I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
,
p
p
.
443
~
453
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
4
4
3
-
4
5
3
443
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
O
ptimi
zing
Ja
v
a
nese script
rec
o
g
ni
tion usin
g
f
in
e
-
t
u
ned
ResNet
-
1
8
and tr
a
nsfer learning
Nur
F
a
t
ea
h
1
,
Su
bh
a
n
2
,
Ya
hy
a
Nur
I
f
riza
2
1
Jav
a
n
e
se
La
n
g
u
a
g
e
a
n
d
L
i
t
e
r
a
t
u
r
e
E
d
u
c
a
t
i
o
n
S
t
u
d
y
P
r
o
g
r
a
m
,
F
a
c
u
l
t
y
o
f
L
a
n
g
u
a
g
e
a
n
d
A
r
t
s,
U
n
i
v
e
r
si
t
a
s N
e
g
e
r
i
S
e
mara
n
g
,
S
e
mara
n
g
,
I
n
d
o
n
e
si
a
2
I
n
f
o
r
mat
i
o
n
S
y
st
e
ms
S
t
u
d
y
P
r
o
g
r
a
m
,
F
a
c
u
l
t
y
o
f
M
a
t
h
e
m
a
t
i
c
s
a
n
d
N
a
t
u
r
a
l
S
c
i
e
n
c
e
s,
U
n
i
v
e
r
s
i
t
a
s Ne
g
e
r
i
S
e
m
a
r
a
n
g
,
S
e
mara
n
g
,
I
n
d
o
n
e
s
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Mar
7
,
2
0
2
5
R
ev
is
ed
Oct
20
,
2
0
2
5
Acc
ep
ted
J
an
10
,
2
0
2
6
Ja
v
a
n
e
se
sc
rip
t,
k
n
o
wn
a
s
a
k
sa
ra
J
a
wa
,
is
a
n
a
n
c
ien
t
sc
ri
p
t
u
se
d
in
h
isto
rica
l
a
n
d
c
u
lt
u
ra
l
tex
ts.
Ho
we
v
e
r,
it
s
c
o
m
p
lex
c
h
a
ra
c
ter
stru
c
tu
re
p
o
se
s c
h
a
ll
e
n
g
e
s
fo
r
a
c
c
u
ra
te
re
c
o
g
n
it
i
o
n
in
m
o
d
e
rn
d
i
g
it
a
l
a
p
p
li
c
a
ti
o
n
s.
T
h
is
stu
d
y
p
r
o
p
o
se
s
a
n
o
p
ti
m
ize
d
c
las
sifica
ti
o
n
a
p
p
r
o
a
c
h
fo
r
a
k
sa
ra
J
a
wa
u
si
n
g
a
fin
e
-
tu
n
e
d
re
sid
u
a
l
n
e
two
r
k
(Re
sN
e
t)
-
1
8
m
o
d
e
l
c
o
m
b
in
e
d
with
th
e
a
d
a
p
ti
v
e
m
o
m
e
n
t
e
stim
a
ti
o
n
(Ad
a
m
)
o
p
ti
m
iza
ti
o
n
a
lg
o
rit
h
m
a
n
d
tran
sfe
r
lea
rn
i
n
g
o
n
t
h
e
h
a
n
a
c
a
ra
k
a
ima
g
e
d
a
tas
e
t.
By
lev
e
ra
g
in
g
t
h
e
re
sid
u
a
l
lea
rn
i
n
g
fra
m
e
wo
rk
o
f
Re
sN
e
t
-
1
8
,
th
e
m
o
d
e
l
e
ffe
c
ti
v
e
ly
c
a
p
tu
re
s
d
e
e
p
sp
a
ti
a
l
fe
a
tu
re
s
o
f
th
e
sc
rip
t
wh
il
e
re
d
u
c
in
g
v
a
n
is
h
in
g
g
ra
d
ie
n
t
issu
e
s.
F
in
e
-
tu
n
in
g
is
a
p
p
li
e
d
t
o
e
n
h
a
n
c
e
m
o
d
e
l
a
d
a
p
tab
il
it
y
,
e
n
su
ri
n
g
ro
b
u
st
fe
a
tu
re
e
x
trac
ti
o
n
sp
e
c
ifi
c
t
o
Ja
v
a
n
e
se
c
h
a
ra
c
ters
.
Ex
p
e
rime
n
tal
re
su
lt
s
d
e
m
o
n
stra
te
th
a
t
th
e
fi
n
e
-
tu
n
e
d
Re
sN
e
t
-
1
8
o
u
t
p
e
rfo
rm
s
c
o
n
v
e
n
ti
o
n
a
l
d
e
e
p
l
e
a
rn
in
g
a
rc
h
it
e
c
tu
re
s
i
n
re
c
o
g
n
izin
g
a
k
sa
r
a
J
a
wa
c
h
a
ra
c
ters
,
a
c
h
iev
in
g
9
3
%
p
re
c
isio
n
,
9
1
%
re
c
a
ll
,
9
1
%
F1
-
sc
o
re
,
a
n
d
9
1
%
a
c
c
u
ra
c
y
.
Th
e
st
u
d
y
fu
rt
h
e
r
e
x
p
l
o
re
s
th
e
imp
a
c
t
o
f
h
y
p
e
r
p
a
ra
m
e
ter
tu
n
i
n
g
,
d
a
ta
a
u
g
m
e
n
tatio
n
,
a
n
d
d
r
o
p
o
u
t
re
g
u
lariz
a
ti
o
n
o
n
m
o
d
e
l
p
e
rfo
rm
a
n
c
e
.
Th
e
fin
d
i
n
g
s
h
ig
h
li
g
h
t
t
h
e
e
ffe
c
ti
v
e
n
e
ss
o
f
tran
sfe
r
l
e
a
rn
in
g
i
n
re
so
u
rc
e
-
li
m
it
e
d
sc
e
n
a
rio
s,
m
a
k
i
n
g
it
a
fe
a
sib
le
so
l
u
ti
o
n
f
o
r
o
p
t
ica
l
c
h
a
ra
c
ter
re
c
o
g
n
it
i
o
n
(OCR)
a
p
p
li
c
a
ti
o
n
s
i
n
Ja
v
a
n
e
se
sc
rip
t
d
i
g
it
iza
ti
o
n
.
Th
i
s
re
se
a
rc
h
c
o
n
tri
b
u
tes
t
o
t
h
e
p
re
se
rv
a
ti
o
n
o
f
c
u
lt
u
ra
l
h
e
rit
a
g
e
th
r
o
u
g
h
a
d
v
a
n
c
e
m
e
n
ts
in
d
e
e
p
lea
rn
in
g
-
b
a
se
d
sc
ri
p
t
re
c
o
g
n
it
io
n
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
J
av
an
ese
s
cr
ip
t
Op
tical
ch
ar
ac
ter
Op
tim
izatio
n
R
esNet
-
18
T
r
an
s
f
er
lear
n
i
n
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Yah
y
a
Nu
r
I
f
r
iza
I
n
f
o
r
m
atio
n
Sy
s
tem
s
Stu
d
y
Pr
o
g
r
am
,
Facu
lty
o
f
Ma
th
e
m
atics a
n
d
Natu
r
al
Scien
ce
s
Un
iv
er
s
itas
Neg
er
i Sem
ar
an
g
Sem
ar
an
g
,
I
n
d
o
n
esia
E
m
ail:
y
ah
y
an
u
r
if
r
iza@ma
il.u
n
n
es.a
c.
id
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
J
av
an
ese
s
cr
ip
t,
also
k
n
o
wn
as
a
ksa
r
a
Ja
w
a
o
r
h
an
ac
a
r
ak
a
,
is
a
cu
lt
u
r
al
h
er
itag
e
o
f
s
ig
n
if
ican
t
h
is
to
r
ical
an
d
id
e
n
tity
v
alu
e
in
I
n
d
o
n
esia,
p
ar
ticu
la
r
ly
o
n
th
e
I
s
lan
d
o
f
J
av
a
[
1
]
,
[
2
]
.
Fo
r
ce
n
tu
r
ies,
it
h
as
b
ee
n
u
s
ed
in
liter
atu
r
e,
h
is
to
r
ical
r
ec
o
r
d
s
,
an
d
o
f
f
icial
d
o
cu
m
en
ts
[
3
]
,
[
4
]
.
Ho
wev
er
,
with
th
e
ad
v
an
ce
m
e
n
t
o
f
m
o
d
er
n
tec
h
n
o
lo
g
y
,
th
e
u
s
e
an
d
in
ter
est
in
lear
n
in
g
th
e
J
av
an
ese
s
cr
ip
t
h
av
e
s
ig
n
if
ica
n
tly
d
ec
lin
ed
.
T
h
is
s
itu
atio
n
u
n
d
er
s
co
r
es
th
e
u
r
g
en
cy
o
f
p
r
eser
v
in
g
a
n
d
e
n
h
a
n
cin
g
t
h
e
r
ec
o
g
n
itio
n
o
f
t
h
e
J
av
an
ese
s
cr
ip
t
t
o
r
em
ain
r
ele
v
an
t
i
n
th
e
d
i
g
ital
er
a.
T
r
a
d
itio
n
al
a
p
p
r
o
ac
h
es,
s
u
ch
as
o
p
tical
c
h
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
,
h
a
v
e
b
ee
n
u
s
ed
to
r
ec
o
g
n
ize
J
av
a
n
ese
s
cr
ip
t
ch
ar
ac
ter
s
.
Ho
we
v
er
,
th
ese
m
eth
o
d
s
f
ac
e
lim
i
tatio
n
s
in
h
an
d
lin
g
v
ar
iatio
n
s
in
h
an
d
wr
itin
g
s
ty
l
es,
d
is
to
r
tio
n
s
,
an
d
s
u
b
o
p
tim
al
im
ag
e
q
u
ality
[
5
]
.
W
ith
th
e
r
ap
id
p
r
o
g
r
ess
in
d
ee
p
lear
n
in
g
tech
n
o
lo
g
y
,
d
e
ep
lear
n
in
g
-
b
ased
ap
p
r
o
ac
h
es
h
av
e
s
h
o
wn
p
r
o
m
is
in
g
p
o
te
n
tial
in
im
p
r
o
v
i
n
g
ch
ar
ac
ter
r
ec
o
g
n
itio
n
ac
cu
r
ac
y
[
6
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
443
-
4
5
3
444
T
r
an
s
f
er
lear
n
in
g
h
as
also
e
m
er
g
ed
as
an
e
f
f
ec
tiv
e
a
p
p
r
o
ac
h
to
e
n
h
an
ce
th
e
p
er
f
o
r
m
a
n
ce
o
f
d
ee
p
lear
n
in
g
m
o
d
els,
p
ar
ticu
lar
l
y
wh
en
th
e
av
ailab
le
tr
ain
in
g
d
ata
is
lim
ited
[
7
]
.
B
y
lev
er
ag
in
g
p
r
e
-
tr
ain
ed
m
o
d
els,
tr
an
s
f
er
lea
r
n
in
g
allo
ws
th
e
k
n
o
wled
g
e
g
ain
e
d
to
a
cc
eler
ate
tr
ain
in
g
a
n
d
im
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
n
ew
m
o
d
els
o
n
s
p
ec
if
ic
task
s
[
8
]
.
Alth
o
u
g
h
r
esear
ch
h
as
b
ee
n
co
n
d
u
cted
o
n
J
av
an
ese
s
cr
ip
t
ch
ar
ac
ter
r
ec
o
g
n
itio
n
,
cu
r
r
e
n
t
ap
p
r
o
ac
h
es
s
till
f
ac
e
s
ev
er
al
lim
itatio
n
s
,
s
u
ch
as
in
s
u
f
f
icien
t
ac
cu
r
ac
y
u
n
d
e
r
v
ar
y
in
g
im
ag
e
co
n
d
itio
n
s
an
d
th
e
n
ee
d
f
o
r
lar
g
e
tr
ain
in
g
d
atasets
[
9
]
.
T
h
er
e
f
o
r
e,
t
h
er
e
is
a
n
ee
d
to
d
ev
elo
p
m
o
r
e
ef
f
ec
tiv
e
an
d
ef
f
icien
t
class
if
icatio
n
m
o
d
els
to
a
d
d
r
ess
th
ese
lim
itatio
n
s
.
E
x
is
tin
g
r
esear
c
h
o
n
a
ksa
r
a
Ja
w
a
r
ec
o
g
n
itio
n
is
s
till
in
it
s
ea
r
ly
s
tag
es,
with
a
lim
ited
n
u
m
b
er
o
f
s
tu
d
ies
f
o
cu
s
in
g
o
n
th
is
s
p
e
cif
ic
s
cr
ip
t.
W
h
ile
s
o
m
e
p
r
o
m
is
in
g
r
esu
lts
h
av
e
b
ee
n
ac
h
iev
ed
u
s
in
g
tech
n
i
q
u
es
lik
e
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NN)
,
s
ev
er
al
k
ey
g
a
p
s
r
em
ain
[
1
0
]
.
Fig
u
r
e
1
s
h
o
ws h
a
n
ac
ar
ak
a.
Fig
u
r
e
1
.
Han
ac
a
r
ak
a
J
av
an
es
e
s
cr
ip
t
T
h
ese
in
clu
d
e
th
e
n
ee
d
f
o
r
lar
g
er
an
d
m
o
r
e
d
iv
er
s
e
d
atasets
,
th
e
d
ev
elo
p
m
en
t
o
f
r
o
b
u
s
t
p
r
e
-
p
r
o
ce
s
s
in
g
tech
n
iq
u
es
to
h
an
d
le
v
ar
iatio
n
s
in
h
an
d
wr
iti
n
g
s
ty
les,
an
d
th
e
ex
p
lo
r
atio
n
o
f
ad
v
an
ce
d
d
ee
p
lear
n
in
g
ar
ch
itectu
r
es
tailo
r
ed
f
o
r
co
m
p
lex
s
cr
ip
ts
lik
e
a
ksa
r
a
Ja
w
a
[
1
1
]
,
[
1
2
]
.
