r
c
h
ta
s
k
s
,
s
uc
h
a
s
e
r
r
or
c
la
s
s
if
ic
a
ti
on
a
nd
pe
r
s
on
id
e
nt
if
ic
a
ti
on
ba
s
e
d
on a
udi
o
-
vi
s
ua
ls
, ope
ns
up
a
w
or
ld
of
pot
e
nt
ia
l
a
ppl
ic
a
ti
ons
a
nd
r
e
s
e
a
r
c
h oppor
tu
ni
ti
e
s
[
25]
.
T
he
c
o
-
tr
a
in
in
g
pr
oc
e
s
s
g
e
ne
r
a
ll
y
in
vol
ve
s
th
r
e
e
m
a
in
s
te
ps
:
v
ie
w
a
c
qui
s
it
io
n,
le
a
r
ne
r
di
f
f
e
r
e
nt
ia
ti
on,
a
nd
la
be
l
c
onf
id
e
nc
e
e
s
ti
m
a
ti
on
[
25]
.
V
ie
w
a
c
qui
s
it
io
n
a
im
s
to
obt
a
in
two
in
de
pe
nde
nt
a
nd
s
uf
f
ic
ie
nt
da
ta
vi
e
w
s
,
e
a
c
h
of
w
hi
c
h
c
a
n
be
u
s
e
d
by
a
di
f
f
e
r
e
nt
le
a
r
ni
ng
m
ode
l.
I
f
in
de
pe
nde
nt
vi
e
w
s
a
r
e
not
na
tu
r
a
ll
y
a
va
il
a
bl
e
,
th
e
y
c
a
n
be
c
ons
tr
uc
te
d
us
in
g
pr
e
-
tr
a
in
e
d
m
ode
ls
[
26]
.
O
nc
e
th
e
vi
e
w
s
a
r
e
a
c
qui
r
e
d,
th
e
ne
xt
s
te
p
is
le
a
r
ne
r
di
f
f
e
r
e
nt
ia
ti
on,
in
w
hi
c
h
two
di
f
f
e
r
e
nt
le
a
r
ni
ng
m
ode
ls
a
r
e
dr
il
le
d
s
e
pa
r
a
te
ly
f
or
e
a
c
h
vi
e
w
.
T
hi
s
a
ppr
oa
c
h
e
ns
ur
e
s
th
a
t
e
a
c
h
m
ode
l
le
a
r
ns
a
uni
que
r
e
pr
e
s
e
nt
a
ti
on
of
th
e
da
ta
,
w
hi
c
h
in
c
r
e
a
s
e
s
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
3
,
J
une
20
25
:
2158
-
2168
2160
m
ode
ls
'
a
bi
li
ty
to
c
om
pl
e
te
a
nd c
or
r
e
c
t
e
a
c
h
ot
he
r
'
s
e
r
r
or
s
[
27]
.
I
n
th
e
la
be
l
c
onf
id
e
nc
e
e
s
ti
m
a
ti
on
s
t
a
ge
,
e
a
c
h
m
ode
l
m
a
ke
s
pr
e
di
c
ti
ons
on
unl
a
be
le
d
da
ta
a
nd
a
s
s
e
s
s
e
s
t
he
m
ode
l'
s
c
onf
id
e
nc
e
in
th
os
e
pr
e
di
c
ti
ons
.
P
r
e
di
c
ti
ons
w
it
h
hi
gh
c
onf
id
e
nc
e
a
r
e
th
e
n
us
e
d
a
s
ps
e
udo
-
la
be
ls
to
tr
a
in
th
e
ot
he
r
m
ode
l,
a
ll
ow
in
g
bot
h
m
ode
ls
to
r
e
in
f
or
c
e
e
a
c
h
ot
he
r
a
nd
im
pr
ove
ove
r
a
ll
pr
e
di
c
t
io
n
a
c
c
ur
a
c
y
[
20]
.
T
hi
s
e
m
pha
s
is
on
hi
gh
-
c
onf
id
e
nc
e
pr
e
di
c
ti
ons
pr
ovi
de
s
a
r
e
a
s
s
ur
in
g f
r
a
m
e
w
or
k f
or
t
he
m
ode
ls
t
o l
e
a
r
n a
nd i
m
pr
ove
.
T
hi
s
a
r
ti
c
le
r
e
vi
e
w
s
th
e
co
-
tr
a
in
in
g
m
e
th
od
of
c
onf
id
e
nc
e
l
a
be
l
e
s
ti
m
a
ti
on,
in
w
hi
c
h
th
e
la
be
l
c
onf
id
e
nc
e
e
s
ti
m
a
ti
on
is
a
n
im
por
ta
nt
s
te
p
in
th
e
co
-
tr
a
in
in
g
m
e
th
od
w
it
h
th
e
a
im
th
a
t
th
e
da
ta
te
s
ti
ng
is
not
m
is
la
be
le
d,
w
hi
c
h
c
a
n
r
e
duc
e
th
e
a
bi
li
ty
of
da
ta
te
s
ti
ng.
A
c
c
or
di
ng
to
th
e
la
be
l
c
onf
id
e
nc
e
e
s
ti
m
a
ti
on
m
e
th
od,
th
e
la
be
l
c
onf
id
e
nc
e
e
s
ti
m
a
ti
on
c
a
n
be
di
vi
de
d
i
nt
o
im
pl
ic
it
a
nd
e
xpl
ic
it
e
s
ti
m
a
ti
ons
.
M
os
t
a
lg
or
it
hm
s
in
im
pl
ic
it
e
s
ti
m
a
ti
on
us
e
th
e
de
gr
e
e
of
di
f
f
e
r
e
nc
e
b
e
twe
e
n
th
e
r
e
s
ul
ts
of
tr
a
in
in
g
da
ta
a
nd
t
e
s
ti
ng
da
ta
to
r
e
f
le
c
t
th
e
c
onf
id
e
nc
e
of
th
e
ps
e
udo
la
be
l.
S
ti
ll
,
if
th
e
r
e
is
unl
a
be
le
d
da
ta
,
if
i
t
is
m
is
la
be
le
d,
it
w
il
l
r
e
duc
e
pe
r
f
or
m
a
nc
e
dur
in
g
it
e
r
a
ti
on,
s
o
th
e
a
c
c
ur
a
c
y
of
im
pl
ic
it
e
s
ti
m
a
ti
on
is
lo
w
e
r
th
a
n
e
xpl
ic
it
e
s
ti
m
a
ti
on
[
28]
.
T
he
a
lg
or
it
hm
in
e
xpl
ic
it
e
s
ti
m
a
ti
on
us
e
s
e
xa
c
t
num
be
r
s
t
o
di
s
pl
a
y
th
e
c
onf
id
e
nc
e
la
b
e
l.
B
a
s
ic
a
ll
y,
th
is
a
lg
or
it
hm
us
e
s
le
a
r
ni
ng
out
put
in
pr
oba
bi
li
s
ti
c
f
or
m
,
th
e
di
f
f
e
r
e
nc
e
in
m
ode
l
a
c
c
ur
a
c
y
b
e
f
or
e
a
nd
a
f
te
r
us
in
g
ps
e
udo
-
la
be
l
e
d
s
a
m
pl
e
s
,
or
th
e
s
im
il
a
r
it
y
be
twe
e
n
th
e
ps
e
udo
la
be
l
of
th
e
c
ur
r
e
nt
un
la
be
le
d
da
ta
a
nd
th
e
s
ur
r
ounding
la
be
l
e
d da
ta
[
25]
.
Co
-
tr
a
in
in
g
us
e
s
im
pl
ic
it
e
s
ti
m
a
ti
on
of
th
e
s
uppor
t
ve
c
to
r
m
a
c
hi
ne
(
S
V
M
)
a
nd
L
S
T
M
a
lg
or
it
hm
s
.
T
he
r
e
a
s
on
f
or
c
hoo
s
in
g
S
V
M
to
be
dr
il
le
d
to
ge
th
e
r
w
it
h
L
S
T
M
i
s
th
a
t
S
V
M
i
s
c
om
m
onl
y
us
e
d
f
or
te
xt
c
la
s
s
if
ic
a
ti
on
[
29]
.
S
V
M
ha
s
a
m
or
e
m
a
tu
r
e
a
nd
m
a
th
e
m
a
ti
c
a
ll
y
c
le
a
r
e
r
c
onc
e
pt
th
a
n
ot
he
r
c
la
s
s
if
ic
a
ti
on
te
c
hni
que
s
.
S
V
M
c
a
n
a
ls
o
s
ol
ve
c
la
s
s
if
ic
a
ti
on
a
nd
r
e
gr
e
s
s
io
n
pr
obl
e
m
s
w
it
h
li
ne
a
r
a
nd
non
-
li
ne
a
r
.
S
V
M
is
a
ls
o
us
e
d
be
c
a
us
e
it
ha
s
a
dva
nt
a
ge
s
in
ha
ndl
in
g
c
la
s
s
if
ic
a
ti
on
pr
obl
e
m
s
,
e
s
pe
c
ia
ll
y
w
he
n
th
e
da
ta
ha
s
hi
gh
di
m
e
ns
io
ns
or
a
n
im
ba
la
n
c
e
be
tw
e
e
n
c
la
s
s
e
s
.
S
V
M
e
f
f
e
c
ti
ve
ly
s
e
pa
r
a
te
s
c
la
s
s
e
s
w
it
h
m
a
xi
m
um
m
a
r
gi
n,
m
a
ki
ng
it
s
ui
ta
bl
e
f
or
c
la
s
s
if
ic
a
ti
on
w
it
h
c
om
pl
e
x
f
e
a
tu
r
e
da
t
a
,
s
uc
h
a
s
m
e
di
c
a
l
im
a
ge
s
or
te
xt
in
a
vi
s
ua
l
c
ont
e
xt
.
I
n
th
e
c
a
s
e
of
im
a
ge
c
la
s
s
if
ic
a
ti
on,
S
V
M
is
of
te
n
r
e
li
e
d
on
be
c
a
us
e
of
it
s
a
bi
li
ty
to
f
in
d
th
e
opt
im
a
l
hype
r
pl
a
ne
th
a
t
s
e
pa
r
a
te
s
da
ta
b
a
s
e
d
on
r
e
le
va
nt
f
e
a
tu
r
e
s
,
w
hi
c
h
ul
ti
m
a
te
ly
he
lp
s
r
e
duc
e
c
la
s
s
if
ic
a
ti
on
e
r
r
or
s
[
30]
.
T
he
L
S
T
M
a
lg
or
it
hm
is
a
pr
e
di
c
ti
ve
m
ode
l
to
pr
oc
e
s
s
s
e
que
nt
ia
l
da
ta
in
vi
r
tu
a
l
le
a
r
ni
ng
e
nvi
r
onm
e
nt
s
(
V
L
E
s
)
.
L
S
T
M
is
a
ppl
ie
d
to
e
va
lu
a
te
a
nd
pr
e
di
c
t
s
tu
de
nt
le
a
r
ni
ng
out
c
om
e
s
ba
s
e
d
on
da
ta
c
ol
le
c
te
d
f
r
om
s
tu
de
nt
in
te
r
a
c
ti
ons
w
it
h
onl
in
e
le
a
r
ni
ng
p
la
tf
or
m
s
.
T
hi
s
m
ode
l
c
a
n
a
na
ly
z
e
s
e
que
nc
e
s
of
s
tu
de
nt
be
ha
vi
or
a
l
da
ta
,
s
uc
h
a
s
lo
gi
n
pa
tt
e
r
ns
,
qui
z
in
te
r
a
c
ti
on
s
,
a
nd
c
our
s
e
a
c
ti
vi
ti
e
s
,
to
id
e
nt
if
y
tr
e
nds
th
a
t
m
a
y
in
di
c
a
te
s
uc
c
e
s
s
or
di
f
f
ic
ul
t
in
le
a
r
ni
ng
[
31]
,
[
32]
.
L
S
T
M
is
a
ls
o
c
ho
s
e
n
be
c
a
us
e
of
it
s
a
bi
li
ty
to
ha
ndl
e
ti
m
e
s
e
r
ie
s
da
ta
,
w
hi
c
h
is
v
e
r
y
im
por
ta
nt
f
or
m
ode
li
ng
le
a
r
ni
ng
da
ta
c
ont
in
uous
ly
ge
ne
r
a
te
d
in
onl
in
e
l
e
a
r
ni
ng
e
nvi
r
onm
e
nt
s
.
T
hi
s
a
lg
or
it
hm
a
ll
ow
s
th
e
s
y
s
te
m
to
r
e
m
e
m
be
r
im
por
ta
nt
in
f
or
m
a
ti
on
in
th
e
lo
ng
te
r
m
,
w
hi
c
h
m
a
ke
s
it
id
e
a
l
f
or
p
r
e
di
c
ti
ng
s
tu
de
nt
pe
r
f
o
r
m
a
nc
e
ba
s
e
d
on
s
tu
de
nt
a
c
ti
vi
ty
hi
s
to
r
y
[
5]
.
I
n
a
not
he
r
a
r
ti
c
le
,
th
e
L
S
T
M
a
lg
or
it
hm
is
u
s
e
d
a
s
pa
r
t
of
a
c
om
bi
ne
d
C
N
N
-
L
S
T
M
m
ode
l
to
de
te
c
t
in
s
ul
ts
in
te
xt
.
L
S
T
M
pl
a
ys
a
r
ol
e
in
pr
oc
e
s
s
in
g
th
e
s
e
que
nc
e
of
w
or
ds
f
r
om
th
e
te
xt
to
u
nde
r
s
ta
nd
th
e
te
m
por
a
l
or
s
e
que
nt
ia
l
c
ont
e
xt
,
w
hi
c
h
is
im
por
ta
nt
in
de
te
c
ti
ng
th
e
m
e
a
ni
ng
of
in
s
ul
t
s
ba
s
e
d
on
th
e
c
ont
e
xt
of
th
e
s
e
nt
e
nc
e
.