Ad
d
iti
o
n
ally
,
r
esear
ch
o
n
r
ea
l
-
tim
e
r
ec
o
g
n
itio
n
s
y
s
tem
s
an
d
in
teg
r
atio
n
with
o
th
er
la
n
g
u
ag
e
tech
n
o
lo
g
ies,
s
u
c
h
as
m
ac
h
in
e
tr
a
n
s
latio
n
an
d
tex
t
-
to
-
s
p
ee
c
h
,
is
s
till
in
it
s
in
f
an
cy
[
1
3
]
.
Ad
d
r
ess
in
g
th
e
s
e
g
ap
s
will
b
e
cr
u
cial
f
o
r
ad
v
an
cin
g
th
e
f
ield
o
f
a
ksa
r
a
Ja
w
a
r
ec
o
g
n
itio
n
an
d
en
s
u
r
in
g
its
p
r
eser
v
atio
n
an
d
ac
ce
s
s
ib
ilit
y
.
E
x
is
tin
g
ap
p
r
o
ac
h
es
to
J
av
a
n
ese
s
cr
ip
t
r
ec
o
g
n
itio
n
h
av
e
f
ac
ed
t
h
r
ee
f
u
n
d
am
en
tal
c
h
allen
g
es:
i)
m
o
s
t
s
tu
d
ies
u
s
ed
b
asic
C
N
N
ar
ch
itectu
r
es
th
at
s
tr
u
g
g
le
with
th
e
s
cr
ip
t'
s
co
m
p
lex
g
r
ap
h
em
es,
ii)
m
eth
o
d
s
r
eq
u
i
r
ed
im
p
r
ac
tically
l
ar
g
e
d
atasets
,
an
d
iii)
s
y
s
tem
s
s
h
o
wed
p
o
o
r
g
e
n
er
aliza
tio
n
to
r
ea
l
-
wo
r
ld
v
a
r
iatio
n
s
in
h
a
n
d
wr
itin
g
q
u
ali
ty
an
d
s
ty
le.
Fo
r
in
s
tan
ce
,
Su
s
an
to
et
a
l.
[
1
4
]
ac
h
iev
ed
o
n
ly
8
5
% a
cc
u
r
ac
y
d
esp
ite
u
s
in
g
a
d
ee
p
1
2
-
lay
er
C
N
N
.
R
esNet
-
1
8
,
a
v
ar
ian
t
o
f
th
e
r
e
s
id
u
al
n
etwo
r
k
(
R
esNet)
ar
ch
itectu
r
e,
h
as
s
ig
n
if
ican
tly
c
o
n
t
r
ib
u
ted
to
ad
v
an
ce
m
e
n
ts
in
d
ee
p
lear
n
i
n
g
,
p
ar
ticu
lar
ly
in
co
m
p
u
ter
v
is
io
n
task
s
s
u
ch
as
im
ag
e
c
lass
if
icatio
n
,
o
b
ject
d
etec
tio
n
,
an
d
m
ed
ical
im
ag
e
an
aly
s
is
.
I
t
co
n
s
is
ts
o
f
1
8
lay
er
s
,
in
clu
d
in
g
co
n
v
o
lu
t
io
n
al
lay
er
s
,
b
atch
n
o
r
m
aliza
tio
n
,
an
d
r
ec
tifie
d
li
n
ea
r
u
n
it
(
R
eL
U
)
ac
tiv
atio
n
f
u
n
ctio
n
s
.
T
h
e
k
e
y
f
ea
tu
r
e
o
f
R
esNet
ar
ch
itectu
r
es
is
th
e
u
s
e
o
f
s
k
ip
co
n
n
ec
tio
n
s
o
r
r
esid
u
al
co
n
n
ec
tio
n
s
,
wh
i
ch
f
ac
ilit
ate
g
r
ad
ie
n
t
f
lo
w
a
n
d
h
elp
m
itig
ate
th
e
v
an
is
h
in
g
g
r
ad
ien
t
p
r
o
b
lem
[
1
5
]
.
T
h
e
ef
f
ec
tiv
e
n
ess
o
f
R
esNet
-
1
8
h
as
b
ee
n
d
em
o
n
s
tr
at
ed
ac
r
o
s
s
v
ar
io
u
s
ap
p
licatio
n
s
.
I
t
h
as
also
b
ee
n
s
u
cc
ess
f
u
lly
ap
p
lied
in
f
r
u
it
im
ag
e
class
if
icatio
n
,
wh
er
e
m
u
ltip
le
R
esNet
-
18
m
o
d
els
en
h
an
ce
d
p
e
r
f
o
r
m
an
c
e
s
ig
n
if
ican
tly
[
1
6
]
.
Ad
d
itio
n
ally
,
it
h
as
b
ee
n
u
tili
ze
d
f
o
r
d
etec
tin
g
m
ed
ical
co
n
d
itio
n
s
s
u
ch
as
b
r
ain
tu
m
o
r
s
an
d
c
o
lo
r
ec
tal
ca
n
ce
r
,
y
ie
ld
in
g
r
eliab
le
an
d
r
ep
r
o
d
u
ci
b
le
r
esu
lts
[
1
7
]
,
[
1
8
]
.
W
e
im
p
lem
en
t
th
e
f
ir
s
t
ap
p
lic
atio
n
o
f
R
esNet
-
1
8
with
s
k
ip
co
n
n
ec
tio
n
s
s
p
ec
if
ically
o
p
ti
m
ized
f
o
r
J
av
an
ese
s
cr
ip
t's
s
tr
u
ctu
r
al
ch
allen
g
es,
o
v
er
co
m
i
n
g
v
a
n
is
h
in
g
g
r
ad
ien
t iss
u
es th
at
p
lag
u
ed
ea
r
lier
C
NN
ap
p
r
o
ac
h
es
.
B
e
y
o
n
d
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
,
R
e
s
N
e
t
-
1
8
h
a
s
b
e
en
ex
p
lo
r
ed
i
n
g
en
o
m
i
c
s
f
o
r
d
i
s
e
a
s
e
d
e
t
ec
t
i
o
n
a
n
d
t
r
e
a
t
m
en
t
o
p
t
im
i
z
a
t
i
o
n
,
d
em
o
n
s
t
r
a
t
in
g
i
t
s
v
e
r
s
a
t
i
l
i
t
y
a
cr
o
s
s
d
i
f
f
er
e
n
t
d
o
m
a
in
s
[
1
9
]
.
I
ts
s
k
i
p
co
n
n
e
c
t
io
n
s
p
l
a
y
a
c
r
u
c
i
a
l
r
o
l
e
i
n
tr
a
i
n
in
g
d
e
ep
e
r
n
e
t
w
o
r
k
s
b
y
a
d
d
r
e
s
s
i
n
g
th
e
v
a
n
i
s
h
i
n
g
g
r
a
d
ie
n
t
p
r
o
b
l
e
m
[
2
0
]
.
T
e
c
h
n
i
q
u
e
s
s
u
c
h
a
s
g
r
ad
i
en
t
a
m
p
l
i
f
i
c
a
ti
o
n
an
d
o
p
t
i
m
ize
d
tr
a
i
n
in
g
s
t
r
a
te
g
i
e
s
c
an
f
u
r
t
h
e
r
e
n
h
an
c
e
i
t
s
p
e
r
f
o
r
m
a
n
c
e
b
y
r
e
d
u
c
i
n
g
tr
ai
n
i
n
g
t
i
m
e
a
n
d
i
m
p
r
o
v
in
g
a
cc
u
r
a
cy
[
1
5
]
,
[
2
1
]
.
A
l
t
h
o
u
g
h
R
e
s
N
e
t
-
1
8
i
s
l
e
s
s
c
o
m
p
l
e
x
t
h
an
d
e
ep
er
m
o
d
e
l
s
l
i
k
e
R
e
s
N
e
t
-
5
0
a
n
d
R
e
s
N
e
t
-
1
0
1
,
i
t
s
t
i
l
l
r
eq
u
ir
e
s
s
u
b
s
t
a
n
t
i
a
l
c
o
m
p
u
t
a
t
io
n
a
l
r
e
s
o
u
r
c
e
s
,
e
s
p
e
c
ia
l
l
y
i
n
m
u
l
t
i
-
m
o
d
e
l
s
e
tu
p
s
.
F
i
n
e
-
t
u
n
i
n
g
p
a
r
a
m
e
t
er
s
l
ik
e
in
p
u
t
r
e
s
o
lu
t
i
o
n
an
d
q
u
a
n
t
iz
a
t
i
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Op
timiz
in
g
Ja
va
n
ese
s
crip
t re
co
g
n
itio
n
u
s
in
g
fin
e
-
tu
n
ed
R
es
N
et
-
1
8
a
n
d
tr
a
n
s
fer lea
r
n
in
g
(
N
u
r
F
a
tea
h
)
445
p
r
e
c
i
s
io
n
i
s
e
s
s
e
n
t
ia
l
f
o
r
o
p
t
im
a
l
p
e
r
f
o
r
m
an
c
e
b
u
t
ca
n
b
e
a
r
e
s
o
u
r
c
e
-
in
t
e
n
s
iv
e
a
n
d
c
o
m
p
l
e
x
p
r
o
ce
s
s
.
A
s
a
p
o
w
er
f
u
l
a
n
d
a
d
ap
t
a
b
le
d
ee
p
l
e
a
r
n
i
n
g
m
o
d
e
l,
R
e
s
N
e
t
-
1
8
h
a
s
a
c
h
ie
v
ed
n
o
t
a
b
le
s
u
c
c
e
s
s
i
n
v
ar
i
o
u
s
a
p
p
l
i
ca
t
i
o
n
s
,
p
ar
t
i
c
u
l
ar
l
y
in
c
o
m
p
u
t
er
v
i
s
i
o
n
a
n
d
m
ed
i
c
a
l
i
m
ag
i
n
g
.
I
t
s
r
e
s
i
d
u
a
l
c
o
n
n
e
c
t
i
o
n
-
b
a
s
e
d
a
r
c
h
i
t
ec
t
u
r
e
s
u
p
p
o
r
t
s
t
h
e
tr
a
in
i
n
g
o
f
d
e
ep
er
n
e
t
w
o
r
k
s
,
w
h
il
e
i
t
s
v
e
r
s
a
t
i
l
i
ty
en
s
u
r
e
s
i
t
s
co
n
t
in
u
ed
p
o
p
u
l
ar
i
t
y
d
e
s
p
i
t
e
i
t
s
c
o
m
p
u
t
a
t
i
o
n
a
l
d
e
m
a
n
d
s
.
T
h
e
r
ec
o
g
n
itio
n
o
f
a
ksa
r
a
J
a
w
a
p
r
esen
ts
a
s
ig
n
if
ican
t
ch
allen
g
e
in
th
e
f
ield
o
f
OC
R
d
u
e
to
its
in
tr
icate
ch
ar
ac
ter
s
tr
u
ctu
r
es
an
d
lim
ited
d
ig
ital
r
eso
u
r
ce
s
[
2
2
]
.
T
h
is
s
tu
d
y
h
ig
h
lig
h
ts
th
e
im
p
o
r
tan
ce
o
f
ap
p
ly
in
g
d
ee
p
lear
n
in
g
tech
n
iq
u
es,
s
p
ec
if
ically
a
f
in
e
-
tu
n
e
d
R
esNet
-
1
8
m
o
d
el,
to
ad
d
r
e
s
s
th
ese
ch
allen
g
es
an
d
en
h
an
ce
t
h
e
d
ig
itizatio
n
o
f
J
av
an
ese
s
cr
ip
t
[
2
3
]
.
B
y
lev
er
ag
in
g
tr
an
s
f
er
lear
n
in
g
a
n
d
a
d
ap
tiv
e
o
p
tim
izatio
n
tech
n
iq
u
es,
th
is
r
esear
ch
co
n
tr
ib
u
tes
to
th
e
p
r
eser
v
atio
n
o
f
cu
ltu
r
al
h
er
itag
e
wh
ile
ad
v
an
cin
g
I
n
d
o
n
esian
lan
g
u
ag
e
tec
h
n
o
lo
g
y
.
T
h
e
s
u
cc
ess
f
u
l
class
i
f
icatio
n
o
f
a
ksa
r
a
Ja
w
a
w
ith
h
ig
h
ac
cu
r
ac
y
d
em
o
n
s
tr
ates
th
e
p
o
ten
tial
o
f
d
ee
p
lear
n
in
g
-
b
ased
OC
R
s
y
s
tem
s
in
au
to
m
atin
g
th
e
r
ec
o
g
n
itio
n
a
n
d
d
ig
ital
d
o
cu
m
e
n
tatio
n
o
f
h
is
to
r
ical
s
cr
ip
ts
,
p
av
in
g
th
e
way
f
o
r
b
r
o
ad
e
r
ap
p
licatio
n
s
in
lin
g
u
is
tic
p
r
eser
v
atio
n
,
ed
u
ca
tio
n
,
a
n
d
c
u
ltu
r
al
s
tu
d
ies.
2.
M
E
T
H
O
D
2
.
1
.
Da
t
a
s
et
des
cr
iptio
n
T
h
is
s
tu
d
y
em
p
lo
y
s
a
f
in
e
-
t
u
n
ed
R
esNet
-
1
8
m
o
d
el
with
ad
ap
tiv
e
m
o
m
e
n
t
esti
m
atio
n
(
Ad
am
)
o
p
tim
izatio
n
an
d
tr
an
s
f
er
lear
n
in
g
to
class
if
y
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
s
.
T
h
e
m
eth
o
d
o
lo
g
y
co
n
s
is
ts
o
f
s
ev
er
al
k
ey
s
tag
es,
in
clu
d
in
g
d
ata
p
r
e
p
r
o
ce
s
s
in
g
,
m
o
d
el
f
in
e
-
tu
n
in
g
,
tr
ain
in
g
,
an
d
ev
alu
atio
n
.
T
h
e
h
an
ac
ar
ak
a
d
ataset
is
a
v
alu
ab
le
r
eso
u
r
ce
f
o
r
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
r
ec
o
g
n
itio
n
r
esear
ch
.
I
t
co
n
s
is
ts
o
f
a
co
llectio
n
o
f
h
an
d
wr
itten
a
ksa
r
a
Ja
w
a
ch
a
r
ac
ter
s
,
s
o
u
r
ce
d
f
r
o
m
v
ar
io
u
s
o
n
lin
e
p
latf
o
r
m
s
an
d
d
ig
ital
s
u
r
v
ey
s
.
T
h
e
d
ataset
is
co
m
p
o
s
ed
o
f
1
5
6
2
im
ag
es,
ea
ch
r
ep
r
esen
tin
g
a
u
n
iq
u
e
ch
ar
ac
ter
f
r
o
m
th
e
a
ksa
r
a
Ja
w
a
s
cr
ip
t.
Mo
r
e
d
etails
ca
n
b
e
s
ee
n
in
Fig
u
r
e
2
.
Fig
u
r
e
2
.
R
esear
ch
f
lo
wc
h
ar
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
443
-
4
5
3
446
T
h
e
r
e
s
e
ar
c
h
f
l
o
w
ch
a
r
t
i
l
lu
s
t
r
a
t
e
s
t
h
e
s
t
ep
-
by
-
s
t
e
p
p
r
o
ce
s
s
o
f
a
k
s
a
r
a
J
a
w
a
ch
a
r
a
c
t
e
r
c
las
s
i
f
i
c
a
t
i
o
n
u
s
i
n
g
a
f
in
e
-
tu
n
e
d
R
e
s
N
e
t
-
1
8
m
o
d
e
l
w
i
t
h
Ad
a
m
o
p
t
i
m
i
za
t
i
o
n
.
T
h
e
p
r
o
c
e
s
s
b
eg
i
n
s
wi
t
h
t
h
e
h
an
a
c
ar
ak
a
i
m
a
g
e
d
a
t
a
s
e
t
,
w
h
i
ch
u
n
d
e
r
g
o
e
s
p
r
e
p
r
o
c
e
s
s
i
n
g
s
t
ep
s
s
u
c
h
a
s
r
e
s
i
z
i
n
g
,
n
o
r
m
a
li
z
a
t
i
o
n
,
an
d
d
a
ta
a
u
g
m
e
n
t
a
t
io
n
t
o
e
n
h
a
n
c
e
i
m
a
g
e
q
u
a
l
i
ty
an
d
i
m
p
r
o
v
e
m
o
d
e
l
g
e
n
er
a
l
i
z
a
t
io
n
.
T
h
e
d
a
t
a
s
e
t
i
s
t
h
en
s
p
l
i
t
i
n
to
t
r
a
i
n
in
g
(
8
0
%
)
,
v
a
l
i
d
a
t
io
n
(
1
0
%
)
,
an
d
t
e
s
t
i
n
g
(
1
0
%
)
s
u
b
s
e
t
s
t
o
e
n
s
u
r
e
r
o
b
u
s
t
m
o
d
e
l
e
v
a
l
u
a
t
io
n
.
T
h
e
t
r
a
i
n
in
g
d
a
ta
i
s
u
s
e
d
t
o
f
in
e
-
t
u
n
e
th
e
R
e
s
N
e
t
-
1
8
m
o
d
e
l
,
w
h
er
e
th
e
f
in
a
l
f
u
l
ly
c
o
n
n
ec
t
e
d
(
F
C
)
l
ay
er
i
s
m
o
d
if
i
e
d
to
m
at
c
h
th
e
n
u
m
b
e
r
o
f
h
an
a
c
ar
a
k
a
c
l
a
s
s
e
s
,
a
n
d
t
h
e
Ad
a
m
o
p
t
im
i
z
e
r
i
s
em
p
l
o
y
e
d
f
o
r
e
f
f
i
c
i
e
n
t
p
a
r
am
e
t
e
r
u
p
d
at
e
s
.
T
h
e
m
o
d
e
l
's
p
e
r
f
o
r
m
a
n
c
e
i
s
th
en
ev
a
l
u
a
te
d
u
s
i
n
g
p
r
e
c
i
s
io
n
,
r
ec
a
l
l
,
F
1
-
s
c
o
r
e
,
a
n
d
a
c
c
u
r
a
c
y
,
en
s
u
r
in
g
th
a
t
th
e
tr
a
i
n
ed
m
o
d
e
l
i
s
o
p
t
i
m
i
z
ed
f
o
r
J
av
an
e
s
e
s
cr
i
p
t
r
e
co
g
n
i
t
i
o
n
.
T
h
e
r
e
s
u
l
t
s
an
d
d
i
s
c
u
s
s
i
o
n
s
h
o
u
l
d
a
n
a
l
y
ze
t
h
e
e
f
f
e
c
t
i
v
e
n
e
s
s
o
f
th
i
s
ap
p
r
o
a
c
h
,
h
ig
h
l
ig
h
t
in
g
im
p
r
o
v
e
m
en
t
s
i
n
a
c
c
u
r
a
cy
an
d
r
o
b
u
s
t
n
e
s
s
in
r
e
c
o
g
n
i
z
in
g
a
k
s
a
r
a
Ja
w
a
ch
a
r
a
c
t
e
r
s
.
2
.
2
.
M
o
del selec
t
io
n a
nd
f
ine
-
t
un
ing
R
esNet
-
1
8
,
a
d
ee
p
C
NN
m
o
d
el
k
n
o
wn
f
o
r
its
r
esid
u
al
le
ar
n
in
g
m
ec
h
a
n
is
m
,
is
ch
o
s
en
d
u
e
to
its
ef
f
icien
cy
in
e
x
tr
ac
tin
g
h
ier
ar
ch
ical
f
ea
tu
r
es.
T
h
e
m
o
d
el
is
p
r
etr
ai
n
ed
o
n
I
m
a
g
eNe
t
an
d
f
in
e
-
tu
n
ed
b
y
r
ep
lacin
g
th
e
FC
lay
er
to
m
at
ch
th
e
n
u
m
b
e
r
o
f
h
a
n
ac
ar
ak
a
ch
ar
ac
ter
class
es.
On
ly
th
e
f
in
al
lay
er
s
ar
e
tr
ai
n
ed
wh
ile
ea
r
lier
co
n
v
o
l
u
tio
n
al
l
ay
er
s
r
em
ain
f
r
o
ze
n
to
r
etai
n
p
r
etr
ain
e
d
k
n
o
wled
g
e.
R
esNet
-
1
8
is
a
C
NN
ar
ch
itectu
r
e
d
esig
n
ed
to
a
d
d
r
ess
th
e
v
an
is
h
in
g
g
r
a
d
ien
t
p
r
o
b
lem
in
d
ee
p
lear
n
in
g
m
o
d
e
ls
[
2
0
]
,
[
2
4
]
.
Un
lik
e
tr
ad
itio
n
al
C
NNs,
R
es
Net
in
tr
o
d
u
ce
s
s
k
ip
co
n
n
ec
tio
n
s
(
s
h
o
r
tcu
t
co
n
n
ec
tio
n
s
)
th
at
allo
w
t
h
e
g
r
ad
ie
n
t
to
f
lo
w
d
ir
ec
tly
th
r
o
u
g
h
lay
er
s
,
p
r
e
v
en
tin
g
d
eg
r
a
d
atio
n
in
d
ee
p
n
etwo
r
k
s
.
R
esNet
-
1
8
co
n
s
is
ts
o
f
1
8
lay
er
s
,
in
clu
d
in
g
co
n
v
o
lu
tio
n
al
lay
er
s
,
b
atch
n
o
r
m
aliza
ti
o
n
,
R
eL
U
ac
tiv
atio
n
,
an
d
r
esid
u
al
b
l
o
ck
s
[
2
5
]
.
T
h
e
r
esid
u
al
lear
n
in
g
f
r
am
ewo
r
k
en
a
b
les
th
e
m
o
d
el
to
lear
n
id
e
n
tity
m
ap
p
in
g
s
,
m
a
k
in
g
it
m
o
r
e
ef
f
icien
t
in
ca
p
tu
r
in
g
h
ier
a
r
ch
ical
f
ea
tu
r
es
f
r
o
m
im
ag
es.
T
h
e
n
e
two
r
k
is
p
r
etr
ain
ed
o
n
lar
g
e
-
s
ca
le
d
atasets
lik
e
I
m
ag
eNe
t,
allo
win
g
it
to
e
x
tr
ac
t
r
o
b
u
s
t
f
ea
t
u
r
es
f
o
r
tr
a
n
s
f
er
lea
r
n
in
g
ap
p
licatio
n
s
.
Ma
th
em
atica
lly
,
a
r
esid
u
al
b
lo
ck
in
R
esNet
ca
n
b
e
ex
p
r
ess
ed
as (
1
)
.
=
(
,
{
}
)
+
(
1
)
W
h
er
e
x
is
th
e
in
p
u
t; F(
x
,
{W
i
})
r
ep
r
esen
ts
th
e
lear
n
ed
tr
a
n
s
f
o
r
m
atio
n
(
s
er
ies o
f
co
n
v
o
lu
ti
o
n
al
lay
er
s
)
; a
n
d
th
e
ad
d
itio
n
o
f
x
allo
ws th
e
m
o
d
el
to
lear
n
an
id
en
tity
m
a
p
p
in
g
,
p
r
ev
en
tin
g
d
eg
r
ad
atio
n
in
d
ee
p
er
n
etwo
r
k
s
.
Du
e
to
its
ef
f
icien
t
f
ea
tu
r
e
ex
tr
ac
tio
n
ca
p
a
b
ilit
ies,
R
esNet
-
1
8
is
c
h
o
s
en
f
o
r
th
is
s
tu
d
y
to
r
ec
o
g
n
ize
th
e
co
m
p
lex
s
h
ap
es
o
f
a
ksa
r
a
Ja
w
a
w
h
ile
m
itig
atin
g
v
an
i
s
h
in
g
g
r
a
d
ien
t
is
s
u
es.
T
h
e
Ad
am
alg
o
r
ith
m
is
an
ad
v
an
ce
d
o
p
tim
izatio
n
tech
n
i
q
u
e
th
at
co
m
b
in
es
th
e
b
en
e
f
its
o
f
m
o
m
en
tu
m
-
b
ased
o
p
ti
m
izatio
n
(
e.
g
.
,
r
o
o
t
m
ea
n
s
q
u
ar
e
p
r
o
p
ag
atio
n
(
R
MSp
r
o
p
)
a
n
d
s
to
ch
asti
c
g
r
ad
ien
t
d
escen
t
(
SGD
)
with
m
o
m
en
tu
m
)
.
Ad
am
is
wid
ely
u
s
ed
d
u
e
t
o
its
ad
ap
tiv
e
lear
n
in
g
r
ates,
wh
ich
h
elp
s
t
ab
ilize
tr
ain
in
g
ac
r
o
s
s
d
if
f
e
r
en
t
d
atasets
ca
n
s
h
o
w
o
n
(
2
)
.
Ad
am
u
p
d
ates th
e
m
o
d
el’
s
p
ar
am
eter
s
u
s
in
g
f
ir
s
t
-
o
r
d
er
an
d
s
ec
o
n
d
-
o
r
d
er
m
o
m
en
t e
s
tim
ates
.
=
1
−
1
+
(
1
−
1
)
2
=
2
−
1
+
(
1
−
2
)
2
=
−
1
−
√
+
(
2
)
W
h
er
e
m
t
is
th
e
ex
p
o
n
en
tially
d
ec
ay
in
g
a
v
er
ag
e
o
f
p
ast
g
r
a
d
ien
ts
;
v
t
is
th
e
ex
p
o
n
en
tially
d
ec
ay
in
g
av
er
a
g
e
o
f
s
q
u
ar
ed
g
r
ad
ien
ts
;
β
1
b
eta
β2
ar
e
h
y
p
er
p
a
r
am
eter
s
(
ty
p
ically
s
et
to
0
.
9
a
n
d
0
.
9
9
9
)
;
η
is
th
e
lear
n
in
g
r
ate;
an
d
ϵ
is
a
s
m
all
co
n
s
tan
t to
p
r
ev
e
n
t d
iv
is
io
n
b
y
ze
r
o
.
Ad
am
is
ch
o
s
en
in
th
is
s
tu
d
y
d
u
e
to
its
f
ast
c
o
n
v
e
r
g
en
ce
an
d
ab
ilit
y
to
h
an
d
le
s
p
ar
s
e
g
r
a
d
ien
ts
,
m
ak
in
g
it
s
u
itab
le
f
o
r
f
in
e
-
t
u
n
in
g
p
r
etr
ai
n
ed
R
esNet
-
1
8
with
o
u
t
ex
ten
s
iv
e
h
y
p
er
p
a
r
a
m
eter
tu
n
i
n
g
.