C
om
bi
ni
ng
C
N
N
f
or
in
it
ia
l
f
e
a
tu
r
e
e
xt
r
a
c
ti
on
a
nd
L
S
T
M
to
unde
r
s
ta
nd
th
e
c
ont
e
xt
s
e
que
nc
e
m
a
ke
s
th
is
m
od
e
l
m
or
e
e
f
f
e
c
ti
ve
in
a
c
c
ur
a
te
ly
d
e
te
c
ti
ng
in
s
ul
t
s
in
te
xt
[
26]
,
[
32]
.
R
e
s
e
a
r
c
h
on a
l
gor
it
hm
s
in
m
a
c
hi
ne
le
a
r
ni
ng
th
a
t
e
xi
s
ts
s
o
f
a
r
is
c
om
pa
r
in
g
th
e
S
V
M
w
it
h
L
S
T
M
a
lg
or
it
hm
a
nd
th
e
d
e
e
p
be
li
e
ve
ne
twor
k
(
D
B
N
)
a
lg
or
it
hm
us
in
g
th
e
f
uz
z
y
lo
gi
c
a
l
r
e
la
ti
ons
hi
ps
(
F
L
R
)
m
ode
l
[
14]
.
R
e
c
e
nt
r
e
s
e
a
r
c
h
on
ps
e
udo
-
la
be
li
ng
in
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
us
in
g
th
e
S
V
M
a
nd
L
S
T
M
c
o
-
tr
a
in
in
g
m
e
th
ods
ha
s
s
how
n
th
a
t
ps
e
udo
-
la
be
li
ng
not
on
ly
im
pr
ove
s
m
ode
l
pe
r
f
or
m
a
nc
e
but
is
a
l
s
o
e
f
f
e
c
ti
ve
in
s
it
ua
ti
ons
in
w
hi
c
h
la
be
le
d
da
ta
is
ve
r
y
li
m
it
e
d.
T
he
s
im
il
a
r
ps
e
udo
la
be
l
e
xpl
oi
ta
ti
on
f
or
s
e
m
i
-
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on
(
S
I
M
P
L
E
)
a
lg
or
it
hm
,
w
hi
c
h
e
f
f
e
c
ti
ve
ly
ut
il
iz
e
s
ps
e
udo
-
la
be
ls
to
e
nha
nc
e
a
c
c
ur
a
c
y
in
s
e
m
i
-
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on,
is
a
r
e
a
l
-
w
or
ld
pr
oof
of
c
onc
e
pt
[
13]
.
T
he
c
o
-
tr
a
in
in
g
m
e
th
od,
w
hi
c
h
c
om
bi
ne
s
s
e
ve
r
a
l
a
lg
or
it
hm
s
to
tr
a
in
m
ode
ls
s
im
ul
ta
ne
ous
ly
,
is
ga
in
in
g
tr
a
c
ti
on,
pa
r
ti
c
ul
a
r
ly
in
a
ddr
e
s
s
in
g
da
ta
im
ba
la
nc
e
pr
obl
e
m
s
.
T
he
c
om
bi
na
ti
on
of
th
e
c
o
-
tr
a
in
in
g
m
e
th
od
w
it
h
m
ode
r
n
a
lg
or
it
hm
s
in
th
e
m
e
di
c
a
l
f
ie
ld
de
m
ons
tr
a
te
s
th
a
t
u
s
in
g
ps
e
udo
-
la
be
l
s
w
it
h
a
hi
gh
le
v
e
l
of
c
onf
id
e
nc
e
(
hi
gh
-
c
onf
id
e
nc
e
ps
e
udo
la
b
e
ls
)
c
a
n
s
ig
ni
f
ic
a
nt
ly
e
nha
nc
e
th
e
a
c
c
ur
a
c
y
a
nd
s
ta
bi
li
ty
of
th
e
m
o
de
l
in
m
e
di
c
a
l
im
a
ge
s
e
gm
e
nt
a
ti
on
[
33]
.
I
n
a
te
xt
ua
l
c
ont
e
xt
,
c
o
-
tr
a
in
in
g
ha
s
pr
ove
n
e
f
f
e
c
ti
ve
in
ove
r
c
om
in
g
th
e
pr
obl
e
m
o
f
unde
r
-
la
be
le
d
da
ta
b
y
c
om
bi
ni
ng
S
V
M
a
nd
L
S
T
M
-
ba
s
e
d
m
od
e
ls
to
opt
im
iz
e
da
ta
s
e
que
nc
e
c
l
a
s
s
if
ic
a
ti
on.
I
n
th
e
vi
s
ua
l
c
ont
e
xt
,
la
be
l
c
lu
s
te
r
in
g
a
nd
c
o
-
tr
a
in
in
g
im
pr
ove
th
e
qua
li
ty
of
ps
e
u
do
-
la
be
ls
a
nd
ove
r
a
ll
m
ode
l
p
e
r
f
or
m
a
nc
e
by
gr
oupi
ng
la
be
ls
ba
s
e
d
on
vi
s
ua
l
s
im
il
a
r
it
y
to
r
e
duc
e
m
is
c
la
s
s
if
i
c
a
ti
on
of
s
im
il
a
r
c
la
s
s
e
s
.
T
hi
s
a
ppr
oa
c
h,
u
s
in
g
a
n
e
m
be
ddi
ng
-
ba
s
e
d
la
be
l
r
e
pr
e
s
e
nt
a
ti
on,
c
r
e
a
te
s
gr
oups
of
vi
s
ua
ll
y
s
im
il
a
r
la
b
e
ls
a
nd
us
e
s
th
e
s
e
gr
oups
in
th
e
ps
e
udo
-
la
be
li
ng
pr
oc
e
s
s
[
30]
.
T
he
f
ut
ur
e
of
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
lo
oks
pr
om
is
in
g
w
it
h
r
e
c
e
nt
te
c
hni
que
s
in
co
-
tr
a
in
in
g
,
s
u
c
h
a
s
c
om
bi
ni
ng
im
pl
ic
it
a
nd
e
xpl
ic
it
c
onf
id
e
nc
e
e
s
ti
m
a
te
s
,
to
im
pr
ove
la
be
li
ng
a
c
c
ur
a
c
y.
E
nt
r
opy
-
ba
s
e
d
a
nd
m
a
r
gi
n
-
ba
s
e
d
c
onf
id
e
nc
e
a
lg
or
it
hm
s
a
r
e
c
ur
r
e
nt
ly
w
id
e
ly
us
e
d
to
e
ns
ur
e
th
e
a
c
c
ur
a
c
y
of
la
be
l
s
on
te
s
t
da
ta
.
F
or
e
xa
m
pl
e
,
th
e
r
ig
ht
s
t
e
p
is
t
o
de
ve
lo
p
c
on
s
is
te
nc
y
-
ba
s
e
d
s
e
m
i
-
s
upe
r
vi
s
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Co
-
tr
ai
ni
ng ps
e
udo
-
la
be
li
ng f
or
t
e
x
t
c
la
s
s
if
ic
at
io
n w
it
h s
uppo
r
t
v
e
c
to
r
m
ac
hi
n
e
…
(
Sr
i
H
anday
ani
)
2161
a
c
ti
ve
le
a
r
ni
ng
te
c
hni
que
s
th
a
t
m
in
im
iz
e
la
be
li
ng
c
os
ts
w
hi
le
im
pr
ovi
ng
m
ode
l
a
c
c
ur
a
c
y
th
r
ough
a
c
c
ur
a
te
c
onf
id
e
nc
e
e
s
ti
m
a
te
s
[
34]
.
T
hi
s
pa
pe
r
of
f
e
r
s
a
uni
que
c
ont
r
ib
ut
io
n
by
c
om
bi
ni
ng
two
di
f
f
e
r
e
nt
m
ode
ls
,
S
V
M
a
nd
L
S
T
M
,
in
a
co
-
tr
a
in
in
g
f
r
a
m
e
w
or
k.
S
V
M
is
us
e
d
to
ha
ndl
e
s
ta
ti
c
f
e
a
tu
r
e
-
ba
s
e
d
c
la
s
s
if
ic
a
ti
on,
w
hi
le
L
S
T
M
is
u
s
e
d
to
m
a
na
ge
a
nd
unde
r
s
ta
nd
te
xt
da
ta
s
e
que
n
c
e
s
.
T
hi
s
in
te
gr
a
ti
on
pr
ovi
de
s
a
dva
nt
a
ge
s
be
c
a
us
e
it
a
ll
ow
s
m
or
e
c
om
pr
e
he
ns
iv
e
m
ode
li
ng of
s
e
que
nt
ia
l
te
xt
da
ta
c
om
pa
r
e
d t
o t
r
a
di
ti
ona
l
co
-
tr
a
in
in
g
m
e
th
ods
t
ha
t
r
e
ly
onl
y o
n
one
m
ode
l
ty
pe
.
F
ur
th
e
r
m
or
e
,
bot
h
m
ode
ls
w
il
l
be
gi
ve
n
a
s
pe
c
if
ic
tr
us
t
th
r
e
s
hol
d:
S
V
M
w
it
h
a
th
r
e
s
hol
d
va
lu
e
of
0.88
a
nd
L
S
T
M
w
it
h
0.5.
T
hi
s
m
e
ti
c
ul
ous
p
a
r
a
m
e
t
e
r
iz
a
ti
on
of
th
e
m
ode
ls
,
a
f
e
a
tu
r
e
th
a
t
i
s
not
a
lwa
ys
e
xpl
a
in
e
d
in
de
ta
il
in
pr
e
vi
ous
s
tu
di
e
s
,
in
s
ti
ll
s
tr
us
t
in
th
e
a
ppl
ic
a
ti
on
of
s
e
m
i
-
s
upe
r
vi
s
e
d
m
ode
ls
.
I
n
a
ddi
ti
on,
a
not
he
r
c
ont
r
ib
ut
io
n
o
f
th
is
a
r
ti
c
le
is
th
a
t
w
hi
le
pr
e
v
io
us
s
tu
di
e
s
ha
ve
f
oc
us
e
d
on
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
a
ppl
ic
a
ti
ons
in
dom
a
in
s
s
uc
h
a
s
im
a
ge
r
e
c
ogni
ti
on
o
r
m
e
di
c
a
l
a
na
ly
s
is
,
th
is
a
r
ti
c
le
f
oc
u
s
e
s
on
it
s
a
ppl
ic
a
ti
on
to
e
duc
a
ti
ona
l
da
ta
,
s
pe
c
if
ic
a
ll
y
to
a
s
s
e
s
s
th
e
c
ondi
ti
on
of
s
tu
de
nt
s
in
hi
ghe
r
e
duc
a
ti
on
in
s
ti
tu
ti
ons
.
T
hi
s
is
a
n
und
e
r
-
e
xpl
or
e
d
a
r
e
a
,
a
nd
th
is
s
tu
dy
s
h
ow
s
how
s
e
m
i
-
s
upe
r
vi
s
e
d
te
c
hni
qu
e
s
c
a
n
b
e
a
ppl
ie
d f
or
m
or
e
e
f
f
e
c
ti
ve
e
duc
a
ti
ona
l
e
va
lu
a
ti
on pur
pos
e
s
.
I
n
co
-
tr
a
in
in
g
m
e
th
ods
,
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
,
la
be
l
c
onf
id
e
nc
e
r
e
f
e
r
s
to
th
e
e
xt
e
nt
to
w
hi
c
h
th
e
m
ode
l
ha
s
c
onf
id
e
nc
e
or
c
e
r
ta
in
ty
a
bout
th
e
la
be
l
gi
v
e
n
to
da
ta
.
T
he
s
e
m
e
th
od
s
,
w
hi
c
h
c
om
bi
ne
m
ode
ls
tr
a
in
e
d
on
two
f
e
a
tu
r
e
s
e
ts
,
a
r
e
pa
r
ti
c
ul
a
r
ly
e
f
f
e
c
ti
ve
in
im
p
r
ovi
ng
pe
r
f
or
m
a
nc
e
.
H
e
r
e
a
r
e
s
om
e
ge
ne
r
a
l
c
onc
e
pt
s
t
ha
t
c
a
n be
us
e
d t
o m
e
a
s
ur
e
l
a
be
l
c
onf
id
e
nc
e
i
n t
he
c
o
nt
e
xt
of
co
-
tr
a
in
in
g
,
−
E
nt
r
opy:
a
p
r
e
va
le
nt
a
ppr
oa
c
h
in
vol
vi
ng
e
nt
r
opy
ut
il
iz
a
ti
on
to
a
s
s
e
s
s
th
e
m
ode
l'
s
unc
e
r
ta
in
ty
r
e
ga
r
di
ng
la
be
ls
.
E
nt
r
opy
qua
nt
if
ie
s
th
e
m
ode
l'
s
pr
oba
bi
li
ty
di
s
tr
ib
ut
io
n
to
r
e
s
e
m
bl
e
a
uni
f
or
m
di
s
c
r
e
te
di
s
tr
ib
ut
io
n.
H
ig
he
r
e
nt
r
opy
va
lu
e
s
in
di
c
a
te
lo
w
e
r
c
onf
id
e
nc
e
le
ve
ls
in
th
e
a
s
s
ig
ne
d
la
be
l
[
34]
.
U
s
in
g
two
m
ode
ls
w
il
l
pr
oduc
e
jo
in
t
e
nt
r
opy
or
r
e
pr
e
s
e
nt
in
g
th
e
e
nt
r
opy
a
s
s
o
c
ia
te
d
w
it
h
th
e
jo
in
t
oc
c
ur
r
e
nc
e
of
c
la
s
s
e
s
a
nd
.
T
hi
s
e
nt
r
opy
va
lu
e
m
e
a
s
ur
e
s
th
e
unc
e
r
ta
in
ty
or
ir
r
e
gul
a
r
it
y
le
ve
l
a
s
s
oc
ia
t
e
d
w
it
h
th
e
two
c
la
s
s
e
s
'
jo
in
t
pr
oba
bi
li
ty
di
s
tr
ib
ut
io
n
[
35]
.