T
h
e
alg
o
r
ith
m
d
y
n
am
ically
ad
ju
s
ts
lear
n
in
g
r
ates
f
o
r
ea
ch
p
ar
am
eter
u
s
in
g
m
o
v
in
g
av
e
r
ag
es
o
f
g
r
ad
ien
ts
an
d
s
q
u
ar
ed
g
r
ad
ie
n
ts
,
wh
ich
s
tab
ilizes tr
ain
in
g
.
T
h
is
ad
ap
tiv
e
ap
p
r
o
ac
h
p
r
o
v
es p
ar
ticu
lar
ly
ef
f
ec
tiv
e
f
o
r
J
av
an
ese
s
cr
ip
t r
ec
o
g
n
itio
n
,
wh
er
e
c
h
ar
ac
ter
co
m
p
lex
ity
d
em
an
d
s
p
r
e
cise w
eig
h
t u
p
d
ates d
u
r
i
n
g
b
a
ck
p
r
o
p
ag
atio
n
.
2
.
3
.
T
ra
ini
ng
pro
ce
s
s
T
h
e
m
o
d
el
is
tr
ain
ed
u
s
in
g
th
e
Ad
am
o
p
tim
izatio
n
alg
o
r
ith
m
with
a
lear
n
in
g
r
ate
o
f
0
.
0
0
1
,
b
atch
s
ize
o
f
3
2
,
a
n
d
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
f
u
n
ctio
n
.
T
h
e
tr
ain
in
g
is
co
n
d
u
cted
f
o
r
5
0
ep
o
c
h
s
,
an
d
ea
r
ly
s
to
p
p
in
g
is
ap
p
lied
to
p
r
e
v
en
t
o
v
er
f
itti
n
g
.
Du
r
in
g
tr
ain
in
g
,
d
r
o
p
o
u
t
r
e
g
u
lar
izati
o
n
is
u
s
ed
to
e
n
h
an
ce
g
e
n
er
a
lizatio
n
,
an
d
b
atch
n
o
r
m
aliza
tio
n
is
ap
p
lied
t
o
s
tab
ilize
lear
n
in
g
.
T
h
e
tr
ai
n
in
g
p
r
o
ce
s
s
in
v
o
l
v
es
f
in
e
-
t
u
n
in
g
a
p
r
et
r
ain
ed
R
esNet
-
1
8
m
o
d
el
f
o
r
a
ksa
r
a
J
a
w
a
class
if
icatio
n
u
s
in
g
th
e
Ad
am
o
p
tim
izer
.
First,
th
e
d
ata
s
et
is
p
r
ep
r
o
ce
s
s
ed
b
y
r
esizin
g
im
a
g
es
to
2
2
4
×
2
2
4
p
ix
els,
n
o
r
m
alizin
g
p
ix
el
v
a
lu
es,
an
d
a
p
p
ly
in
g
au
g
m
en
tati
o
n
tech
n
i
q
u
es
lik
e
r
o
tatio
n
an
d
co
n
tr
ast
ad
ju
s
tm
e
n
t.
T
h
e
d
ataset
is
s
p
lit
in
to
tr
ain
in
g
(
8
0
%),
v
alid
atio
n
(
1
0
%),
an
d
test
in
g
(
1
0
%)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Op
timiz
in
g
Ja
va
n
ese
s
crip
t re
co
g
n
itio
n
u
s
in
g
fin
e
-
tu
n
ed
R
es
N
et
-
1
8
a
n
d
tr
a
n
s
fer lea
r
n
in
g
(
N
u
r
F
a
tea
h
)
447
s
u
b
s
ets
to
en
s
u
r
e
r
o
b
u
s
t
lear
n
in
g
.
Du
r
in
g
t
r
ain
in
g
,
th
e
R
e
s
Net
-
1
8
m
o
d
el
u
n
d
e
r
g
o
es
f
o
r
war
d
p
r
o
p
ag
atio
n
,
wh
er
e
in
p
u
t
im
ag
es
p
ass
th
r
o
u
g
h
c
o
n
v
o
lu
tio
n
al
lay
e
r
s
f
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
f
o
llo
wed
b
y
class
if
icatio
n
u
s
in
g
a
m
o
d
if
ied
FC
lay
er
.
T
h
e
lo
s
s
f
u
n
ctio
n
,
cr
o
s
s
-
en
tr
o
p
y
lo
s
s
,
ev
alu
ates
p
r
ed
ictio
n
er
r
o
r
s
,
an
d
b
ac
k
p
r
o
p
ag
atio
n
u
p
d
ates
weig
h
ts
u
s
in
g
th
e
Ad
am
o
p
tim
izer
,
wh
ich
ad
ju
s
ts
l
ea
r
n
in
g
r
ates
ad
ap
tiv
el
y
f
o
r
ef
f
icien
t
co
n
v
er
g
e
n
ce
.
R
eg
u
lar
izatio
n
tech
n
iq
u
es
lik
e
d
r
o
p
o
u
t
ar
e
ap
p
lied
to
p
r
ev
en
t
o
v
er
f
itti
n
g
,
a
n
d
th
e
m
o
d
el
is
tr
ain
ed
f
o
r
u
p
to
5
0
ep
o
ch
s
with
ea
r
ly
s
to
p
p
in
g
to
p
r
ev
e
n
t
u
n
n
ec
ess
ar
y
co
m
p
u
tatio
n
s
.
Per
f
o
r
m
a
n
ce
i
s
ev
alu
ated
u
s
in
g
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e
,
an
d
ac
cu
r
ac
y
,
en
s
u
r
in
g
o
p
tim
al
r
ec
o
g
n
itio
n
o
f
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
s
.
Fin
ally
,
th
e
b
est
-
p
er
f
o
r
m
in
g
m
o
d
el
i
s
s
elec
ted
f
o
r
test
in
g
a
n
d
later
d
ep
lo
y
ed
f
o
r
OC
R
ap
p
licatio
n
s
,
en
ab
lin
g
au
to
m
ated
r
ec
o
g
n
itio
n
o
f
J
av
a
n
ese
s
cr
ip
t.
2
.
4
.
E
v
a
lua
t
i
o
n m
et
rics
T
h
e
tr
ain
e
d
m
o
d
el
is
ev
al
u
ated
o
n
th
e
test
s
et
u
s
in
g
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
a
n
d
a
cc
u
r
ac
y
t
o
m
ea
s
u
r
e
its
ef
f
ec
tiv
e
n
ess
in
r
ec
o
g
n
izin
g
J
av
an
ese
s
cr
ip
t
.
T
o
ass
ess
th
e
p
er
f
o
r
m
an
ce
o
f
th
e
f
in
e
-
tu
n
ed
R
esNet
-
1
8
m
o
d
el
f
o
r
a
ksa
r
a
Ja
w
a
class
if
icatio
n
,
s
ev
er
al
e
v
alu
atio
n
m
etr
ics
ar
e
u
s
ed
,
i
n
clu
d
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e.
T
h
ese
m
etr
ics
p
r
o
v
id
e
a
co
m
p
r
eh
en
s
iv
e
u
n
d
er
s
tan
d
i
n
g
o
f
t
h
e
m
o
d
el’
s
ef
f
ec
tiv
en
ess
in
d
is
tin
g
u
is
h
i
n
g
b
etwe
en
d
if
f
er
e
n
t
J
av
an
ese
ch
ar
ac
ter
s
.
Acc
u
r
ac
y
m
ea
s
u
r
es
th
e
o
v
er
all
co
r
r
ec
tn
ess
o
f
th
e
m
o
d
el’
s
p
r
ed
ictio
n
s
,
ca
lcu
lated
as
th
e
r
atio
o
f
co
r
r
ec
tly
class
if
ied
s
am
p
les
to
th
e
to
tal
n
u
m
b
er
o
f
s
am
p
les as sh
o
wn
i
n
(
3
)
.
A
c
c
ura
c
y
=
(
+
)
(
+
+
)
(
3
)
W
h
er
e
tr
u
e
p
o
s
itiv
es
(
T
P):
co
r
r
ec
tly
p
r
ed
icted
a
ksa
r
a
J
a
w
a
ch
ar
ac
ter
s
;
tr
u
e
n
e
g
ativ
es
(
T
N)
:
co
r
r
ec
tly
p
r
ed
icted
n
o
n
-
tar
g
et
c
h
ar
ac
ter
s
;
f
alse
p
o
s
itiv
es
(
FP
)
:
in
co
r
r
e
ctly
class
if
ied
ch
ar
ac
ter
s
th
at
d
o
n
o
t
b
el
o
n
g
to
th
e
class
; a
n
d
f
alse n
eg
ativ
es (
FN)
: m
is
class
if
ied
ch
ar
ac
ter
s
th
at
s
h
o
u
ld
b
elo
n
g
to
th
e
class
.
Pre
cisi
o
n
in
d
icate
s
h
o
w
m
an
y
o
f
th
e
p
r
ed
icted
p
o
s
itiv
e
class
if
icatio
n
s
wer
e
co
r
r
ec
t,
m
ea
s
u
r
in
g
th
e
m
o
d
el'
s
ab
ilit
y
to
av
o
id
FP
.
I
t
is
ca
lcu
lated
as
th
e
r
atio
o
f
TP
(
co
r
r
ec
tly
class
if
ied
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
s
)
to
all
p
r
ed
icted
p
o
s
itiv
es
(
TP
+
FP
)
.
T
h
is
m
etr
ic
is
p
ar
ticu
lar
l
y
cr
u
cial
f
o
r
J
av
an
ese
s
cr
ip
t
r
ec
o
g
n
itio
n
,
wh
er
e
m
is
class
if
y
in
g
s
im
ilar
-
lo
o
k
in
g
ch
ar
ac
ter
s
(
e.
g
.
,
'
h
a'
v
s
'
n
a
'
)
co
u
ld
d
is
to
r
t h
is
to
r
ical
d
o
cu
m
e
n
t in
ter
p
r
etatio
n
s
.
I
t
is
ca
lcu
lated
as (
4
)
.
Pr
e
c
ision
=
(
+
)
(
4
)
A
h
ig
h
er
p
r
ec
is
io
n
m
ea
n
s
th
at
th
e
m
o
d
el
p
r
o
d
u
ce
s
f
ewe
r
FP
,
wh
ich
is
cr
u
cial
wh
en
m
is
class
if
icatio
n
ca
n
lead
to
in
c
o
r
r
ec
t
OC
R
r
ec
o
g
n
itio
n
.
I
n
th
e
c
o
n
tex
t
o
f
J
av
an
ese
s
cr
ip
t
d
ig
itizatio
n
,
h
ig
h
p
r
ec
is
io
n
en
s
u
r
es
th
at
ch
ar
ac
ter
s
lik
e
"d
a"
a
n
d
"d
h
a"
⎯
wh
ich
d
if
f
er
b
y
o
n
ly
a
s
in
g
le
s
tr
o
k
e
⎯
ar
e
n
o
t
co
n
f
u
s
ed
d
u
r
in
g
tr
an
s
cr
ip
tio
n
.
T
h
is
m
etr
ic
b
e
co
m
es
esp
ec
ially
im
p
o
r
ta
n
t
wh
en
p
r
o
ce
s
s
in
g
h
is
to
r
ical
m
an
u
s
cr
ip
ts
wh
er
e
p
r
eser
v
atio
n
ac
c
u
r
ac
y
is
p
a
r
a
m
o
u
n
t.
I
t is ca
lcu
lated
as (
5
)
.
R
e
c
a
l
l
=
(
+
)
(
5
)
A
h
ig
h
r
ec
all
v
alu
e
m
ea
n
s
th
e
m
o
d
el
co
r
r
ec
tly
id
en
tifie
s
m
o
s
t
o
f
th
e
tr
u
e
ch
ar
ac
ter
s
b
u
t
m
ay
s
till
m
is
class
if
y
s
o
m
e
as
n
eg
ativ
es
.
T
h
is
is
p
ar
ticu
lar
ly
im
p
o
r
ta
n
t
f
o
r
p
r
eser
v
in
g
r
ar
e
J
av
an
ese
s
cr
ip
t
v
ar
ian
ts
th
at
ap
p
ea
r
i
n
f
r
eq
u
en
tly
in
h
is
to
r
ic
al
d
o
cu
m
e
n
ts
.
W
h
ile
h
ig
h
r
ec
all
en
s
u
r
es
co
m
p
r
eh
en
s
iv
e
ch
ar
ac
ter
d
etec
tio
n
,
it
m
ay
co
m
e
at
t
h
e
co
s
t o
f
ac
ce
p
tin
g
s
o
m
e
FP
in
th
e
r
esu
lts
.
I
t i
s
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
an
d
r
ec
all
.
F1
−
s
c
or
e
=
2
×
(
×
)
(
+
)
(
6
)
A
h
ig
h
F1
-
s
co
r
e
i
n
d
icate
s
a
g
o
o
d
b
alan
ce
b
etwe
en
id
en
tif
y
in
g
c
o
r
r
ec
t
c
h
ar
ac
ter
s
an
d
m
in
im
izin
g
f
alse
class
if
icatio
n
s
.
T
h
is
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
a
n
d
r
ec
all
b
ec
o
m
es
p
ar
ticu
lar
ly
v
alu
ab
le
wh
e
n
ev
alu
atin
g
J
av
a
n
ese
s
cr
ip
t
r
ec
o
g
n
itio
n
.
W
h
er
e
b
o
th
ac
c
u
r
ate
id
en
tific
atio
n
(
p
r
ec
is
io
n
)
an
d
co
m
p
lete
d
etec
tio
n
(
r
ec
all)
o
f
c
h
ar
ac
ter
s
ar
e
e
q
u
al
ly
im
p
o
r
tan
t.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
Data
p
r
ep
r
o
ce
s
s
in
g
p
lay
e
d
a
v
ital
r
o
le
in
en
h
an
cin
g
th
e
p
e
r
f
o
r
m
a
n
ce
o
f
th
e
R
esNet
-
1
8
m
o
d
el
f
o
r
a
ksa
r
a
Ja
w
a
class
if
icatio
n
.
A
s
o
u
tlin
ed
in
Alg
o
r
ith
m
1
,
th
e
d
ataset
u
n
d
er
wen
t
m
u
ltip
le
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
,
in
clu
d
in
g
im
a
g
e
r
esizin
g
(
2
2
4
×2
2
4
p
ix
els),
n
o
r
m
aliza
tio
n
,
an
d
au
g
m
en
tatio
n
s
u
ch
as
r
o
tatio
n
an
d
c
o
n
tr
ast
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
443
-
4
5
3
448
ad
ju
s
tm
en
t.
T
h
ese
s
tep
s
en
s
u
r
ed
co
n
s
is
ten
cy
in
in
p
u
t d
ata,
r
ed
u
ce
d
n
o
is
e,
an
d
im
p
r
o
v
ed
th
e
m
o
d
el’
s
ab
ilit
y
to
g
en
er
alize
ac
r
o
s
s
d
if
f
er
e
n
t
ch
ar
ac
ter
v
ar
iatio
n
s
.
Au
g
m
en
tat
io
n
tech
n
iq
u
es
h
elp
ed
ad
d
r
ess
th
e
ch
allen
g
e
o
f
lim
ited
tr
ain
in
g
d
ata
b
y
ar
tific
ially
ex
p
an
d
in
g
th
e
d
ataset
an
d
in
tr
o
d
u
cin
g
v
a
r
iatio
n
s
in
ch
ar
ac
ter
ap
p
ea
r
an
ce
,
m
ak
in
g
th
e
m
o
d
el
m
o
r
e
r
o
b
u
s
t.
Alg
o
r
ith
m
1
.
Fin
e
-
tu
n
e
d
R
esN
et
-
1
8
with
Ad
am
o
p
tim
izatio
n
I
n
p
u
t: X_
tr
ai
n
,
Y_
tr
ain
,
X_
test
,
Y_
test
(
T
esti
n
g
d
ataset)
,
lr
,
ep
o
ch
s
,
b
atch
_
s
ize
Ou
tp
u
t: R
esNet1
8
*
(
Fin
e
-
tu
n
ed
R
esNet
-
1
8
m
o
d
el)
B
eg
in
L
o
a
d
p
r
etr
ai
n
ed
R
esNet
-
18
Fre
ez
e
in
itial c
o
n
v
o
lu
tio
n
al
lay
er
s
R
ep
lace
th
e
f
u
lly
co
n
n
ec
te
d
(
FC
)
lay
er
:
FC
_
in
p
u
t =
R
esNet1
8
.
f
c.
in
_
f
e
atu
r
es
R
esNet1
8
.
f
c
=
Fu
lly
C
o
n
n
ec
ted
(
FC
_
in
p
u
t,
n
u
m
_
class
es)
Pre
p
r
o
ce
s
s
Data
s
et
-
R
esize
im
ag
es to
2
2
4
×
2
2
4
-
No
r
m
alize
p
ix
el
v
alu
es
-
Ap
p
l
y
d
ata
au
g
m
en
tatio
n
(
r
o
tatio
n
,
f
lip
p
in
g
,
co
n
tr
ast ad
j
u
s
tm
en
t)
-
Sp
lit in
to
tr
ain
(
8
0
%),
v
ali
d
atio
n
(
1
0
%),
test
(
1
0
%)
Def
in
e
tr
ain
in
g
co
m
p
o
n
e
n
ts
-
L
o
s
s
f
u
n
ctio
n
: Cro
s
s
E
n
tr
o
p
y
L
o
s
s
(
)
-
Op
tim
izer
: A
d
am
(
R
esNet
1
8
.
p
ar
a
m
eter
s
(
)
,
lr
)
W
h
ile
s
to
p
p
in
g
cr
iter
io
n
n
o
t m
et
d
o
Fo
r
ea
ch
b
atch
(
X_
b
,
Y_
b
)
in
tr
ain
i
n
g
s
et:
-
Fo
r
war
d
p
ass
: o
u
tp
u
t =
R
esNet1
8
(
X_
b
)
-
C
o
m
p
u
te
lo
s
s
: lo
s
s
=
C
r
o
s
s
E
n
tr
o
p
y
L
o
s
s
(
o
u
tp
u
t,
Y_
b
)
-
Z
er
o
g
r
ad
ien
ts
:
o
p
tim
izer
.
ze
r
o
_
g
r
a
d
(
)
-
B
ac
k
p
r
o
p
a
g
atio
n
: lo
s
s
.
b
ac
k
war
d
(
)
-
Up
d
ate
weig
h
ts
: o
p
tim
izer
.
s
t
ep
(
)
-
Valid
ate
m
o
d
el
o
n
v
alid
atio
n
s
et
E
v
alu
ate
m
o
d
el
-
C
o
m
p
u
te
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
ac
cu
r
ac
y
o
n
test
s
et
R
etu
r
n
f
in
e
-
tu
n
ed
m
o
d
el
R
e
s
Net1
8
*
E
n
d
T
h
e
d
a
t
a
s
e
t
,
co
n
s
i
s
t
i
n
g
o
f
1
,
5
6
2
i
m
a
g
e
s
o
f
a
k
s
a
r
a
Ja
w
a
ch
a
r
a
c
t
er
s
,
w
a
s
s
p
l
i
t
i
n
to
th
r
e
e
s
u
b
s
e
t
s
t
o
e
n
s
u
r
e
e
f
f
e
c
t
iv
e
tr
a
i
n
in
g
a
n
d
ev
a
l
u
a
t
io
n
o
f
th
e
R
e
s
N
e
t
-
1
8
m
o
d
e
l
.
F
o
l
l
o
w
in
g
a
s
t
an
d
a
r
d
d
a
t
a
d
i
v
i
s
i
o
n
s
t
r
a
t
e
g
y
,
8
0
%
(
1
,
2
5
0
i
m
ag
e
s
)
w
e
r
e
a
l
lo
c
a
t
ed
f
o
r
t
r
a
in
i
n
g
,
a
l
l
o
w
in
g
t
h
e
m
o
d
e
l
t
o
le
a
r
n
c
h
ar
a
c
t
er
p
a
t
t
e
r
n
s
e
f
f
i
c
i
en
t
l
y
.
A
s
e
p
a
r
a
t
e
1
0
%
(
1
5
6
im
a
g
e
s
)
w
a
s
r
e
s
e
r
v
ed
f
o
r
v
a
l
i
d
a
t
i
o
n
,
en
a
b
l
in
g
th
e
f
in
e
-
tu
n
in
g
o
f
h
y
p
e
r
p
a
r
am
e
t
e
r
s
a
n
d
p
r
ev
e
n
t
i
n
g
o
v
er
f
i
t
t
in
g
b
y
m
o
n
i
to
r
in
g
m
o
d
e
l
p
e
r
f
o
r
m
an
c
e
d
u
r
in
g
tr
a
i
n
in
g
.
T
h
e
r
e
m
a
in
i
n
g
1
0
%
(
1
5
6
i
m
a
g
e
s
)
w
a
s
d
e
s
i
g
n
a
t
ed
f
o
r
t
e
s
t
i
n
g
,
p
r
o
v
id
i
n
g
an
u
n
b
i
a
s
e
d
a
s
s
e
s
s
m
e
n
t
o
f
th
e
m
o
d
e
l
’
s
g
e
n
er
a
l
i
z
a
t
io
n
a
b
i
l
i
ty
o
n
u
n
s
ee
n
d
a
t
a
.
A
f
t
er
s
p
l
it
t
i
n
g
t
h
e
d
at
a
s
e
t
,
t
h
e
R
e
s
N
e
t
-
1
8
m
o
d
e
l
wa
s
i
n
i
t
i
a
l
i
z
ed
w
i
th
p
r
e
tr
a
i
n
ed
w
e
ig
h
t
s
f
r
o
m
I
m
a
g
e
N
e
t
to
le
v
er
a
g
e
i
t
s
ex
i
s
t
i
n
g
f
e
a
tu
r
e
e
x
tr
a
c
t
io
n
c
ap
ab
i
l
i
t
i
e
s
.
T
h
e
f
i
r
s
t
c
o
n
v
o
lu
t
i
o
n
a
l
l
ay
e
r
s
we
r
e
f
r
o
z
e
n
to
r
e
ta
i
n
p
r
ev
i
o
u
s
l
y
le
a
r
n
e
d
l
o
w
-
l
ev
e
l
f
e
a
t
u
r
e
s
s
u
ch
a
s
e
d
g
e
s
an
d
t
e
x
tu
r
e
s
,
w
h
i
l
e
t
h
e
F
C
l
a
y
er
w
a
s
r
ep
l
ac
e
d
t
o
m
a
t
c
h
th
e
n
u
m
b
e
r
o
f
a
k
s
a
r
a
J
a
w
a
c
la
s
s
e
s
.
T
h
e
in
p
u
t
i
m
a
g
e
s
,
r
e
s
i
z
e
d
to
2
2
4
×
2
2
4
p
i
x
e
l
s
,
p
a
s
s
e
d
th
r
o
u
g
h
m
u
l
t
ip
l
e
co
n
v
o
lu
t
i
o
n
a
l
la
y
e
r
s
,
b
a
t
c
h
n
o
r
m
a
l
i
z
a
t
io
n
,
a
n
d
R
e
L
U
a
c
t
i
v
a
t
io
n
s
,
e
n
a
b
l
in
g
d
e
e
p
h
ie
r
ar
c
h
i
ca
l
f
e
a
t
u
r
e
ex
t
r
a
c
t
io
n
.
Ma
x
p
o
o
l
in
g
l
a
y
er
s
h
e
l
p
ed
r
ed
u
ce
d
im
e
n
s
i
o
n
a
l
i
t
y
wh
i
l
e
p
r
e
s
e
r
v
i
n
g
cr
i
t
i
c
a
l
f
e
a
tu
r
e
s
.
T
h
e
m
o
d
i
f
i
ed
F
C
l
a
y
er
u
ti
l
iz
e
d
S
o
f
t
M
ax
a
c
t
iv
a
t
i
o
n
to
cl
a
s
s
i
f
y
i
m
ag
e
s
i
n
to
r
e
s
p
e
c
t
iv
e
a
k
s
a
r
a
Ja
w
a
ch
a
r
a
c
t
e
r
s
.
T
h
i
s
i
n
i
t
i
a
l
p
h
a
s
e
p
r
ep
a
r
ed
t
h
e
m
o
d
e
l
f
o
r
o
p
t
i
m
i
za
t
i
o
n
,
e
n
s
u
r
i
n
g
th
a
t
t
h
e
b
a
c
k
b
o
n
e
ar
c
h
i
t
ec
t
u
r
e
w
as
e
f
f
e
c
t
iv
e
l
y
a
d
ap
t
ed
t
o
th
e
n
e
w
d
a
t
a
s
e
t
b
e
f
o
r
e
a
p
p
l
y
in
g
A
d
a
m
o
p
t
im
i
z
a
t
i
o
n
t
o
f
in
e
-
tu
n
e
t
h
e
m
o
d
e
l’
s
p
a
r
am
e
t
e
r
s
.
T
h
e
m
o
d
el
u
n
d
er
g
o
es
f
o
r
war
d
p
r
o
p
ag
atio
n
,
wh
e
r
e
in
p
u
t
i
m
ag
es
p
ass
th
r
o
u
g
h
co
n
v
o
lu
ti
o
n
al,
b
atc
h
n
o
r
m
aliza
tio
n
,
an
d
p
o
o
lin
g
l
ay
er
s
to
e
x
tr
ac
t
h
ier
a
r
ch
ical
f
ea
tu
r
es.
T
h
e
f
in
al
class
if
icatio
n
lay
er
ap
p
lies
So
f
tMa
x
ac
tiv
atio
n
to
ass
ig
n
p
r
o
b
a
b
ilit
ies
to
ea
ch
c
h
ar
ac
te
r
.
T
h
e
m
o
d
el
is
tr
ain
e
d
u
s
in
g
cr
o
s
s
-
en
tr
o
p
y
l
o
s
s
,
with
weig
h
t
u
p
d
ates
p
e
r
f
o
r
m
ed
u
s
in
g
th
e
A
d
am
o
p
tim
izat
io
n
alg
o
r
ith
m
,
wh
ich
ad
ap
tiv
ely
ad
ju
s
ts
lear
n
in
g
r
ates
f
o
r
f
aster
co
n
v
er
g
en
ce
.
T
r
ain
in
g
co
n
tin
u
es
f
o
r
a
m
ax
im
u
m
o
f
5
0
ep
o
ch
s
,
in
co
r
p
o
r
atin
g
ea
r
ly
s
to
p
p
in
g
to
p
r
ev
en
t
o
v
er
f
itti
n
g
.
On
ce
tr
ain
ed
,
th
e
m
o
d
el
is
ev
alu
ated
o
n
th
e
test
s
et
u
s
in
g
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e,
en
s
u
r
in
g
its
ef
f
ec
tiv
en
ess
in
r
ec
o
g
n
izin
g
a
ksa
r
a
J
a
w
a
ch
ar
ac
ter
s
.
I
t
ca
n
b
e
s
ee
n
in
Fig
u
r
e
3
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Op
timiz
in
g
Ja
va
n
ese
s
crip
t re
co
g
n
itio
n
u
s
in
g
fin
e
-
tu
n
ed
R
es
N
et
-
1
8
a
n
d
tr
a
n
s
fer lea
r
n
in
g
(
N
u
r
F
a
tea
h
)
449
Fig
u
r
e
3
.