T
he
hi
ghe
r
th
e
e
nt
r
opy
va
lu
e
,
th
e
gr
e
a
te
r
th
e
unc
e
r
ta
in
ty
in
th
e
oc
c
ur
r
e
nc
e
of
th
e
two
c
l
a
s
s
e
s
s
i
m
ul
ta
ne
ous
ly
.
T
h
e
jo
in
t
e
nt
r
opy
c
a
n
be
c
a
lc
ul
a
te
d us
in
g t
he
f
or
m
ul
a
,
=
−
=
1
(
)
l
o
g
2
(
)
=
1,…,n
(
1)
in
w
hi
c
h
,
is
th
is
r
e
pr
e
s
e
nt
in
g
th
e
j
oi
nt
e
nt
r
opy f
or
t
he
-
th
c
a
te
gor
y.
is
to
ta
l
num
be
r
of
c
a
te
gor
ie
s
or
c
la
s
s
e
s
i
n t
he
-
th
va
r
ia
bl
e
be
in
g c
ons
id
e
r
e
d
.
is
th
e
pr
ob
a
bi
li
t
y of
j
oi
nt
o
c
c
ur
r
e
nc
e
of
t
w
o
c
la
s
s
e
s
, na
m
e
ly
c
la
s
s
a
nd
c
la
s
s
,
in
w
hi
c
h
r
e
pr
e
s
e
nt
s
th
e
f
ir
s
t
c
l
a
s
s
a
nd
r
e
pr
e
s
e
nt
s
t
h
e
s
e
c
o
nd
c
la
s
s
,
a
nd
l
o
g
2
(
)
T
h
is
i
s
th
e
l
og
a
r
it
hm
ba
s
e
2
of
t
h
e
jo
i
nt
pr
o
ba
bi
l
it
y
.
n
is
t
he
to
ta
l
num
be
r
of
c
la
s
s
e
s
.
T
he
m
in
us
s
ig
n
(
-
)
in
th
e
jo
in
t
e
nt
r
opy
f
or
m
ul
a
is
be
c
a
u
s
e
e
nt
r
opy
m
e
a
s
ur
e
s
unc
e
r
ta
in
ty
in
a
pr
oba
bi
li
ty
di
s
tr
ib
ut
io
n.
T
he
m
in
us
s
ig
n
e
ns
ur
e
s
th
a
t
th
e
e
nt
r
op
y
va
lu
e
is
a
lwa
ys
pos
it
iv
e
.
T
he
lo
ga
r
it
hm
of
a
num
be
r
be
twe
e
n 0 a
nd 1 pr
oduc
e
s
a
ne
ga
ti
ve
va
lu
e
. T
he
r
e
f
or
e
,
th
e
m
in
us
s
ig
n e
ns
ur
e
s
t
ha
t
e
nt
r
opy is
pos
it
iv
e
be
c
a
us
e
un
c
e
r
ta
in
ty
(
e
nt
r
opy)
c
a
nnot
be
ne
ga
ti
ve
. I
n t
he
c
ont
e
xt
of
t
he
C
o
-
tr
a
in
in
g
m
e
th
od,
e
nt
r
opy is
us
e
d t
o
e
va
lu
a
te
t
he
m
ode
l'
s
c
onf
id
e
nc
e
i
n ps
e
udo
-
la
be
l
s
.
−
M
a
r
gi
n
-
ba
s
e
d
c
onf
id
e
nc
e
:
la
be
l
c
onf
id
e
nc
e
c
a
n
a
ls
o
be
m
e
a
s
ur
e
d
ba
s
e
d
on
th
e
m
a
r
gi
n
be
twe
e
n
th
e
pr
oba
bi
li
ty
of
th
e
pr
im
a
r
y
la
be
l
a
nd
th
e
s
e
c
ond
m
os
t
li
ke
ly
la
be
l.
T
he
la
r
ge
r
th
e
m
a
r
gi
n,
th
e
hi
ghe
r
th
e
m
ode
l'
s
c
onf
id
e
nc
e
i
n t
he
s
e
l
e
c
te
d l
a
be
l
[
36]
.
−
V
ot
e
-
ba
s
e
d
c
onf
id
e
nc
e
:
in
c
o
-
tr
a
in
in
g
, da
ta
c
a
n be
la
be
l
ed
by both m
ode
ls
. I
f
bot
h m
ode
ls
gi
ve
t
he
s
a
m
e
la
be
l,
t
he
n t
he
c
onf
id
e
nc
e
i
n t
he
l
a
be
l
i
s
hi
ghe
r
t
ha
n i
f
t
he
y gi
ve
di
f
f
e
r
e
nt
la
be
ls
[
20]
.
−
C
onf
id
e
nc
e
m
e
a
s
ur
e
s
c
om
bi
na
ti
on:
s
om
e
ti
m
e
s
,
c
om
bi
ni
ng
m
ul
ti
pl
e
c
onf
id
e
nc
e
m
e
a
s
ur
e
s
c
a
n
pr
ovi
de
a
m
or
e
c
om
pr
e
he
ns
iv
e
pi
c
tu
r
e
of
t
he
m
ode
l'
s
c
onf
id
e
nc
e
i
n a
gi
v
e
n l
a
be
l
[
33]
.
2.
M
E
T
H
O
D
Co
-
tr
a
in
in
g
is
a
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng
a
ppr
oa
c
h
in
c
lu
di
ng
two
m
ode
ls
:
S
V
M
a
nd
L
S
T
M
.
T
he
y
m
ut
ua
ll
y
ove
r
s
e
e
e
a
c
h
ot
he
r
a
nd
r
e
f
in
e
th
e
m
s
e
lv
e
s
by
le
ve
r
a
gi
ng
in
s
ig
ht
s
f
r
om
unl
a
be
le
d
da
ta
[
37]
.
T
he
c
o
-
tr
a
in
in
g a
lg
or
it
hm
pr
e
s
e
nt
e
d i
s
a
s
A
lg
or
it
hm
1
.
A
lg
or
it
hm
1:
C
o
-
t
r
a
in
in
g A
lg
o
r
it
hm
1
L
D
L
a
be
le
d
d
a
ta
2
UD
U
nl
a
be
le
d
d
a
ta
3
RD
R
a
ndom
d
a
ta
U
D
→
UD’
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
3
,
J
une
20
25
:
2158
-
2168
2162
4
F
a
il
e
dl
te
r
0
5
W
hi
le
(
F
a
il
e
dl
te
r
<
n)
6
h1 (
S
V
M
)
T
r
a
in
in
g
F
ir
s
t
L
D
7
h2
(
L
S
T
M
)
T
r
a
in
in
g
S
e
c
ond L
D
8
U
D
’
1
U
D
’
C
la
s
s
if
ic
a
ti
on w
it
h h1
9
U
D
’
2
U
D
’
C
la
s
s
if
ic
a
ti
on w
it
h h2
10
U
D
’
1
L
a
be
l
be
s
t
c
la
s
s
if
ic
a
ti
on e
x
a
m
pl
e
11
U
D
’
2
L
a
be
l
be
s
t
c
la
s
s
if
ic
a
ti
on e
x
a
m
pl
e
12
LD
A
ppe
nd U
D
’
1 a
nd U
D
’
2
13
if
N
o
e
xa
m
pl
e
a
dde
d
14
F
a
il
dl
te
r
+
+
15
e
ls
e
16
F
a
il
dl
te
r
0
17
e
nd
18
E
nd
I
n
th
e
c
o
-
t
r
a
in
in
g
pr
oc
e
s
s
,
th
e
unl
a
be
le
d
da
ta
(
U
D
)
is
la
be
le
d
w
he
n
bot
h
c
la
s
s
if
ie
r
s
(
L
S
T
M
a
nd
S
V
M
)
in
de
pe
nde
nt
ly
a
gr
e
e
to
a
s
s
ig
n
a
la
be
l.
T
he
f
ir
s
t
c
la
s
s
if
ie
r
(
h1
or
S
V
M
)
a
s
s
ig
n
s
a
la
be
l
w
it
h
a
c
onf
id
e
nc
e
le
ve
l
gr
e
a
te
r
th
a
n
a
c
e
r
ta
in
th
r
e
s
hol
d
va
lu
e
X
,
in
di
c
a
ti
ng
a
hi
gh
le
ve
l
of
c
e
r
ta
in
ty
.
S
im
il
a
r
ly
,
th
e
s
e
c
ond
c
la
s
s
if
ie
r
(
h2
or
L
S
T
M
)
a
s
s
ig
ns
a
la
be
l
w
it
h
a
c
onf
id
e
nc
e
le
ve
l
gr
e
a
te
r
th
a
n
a
th
r
e
s
hol
d
va
lu
e
of
Y
.
T
he
co
-
tr
a
in
in
g
pr
oc
e
s
s
is
h
a
lt
e
d
w
he
n
bot
h
c
l
a
s
s
if
ie
r
s
f
a
il
to
a
gr
e
e
on
th
e
s
a
m
e
la
be
l
dur
in
g a
n
it
e
r
a
ti
on
.
T
he
co
-
tr
a
in
in
g
a
lg
or
it
hm
in
A
lg
or
it
h
m
1
te
s
ts
on
100
da
ta
s
e
ts
w
it
h
R
D
=
5
s
e
t
a
s
th
e
f
ix
e
d
va
lu
e
.
T
he
s
e
da
ta
s
e
ts
in
c
lu
de
90
unl
a
be
le
d
te
s
ti
ng
da
ta
poi
nt
s
a
nd
10
la
be
le
d
tr
a
in
in
g
da
ta
poi
nt
s
.
T
hi
s
pr
opor
ti
on
o
f
la
be
le
d
a
nd
unl
a
be
le
d
da
ta
i
s
us
e
d
to
a
na
ly
z
e
th
e
pe
r
f
or
m
a
nc
e
of
co
-
t
r
a
in
in
g.
N
ig
a
m
a
nd
G
ha
ni
[
38]
ha
ve
a
r
gue
d
th
a
t
a
lt
hough
th
e
gr
e
a
te
s
t
pe
r
f
or
m
a
nc
e
im
pr
ove
m
e
nt
oc
c
ur
s
w
he
n
th
e
la
be
l
ed
da
ta
i
s
ve
r
y
s
m
a
ll
,
a
ddi
ng
f
ur
th
e
r
la
be
l
ed
da
ta
pr
ovi
de
s
a
s
m
a
ll
e
r
i
m
pr
ove
m
e
nt
.
T
he
co
-
tr
a
in
in
g
a
lg
or
it
hm
f
lo
w
in
A
lg
or
it
hm
1
be
gi
n
s
w
it
h
th
e
in
it
ia
li
z
a
ti
on
pr
oc
e
s
s
of
L
D
a
nd
U
D
.
A
t
th
is
in
it
ia
li
z
a
ti
on
s
ta
g
e
,
th
e
c
onf
id
e
nc
e
s
c
or
e
th
r
e
s
hol
d
f
or
S
V
M
a
nd
L
S
T
M
is
de
te
r
m
in
e
d,
a
nd
th
e
la
be
le
d
da
ta
s
e
t
to
tr
a
in
e
a
c
h
m
ode
l
a
nd
th
e
unl
a
b
e
le
d
da
ta
s
e
t
(
R
D
)
a
r
e
de
te
r
m
in
e
d.
I
n
li
ne
6,
it
e
r
a
ti
ons
w
il
l
be
c
a
r
r
ie
d
out
on
th
e
tr
a
in
in
g
pr
oc
e
s
s
of
th
e
S
V
M
c
la
s
s
if
ie
r
.
L
in
e
8
pr
oduc
e
s
pr
e
di
c
ti
ons
w
it
h
S
V
M
(
U
D
'
1)
,
a
nd
in
li
ne
10,
e
a
c
h
pr
e
di
c
ti
on
m
a
de
by
th
e
S
V
M
m
od
e
l
is
a
c
c
om
pa
ni
e
d
by
a
c
onf
id
e
nc
e
s
c
or
e
.
T
hi
s
s
c
or
e
in
di
c
a
te
s
th
e
le
ve
l
of
c
onf
id
e
nc
e
of
th
e
m
ode
l
in
th
e
c
la
s
s
if
ic
a
ti
on
de
c
is
io
n
m
a
de
by
th
e
S
V
M
,
in
w
hi
c
h
th
e
c
onf
id
e
nc
e
s
c
or
e
w
il
l
be
th
e
th
r
e
s
hol
d
f
or
unl
a
be
le
d
d
a
ta
w
h
e
n
gi
ve
n
a
ps
e
udo
-
la
b
e
l
s
o
th
a
t
th
e
unl
a
be
le
d
da
ta
ge
ts
a
hi
ghe
r
c
onf
id
e
nc
e
s
c
or
e
th
a
n
th
e
S
V
M
th
r
e
s
hol
d
d
ur
in
g
th
e
tr
a
in
in
g
s
e
t.
I
n
li
n
e
7,
it
e
r
a
ti
ons
w
il
l
be
c
a
r
r
ie
d
out
on
th
e
tr
a
in
in
g
pr
oc
e
s
s
of
th
e
L
S
T
M
c
la
s
s
if
ie
r
to
pr
oduc
e
pr
e
di
c
ti
ons
w
it
h
L
S
T
M
(
U
D
'
2)
a
nd
c
onf
id
e
nc
e
s
c
or
e
s
on unla
b
e
le
d da
ta
, t
he
n a
dd p
s
e
udo
-
la
be
ls
w
i
th
hi
ghe
r
c
onf
id
e
nc
e
t
ha
n t
he
L
S
T
M
t
hr
e
s
hol
d
to
th
e
tr
a
in
in
g
s
e
t.