T
r
ain
in
g
ac
c
u
r
ac
y
g
r
ap
h
T
h
e
f
in
e
-
tu
n
ed
R
e
s
N
e
t
-
1
8
a
lg
o
r
i
th
m
f
o
r
a
k
s
a
r
a
J
a
w
a
c
la
s
s
i
f
i
c
a
t
i
o
n
f
o
l
lo
w
s
a
s
t
r
u
c
t
u
r
ed
p
i
p
e
l
in
e
,
b
e
g
in
n
in
g
w
i
t
h
d
a
t
a
p
r
e
p
r
o
c
e
s
s
i
n
g
,
w
h
e
r
e
i
m
a
g
e
s
a
r
e
r
e
s
i
z
ed
t
o
2
2
4
×
2
2
4
p
i
x
e
l
s
,
n
o
r
m
a
l
i
z
ed
,
an
d
a
u
g
m
e
n
t
ed
to
e
n
h
a
n
ce
g
e
n
er
a
l
i
z
a
t
i
o
n
.
T
h
e
d
a
t
a
s
e
t
o
f
1
,
5
6
2
i
m
a
g
e
s
i
s
t
h
e
n
s
p
l
i
t
in
to
t
r
a
i
n
in
g
(
8
0
%
)
,
v
a
l
i
d
a
t
io
n
(
1
0
%
)
,
an
d
t
e
s
t
i
n
g
(
1
0
%)
s
e
t
s
to
en
s
u
r
e
b
a
l
an
c
e
d
l
e
ar
n
in
g
a
n
d
ev
a
lu
a
t
i
o
n
.
T
h
e
R
e
s
N
e
t
-
18
m
o
d
e
l
,
p
r
e
t
r
a
i
n
ed
o
n
I
m
a
g
e
Ne
t
,
i
s
m
o
d
i
f
i
ed
b
y
f
r
e
e
z
in
g
th
e
i
n
i
t
i
a
l
co
n
v
o
lu
t
i
o
n
a
l
l
ay
e
r
s
w
h
i
l
e
r
ep
l
a
c
in
g
th
e
F
C
l
a
y
e
r
to
m
a
t
ch
th
e
n
u
m
b
e
r
o
f
a
k
s
a
r
a
Ja
w
a
c
l
a
s
s
e
s
.
T
r
a
i
n
in
g
a
c
cu
r
ac
y
g
r
a
p
h
i
l
l
u
s
t
r
a
t
e
s
t
h
e
m
o
d
e
l
's
l
e
a
r
n
in
g
p
r
o
g
r
e
s
s
i
o
n
o
v
e
r
m
u
l
t
i
p
l
e
ep
o
ch
s
,
s
h
o
w
in
g
b
o
t
h
t
h
e
tr
a
i
n
in
g
a
c
cu
r
a
c
y
(
b
l
u
e
li
n
e
)
a
n
d
v
a
l
id
a
t
io
n
a
c
c
u
r
a
c
y
(
o
r
an
g
e
l
in
e
)
.
I
n
i
t
i
al
l
y
,
b
o
th
c
u
r
v
e
s
e
x
h
i
b
i
t
a
s
t
ee
p
u
p
w
ar
d
t
r
en
d
,
in
d
i
ca
t
i
n
g
th
a
t
th
e
R
e
s
N
e
t
-
1
8
m
o
d
e
l
i
s
q
u
i
c
k
ly
l
e
ar
n
in
g
an
d
i
m
p
r
o
v
in
g
i
t
s
ab
i
l
i
ty
to
c
l
a
s
s
i
f
y
a
k
s
a
r
a
J
a
w
a
ch
a
r
ac
t
e
r
s
.
Ho
w
ev
e
r
,
th
e
v
a
l
i
d
a
t
io
n
ac
c
u
r
a
c
y
f
l
u
c
tu
a
t
es
,
s
u
g
g
e
s
t
in
g
t
h
a
t
th
e
m
o
d
e
l
o
c
c
a
s
i
o
n
a
l
l
y
o
v
er
f
i
t
s
t
o
th
e
t
r
a
i
n
i
n
g
d
a
t
a,
l
e
a
d
in
g
t
o
p
e
r
f
o
r
m
a
n
ce
v
a
r
i
a
t
io
n
s
o
n
u
n
s
e
en
v
a
l
i
d
a
t
i
o
n
s
a
m
p
l
e
s
.
A
r
o
u
n
d
e
p
o
ch
1
2
o
n
w
ar
d
,
b
o
t
h
c
u
r
v
e
s
s
t
ab
i
l
i
ze
a
n
d
co
n
v
e
r
g
e
n
ea
r
0
.
9
(
9
0
%)
a
c
c
u
r
a
c
y
,
d
e
m
o
n
s
tr
a
t
i
n
g
t
h
a
t
t
h
e
m
o
d
e
l
h
a
s
ef
f
e
c
t
iv
e
ly
le
a
r
n
e
d
t
h
e
p
a
t
t
er
n
s
i
n
t
h
e
d
a
t
a
s
e
t
.
T
h
e
f
l
u
c
tu
a
t
io
n
s
in
th
e
v
a
l
i
d
a
t
i
o
n
cu
r
v
e
in
d
ic
a
te
s
e
n
s
i
t
i
v
i
ty
to
s
p
e
c
i
f
ic
b
a
t
ch
e
s
,
w
h
i
ch
c
o
u
l
d
b
e
m
i
t
i
g
a
t
ed
b
y
r
eg
u
l
ar
i
z
a
t
io
n
te
c
h
n
iq
u
e
s
s
u
ch
a
s
d
r
o
p
o
u
t
o
r
f
u
r
t
h
er
h
y
p
e
r
p
a
r
am
e
t
er
t
u
n
i
n
g
.
Ov
e
r
a
l
l,
th
e
g
r
ap
h
co
n
f
i
r
m
s
th
e
m
o
d
e
l
's
s
t
r
o
n
g
g
e
n
er
a
l
i
z
a
t
io
n
ab
i
l
i
ty
,
w
i
t
h
c
o
n
s
i
s
t
e
n
t
t
r
a
i
n
in
g
an
d
v
a
l
i
d
a
t
io
n
a
c
c
u
r
a
cy
a
f
t
e
r
s
u
f
f
i
c
i
e
n
t
e
p
o
ch
s
.
I
t
c
an
b
e
s
e
en
i
n
F
i
g
u
r
e
4
.
Fig
u
r
e
4
.
T
r
ain
in
g
l
o
s
s
g
r
ap
h
T
r
ain
in
g
lo
s
s
g
r
ap
h
d
ep
icts
th
e
r
ed
u
ctio
n
in
tr
ain
i
n
g
(
b
l
u
e
lin
e)
an
d
v
alid
atio
n
lo
s
s
(
o
r
an
g
e
lin
e)
o
v
e
r
m
u
ltip
le
ep
o
c
h
s
,
d
em
o
n
s
tr
atin
g
h
o
w
well
th
e
R
esNet
-
1
8
m
o
d
el
m
i
n
im
izes
class
if
ica
tio
n
er
r
o
r
s
d
u
r
in
g
tr
ain
in
g
.
I
n
itially
,
b
o
th
lo
s
s
es
ar
e
h
ig
h
,
in
d
icatin
g
th
at
th
e
m
o
d
el
is
lear
n
in
g
to
ex
tr
ac
t
m
ea
n
in
g
f
u
l
f
ea
tu
r
es
f
r
o
m
th
e
a
ksa
r
a
J
a
w
a
d
ataset.
A
s
h
ar
p
d
ec
lin
e
in
th
e
f
ir
s
t
f
e
w
ep
o
ch
s
s
u
g
g
ests
r
a
p
id
im
p
r
o
v
em
en
t,
f
o
llo
we
d
b
y
a
m
o
r
e
g
r
ad
u
al
d
ec
r
ea
s
e
as
th
e
m
o
d
el
co
n
v
er
g
es.
T
h
e
v
alid
atio
n
lo
s
s
f
lu
ctu
ates,
r
e
f
lectin
g
o
cc
asio
n
al
o
v
er
f
itti
n
g
,
wh
er
e
t
h
e
m
o
d
el
lear
n
s
p
atter
n
s
in
th
e
tr
ain
i
n
g
d
ata
t
h
at
d
o
n
o
t
g
en
er
ali
ze
well
to
u
n
s
ee
n
v
alid
atio
n
d
ata.
Ho
wev
er
,
af
ter
ap
p
r
o
x
im
ately
e
p
o
ch
1
2
,
b
o
th
c
u
r
v
es
s
tab
ilize
n
ea
r
a
lo
w
lo
s
s
v
alu
e,
co
n
f
ir
m
in
g
th
at
th
e
m
o
d
el
h
a
s
ac
h
iev
ed
o
p
tim
al
co
n
v
er
g
e
n
ce
.
T
h
e
r
em
ain
in
g
f
lu
ctu
atio
n
s
in
v
alid
atio
n
lo
s
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
443
-
4
5
3
450
co
u
ld
b
e
m
itig
ated
with
ad
d
it
io
n
al
r
eg
u
lar
izatio
n
tech
n
iq
u
e
s
o
r
in
cr
ea
s
ed
t
r
ain
in
g
d
ata.
Ov
er
all,
th
is
g
r
ap
h
d
em
o
n
s
tr
ates th
at
th
e
m
o
d
el
is
ef
f
ec
tiv
ely
lear
n
in
g
wh
ile
a
v
o
id
in
g
ex
ce
s
s
iv
e
o
v
e
r
f
itti
n
g
.
E
v
a
l
u
a
t
i
o
n
m
e
t
r
i
c
s
p
r
e
s
e
n
t
e
d
in
t
h
e
F
i
g
u
r
e
5
,
w
h
i
c
h
i
n
cl
u
d
e
s
a
c
c
u
r
a
c
y
,
p
r
e
c
is
i
o
n
,
r
e
c
a
ll
,
a
n
d
F
1
-
s
c
o
r
e
.
T
h
ese
m
etr
ics
ar
e
u
s
ed
t
o
ass
ess
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
f
i
n
e
-
tu
n
ed
R
esNet
-
1
8
m
o
d
el
i
n
class
if
y
in
g
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
s
.
Acc
u
r
ac
y
r
ep
r
esen
ts
th
e
p
r
o
p
o
r
tio
n
o
f
co
r
r
ec
tly
class
if
ied
im
ag
es
am
o
n
g
th
e
to
tal
test
s
am
p
les,
p
r
o
v
id
in
g
a
n
o
v
er
al
l
p
er
f
o
r
m
a
n
ce
m
ea
s
u
r
e.
Pre
c
is
io
n
in
d
icate
s
th
e
r
atio
o
f
co
r
r
ec
tly
p
r
e
d
icted
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
s
to
th
e
to
tal
n
u
m
b
er
o
f
p
r
ed
icted
in
s
t
an
ce
s
,
en
s
u
r
in
g
m
i
n
im
al
FP
.
R
ec
all
m
ea
s
u
r
es
th
e
m
o
d
el’
s
ab
ilit
y
to
c
o
r
r
ec
tly
id
en
tify
tr
u
e
in
s
tan
ce
s
o
f
ea
ch
ch
ar
a
cter
,
h
ig
h
lig
h
ti
n
g
its
ef
f
ec
tiv
en
ess
in
m
in
im
izin
g
FN
.
F1
-
s
co
r
e,
wh
i
ch
is
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
an
d
r
ec
all,
p
r
o
v
id
es
b
a
lan
ce
d
ass
ess
m
en
t
o
f
th
e
m
o
d
el’
s
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
.
Fig
u
r
e
6
p
r
esen
ts
th
ese
m
etr
ics to
g
iv
e
a
c
lear
u
n
d
er
s
tan
d
in
g
o
f
th
e
m
o
d
el’
s
r
eliab
ilit
y
an
d
ef
f
ec
ti
v
en
ess
b
ef
o
r
e
p
r
o
ce
e
d
in
g
to
a
d
etailed
d
is
cu
s
s
io
n
o
f
th
e
r
esu
lts
.
Fig
u
r
e
5
.
E
v
alu
atio
n
o
f
th
e
f
in
e
-
tu
n
ed
R
esNet
-
18
Fig
u
r
e
6
.
C
o
n
f
u
s
io
n
m
atr
i
x
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Op
timiz
in
g
Ja
va
n
ese
s
crip
t re
co
g
n
itio
n
u
s
in
g
fin
e
-
tu
n
ed
R
es
N
et
-
1
8
a
n
d
tr
a
n
s
fer lea
r
n
in
g
(
N
u
r
F
a
tea
h
)
451
T
h
e
co
n
f
u
s
i
o
n
m
a
tr
i
x
p
r
o
v
id
e
s
a
d
e
ta
i
l
e
d
b
r
ea
k
d
o
wn
o
f
t
h
e
R
e
s
N
e
t
-
1
8
m
o
d
e
l
's
c
l
a
s
s
i
f
i
c
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
o
n
t
h
e
a
k
s
a
r
a
Ja
w
a
d
a
t
a
s
e
t
b
y
d
i
s
p
l
ay
i
n
g
t
h
e
n
u
m
b
er
o
f
c
o
r
r
e
c
t
an
d
in
c
o
r
r
e
c
t
p
r
ed
i
c
t
i
o
n
s
f
o
r
e
a
c
h
c
h
ar
a
c
t
er
c
l
a
s
s
.
T
h
e
d
i
a
g
o
n
a
l
v
a
l
u
e
s
r
ep
r
e
s
en
t
th
e
co
r
r
e
c
t
c
l
a
s
s
i
f
i
c
a
t
io
n
s
,
w
h
e
r
e
t
h
e
m
o
d
e
l
s
u
c
c
e
s
s
f
u
l
ly
i
d
e
n
t
i
f
ie
d
th
e
c
h
a
r
ac
t
e
r
s
,
w
i
th
m
o
s
t
v
a
l
u
es
b
e
i
n
g
h
ig
h
(
1
0
-
1
2
o
c
cu
r
r
e
n
c
e
s
p
e
r
c
la
s
s
)
,
i
n
d
i
c
a
t
in
g
s
tr
o
n
g
c
l
a
s
s
i
f
i
ca
t
i
o
n
a
c
cu
r
a
cy
.
T
h
e
o
f
f
-
d
i
ag
o
n
a
l
v
a
l
u
e
s
r
e
p
r
e
s
e
n
t
m
i
s
c
la
s
s
i
f
i
c
at
i
o
n
s
,
wh
e
r
e
s
o
m
e
c
h
a
r
ac
t
e
r
s
w
er
e
c
o
n
f
u
s
e
d
wi
t
h
o
th
e
r
s
,
o
f
t
en
d
u
e
t
o
v
i
s
u
a
l
s
i
m
i
l
a
r
i
t
ie
s
b
e
t
w
e
en
c
er
t
a
i
n
a
k
s
a
r
a
Ja
w
a
s
y
m
b
o
l
s
.
M
i
s
c
l
a
s
s
i
f
i
c
a
t
io
n
in
s
t
a
n
c
e
s
a
r
e
m
a
r
k
e
d
w
i
t
h
l
ig
h
t
e
r
g
r
e
e
n
s
h
a
d
e
s
,
s
h
o
w
in
g
lo
w
e
r
e
r
r
o
r
c
o
u
n
t
s
,
w
i
t
h
o
cc
a
s
i
o
n
a
l
er
r
o
r
s
i
n
th
e
'h
a
',
'j
a
',
a
n
d
's
a
'
c
l
a
s
s
e
s
d
u
e
t
o
t
h
e
ir
s
t
r
u
c
t
u
r
a
l
r
e
s
e
m
b
l
an
c
e
to
o
t
h
er
c
h
a
r
ac
t
e
r
s
.
T
h
e
m
a
t
r
ix
s
u
g
g
es
t
s
t
h
a
t
th
e
m
o
d
e
l
h
a
s
h
ig
h
p
r
e
c
i
s
i
o
n
a
n
d
r
e
c
a
l
l,
b
u
t
f
u
r
th
e
r
r
ef
i
n
em
e
n
t
—
s
u
c
h
a
s
a
d
d
i
t
io
n
a
l
tr
a
i
n
in
g
d
a
t
a,
f
e
a
t
u
r
e
e
x
tr
a
c
t
i
o
n
en
h
an
c
em
en
t
s
,
o
r
t
u
n
i
n
g
h
y
p
er
p
ar
a
m
e
te
r
s
—
c
o
u
ld
r
ed
u
c
e
m
i
s
c
l
a
s
s
i
f
i
c
a
t
i
o
n
r
a
te
s
a
n
d
f
u
r
t
h
er
i
m
p
r
o
v
e
th
e
m
o
d
e
l’
s
r
o
b
u
s
t
n
e
s
s
.
Mo
r
e
d
e
t
a
i
l
s
in
g
r
ap
h
i
c
f
o
r
m
c
an
b
e
s
e
e
n
i
n
F
i
g
u
r
e
5
.
E
v
a
l
u
a
t
i
o
n
o
f
t
h
e
f
i
n
e
-
t
u
n
e
d
R
e
s
N
e
t
-
1
8
m
o
d
e
l
f
o
r
a
k
s
a
r
a
J
a
w
a
c
l
a
s
s
i
f
i
c
a
t
i
o
n
d
e
m
o
n
s
t
r
a
t
e
s
i
t
s
h
i
g
h
r
e
l
i
a
b
i
l
i
t
y
a
n
d
b
a
l
a
n
c
e
d
p
e
r
f
o
r
m
a
n
c
e
a
c
r
o
s
s
k
e
y
m
e
t
r
i
c
s
:
p
r
e
c
i
s
i
o
n
,
r
e
c
a
l
l
,
F
1
-
s
c
o
r
e
,
a
n
d
a
c
c
u
r
a
c
y
.
T
h
e
m
o
d
e
l
a
c
h
i
e
v
e
d
a
n
a
c
c
u
r
a
c
y
o
f
9
1
%
,
i
n
d
i
c
a
t
i
n
g
i
t
s
o
v
e
r
a
l
l
e
f
f
e
c
t
i
v
e
n
e
s
s
i
n
c
o
r
r
e
c
t
l
y
c
l
a
s
s
i
f
y
i
n
g
c
h
a
r
a
c
t
e
r
s
.
T
h
e
p
r
e
c
i
s
i
o
n
s
c
o
r
e
o
f
9
3
%
r
e
f
l
e
c
t
s
t
h
e
m
o
d
e
l
’
s
a
b
i
l
i
t
y
t
o
m
i
n
i
m
i
z
e
FP
,
e
n
s
u
r
i
n
g
t
h
a
t
p
r
e
d
i
c
t
e
d
c
h
a
r
a
c
t
e
r
s
a
r
e
h
i
g
h
l
y
a
c
c
u
r
a
t
e
.
T
h
e
r
e
c
a
l
l
s
c
o
r
e
o
f
9
1
%
d
e
m
o
n
s
t
r
a
t
e
s
t
h
a
t
t
h
e
m
o
d
e
l
e
f
f
e
c
t
i
v
e
l
y
i
d
e
n
t
i
f
i
e
s
t
h
e
c
o
r
r
e
c
t
c
h
a
r
a
c
t
e
r
s
w
h
i
l
e
m
i
n
i
m
i
z
i
n
g
FN
.
T
h
e
F
1
-
s
c
o
r
e
o
f
9
1
%
,
w
h
i
c
h
b
a
l
a
n
c
e
s
p
r
e
c
i
s
i
o
n
a
n
d
r
e
c
a
l
l
,
c
o
n
f
i
r
m
s
t
h
a
t
t
h
e
m
o
d
e
l
m
a
i
n
t
a
i
n
s
s
t
r
o
n
g
c
l
a
s
s
i
f
i
c
a
t
i
o
n
p
e
r
f
o
r
m
a
n
c
e
a
c
r
o
s
s
a
l
l
c
h
a
r
a
c
t
e
r
c
l
a
s
s
e
s
.
T
h
e
s
e
m
e
t
r
i
c
s
c
o
l
l
e
c
t
i
v
e
l
y
i
n
d
i
c
a
t
e
t
h
a
t
t
h
e
R
e
s
N
e
t
-
1
8
m
o
d
e
l
,
o
p
t
i
m
i
z
e
d
w
i
t
h
t
h
e
A
d
a
m
a
l
g
o
r
i
t
h
m
,
e
f
f
e
c
t
i
v
e
l
y
g
e
n
e
r
a
l
i
z
e
s
a
c
r
o
s
s
d
i
f
f
e
r
e
n
t
a
k
s
a
r
a
J
a
w
a
c
h
a
r
a
c
t
e
r
s
,
e
n
s
u
r
i
n
g
h
i
g
h
r
e
c
o
g
n
i
t
i
o
n
a
c
c
u
r
a
c
y
f
o
r
O
C
R
a
p
p
l
i
c
a
t
i
o
n
s
.
4.
CO
NCLU
SI
O
N
T
h
is
s
tu
d
y
s
u
cc
ess
f
u
lly
im
p
l
em
en
ted
an
d
ev
alu
ate
d
a
f
in
e
-
tu
n
ed
R
esNet
-
1
8
m
o
d
el
wi
th
Ad
am
o
p
tim
izatio
n
f
o
r
a
ksa
r
a
Ja
w
a
ch
ar
ac
ter
class
if
icatio
n
.
T
h
e
m
o
d
el
d
em
o
n
s
tr
ated
h
ig
h
p
er
f
o
r
m
an
ce
,
ac
h
iev
in
g
9
1
%
ac
cu
r
ac
y
,
9
3
%
p
r
ec
is
io
n
,
9
1
%
r
ec
all,
a
n
d
9
1
%
F1
-
s
co
r
e,
co
n
f
ir
m
in
g
its
ef
f
ec
tiv
e
n
ess
in
r
ec
o
g
n
izin
g
co
m
p
lex
J
av
an
ese
s
cr
ip
t.
T
h
e
d
ata
p
r
ep
r
o
ce
s
s
in
g
s
tep
s
,
in
clu
d
in
g
r
esizin
g
,
n
o
r
m
aliza
tio
n
,
an
d
au
g
m
e
n
tatio
n
,
s
ig
n
if
ican
tly
co
n
tr
ib
u
ted
to
th
e
m
o
d
el'
s
r
o
b
u
s
tn
ess
.
Fin
e
-
tu
n
in
g
th
e
p
r
etr
ain
ed
R
esNet
-
1
8
b
y
f
r
ee
zin
g
in
itial
lay
er
s
an
d
m
o
d
if
y
in
g
t
h
e
FC
lay
er
en
a
b
led
ef
f
icien
t
f
ea
tu
r
e
ex
tr
ac
tio
n
,
wh
ile
th
e
Ad
am
o
p
tim
izer
p
r
o
v
id
ed
s
tab
le
an
d
ad
ap
tiv
e
weig
h
t
u
p
d
ates,
en
s
u
r
in
g
f
aster
co
n
v
e
r
g
en
ce
an
d
r
ed
u
ce
d
o
v
er
f
itti
n
g
.
T
h
e
c
o
n
f
u
s
io
n
m
atr
ix
an
d
lo
s
s
an
al
y
s
is
f
u
r
t
h
er
v
ali
d
ated
th
at
th
e
m
o
d
el
ef
f
ec
tiv
ely
d
is
tin
g
u
is
h
es
b
etwe
en
a
ksa
r
a
J
a
w
a
ch
ar
ac
ter
s
,
with
m
i
n
o
r
m
is
class
if
icatio
n
s
o
cc
u
r
r
in
g
in
v
is
u
ally
s
im
ilar
class
es.
Desp
it
e
th
ese
p
r
o
m
is
in
g
r
esu
lts
,
s
o
m
e
ch
allen
g
es
r
e
m
ain
,
p
a
r
ticu
lar
ly
in
ca
s
es
wh
er
e
s
im
ilar
c
h
ar
ac
ter
s
tr
u
ctu
r
es
lead
t
o
m
is
class
if
icatio
n
er
r
o
r
s
.
Fu
tu
r
e
r
esear
ch
ca
n
f
o
c
u
s
o
n
en
h
an
cin
g
th
e
m
o
d
el
b
y
in
co
r
p
o
r
ati
n
g
lar
g
er
an
d
m
o
r
e
d
iv
er
s
e
d
atasets
,
ex
p
er
im
en
ti
n
g
with
d
ee
p
e
r
ar
ch
itectu
r
es
lik
e
R
esNet
-
3
4
o
r
R
esNet
-
5
0
,
an
d
in
teg
r
atin
g
d
en
o
is
in
g
tec
h
n
iq
u
es
to
im
p
r
o
v
e
im
a
g
e
clar
ity
.
Ad
d
itio
n
a
lly
,
r
ea
l
-
wo
r
l
d
d
ep
lo
y
m
en
t
c
an
b
e
ex
p
l
o
r
ed
b
y
d
ev
elo
p
in
g
a
m
o
b
ile
o
r
web
-
b
ased
OC
R
ap
p
licatio
n
,
en
ab
lin
g
au
to
m
atic
d
ig
itizatio
n
o
f
J
av
an
ese
s
cr
ip
t
f
o
r
h
is
to
r
ical
an
d
cu
ltu
r
al
p
r
eser
v
atio
n
.
T
h
ese
ad
v
an
ce
m
e
n
ts
will
f
u
r
th
er
r
ef
in
e
th
e
ac
cu
r
a
cy
an
d
u
s
ab
ilit
y
o
f
d
ee
p
lear
n
i
n
g
m
o
d
els
f
o
r
a
ks
a
r
a
J
a
w
a
r
ec
o
g
n
itio
n
,
co
n
tr
ib
u
tin
g
to
th
e
b
r
o
ad
e
r
f
ield
o
f
I
n
d
o
n
esian
la
n
g
u
a
g
e
tech
n
o
lo
g
y
an
d
d
ig
ital h
er
ita
g
e
co
n
s
er
v
atio
n
.
ACK
NO
WL
E
DG
M
E
N
T
S
W
e
wo
u
ld
lik
e
to
e
x
p
r
ess
o
u
r
g
r
atitu
d
e
t
o
L
PP
M
UNNE
S
wh
ich
th
r
o
u
g
h
t
h
e
lectu
r
er
r
esear
c
h
p
r
o
g
r
a
m
with
ag
r
ee
m
en
t
n
u
m
b
er
:
7
0
.
2
6
.
2
/UN3
7
/PP
K.
1
0
/2
0
2
4
h
as
p
r
o
v
id
e
d
v
er
y
m
e
an
in
g
f
u
l
f
i
n
an
cial
s
u
p
p
o
r
t in
t
h
e
im
p
lem
e
n
tatio
n
o
f
th
is
r
esear
ch
s
o
th
at
o
u
r
r
esear
ch
ca
n
r
u
n
s
m
o
o
th
ly
a
n
d
s
u
cc
ess
f
u
lly
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
is
wo
r
k
was su
p
p
o
r
ted
b
y
t
h
e
L
PP
M
UNNE
S u
n
d
er
Gr
an
t N
o
.