F
or
th
e
S
V
M
a
nd
L
S
T
M
a
lg
or
it
hm
s
th
a
t
a
r
e
be
in
g
te
s
te
d,
th
e
co
-
tr
a
in
in
g
a
ppr
oa
c
h
a
im
s
to
de
te
r
m
in
e
th
e
c
onf
id
e
nc
e
le
ve
l
th
r
e
s
hol
d
th
a
t
is
us
e
d
dur
in
g
th
e
co
-
tr
a
in
in
g
pr
oc
e
dur
e
.
F
or
th
e
S
V
M
a
nd
L
S
T
M
a
lg
or
it
hm
s
be
in
g
te
s
te
d,
th
e
co
-
tr
a
in
in
g
a
ppr
oa
c
h
a
im
s
to
de
te
r
m
in
e
th
e
c
onf
id
e
nc
e
le
ve
l
th
r
e
s
hol
d
us
e
d dur
in
g t
he
co
-
tr
a
in
in
g
pr
oc
e
dur
e
.
T
he
pha
s
e
s
in
th
e
co
-
t
r
a
in
in
g
pr
oc
e
s
s
f
lo
w
a
r
e
de
ta
il
e
d
in
F
ig
ur
e
s
1(
a
)
a
nd
1(
b)
.
F
ig
ur
e
1(
a
)
s
how
s
th
a
t
th
e
te
c
hni
que
be
gi
ns
by
lo
a
di
ng
a
da
ta
s
e
t
f
r
om
a
f
il
e
(
e
.g.,
F
il
e
.c
s
v)
.
T
he
da
ta
s
e
t
is
th
e
n
pr
e
-
pr
oc
e
s
s
e
d
to
c
le
a
n
a
nd
pr
e
pa
r
e
th
e
da
ta
f
or
tr
a
in
in
g.
O
nc
e
pr
e
pr
oc
e
s
s
in
g
is
c
om
pl
e
te
,
th
e
w
or
kf
lo
w
c
he
c
ks
w
he
th
e
r
tr
a
in
in
g
da
ta
is
a
va
il
a
bl
e
.
I
f
no
tr
a
in
in
g
da
ta
is
pr
e
s
e
nt
,
th
e
pr
oc
e
s
s
te
r
m
in
a
te
s
.
I
f
tr
a
in
in
g
da
ta
is
a
va
il
a
bl
e
, a
n
S
V
M
c
la
s
s
if
ie
r
is
a
ppl
ie
d
w
it
h
ps
e
udo
-
la
be
l
in
g
to
ge
ne
r
a
te
la
b
e
ls
f
or
th
e
un
la
be
l
ed
da
ta
. T
he
s
e
ps
e
udo
-
la
b
e
ls
a
r
e
th
e
n
us
e
d
to
c
r
e
a
te
a
ne
w
c
la
s
s
if
ic
a
ti
on
m
ode
l.
T
he
upd
a
te
d
c
la
s
s
if
ic
a
ti
on
m
ode
l
is
ut
il
iz
e
d
to
g
e
ne
r
a
te
pr
e
di
c
ti
ons
f
or
t
he
da
ta
. T
he
w
or
kf
lo
w
t
he
n e
va
lu
a
te
s
t
he
ne
w
pr
e
di
c
ti
on mode
l
to
de
te
r
m
in
e
i
ts
pe
r
f
or
m
a
nc
e
.
I
f
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
m
e
e
ts
th
e
de
s
ir
e
d
c
r
it
e
r
ia
,
th
e
pr
oc
e
s
s
m
a
y
it
e
r
a
te
a
ga
in
to
r
e
f
in
e
th
e
pr
e
di
c
ti
ons
f
ur
th
e
r
.
T
he
pr
oc
e
s
s
c
on
c
lu
de
s
if
th
e
pe
r
f
or
m
a
nc
e
is
uns
a
ti
s
f
a
c
t
or
y
or
f
ur
th
e
r
it
e
r
a
ti
ons
a
r
e
not
r
e
qui
r
e
d.
T
hi
s
it
e
r
a
ti
ve
c
yc
le
a
im
s
t
o i
m
pr
ove
t
he
c
la
s
s
if
ic
a
ti
on mode
l
by i
nt
e
gr
a
ti
ng
ps
e
udo
-
la
be
l
in
g
w
it
h S
V
M
.
S
im
il
a
r
ly
,
th
e
c
o
-
tr
a
in
in
g
m
e
th
odol
ogy
de
pi
c
te
d
in
F
ig
ur
e
1(
b)
s
ta
r
ts
by
lo
a
di
ng
a
da
ta
s
e
t
f
r
om
a
f
il
e
(
e
.g.,
F
il
e
.c
s
v)
.
O
nc
e
th
e
da
ta
s
e
t
is
lo
a
de
d,
it
unde
r
goe
s
pr
e
pr
oc
e
s
s
in
g
to
c
le
a
n
a
nd
pr
e
pa
r
e
th
e
da
ta
f
or
f
ur
th
e
r
us
e
,
s
uc
h
a
s
ha
ndl
in
g
m
is
s
in
g
v
a
lu
e
s
,
nor
m
a
li
z
in
g
f
e
a
t
ur
e
s
,
or
te
xt
to
ke
ni
z
a
ti
on.
A
f
te
r
pr
e
pr
oc
e
s
s
in
g,
th
e
s
ys
te
m
c
he
c
ks
if
tr
a
in
in
g
da
ta
is
a
v
a
il
a
bl
e
.
I
f
no
tr
a
in
in
g
da
ta
is
pr
e
s
e
nt
,
th
e
pr
oc
e
s
s
e
nd
s
.
H
ow
e
v
e
r
,
if
tr
a
in
in
g
da
ta
is
a
va
il
a
bl
e
,
th
e
ne
xt
s
te
p
in
vol
ve
s
us
in
g
a
n
L
S
T
M
m
ode
l
w
it
h
ps
e
udo
-
la
be
li
ng.
T
he
L
S
T
M
m
ode
l
is
tr
a
in
e
d
on
th
e
e
xi
s
ti
ng
la
be
le
d
da
ta
,
a
nd
ps
e
udo
-
la
be
ls
a
r
e
ge
ne
r
a
te
d
f
or
th
e
unl
a
be
le
d
da
ta
w
it
h
a
c
onf
id
e
nc
e
th
r
e
s
hol
d
to
e
n
s
ur
e
qua
li
ty
.
T
h
e
s
e
p
s
e
udo
-
la
be
le
d
s
a
m
pl
e
s
a
r
e
th
e
n
us
e
d
to
upda
te
th
e
tr
a
in
in
g
da
ta
s
e
t,
w
hi
c
h
a
ll
ow
s
f
or
c
r
e
a
ti
ng
a
ne
w
c
la
s
s
if
ic
a
ti
on
m
ode
l.
T
hi
s
upda
te
d
m
ode
l
is
us
e
d
to
ge
ne
r
a
te
n
e
w
pr
e
di
c
ti
ons
.
T
h
e
ne
w
ly
tr
a
in
e
d
m
ode
l'
s
pe
r
f
or
m
a
nc
e
is
e
v
a
lu
a
te
d
ba
s
e
d
on
pr
e
de
f
in
e
d
c
r
it
e
r
ia
.
I
f
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Co
-
tr
ai
ni
ng ps
e
udo
-
la
be
li
ng f
or
t
e
x
t
c
la
s
s
if
ic
at
io
n w
it
h s
uppo
r
t
v
e
c
to
r
m
ac
hi
n
e
…
(
Sr
i
H
anday
ani
)
2163
e
va
lu
a
ti
on
in
di
c
a
te
s
th
e
m
ode
l'
s
pe
r
f
or
m
a
nc
e
is
s
a
ti
s
f
a
c
to
r
y,
th
e
pr
oc
e
s
s
c
a
n
be
r
e
pe
a
te
d
to
r
e
f
in
e
th
e
m
ode
l
w
it
h
a
ddi
ti
ona
l
ps
e
udo
-
la
be
li
ng
it
e
r
a
ti
ons
.
T
he
pr
oc
e
s
s
c
onc
lu
de
s
if
th
e
e
va
lu
a
ti
on
f
a
il
s
to
m
e
e
t
th
e
r
e
qui
r
e
d
s
ta
nda
r
ds
or
if
f
ur
th
e
r
it
e
r
a
ti
ons
a
r
e
unn
e
c
e
s
s
a
r
y.
T
hi
s
it
e
r
a
ti
ve
w
or
kf
lo
w
e
nha
nc
e
s
th
e
c
la
s
s
if
ic
a
ti
on
m
ode
l'
s
a
c
c
ur
a
c
y by int
e
gr
a
ti
ng
ps
e
udo
-
la
be
li
ng a
nd L
S
T
M
f
or
s
e
que
nt
ia
l
da
ta
.
(
a
)
(
b)
F
ig
ur
e
1. F
lo
w
c
ha
r
t
c
o
-
tr
a
in
in
g pr
oc
e
s
s
f
or
(
a
)
S
V
M
a
nd (
b)
L
S
T
M
3.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
he
im
pl
e
m
e
nt
a
ti
on
of
co
-
tr
a
in
in
g
to
p
r
oduc
e
a
n
in
f
o
r
m
a
ti
o
n
f
r
a
m
e
w
or
k
f
oc
us
e
s
on
le
ve
r
a
gi
ng
la
be
le
d
a
nd
unl
a
be
le
d
da
ta
to
e
nha
n
c
e
th
e
a
c
c
ur
a
c
y
a
nd
r
e
l
ia
bi
li
ty
of
pr
e
di
c
ti
ve
m
ode
ls
.
T
hi
s
a
ppr
oa
c
h
in
te
gr
a
te
s
c
om
pl
e
m
e
nt
a
r
y
a
lg
or
it
hm
s
,
s
uc
h
a
s
S
V
M
a
nd
L
S
T
M
,
to
ge
ne
r
a
te
ps
e
udo
-
la
be
ls
it
e
r
a
ti
ve
ly
,
e
na
bl
in
g
th
e
de
ve
lo
pm
e
nt
of
a
r
obus
t
f
r
a
m
e
w
or
k
th
a
t
s
uppor
ts
da
ta
c
la
s
s
if
ic
a
ti
on,
de
c
is
io
n
-
m
a
ki
ng,
a
nd
pe
r
f
or
m
a
nc
e
e
va
lu
a
ti
on
[
33]
.
T
he
in
f
or
m
a
ti
on
s
ys
te
m
f
r
a
m
e
w
o
r
k
pr
ovi
de
s
a
n
out
li
ne
of
a
s
im
pl
e
in
f
or
m
a
ti
on
s
ys
te
m
.
I
t
s
how
s
th
e
m
a
in
pa
r
ts
th
a
t
a
r
e
th
e
s
a
m
e
in
a
ll
in
f
or
m
a
ti
on
s
to
r
a
ge
a
nd
r
e
tr
ie
va
l
in
s
ti
tu
ti
ons
,
s
uc
h
a
s
li
br
a
r
ie
s
,
a
r
c
hi
ve
s
,
doc
um
e
nt
a
ti
on
a
nd
in
f
or
m
a
ti
on
c
e
nt
e
r
s
,
r
e
ga
r
dl
e
s
s
of
th
e
l
e
ve
l
of
m
e
c
h
a
ni
s
m
or
ty
pe
of
in
f
or
m
a
ti
on
m
a
na
ge
d
by
in
s
ti
tu
ti
ons
.
B
y
lo
oki
ng
a
t
th
e
in
f
o
r
m
a
ti
on
s
ys
te
m
f
r
a
m
e
w
or
k,
it
is
e
xpe
c
te
d
to
lo
ok
th
e
in
f
or
m
a
ti
on
s
ys
te
m
f
r
a
m
e
w
or
k
w
ha
t
c
om
pone
nt
s
in
f
or
m
a
ti
on
uni
ts
c
a
n
be
unde
r
s
to
od
a
nd
w
h
a
t
pr
oc
e
s
s
s
houl
d
oc
c
ur
[
37]
.
T
he
a
ut
hor
s
w
il
l
u
s
e
la
b
e
le
d
in
put
da
ta
(
tr
a
in
in
g
da
ta
)
,
w
hi
c
h
m
a
ke
s
up
a
r
ound
5%
of
th
e
e
nt
ir
e
da
ta
s
e
t,
to
te
s
t
th
e
s
ugge
s
te
d
in
f
or
m
a
ti
on
s
ys
te
m
f
r
a
m
e
w
or
k.
T
hi
s
s
e
le
c
ti
on
of
th
e
pe
r
c
e
nt
a
g
e
of
tr
a
in
in
g
da
ta
is
c
ons
is
te
nt
w
it
h
s
tu
dy
li
te
r
a
tu
r
e
pr
a
c
ti
c
e
s
,
w
he
r
e
th
e
pe
r
c
e
nt
a
ge
of
tr
a
in
in
g
da
ta
us
ua
ll
y
f
a
ll
s
be
twe
e
n
1%
a
nd
50%
of
th
e
to
ta
l
d
a
ta
s
e
t,
de
pe
ndi
ng
on
t
he
pa
r
ti
c
ul
a
r
pr
obl
e
m
dom
a
in
a
nd
r
e
s
our
c
e
s
a
va
il
a
bl
e
, i
n a
ddi
ti
on t
o t
e
s
ti
ng da
ta
(
unl
a
be
le
d da
t
a
)
[
38]
.
To
de
te
r
m
in
e
th
e
th
r
e
s
hol
d
va
lu
e
of
th
e
la
be
l
c
onf
id
e
nc
e
l
e
ve
l
of
th
e
S
V
M
c
la
s
s
if
ie
r
a
nd
th
e
L
S
T
M
c
la
s
s
if
ie
r
in
th
e
co
-
tr
a
in
in
g
m
e
th
od,
th
e
s
te
ps
a
r
e
to
tr
a
in
th
e
f
ir
s
t
a
nd
s
e
c
ond
c
la
s
s
if
ic
a
ti
on
m
ode
ls
u
s
in
g
th
e
da
ta
la
be
l
e
d a
s
de
s
ir
e
d i
n t
he
ps
e
udo
-
la
be
l
in
g
pr
oc
e
s
s
. F
or
t
he
f
ir
s
t
c
la
s
s
if
ic
a
ti
on mode
l
tr
a
in
in
g, t
he
r
e
s
ul
ts
of
th
e
la
be
l
e
d
da
ta
c
la
s
s
if
ic
a
ti
on
a
r
e
us
e
d
to
c
la
s
s
if
y
un
la
be
l
ed
d
a
ta
s
o
th
a
t
th
e
l
a
be
l
c
onf
id
e
nc
e
l
e
ve
l
of
e
a
c
h
pr
e
di
c
ti
on
w
il
l
be
obt
a
in
e
d.
L
ik
e
w
is
e
,
f
or
th
e
s
e
c
ond
c
la
s
s
if
ic
a
ti
on
m
ode
l
tr
a
in
in
g,
th
e
r
e
s
ul
ts
of
th
e
la
be
l
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
3
,
J
une
20
25
:
2158
-
2168
2164
da
ta
c
la
s
s
if
ic
a
ti
on
w
il
l
b
e
us
e
d
to
c
la
s
s
if
y
un
la
be
l
ed
d
a
ta
t
o
obt
a
in
th
e
la
b
e
l
c
onf
id
e
nc
e
le
ve
l
of
e
a
c
h
pr
e
di
c
ti
on i
n t
he
s
e
c
ond c
la
s
s
if
ic
a
ti
on.
T
he
di
a
gr
a
m
in
F
ig
ur
e
2
il
lu
s
tr
a
te
s
th
e
s
ys
te
m
f
r
a
m
e
w
or
k
f
o
r
co
-
tr
a
in
in
g
us
in
g
S
V
M
a
nd
L
S
T
M
w
it
h
a
ps
e
udo
-
la
be
l
in
g
a
ppr
oa
c
h
to
pr
oc
e
s
s
a
nd
c
la
s
s
if
y
s
tu
de
nt
da
ta
.
T
he
s
y
s
te
m
be
gi
ns
w
it
h
two
da
ta
s
e
ts
:
tr
a
in
in
g
da
ta
,
w
hi
c
h
c
ons
is
t
s
of
a
s
m
a
ll
s
e
t
of
s
upe
r
vi
s
e
d
da
ta
w
it
h
la
be
l
ed
s
a
m
pl
e
s
,
a
nd
te
s
ti
ng
da
ta
,
w
hi
c
h
c
ont
a
in
s
m
a
ny
uns
up
e
r
vi
s
e
d,
un
la
b
e
l
ed
d
a
ta
.
T
he
L
S
T
M
m
ode
l
f
ir
s
t
pr
oc
e
s
s
e
s
th
e
tr
a
in
in
g
d
a
ta
a
nd
ge
n
e
r
a
te
s
ps
e
udo
-
la
be
ls
f
or
th
e
te
s
ti
ng
d
a
ta
ba
s
e
d
on
it
s
le
a
r
ni
ng
f
r
om
t
he
la
be
l
ed
s
a
m
pl
e
s
.
S
im
ul
ta
ne
ous
ly
,
th
e
S
V
M
m
ode
l
is
a
ppl
ie
d
to
th
e
te
s
ti
ng
da
ta
to
ge
ne
r
a
te
it
s
ps
e
udo
-
la
be
ls
.
T
he
s
e
ps
e
udo
-
la
be
ls
f
r
om
bot
h
m
ode
ls
a
r
e
th
e
n c
om
pa
r
e
d i
n t
hr
e
e
ke
y s
ta
ge
s
.
−
S
ig
ni
f
ic
a
nc
e
of
c
om
pa
r
in
g
ps
e
udo
-
la
be
li
ng
r
e
s
ul
ts
:
th
e
r
e
s
ul
ts
of
ps
e
udo
-
la
be
l
in
g
,
a
c
r
uc
ia
l
s
te
p
in
m
ode
l
e
va
lu
a
ti
on pe
r
f
or
m
e
d by both L
S
T
M
a
nd S
V
M
, a
r
e
c
om
pa
r
e
d w
it
h t
he
te
s
t
da
ta
. T
hi
s
c
om
pa
r
is
on
is
vi
ta
l
in
e
va
lu
a
ti
ng how w
e
ll
t
he
m
ode
ls
p
e
r
f
or
m
on t
he
un
la
be
l
ed
da
ta
s
e
t.
−
C
onf
id
e
nc
e
pr
e
di
c
ti
on
c
om
pa
r
is
on:
th
e
pr
e
di
c
ti
ons
m
a
d
e
by
th
e
L
S
T
M
a
nd
S
V
M
m
ode
l
s
a
r
e
c
om
p
a
r
e
d
in
te
r
m
s
of
th
e
ir
c
onf
id
e
nc
e
le
ve
l
s
.
T
hi
s
s
te
p
is
in
va
lu
a
bl
e
in
id
e
nt
if
yi
ng
w
hi
c
h
m
ode
l
pr
oduc
e
s
m
or
e
r
e
li
a
bl
e
ps
e
udo
-
la
be
ls
.
−
C
om
pa
r
is
on
w
it
h
tr
a
in
in
g
da
ta
:
th
e
ps
e
udo
-
la
be
l
in
g
r
e
s
ul
ts
f
or
th
e
te
s
ti
ng
da
ta
a
r
e
c
om
pa
r
e
d
a
ga
in
s
t
th
e
or
ig
in
a
l
la
be
l
ed
tr
a
in
in
g da
ta
to
e
va
lu
a
te
c
ons
i
s
te
nc
y a
nd
a
li
gnm
e
nt
w
it
h t
he
i
ni
ti
a
l
da
ta
s
e
t.
F
ig
ur
e
2.
Co
-
tr
a
in
in
g m
e
th
od i
nf
o
r
m
a
ti
on s
ys
te
m
f
r
a
m
e
w
or
k
T
he
s
ys
te
m
'
s
pe
r
f
o
r
m
a
nc
e
is
t
he
n
a
s
s
e
s
s
e
d
us
in
g
a
c
c
u
r
a
c
y,
pr
e
c
is
i
on
,
r
e
c
a
l
l,
a
nd
F
1
-
s
c
o
r
e
m
e
tr
ic
s
.
B
a
s
e
d
o
n
t
hi
s
e
va
l
ua
t
io
n,
th
e
s
ys
te
m
p
r
o
vi
de
s
a
c
ti
ona
bl
e
out
pu
ts
,
in
c
lu
di
n
g
de
c
is
io
ns
,
s
u
gge
s
ti
ons
,
pol
ic
i
e
s
,
a
nd
r
e
c
o
m
m
e
nd
a
t
io
ns
f
o
r
f
u
r
th
e
r
im
p
r
ove
m
e
nt
s
.
T
hi
s
i
te
r
a
t
iv
e
f
r
a
m
e
w
o
r
k
e
ns
u
r
e
s
r
ob
us
t
c
la
s
s
i
f
ic
a
t
io
n
o
f
s
tu
de
n
t
da
ta
b
y
le
v
e
r
a
gi
ng
b
ot
h
m
ode
ls
'
s
t
r
e
ng
th
s
a
n
d
r
e
f
in
in
g
pr
e
di
c
t
io
ns
th
r
o
ugh
ps
e
ud
o
-
la
be
l
in
g
.
B
y
le
ve
r
a
g
in
g
t
he
r
o
bus
t
c
a
p
a
bi
li
ti
e
s
o
f
G
oo
gl
e
C
o
la
b
,
t
he
tr
ia
l
i
m
p
le
m
e
n
ta
t
io
n
o
f
c
o
-
t
r
a
i
ni
ng
be
tw
e
e
n
S
V
M
a
nd
L
S
T
M
y
ie
ld
e
d
a
th
r
e
s
h
ol
d
la
be
l
va
lu
e
f
or
S
V
M
of
0.
88
a
nd
a
t
hr
e
s
ho
ld
la
b
e
l
va
l
ue
f
o
r
L
S
T
M
o
f
0.5
,
a
s
de
pi
c
te
d
i
n
F
i
gu
r
e
3(
a
)
.
I
t
de
m
ons
t
r
a
te
s
t
h
a
t
th
e
le
ve
l
o
f
la
be
l
c
on
f
i
de
nc
e
f
o
r
S
V
M
is
>
=
0.8
8.
F
i
gu
r
e
3
(
b
)
de
m
ons
tr
a
te
s
th
a
t
th
e
l
e
ve
l
o
f
l
a
be
l
c
o
nf
id
e
nc
e
f
o
r
L
S
T
M
is
>
=
0
.5
.
L
a
b
e
l
c
o
nf
id
e
nc
e
e
s
t
im
a
ti
on
is
a
n
i
m
p
or
ta
nt
s
te
p
in
a
s
e
m
i
-
s
u
pe
r
vi
s
e
d
a
lg
o
r
i
th
m
[
2
5]
.
L
a
be
l
c
on
f
i
de
nc
e
e
s
t
im
a
ti
on
a
im
s
to
p
r
e
ve
nt
un
la
be
le
d
da
ta
f
r
om
be
in
g
m
is
l
a
be
le
d
,
w
h
ic
h
c
a
n
lo
w
e
r
th
e
la
be
l
th
r
e
s
h
ol
d
.
A
c
c
or
di
ng
t
o
th
e
la
b
e
l
c
on
f
i
de
nc
e
e
s
ti
m
a
ti
o
n
m
e
th
od,
it
c
a
n
be
d
iv
i
de
d
in
to
e
x
pl
ic
i
t
e
s
t
im
a
t
io
n
a
nd
im
pl
ic
i
t
e
s
ti
m
a
ti
o
n.
I
m
p
li
c
it
e
s
t
im
a
t
io
n
is
hi
gh
ly
de
p
e
nde
nt
on
th
e
t
r
a
in
in
g
da
ta
(
la
be
le
d
da
ta
)
;
unl
a
be
le
d
d
a
ta
m
a
y
be
m
is
la
be
l
e
d,
a
nd
it
w
i
ll
d
e
g
r
a
de
th
e
l
e
a
r
n
e
r
'
s
pe
r
f
o
r
m
a
nc
e
du
r
i
ng
it
e
r
a
ti
on,
s
o
t
he
a
c
c
u
r
a
c
y
of
im
pl
ic
it
e
s
ti
m
a
t
io
n
is
lo
w
e
r
th
a
n
e
x
pl
ic
it
e
s
ti
m
a
ti
on.
H
ow
e
ve
r
,
th
e
c
os
t
r
e
qu
ir
e
d
is
s
m
a
ll
e
r
th
a
n
t
he
e
xp
li
c
i
t
e
s
ti
m
a
t
io
n.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Co
-
tr
ai
ni
ng ps
e
udo
-
la
be
li
ng f
or
t
e
x
t
c
la
s
s
if
ic
at
io
n w
it
h s
uppo
r
t
v
e
c
to
r
m
ac
hi
n
e
…
(
Sr
i
H
anday
ani
)
2165
M
os
t
a
lg
or
it
hm
s
in
im
pl
ic
it
e
s
ti
m
a
ti
on
us
e
th
e
d
e
gr
e
e
of
di
f
f
e
r
e
nc
e
in
th
e
r
e
s
ul
ts
to
r
e
f
le
c
t
th
e
c
ur
r
e
nt
ps
e
udo
-
la
be
l
c
onf
id
e
nc
e
.
A
lg
or
it
hm
s
in
e
xpl
ic
it
e
s
ti
m
a
ti
on
us
e
e
xa
c
t
num
be
r
s
to
r
e
pr
e
s
e
nt
th
e
la
be
l
c
onf
id
e
nc
e
.
T
ypi
c
a
ll
y,
th
e
s
e
a
lg
or
it
hm
s
us
e
le
a
r
ni
ng
out
put
s
i
n
pr
oba
bi
li
s
ti
c
f
or
m
,
th
e
di
f
f
e
r
e
nc
e
in
m
ode
l
a
c
c
ur
a
c
y
be
f
or
e
a
nd
a
f
te
r
us
in
g
ps
e
udo
-
la
be
le
d
da
ta
,
or
th
e
s
im
il
a
r
it
y
be
twe
e
n
th
e
ps
e
udo
-
la
be
ls
of
th
e
c
ur
r
e
nt
unl
a
be
le
d da
ta
a
nd t
he
s
ur
r
ounding l
a
be
le
d da
t
a
.
F
ig
ur
e
3. T
he
r
e
la
ti
on of
da
ta
a
nd t
hr
e
s
hol
d f
or
(
a
)
S
V
M
a
nd (
b)
L
S
T
M
E
xpl
ic
it
e
s
ti
m
a
ti
on
is
a
ppl
ie
d
in
th
is
S
V
M
a
nd
L
S
T
M
C
o
-
tr
a
in
in
g
tr
ia
l.
A
s
th
e
f
ir
s
t
c
la
s
s
if
ie
r
,
S
V
M
is
us
e
d f
or
pr
e
di
c
ti
on,
a
nd L
S
T
M
, a
s
t
he
s
e
c
ond c
la
s
s
if
ie
r
, i
s
us
e
d f
or
unde
r
s
ta
ndi
ng t
he
s
e
que
nc
e
of
t
e
xt
da
ta
in
th
is
c
o
-
tr
a
in
in
g.
I
n
th
e
f
ir
s
t
c
la
s
s
if
ic
a
ti
on
(
S
V
M
)
,
w
h
e
n
th
e
la
be
l
gi
v
e
n
to
th
e
te
s
ti
ng
da
ta
h
a
s
a
la
b
e
l
c
onf
id
e
nc
e
le
ve
l
(
th
r
e
s
hol
d)
be
lo
w
0.88,
th
e
tr
a
in
in
g
da
ta
w
il
l
be
it
e
r
a
te
d
a
ga
in
s
o
th
a
t
th
e
la
be
l
th
a
t
w
il
l
be
gi
ve
n
to
th
e
te
s
ti
ng
da
ta
ha
s
a
la
be
l
c
onf
id
e
nc
e
le
ve
l
(
t
hr
e
s
hol
d)
>
=
0.88.
L
ik
e
w
is
e
,
in
th
e
s
e
c
ond
c
la
s
s
if
ic
a
ti
on
(
L
S
T
M
)
,
w
he
n
th
e
la
be
l
gi
v
e
n
to
th
e
te
s
ti
ng
da
ta
ha
s
a
la
b
e
l
c
onf
id
e
nc
e
le
ve
l
(
th
r
e
s
hol
d)
be
lo
w
0.5,
th
e
tr
a
in
in
g
da
ta
w
il
l
be
it
e
r
a
te
d
a
ga
in
s
o
th
a
t
th
e
la
be
l
th
a
t
w
il
l
be
gi
ve
n
to
th
e
t
e
s
ti
ng
da
t
a
ha
s
a
la
be
l
c
onf
id
e
nc
e
le
ve
l
(
th
r
e
s
hol
d)
>
=
0.5.
T
he
c
on
c
e
pt
of
m
e
a
s
ur
in
g
th
e
la
be
l
c
onf
id
e
nc
e
le
ve
l
(
th
r
e
s
hol
d)
in
th
is
a
r
ti
c
le
c
om
bi
ne
s
th
e
la
be
l
c
onf
id
e
nc
e
m
e
a
s
ur
e
s
f
r
om
c
la
s
s
if
ic
a
ti
on
1
a
nd
2
to
pr
ov
id
e
a
m
or
e
c
om
pr
e
he
ns
iv
e
vi
e
w
t
o w
hi
c
h t
he
m
ode
l
is
c
onf
id
e
nt
w
it
h t
he
l
a
be
l
gi
ve
n.
S
om
e
s
tu
di
e
s
a
li
gn
w
it
h
th
e
m
e
th
od
th
e
a
ut
hor
s
a
ppl
ie
d,
us
i
ng
two
di
f
f
e
r
e
nt
m
ode
ls
(
S
V
M
a
nd
L
S
T
M
)
in
th
e
c
o
-
tr
a
in
in
g
f
r
a
m
e
w
or
k.
A
c
ol
la
bor
a
ti
ve
a
ppr
oa
c
h
in
w
hi
c
h e
a
c
h
m
ode
l
pr
oduc
e
s
pr
e
di
c
ti
ons
on
unl
a
be
le
d
da
ta
a
nd
hi
ghl
y
c
onf
id
e
nt
pr
e
di
c
ti
ons
a
r
e
u
s
e
d
a
s
ps
e
udo
-
la
be
ls
to
tr
a
in
th
e
ot
he
r
m
ode
l.
F
or
e
xa
m
pl
e
,
r
e
s
e
a
r
c
h
th
a
t
c
om
bi
ne
s
two
de
e
p
le
a
r
ni
ng
m
ode
ls
tr
a
in
e
d
on
two
di
f
f
e
r
e
nt
r
e
pr
e
s
e
nt
a
ti
ons
of
te
xt
da
ta
f
or
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on.
E
a
c
h
m
ode
l
pr
oduc
e
s
pr
e
di
c
ti
ons
on
unl
a
be
le
d
da
ta
,
a
nd
hi
ghl
y
c
onf
id
e
nt
pr
e
di
c
ti
ons
a
r
e
us
e
d
a
s
p
s
e
udo
-
la
be
ls
to
tr
a
in
th
e
ot
he
r
m
ode
l.
T
hi
s
a
ppr
oa
c
h
im
pr
ove
s
th
e
a
c
c
ur
a
c
y
of
s
e
nt
im
e
nt
c
la
s
s
if
ic
a
ti
on
on
m
a
s
s
iv
e
ope
n
onl
in
e
c
our
s
e
s
(
M
O
O
C
)
or
c
a
ll
s
f
or
la
r
ge
-
s
c
a
le
onl
in
e
c
our
s
e
s
ope
n
to
th
e
publ
ic
,
u
s
ua
ll
y
pr
ovi
de
d
by
uni
ve
r
s
it
ie
s
or
onl
in
e
le
a
r
ni
ng
pl
a
tf
or
m
s
s
uc
h
a
s
C
our
s
e
r
a
,
e
dX
,
or
f
ut
ur
e
le
a
r
n
[
39]
.
F
u
r
th
e
r
m
or
e
,
s
tu
di
e
s
in
tr
oduc
in
g
M
e
ta
c
o
-
tr
a
in
in
g
ha
ve
th
e
pr
in
c
ip
le
th
a
t
two
m
ode
ls
t
r
a
in
e
d
in
de
pe
nde
nt
ly
on
di
f
f
e
r
e
nt
da
ta
r
e
pr
e
s
e
nt
a
ti
ons
c
a
n
c
or
r
e
c
t
e
a
c
h
ot
he
r
'
s
e
r
r
or
s
a
nd
im
pr
ove
th
e
ove
r
a
ll
a
c
c
ur
a
c
y,
or
two
pr
e
-
tr
a
in
e
d
m
ode
ls
c
a
n
be
c
om
bi
ne
d
on
di
f
f
e
r
e
nt
r
e
pr
e
s
e
nt
a
ti
ons
of
th
e
s
a
m
e
da
ta
.
E
a
c
h
m
ode
l
pr
oduc
e
s
pr
e
di
c
ti
ons
on
unl
a
be
le
d
da
ta
,
a
nd
pr
e
di
c
ti
ons
w
it
h
a
hi
gh
c
onf
id
e
nc
e
le
ve
l
a
r
e
us
e
d
a
s
ps
e
udo
-
la
be
ls
to
tr
a
in
th
e
ot
he
r
m
ode
l
[
26]
.
T
he
a
ppr
oa
c
h
ta
ke
n
by
th
e
a
ut
hor
s
us
e
s
two
di
f
f
e
r
e
nt
m
ode
ls
in
th
e
c
o
-
tr
a
in
in
g f
r
a
m
e
w
or
k, o
r
t
he
a
ppr
oa
c
h us
in
g t
w
o l
e
a
r
ni
ng
m
ode
ls
f
or
t
he
M
O
O
C
f
or
um
a
nd t
he
M
e
ta
c
o
-
tr
a
in
in
g
te
c
hni
que
.
T
he
s
e
m
ode
l
s
s
how
e
d
im
pr
ove
d
pe
r
f
or
m
a
nc
e
on
s
e
m
i
-
s
upe
r
vi
s
e
d
c
la
s
s
if
ic
a
ti
on
ta
s
k
s
,
w
hi
c
h s
how
s
t
ha
t
th
is
s
tr
a
te
gy e
f
f
e
c
ti
ve
ly
i
m
pr
ove
s
m
ode
l
pe
r
f
or
m
a
nc
e
on va
r
io
us
m
a
c
hi
ne
-
le
a
r
ni
ng t
a
s
k
s
.
4.
C
O
N
C
L
U
S
I
O
N
I
n
th
is
s
tu
dy,
G
oogl
e
C
ol
a
b
w
a
s
us
e
d
to
f
a
c
il
it
a
te
th
e
a
ppl
ic
a
ti
o
n
of
th
e
S
V
M
a
nd
L
S
T
M
co
-
t
r
a
in
in
g
te
c
hni
que
s
in
th
e
c
ont
e
xt
of
s
e
m
i
-
s
upe
r
vi
s
e
d
le
a
r
ni
ng.
T
he
f
ir
s
t
da
ta
s
e
ts
us
e
d
to
c
a
lc
ul
a
te
la
be
l
c
onf
id
e
nc
e
le
ve
ls
f
r
om
S
V
M
a
nd
L
S
T
M
w
e
r
e
obt
a
in
e
d
f
r
om
s
tu
de
nt
r
e
s
pons
e
s
to
que
s
ti
onn
a
ir
e
s
a
t
th
r
e
e
C
e
nt
r
a
l
J
a
va
uni
ve
r
s
it
ie
s
,
w
hi
c
h
w
e
r
e
c
ol
le
c
te
d
vi
a
G
oogl
e
F
or
m
s
ubm
is
s
io
ns
.
T
he
de
te
r
m
in
e
d
th
r
e
s
hol
d
la
b
e
ls
f
or
S
V
M
a
nd
L
S
T
M
a
r
e
≥0.88
a
nd
≥0.5,
r
e
s
pe
c
ti
ve
ly
.
C
la
s
s
if
ie
r
1
(
S
V
M
)
a
nd
c
la
s
s
if
ie
r
2
(
L
S
T
M
)
th
r
e
s
hol
ds
a
r
e
c
r
uc
ia
l
in
de
f
in
in
g
th
e
f
ol
lo
w
in
g
it
e
r
a
ti
ons
,
e
s
pe
c
ia
ll
y
w
he
n
a
ppl
yi
ng
ps
e
udo
la
be
ls
to
te
s
ti
ng
da
ta
.
I
n
pa
r
ti
c
ul
a
r
,
it
e
r
a
ti
ons
ha
ppe
n
w
h
e
n
th
e
p
s
e
udo
-
la
be
l
f
or
th
e
t
e
s
t
da
ta
is
le
s
s
th
a
n
th
e
c
or
r
e
s
ponding
th
r
e
s
hol
d
va
lu
e
s
of
0.5 f
or
L
S
T
M
a
nd 0.88 f
or
S
V
M
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2252
-
8938
I
nt
J
A
r
ti
f
I
nt
e
ll
,
V
ol
.
14
, N
o.
3
,
J
une
20
25
:
2158
-
2168
2166
A
C
K
N
O
WL
E
D
G
M
E
N
T
S
T
he
a
ut
hor
w
oul
d
li
ke
to
th
a
nk
T
r
i
E
nda
h
R
e
tn
o
K
us
um
a
ni
ngr
um
,
S
.S
os
.,
M
.I
.K
om
.,
f
or
he
r
va
lu
a
bl
e
a
s
s
is
ta
nc
e
dur
in
g he
r
vi
s
it
t
o T
id
a
r
U
ni
ve
r
s
it
y i
n M
a
ge
la
ng f
or
da
ta
c
ol
le
c
ti
on.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
T
he
r
e
s
e
a
r
c
h de
s
c
r
ib
e
d i
n t
hi
s
a
r
ti
c
le
w
a
s
f
unde
d by S
e
m
a
r
a
ng
U
ni
ve
r
s
it
y.
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
hi
s
jo
ur
na
l
us
e
s
th
e
C
ont
r
ib
ut
or
R
ol
e
s
T
a
xonomy
(
C
R
e
di
T
)
to
r
e
c
ogni
z
e
in
di
vi
dua
l
a
ut
hor
c
ont
r
ib
ut
io
ns
, r
e
duc
e
a
ut
hor
s
hi
p di
s
put
e
s
,
a
nd f
a
c
il
it
a
te
c
ol
la
bo
r
a
ti
on.
N
am
e
o
f
A
u
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
S
r
i
H
a
nda
ya
ni
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
R
iz
a
l
I
s
na
nt
o
✓
✓
✓
✓
✓
✓
B
udi
W
a
r
s
it
o
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
l
i
z
a
t
i
on
M
:
M
e
t
hodol
ogy
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
da
t
i
on
Fo
:
Fo
r
m
a
l
a
na
l
ys
i
s
I
:
I
nve
s
t
i
ga
t
i
on
R
:
R
e
s
our
c
e
s
D
:
D
a
t
a
C
ur
a
t
i
on
O
:
W
r
i
t
i
ng
-
O
r
i
gi
na
l
D
r
a
f
t
E
:
W
r
i
t
i
ng
-
R
e
vi
e
w
&
E
di
t
i
ng
Vi
:
Vi
s
ua
l
i
z
a
t
i
on
Su
:
Su
pe
r
vi
s
i
on
P
:
P
r
oj
e
c
t
a
dm
i
ni
s
t
r
a
t
i
on
Fu
:
Fu
ndi
ng a
c
qui
s
i
t
i
on
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
T
he
a
ut
hor
s
d
e
c
la
r
e
th
a
t
th
e
y
ha
ve
no
known
f
in
a
nc
ia
l
c
onf
li
c
t
s
of
in
te
r
e
s
t
or
pe
r
s
ona
l
r
e
la
ti
ons
hi
ps
th
a
t
c
oul
d ha
ve
i
nf
lu
e
nc
e
d t
he
w
or
k r
e
por
te
d i
n t
hi
s
pa
pe
r
.
I
N
F
O
R
M
E
D
C
O
N
S
E
N
T
W
e
ha
ve
obt
a
in
e
d i
nf
or
m
e
d c
ons
e
nt
f
r
om
a
ll
i
ndi
vi
dua
ls
i
nc
lu
de
d i
n t
hi
s
s
tu
dy.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
T
he
da
t
a
s
uppor
ti
ng
th
e
f
in
di
ngs
of
th
is
s
tu
dy
a
r
e
a
va
il
a
bl
e
f
r
om
U
ni
ve
r
s
it
a
s
S
e
m
a
r
a
ng.
A
c
c
e
s
s
to
th
e
s
e
da
t
a
is
r
e
s
tr
ic
te
d
a
nd
c
a
n
onl
y
be
obt
a
in
e
d
w
it
h
pe
r
m
is
s
i
on
f
r
om
U
ni
ve
r
s
it
a
s
S
e
m
a
r
a
ng.
D
a
ta
r
e
que
s
ts
c
a
n be
m
a
de
t
o t
he
a
ut
hor
a
t
s
r
i@us
m
.a
c
.i
d, w
it
h pe
r
m
is
s
io
n f
r
om
U
ni
ve
r
s
it
a
s
S
e
m
a
r
a
ng.
R
E
F
E
R
E
N
C
E
S
[
1]
W
. L
i
u, X
.
Z
he
ng, M
.
H
u, a
nd
C
.
C
he
n,
“
C
ol
l
a
bor
a
t
i
ve
f
i
l
t
e
r
i
ng w
i
t
h a
t
t
r
i
but
i
on a
l
i
gnm
e
nt
f
or
r
e
vi
e
w
-
ba
s
e
d non
-
ove
r
l
a
ppe
d
c
r
os
s
dom
a
i
n
r
e
c
om
m
e
nda
t
i
on,”
W
W
W
2022
-
P
r
oc
e
e
di
ngs
of
t
he
A
C
M
W
e
b
C
onf
e
r
e
nc
e
2022
,
pp.
1181
–
1190,
2022,
doi
:
10.1145/
3485447.3512166.
[
2]
W
.
L
i
u,
X
.
Z
he
ng,
J
.
S
u,
M
.
H
u,
Y
.
T
a
n,
a
nd
C
.
C
he
n,
“
E
xpl
oi
t
i
ng
va
r
i
a
t
i
ona
l
dom
a
i
n
-
i
nva
r
i
a
nt
us
e
r
e
m
be
ddi
ng
f
or
pa
r
t
i
a
l
l
y
ove
r
l
a
ppe
d
c
r
os
s
dom
a
i
n
r
e
c
om
m
e
nda
t
i
on,”
SI
G
I
R
2022
-
P
r
oc
e
e
di
ngs
of
t
h
e
45t
h
I
nt
e
r
nat
i
onal
A
C
M
SI
G
I
R
C
onf
e
r
e
n
c
e
o
n
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
, pp. 312
–
321, 2022, doi
:
1
0.1145/
3477495.3531975.
[
3]
S
.
S
t
i
e
gl
i
t
z
,
M
.
M
i
r
ba
ba
i
e
,
B
.
R
os
s
,
a
nd
C
.
N
e
ube
r
ge
r
,
“
S
oc
i
a
l
m
e
di
a
a
na
l
yt
i
c
s
-
c
ha
l
l
e
nge
s
i
n
t
opi
c
di
s
c
ove
r
y,
da
t
a
c
ol
l
e
c
t
i
on,
a
n
d
da
t
a
pr
e
pa
r
a
t
i
on,”
I
nt
e
r
nat
i
onal
J
ou
r
nal
of
I
nf
or
m
at
i
on
M
ana
ge
m
e
nt
,
vol
.
39,
pp.
156
–
168,
2018,
doi
:
10.1016/
j
.i
j
i
nf
om
gt
.2017.12.002.
[
4]
J
. L
i
e
t
al
.
, “
U
ns
upe
r
vi
s
e
d be
l
i
e
f
r
e
pr
e
s
e
nt
a
t
i
on l
e
a
r
ni
ng w
i
t
h i
nf
or
m
a
t
i
on
-
t
he
or
e
t
i
c
va
r
i
a
t
i
ona
l
gr
a
ph a
ut
o
-
e
nc
ode
r
s
,”
SI
G
I
R
2022
-
P
r
oc
e
e
di
ngs
of
t
he
45t
h
I
nt
e
r
nat
i
onal
A
C
M
SI
G
I
R
C
onf
e
r
e
nc
e
on
R
e
s
e
a
r
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
,
pp. 1728
–
1738, 2022, doi
:
10.1145/
3477495.3532072.
[
5]
S
.
S
ha
i
k
h,
S
.
M
.
D
a
u
dp
ot
a
, A
. S
.
I
m
r
a
n,
a
nd
Z
.
K
a
s
t
r
a
t
i
, “
T
o
w
a
r
ds
i
m
pr
ove
d c
l
a
s
s
i
f
i
c
a
t
i
on a
c
c
u
r
a
c
y o
n
hi
g
hl
y
i
m
b
a
l
a
nc
e
d t
e
xt
da
t
a
s
e
t
us
i
n
g d
e
e
p
ne
ur
a
l
l
a
ngu
a
ge
m
ode
l
s
,”
A
pp
l
i
e
d
Sc
i
e
nc
e
s
, v
ol
.
11,
n
o.
2,
pp
.
1
–
2
0,
2
021
, d
oi
:
10.
33
90
/
a
p
p11
02
086
9.
[
6]
A
.
H
a
di
f
a
r
,
L
.
S
t
e
r
c
kx,
T
.
D
e
m
e
e
s
t
e
r
,
a
nd
C
.
D
e
ve
l
d
e
r
,
“
A
s
e
l
f
-
t
r
a
i
ni
ng
a
ppr
oa
c
h
f
or
s
hor
t
t
e
xt
c
l
u
s
t
e
r
i
ng,”
A
C
L
2019
-
4t
h
W
or
k
s
hop
on
R
e
pr
e
s
e
nt
at
i
on
L
e
ar
ni
ng
f
or
N
L
P
,
R
e
pL
4N
L
P
2019
-
P
r
oc
e
e
di
ngs
of
t
he
W
or
k
s
hop
,
pp.
194
–
199,
2019
,
doi
:
10.18653/
v1/
w
19
-
4322.
[
7]
D
.
Z
ha
ng
e
t
al
.
,
“
S
uppor
t
i
ng
c
l
u
s
t
e
r
i
ng
w
i
t
h
c
ont
r
a
s
t
i
ve
l
e
a
r
ni
ng,”
N
A
A
C
L
-
H
L
T
2021
-
2021
C
onf
e
r
e
nc
e
of
t
he
N
or
t
h
A
m
e
r
i
c
a
n
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
ngui
s
t
i
c
s
:
H
um
an
L
angua
ge
T
e
c
hnol
ogi
e
s
,
P
r
oc
e
e
di
ng
s
of
t
he
C
onf
e
r
e
nc
e
,
pp. 5419
–
5430, 2021, doi
:
10.18653/
v1/
2021.na
a
c
l
-
m
a
i
n.427.
[
8]
X
.
J
i
,
A
.
V
e
da
l
di
,
a
nd
J
.
H
e
nr
i
que
s
,
“
I
nva
r
i
a
nt
i
nf
or
m
a
t
i
on
c
l
us
t
e
r
i
ng
f
or
uns
upe
r
vi
s
e
d
i
m
a
ge
c
l
a
s
s
i
f
i
c
a
t
i
on
a
nd
s
e
gm
e
nt
a
t
i
on
,”
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
A
r
ti
f
I
nt
e
ll
I
S
S
N
:
2252
-
8938
Co
-
tr
ai
ni
ng ps
e
udo
-
la
be
li
ng f
or
t
e
x
t
c
la
s
s
if
ic
at
io
n w
it
h s
uppo
r
t
v
e
c
to
r
m
ac
hi
n
e
…
(
Sr
i
H
anday
ani
)
2167
2019
I
E
E
E
/
C
V
F
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
(
I
C
C
V
)
,
S
e
oul
,
K
or
e
a
(
S
out
h)
,
2019,
pp.
9864
-
9873,
doi
:
10.1109/
I
C
C
V
.2019.00996.
[
9]
T
.
E
. T
a
l
l
o
a
nd
A
.
M
us
dhol
i
f
a
h,
“
T
he
i
m
pl
e
m
e
nt
a
t
i
on
of
ge
ne
t
i
c
a
l
gor
i
t
h
m
i
n
s
m
ot
e
(
S
ynt
he
t
i
c
m
i
no
r
i
t
y
ove
r
s
a
m
pl
i
ng
t
e
c
hni
que
)
f
or
ha
ndl
i
ng
i
m
ba
l
a
nc
e
d
da
t
a
s
e
t
pr
obl
e
m
,”
2018
4t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
Sc
i
e
nc
e
and
T
e
c
hnol
ogy
(
I
C
ST
)
,
Y
ogya
ka
r
t
a
,
I
ndone
s
i
a
, 2018, pp. 1
-
4, doi
:
10.1109/
I
C
S
T
C
.2018.8528591.
[
10]
C
.
G
ong,
D
.
T
a
o,
S
.
J
.
M
a
yba
nk,
W
.
L
i
u,
G
.
K
a
ng,
a
nd
J
.
Y
a
ng,
“
M
ul
t
i
-
m
oda
l
c
ur
r
i
c
ul
um
l
e
a
r
ni
ng
f
or
s
e
m
i
-
s
upe
r
vi
s
e
d
i
m
a
ge
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 I
m
age
P
r
oc
e
s
s
i
ng
, vol
. 25, no. 7, pp. 324
9
–
3260, 2016, doi
:
10.1109/
T
I
P
.2016.2563981.
[
11]
J
.
E
ngue
ha
r
d,
P
.
O
’
H
a
l
l
or
a
n,
a
nd
A
.
G
hol
i
pour
,
“
S
e
m
i
-
s
upe
r
vi
s
e
d
l
e
a
r
ni
ng
w
i
t
h
de
e
p
e
m
be
dde
d
c
l
us
t
e
r
i
ng
f
or
i
m
a
g
e
c
l
a
s
s
i
f
i
c
a
t
i
on a
nd s
e
gm
e
nt
a
t
i
on,”
I
E
E
E
A
c
c
e
s
s
, vol
. 7, pp. 11093
–
11104, 2019, doi
:
10.1109/
A
C
C
E
S
S
.2019.2891970.
[
12]
M
.
K
.
X
i
e
,
J
.
H
.
X
i
a
o,
H
.
Z
.
L
i
u,
G
.
N
i
u,
M
.
S
ug
i
ya
m
a
,
a
nd
S
.
J
.
H
ua
ng,
“
C
l
a
s
s
-
di
s
t
r
i
but
i
on
-
a
w
a
r
e
ps
e
udo
-
l
a
be
l
i
ng
f
or
s
e
m
i
-
s
upe
r
vi
s
e
d m
ul
t
i
-
l
a
be
l
l
e
a
r
ni
ng,”
A
dv
anc
e
s
i
n N
e
u
r
al
I
nf
or
m
at
i
on P
r
oc
e
s
s
i
ng S
y
s
t
e
m
s
, vol
. 36, 2023.
[
13]
Z
.
H
u,
Z
.
Y
a
ng,
X
.
H
u,
a
nd
R
.
N
e
va
t
i
a
,
“
S
i
m
pl
e
:
s
i
m
i
l
a
r
ps
e
udo
l
a
be
l
e
xpl
oi
t
a
t
i
on
f
or
s
e
m
i
-
s
upe
r
vi
s
e
d
c
l
a
s
s
i
f
i
c
a
t
i
on,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
n
c
e
on
C
o
m
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
pp.
15094
–
15103,
2021,
doi
:
10.1109/
C
V
P
R
46437.2021.01485.
[
14]
S
.
P
a
ni
gr
a
hi
a
nd
H
.
S
.
B
e
he
r
a
,
“
A
s
t
udy
on
l
e
a
di
ng
m
a
c
hi
ne
l
e
a
r
ni
ng
t
e
c
hni
que
s
f
or
hi
gh
or
de
r
f
uz
z
y
t
i
m
e
s
e
r
i
e
s
f
or
e
c
a
s
t
i
ng,”
E
ngi
ne
e
r
i
ng A
ppl
i
c
at
i
ons
of
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
, vol
. 87, 2020, doi
:
10.1016/
j
.e
nga
ppa
i
.2019.103245.
[
15]
S
.
E
m
t
i
ya
z
a
nd
M
.
R
.
K
e
yv
a
npour
,
“
C
us
t
om
e
r
s
be
ha
vi
or
m
ode
l
i
ng
by
s
e
m
i
-
s
upe
r
vi
s
e
d
l
e
a
r
ni
ng
i
n
c
us
t
om
e
r
r
e
l
a
t
i
ons
hi
p
m
a
na
ge
m
e
nt
,”
A
dv
anc
e
s
i
n
I
nf
or
m
at
i
on
Sc
i
e
nc
e
s
and
Se
r
v
i
c
e
Sc
i
e
nc
e
s
,
vol
.
3,
no.
9,
pp.
229
–
236,
2011,
doi
:
10.4156/
A
I
S
S
.vol
3.i
s
s
ue
9.31.
[
16]
Z
. A
z
i
z
a
h, T
. O
hya
m
a
,
X
. Z
ha
o, Y
.
O
hka
w
a
,
a
nd T
.
M
i
t
s
ui
s
hi
,
“
P
r
e
di
c
t
i
ng a
t
-
r
i
s
k s
t
ude
nt
s
i
n t
he
e
a
r
l
y
s
t
a
ge
of
a
bl
e
nde
d
l
e
a
r
ni
ng
c
our
s
e
vi
a
m
a
c
hi
ne
l
e
a
r
ni
ng
us
i
ng
l
i
m
i
t
e
d
da
t
a
,”
C
om
put
e
r
s
and
E
d
uc
at
i
on:
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
,
vol
.
7,
2024,
doi
:
10.1016/
j
.c
a
e
a
i
.2024.100261.
[
17]
G
.
A
l
-
T
a
m
e
e
m
i
,
J
.
X
ue
,
I
.
H
.
A
l
i
,
a
nd
S
.
A
j
i
t
,
“
A
hybr
i
d
m
a
c
hi
n
e
l
e
a
r
ni
ng
a
ppr
oa
c
h
f
or
pr
e
di
c
t
i
ng
s
t
ude
nt
p
e
r
f
or
m
a
nc
e
us
i
ng
m
ul
t
i
-
c
l
a
s
s
e
duc
a
t
i
ona
l
da
t
a
s
e
t
s
,”
P
r
o
c
e
di
a C
om
put
e
r
Sc
i
e
nc
e
, vol
. 238, pp. 88
8
–
895, 2024, doi
:
10.1016/
j
.pr
oc
s
.2024.06.108.
[
18]
A
.
M
.
R
a
be
l
o
a
nd
L
.
E
.
Z
á
r
a
t
e
,
“
A
m
od
e
l
f
or
pr
e
di
c
t
i
ng
dr
opout
of
hi
ghe
r
e
duc
a
t
i
on
s
t
ude
nt
s
,
”
D
at
a
Sc
i
e
n
c
e
and
M
anage
m
e
nt
,
2024, doi
:
10.1016/
j
.ds
m
.2024.07.001.
[
19]
V
.
C
hr
i
s
t
ou
e
t
al
.
,
“
P
e
r
f
or
m
a
nc
e
a
nd
e
a
r
l
y
dr
op
pr
e
di
c
t
i
on f
or
hi
ghe
r
e
duc
a
t
i
on
s
t
ude
nt
s
us
i
ng
m
a
c
hi
ne
l
e
a
r
ni
ng,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
i
t
h A
ppl
i
c
at
i
ons
, vol
. 225, 2023, doi
:
10.1016/
j
.e
s
w
a
.2023.120079.
[
20]
I
. N
a
s
s
a
r
, S
. H
e
r
a
t
h, E
. A
bba
s
ne
j
a
d,
W
. B
unt
i
ne
,
a
nd G
.
H
a
f
f
a
r
i
, “
A
l
l
l
a
be
l
s
a
r
e
not
c
r
e
a
t
e
d e
qua
l
:
e
nha
nc
i
ng
s
e
m
i
-
s
upe
r
vi
s
i
on vi
a
l
a
be
l
gr
oupi
ng
a
nd
c
o
-
t
r
a
i
ni
ng,”
P
r
oc
e
e
di
ngs
of
t
he
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
ogni
t
i
on
, pp. 7237
–
7246, 2021, doi
:
10.1109/
C
V
P
R
46437.2021.00716.
[
21]
X
.
X
u,
W
.
L
i
,
D
.
X
u,
a
nd
I
.
W
.
T
s
a
ng,
“
C
o
-
l
a
be
l
i
ng
f
or
m
ul
t
i
-
vi
e
w
w
e
a
kl
y
l
a
be
l
e
d
l
e
a
r
ni
ng,”
I
E
E
E
T
r
ans
ac
t
i
ons
on
P
at
t
e
r
n
A
nal
y
s
i
s
and M
ac
hi
ne
I
nt
e
l
l
i
ge
nc
e
, vol
. 38, no. 6, pp. 1113
–
1125, 2016, doi
:
10.1109/
T
P
A
M
I
.2015.2476813.
[
22]
W
.
Z
ha
n
a
nd
M
.
L
.
Z
ha
ng,
“
I
nduc
t
i
ve
s
e
m
i
-
s
upe
r
vi
s
e
d
m
ul
t
i
-
l
a
be
l
l
e
a
r
ni
ng
w
i
t
h
c
o
-
t
r
a
i
ni
ng,”
P
r
oc
e
e
di
ngs
of
t
he
A
C
M
SI
G
K
D
D
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
K
now
l
e
dg
e
D
i
s
c
ov
e
r
y
and
D
at
a
M
i
ni
ng
,
vol
.
P
a
r
t
F
129685,
pp.
1305
–
1314,
2017,
doi
:
10.1145/
3097983.3098141.
[
23]
Z
.
H
.
Z
hou,
“
A
br
i
e
f
i
nt
r
oduc
t
i
on
t
o
w
e
a
kl
y
s
upe
r
vi
s
e
d
l
e
a
r
ni
ng,”
N
at
i
onal
Sc
i
e
nc
e
R
e
v
i
e
w
,
vol
.
5,
no.
1,
pp.
44
–
53,
2018
,
doi
:
10.1093/
ns
r
/
nw
x106.
[
24]
F
.
Z
hua
ng
e
t
al
.
,
“
A
c
om
pr
e
he
ns
i
ve
s
ur
ve
y
on
t
r
a
ns
f
e
r
l
e
a
r
ni
ng,”
P
r
oc
e
e
di
n
gs
of
t
he
I
E
E
E
,
vol
.
109,
no.
1,
pp.
43
–
76,
2021,
doi
:
10.1109/
J
P
R
O
C
.2020.3004555.
[
25]
V
.
P
i
c
c
i
a
l
l
i
a
nd
M
.
S
c
i
a
ndr
one
,
“
N
onl
i
ne
a
r
opt
i
m
i
z
a
t
i
on
a
nd
s
uppor
t
ve
c
t
or
m
a
c
hi
ne
s
,”
A
nnal
s
of
O
pe
r
at
i
ons
R
e
s
e
a
r
c
h
,
vol
.
314,
no. 1, pp. 15
–
47, 2022, doi
:
10.1007/
s
10479
-
022
-
04655
-
x.
[
26]
J
.
C
.
R
ot
he
nbe
r
ge
r
a
nd
D
.
I
.
D
i
oc
hnos
,
“
M
e
t
a
c
o
-
t
r
a
i
ni
ng:
t
w
o
vi
e
w
s
a
r
e
be
t
t
e
r
t
ha
n
one
,”
ar
X
i
v
-
C
om
put
e
r
Sc
i
e
nc
e
,
pp.
1
-
17,
2023, doi
:
10.1145/
3474085.3475622.
[
27]
C
.
T
a
n,
J
.
X
i
a
,
L
.
W
u,
a
nd
S
.
Z
.
L
i
,
“
C
o
-
l
e
a
r
ni
ng:
l
e
a
r
ni
ng
f
r
om
noi
s
y
l
a
be
l
s
w
i
t
h
s
e
l
f
-
s
upe
r
vi
s
i
on,”
i
n
M
M
2021
-
P
r
oc
e
e
di
ngs
o
f
t
he
29t
h A
C
M
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on M
ul
t
i
m
e
di
a
, pp. 1405
–
1413,
2021,
doi
:
10.1145/
3474085.3475622.
[
28]
D
.
T
.
N
guye
n,
C
.
K
.
M
um
m
a
di
,
T
.
P
.
N
.
N
go,
T
.
H
. P
.
N
guye
n,
L
.
B
e
gge
l
,
a
n
d
T
.
B
r
ox,
“
S
e
l
f
:
l
e
a
r
ni
ng
t
o
f
i
l
t
e
r
noi
s
y
l
a
be
l
s
w
i
t
h
s
e
l
f
-
e
ns
e
m
bl
i
ng,”
8t
h I
nt
e
r
nat
i
onal
C
onf
e
r
e
n
c
e
on L
e
ar
ni
ng R
e
pr
e
s
e
nt
at
i
ons
, I
C
L
R
2020
, 2020.
[
29]
M
.
G
oudj
i
l
,
M
.
K
oudi
l
,
M
.
B
e
dda
,
a
nd
N
.
G
hogga
l
i
,
“
A
nove
l
a
c
t
i
ve
l
e
a
r
ni
ng
m
e
t
hod
us
i
ng
s
vm
f
or
t
e
xt
c
l
a
s
s
i
f
i
c
a
t
i
on,”
I
nt
e
r
nat
i
onal
J
our
nal
of
A
ut
om
at
i
on and C
om
put
i
ng
, vol
. 15, no. 3, pp. 290
–
298, 2018, doi
:
10.1007/
s
11633
-
015
-
0912
-
z.
[
30]
M
.
H
.
S
.
E
na
s
,
H
.
A
.
S
a
l
e
h,
a
nd
E
.
A
.
K
ha
l
e
l
,
“
C
l
a
s
s
i
f
i
c
a
t
i
on
of
m
a
m
m
ogr
a
m
s
ba
s
e
d
on
f
e
a
t
ur
e
s
e
xt
r
a
c
t
i
on
t
e
c
hni
que
s
us
i
n
g
s
uppor
t
ve
c
t
or
m
a
c
hi
ne
,”
C
om
put
e
r
S
c
i
e
nc
e
and
I
nf
or
m
at
i
on
T
e
c
hno
l
ogi
e
s
,
vol
.
2,
no.
3,
pp.
121
–
131,
2021,
doi
:
10.11591/
c
s
i
t
.v2i
3.pp121
-
131.
[
31]
E
.
I
s
m
a
n
t
o
a
n
d N
. E
f
f
e
nd
i
, “
A
n
L
S
T
M
-
ba
s
e
d
p
r
e
d
i
c
t
i
o
n m
o
de
l
f
o
r
g
r
a
d
i
e
nt
-
de
s
c
e
n
d
i
ng o
pt
i
m
i
z
a
t
i
o
n i
n vi
r
t
ua
l
l
e
a
r
n
i
n
g e
n
v
i
r
o
n
m
e
nt
s
,”
C
om
p
ut
e
r
Sc
i
e
nc
e
a
nd
I
n
f
or
m
a
t
i
o
n
T
e
c
h
n
ol
og
i
e
s
,
vo
l
.
4,
n
o
.
3,
pp
.
1
99
–
2
0
7,
2
0
24
, d
o
i
:
10
.
11
59
1
/
c
s
i
t
.
v4
i
3.
p
p1
99
-
2
07
.
[
32]
M
.
M
.
B
e
n
I
s
m
a
i
l
,
“
I
n
s
ul
t
de
t
e
c
t
i
on
us
i
ng
a
pa
r
t
i
t
i
ona
l
C
N
N
-
L
S
T
M
m
od
e
l
,”
I
nt
e
r
nat
i
onal
J
our
nal
of
D
at
a
A
nal
y
s
i
s
T
e
c
hni
que
s
and St
r
at
e
gi
e
s
, vol
. 14, no. 4, pp. 336
–
349, 2023, doi
:
10.1504/
I
J
D
A
T
S
.2022.1
29175.
[
33]
Z
.
S
he
n,
P
.
C
a
o,
H
.
Y
a
ng,
X
.
L
i
u,
J
.
Y
a
ng,
a
nd
O
.
R
.
Z
a
i
a
ne
,
“
C
o
-
t
r
a
i
ni
ng
w
i
t
h
hi
gh
-
c
onf
i
de
nc
e
ps
e
udo
l
a
be
l
s
f
or
s
e
m
i
-
s
upe
r
vi
s
e
d
m
e
di
c
a
l
i
m
a
ge
s
e
gm
e
nt
a
t
i
on,”
I
J
C
A
I
I
nt
e
r
nat
i
onal
J
oi
nt
C
o
nf
e
r
e
nc
e
on
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
nc
e
,
vol
.
2023,
pp. 4199
–
4207, 2023, doi
:
10.24963/
i
j
c
a
i
.2023/
467.
[
34]
M
.
G
a
o,
Z
.
Z
ha
ng,
G
.
Y
u,
S
.
A
r
ı
k,
L
.
S
.
D
a
vi
s
,
a
nd
T
.
P
f
i
s
t
e
r
,
“
C
ons
i
s
t
e
nc
y
-
ba
s
e
d
s
e
m
i
-
s
upe
r
vi
s
e
d
a
c
t
i
ve
l
e
a
r
ni
ng:
t
ow
a
r
ds
m
i
ni
m
i
z
i
ng l
a
be
l
i
ng c
os
t
,”
C
om
put
e
r
V
i
s
i
on
–
E
C
C
V
2020
, pp. 510
–
526, 2020,
doi
:
10.1007/
978
-
3
-
030
-
58607
-
2_30.
[
35]
W
.
C
he
n,
H
.
R
.
P
our
gha
s
e
m
i
,
a
nd
S
.
A
.
N
a
ghi
bi
,
“
A
c
om
pa
r
a
t
i
ve
s
t
udy
of
l
a
nds
l
i
de
s
us
c
e
pt
i
bi
l
i
t
y
m
a
ps
pr
oduc
e
d
us
i
ng
s
uppor
t
ve
c
t
or
m
a
c
hi
ne
w
i
t
h
di
f
f
e
r
e
nt
ke
r
ne
l
f
unc
t
i
ons
a
nd
e
nt
r
opy
da
t
a
m
i
ni
ng
m
ode
l
s
i
n
C
hi
na
,”
B
ul
l
e
t
i
n
of
E
ngi
ne
e
r
i
ng
G
e
ol
ogy
an
d
t
he
E
nv
i
r
onm
e
nt
, vol
. 77, no. 2, pp. 647
–
664, 2018, doi
:
10.1007/
s
10064
-
017
-
1010
-
y.
[
36]
G
.
G
i
ge
r
e
nz
e
r
,
U
.
H
of
f
r
a
ge
,
a
nd
H
.
K
l
e
i
nbol
t
i
ng,
“
P
r
oba
bi
l
i
s
t
i
c
m
e
nt
a
l
m
ode
l
s
:
a
br
uns
w
i
ki
a
n
t
he
or
y
of
c
onf
i
de
nc
e
,
”
P
s
y
c
hol
ogi
c
al
R
e
v
i
e
w
, vol
. 98, no. 4, pp. 506
–
528, 1991, doi
:
10.1037/
0033
-
295X
.98.4.506.
[
37]
J
.
M
.
G
r
i
f
f
i
t
hs
a
nd
D
.
W
.
K
i
ng,
“
U
S
i
n
f
or
m
a
t
i
on
r
e
t
r
i
e
va
l
s
ys
t
e
m
e
vol
ut
i
on
a
nd
e
va
l
ua
t
i
on
(
1945
-
1975)
,”
I
E
E
E
A
nnal
s
of
t
he
H
i
s
t
or
y
of
C
om
put
i
ng
, vol
. 24, no. 3, pp. 35
–
55, 2002, doi
:
10.1109/
M
A
H
C
.20
02.1024761.
[
38]
K
.
N
i
ga
m
a
nd
R
.
G
ha
ni
,
“
A
na
l
yz
i
ng
t
he
e
f
f
e
c
t
i
ve
ne
s
s
a
nd
a
ppl
i
c
a
bi
l
i
t
y
of
c
o
-
t
r
a
i
ni
ng,”
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
I
nf
or
m
at
i
o
n
and K
now
l
e
dge
M
anage
m
e
nt
, P
r
oc
e
e
di
ngs
, vol
. 2000
-
J
a
nua
, pp. 86
–
93, 2000, doi
:
10.1145/
354756.354805.
[
39]
J
.
C
he
n,
J
.
F
e
ng,
X
.
S
un,
a
nd
Y
.
L
i
u,
“
C
o
-
t
r
a
i
ni
ng
s
e
m
i
-
s
upe
r
vi
s
e
d
d
e
e
p
l
e
a
r
ni
ng
f
or
s
e
nt
i
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
of
M
O
O
C
f
or
um
pos
t
s
,”
Sy
m
m
e
t
r
y
, vol
. 12, no. 1, 2020, doi
:
10.3390/
S
Y
M
12010008.
Evaluation Warning : The document was created with Spire.PDF for Python.