7
0
.
2
6
.
2
/UN3
7
/PP
K.
1
0
/2
0
2
4
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
1
,
Feb
r
u
ar
y
20
26
:
443
-
4
5
3
452
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Nu
r
Fateah
✓
✓
✓
✓
✓
✓
✓
✓
Su
b
h
an
✓
✓
✓
✓
✓
✓
✓
✓
Yah
y
a
Nu
r
I
f
r
iza
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
t
h
o
r
s
d
ec
lar
e
th
at
th
e
y
h
av
e
n
o
k
n
o
wn
c
o
m
p
etin
g
f
in
an
cial
in
ter
ests
o
r
p
er
s
o
n
al
r
el
atio
n
s
h
ip
s
th
at
co
u
ld
h
av
e
ap
p
ea
r
ed
t
o
in
f
lu
en
ce
th
e
wo
r
k
r
e
p
o
r
te
d
in
t
h
is
p
ap
er
.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
av
ailab
l
e
f
r
o
m
Kag
g
le.
R
estrictio
n
s
ap
p
ly
to
th
e
av
ailab
ilit
y
o
f
th
ese
d
ata,
wh
ich
wer
e
u
s
ed
u
n
d
er
licen
s
e
f
o
r
th
is
s
tu
d
y
.
Data
ar
e
a
v
ailab
le
at
h
ttp
s
://www.
k
ag
g
le.
co
m
/d
atasets
/v
zr
en
g
g
am
an
i/h
a
n
ac
ar
ak
a/
d
ata
with
th
e
p
er
m
is
s
io
n
o
f
K
ag
g
le.
RE
F
E
R
E
NC
E
S
[
1
]
N
.
N
a
b
i
l
a
h
a
n
d
F
.
N
i
k
m
a
h
,
“
Th
e
r
e
l
a
t
i
o
n
s
h
i
p
o
f
a
k
s
a
r
a
J
a
w
a
a
s
l
o
c
a
l
f
o
l
k
l
o
r
e
w
i
t
h
m
o
d
e
r
a
t
e
a
n
d
p
r
o
g
r
e
ss
i
v
e
i
s
l
a
mi
c
e
d
u
c
a
t
i
o
n
,
”
An
n
u
a
l
I
n
t
e
r
n
a
t
i
o
n
a
l
C
O
n
f
e
r
e
n
c
e
o
n
I
sl
a
m
i
c
Ed
u
c
a
t
i
o
n
f
o
r
S
t
u
d
e
n
t
s
,
v
o
l
.
1
,
n
o
.
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
8
3
2
6
/
a
i
c
o
i
e
s.
v
1
i
1
.
2
5
7
.
[
2
]
B
.
S
e
t
i
a
w
a
n
,
P
.
W
i
b
o
w
o
,
F
.
A
z
z
a
h
r
a
h
,
P
.
H
a
r
j
a
n
i
,
a
n
d
K
.
S
a
d
d
h
o
n
o
,
“
S
t
u
d
y
o
f
Ja
v
a
n
e
se
s
c
r
i
p
t
w
r
i
t
i
n
g
l
e
a
r
n
i
n
g
i
n
se
n
i
o
r
h
i
g
h
sch
o
o
l
s i
n
B
o
y
o
l
a
l
i
R
e
g
e
n
c
y
,
”
i
n
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
L
o
c
a
l
W
i
sd
o
m
,
I
N
C
O
L
WIS
2
0
1
9
,
A
u
g
u
s
t
2
9
-
3
0
,
2
0
1
9
,
P
a
d
a
n
g
,
We
st
S
u
m
a
t
e
ra
,
I
n
d
o
n
e
si
a
,
2
0
1
9
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
4
1
0
8
/
e
a
i
.
2
9
-
8
-
2
0
1
9
.
2
2
8
8
9
0
6
.
[
3
]
S
.
R
o
b
s
o
n
,
“
Ja
v
a
n
e
s
e
scri
p
t
a
s
c
u
l
t
u
r
a
l
a
r
t
i
f
a
c
t
:
h
i
st
o
r
i
c
a
l
b
a
c
k
g
r
o
u
n
d
,
”
RI
MA
:
Re
v
i
e
w
o
f
I
n
d
o
n
e
s
i
a
n
a
n
d
M
a
l
a
y
s
i
a
n
A
f
f
a
i
rs
,
v
o
l
.
4
5
,
n
o
.
1
/
2
,
p
p
.
9
–
3
6
,
2
0
1
1
.
[
4
]
S
.
T.
W
i
d
o
d
o
a
n
d
A
.
P
u
r
w
a
n
t
o
r
o
,
“
T
h
e
c
o
n
c
e
p
t
o
f
e
n
v
i
r
o
n
me
n
t
a
l
c
o
n
s
e
r
v
a
t
i
o
n
i
n
t
h
e
Ja
v
a
n
e
s
e
c
o
mm
u
n
i
t
y
i
n
t
h
e
1
9
t
h
c
e
n
t
u
r
y
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
n
ser
v
a
t
i
o
n
S
c
i
e
n
c
e
,
v
o
l
.
1
0
,
n
o
.
3
,
p
p
.
4
4
9
–
4
5
8
,
2
0
1
9
.
[
5
]
T.
T.
H
.
N
g
u
y
e
n
,
A
.
J
a
t
o
w
t
,
M
.
C
o
u
s
t
a
t
y
,
a
n
d
A
.
D
o
u
c
e
t
,
“
S
u
r
v
e
y
o
f
p
o
s
t
-
O
C
R
p
r
o
c
e
ssi
n
g
a
p
p
r
o
a
c
h
e
s,
”
AC
M
C
o
m
p
u
t
i
n
g
S
u
r
v
e
y
s
,
v
o
l
.
5
4
,
n
o
.
6
,
p
p
.
1
–
3
7
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
5
3
4
7
6
.
[
6
]
S
.
M
a
h
a
d
e
v
k
a
r
,
S
.
P
a
t
i
l
,
a
n
d
K
.
K
o
t
e
c
h
a
,
“
E
n
h
a
n
c
e
m
e
n
t
o
f
h
a
n
d
w
r
i
t
t
e
n
t
e
x
t
r
e
c
o
g
n
i
t
i
o
n
u
si
n
g
A
I
-
b
a
sed
h
y
b
r
i
d
a
p
p
r
o
a
c
h
,
”
Me
t
h
o
d
s
X
,
v
o
l
.
1
2
,
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
me
x
.
2
0
2
4
.
1
0
2
6
5
4
.
[
7
]
J.
S
.
K
u
mar,
S
.
A
n
u
a
r
,
a
n
d
N
.
H
.
H
a
ssan
,
“
Tr
a
n
sf
e
r
l
e
a
r
n
i
n
g
b
a
se
d
p
e
r
f
o
r
man
c
e
c
o
m
p
a
r
i
s
o
n
o
f
t
h
e
p
r
e
-
t
r
a
i
n
e
d
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Ad
v
a
n
c
e
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
1
3
,
n
o
.
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
4
5
6
9
/
i
j
a
c
sa
.
2
0
2
2
.
0
1
3
0
1
9
3
.
[
8
]
G
.
N
.
A
.
K
e
sa
u
l
y
a
,
A
.
F
a
r
i
z
a
,
a
n
d
T
.
K
a
r
l
i
t
a
,
“
J
a
v
a
n
e
se
scr
i
p
t
t
e
x
t
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
u
s
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
2
0
2
2
I
n
t
e
r
n
a
t
i
o
n
a
l
El
e
c
t
ro
n
i
c
s
S
y
m
p
o
si
u
m
(
I
E
S
)
,
2
0
2
2
,
p
p
.
5
3
4
–
5
3
9
,
d
o
i
:
1
0
.
1
1
0
9
/
i
e
s
5
5
8
7
6
.
2
0
2
2
.
9
8
8
8
5
2
7
.
[
9
]
M
.
G
.
P
r
a
d
a
n
a
,
H
.
B
.
S
e
t
a
,
N
.
I
r
z
a
v
i
k
a
,
P
.
H
.
S
a
p
u
t
r
o
,
a
n
d
R
.
R
u
s
i
y
o
n
o
,
“
Le
v
e
n
s
h
t
e
i
n
d
i
st
a
n
c
e
a
l
g
o
r
i
t
h
m
i
n
J
a
v
a
n
e
se
c
h
a
r
a
c
t
e
r
t
r
a
n
s
l
a
t
i
o
n
m
a
c
h
i
n
e
b
a
s
e
d
o
n
o
p
t
i
c
a
l
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
,
”
J
O
I
V
:
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
n
I
n
f
o
rm
a
t
i
c
s
Vi
s
u
a
l
i
za
t
i
o
n
,
v
o
l
.
9
,
n
o
.
4
,
p
p
.
1
4
1
1
–
1
4
1
8
,
2
0
2
5
,
d
o
i
:
1
0
.
6
2
5
2
7
/
j
o
i
v
.
9
.
4
.
3
1
5
1
.
[
1
0
]
M
.
N
e
v
i
n
,
I
.
K
.
A
.
A
.
P
u
t
r
a
,
D
.
B
.
A
n
so
r
i
,
R
.
S
a
r
n
o
,
a
n
d
A
.
H
a
r
y
o
n
o
,
“
A
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
f
o
r
w
o
r
d
se
g
me
n
t
a
t
i
o
n
i
n
Jav
a
n
e
se
l
e
t
t
e
r
ma
n
u
scri
p
t
t
r
a
n
sl
i
t
e
r
a
t
i
o
n
,
”
i
n
2
0
2
4
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
T
e
c
h
n
o
l
o
g
y
I
n
n
o
v
a
t
i
o
n
a
n
d
I
t
s
Ap
p
l
i
c
a
t
i
o
n
s
(
I
C
T
I
I
A)
,
2
0
2
4
,
p
p
.
1
–
6
,
d
o
i
:
1
0
.
1
1
0
9
/
i
c
t
i
i
a
6
1
8
2
7
.
2
0
2
4
.
1
0
7
6
1
1
6
7
.
[
1
1
]
S
.
A
r
o
r
a
,
L
.
M
a
l
i
k
,
S
.
G
o
y
a
l
,
D
.
B
h
a
t
t
a
c
h
a
r
j
e
e
,
M
.
N
a
si
p
u
r
i
,
a
n
d
O
.
K
r
e
j
c
a
r
,
“
D
e
v
a
n
a
g
a
r
i
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
:
a
c
o
mp
r
e
h
e
n
s
i
v
e
l
i
t
e
r
a
t
u
r
e
r
e
v
i
e
w
,
”
I
E
EE
Ac
c
e
ss
,
v
o
l
.
1
3
,
p
p
.
1
2
4
9
–
1
2
8
4
,
2
0
2
5
,
d
o
i
:
1
0
.
1
1
0
9
/
a
c
c
e
ss.
2
0
2
4
.
3
5
2
0
2
4
8
.
[
1
2
]
K
.
D
u
t
t
a
,
P
.
K
r
i
s
h
n
a
n
,
M
.
M
a
t
h
e
w
,
a
n
d
C
.
V
.
Jawa
h
a
r
,
“
O
f
f
l
i
n
e
h
a
n
d
w
r
i
t
i
n
g
r
e
c
o
g
n
i
t
i
o
n
o
n
d
e
v
a
n
a
g
a
r
i
u
s
i
n
g
a
n
e
w
b
e
n
c
h
m
a
r
k
d
a
t
a
se
t
,
”
i
n
2
0
1
8
1
3
t
h
I
APR
I
n
t
e
r
n
a
t
i
o
n
a
l
Wo
r
k
sh
o
p
o
n
D
o
c
u
m
e
n
t
A
n
a
l
y
si
s
S
y
st
e
m
s
(
D
A
S
)
,
2
0
1
8
,
p
p
.
2
5
–
30
,
d
o
i
:
1
0
.
1
1
0
9
/
d
a
s.
2
0
1
8
.
6
9
.
[
1
3
]
İ
.
U
mu
t
a
n
d
Ü
.
C
.
K
u
m
d
e
r
e
l
i
,
“
N
o
v
e
l
w
e
a
r
a
b
l
e
s
y
s
t
e
m
t
o
r
e
c
o
g
n
i
z
e
si
g
n
l
a
n
g
u
a
g
e
i
n
r
e
a
l
t
i
m
e
,
”
S
e
n
s
o
rs
,
v
o
l
.
2
4
,
n
o
.
1
4
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
s
2
4
1
4
4
6
1
3
.
[
1
4
]
A
.
S
u
sa
n
t
o
,
I
.
U
.
W
.
M
u
l
y
o
n
o
,
C
.
A
.
S
a
r
i
,
E
.
H
.
R
a
c
h
maw
a
n
t
o
,
D
.
R
.
I
.
M
.
S
e
t
i
a
d
i
,
a
n
d
M
.
K
.
S
a
r
k
e
r
,
“
H
a
n
d
w
r
i
t
t
e
n
J
a
v
a
n
e
se
scri
p
t
r
e
c
o
g
n
i
t
i
o
n
me
t
h
o
d
b
a
s
e
d
1
2
-
l
a
y
e
r
s
d
e
e
p
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
a
n
d
d
a
t
a
a
u
g
me
n
t
a
t
i
o
n
,
”
I
AE
S
I
n
t
e
rn
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
2
,
n
o
.
3
,
p
p
.
1
4
4
8
–
1
4
5
8
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
a
i
.
v
1
2
.
i
3
.
p
p
1
4
4
8
-
1
4
5
8
.
[
1
5
]
S
.
B
a
s
o
d
i
,
C
.
Ji
,
H
.
Z
h
a
n
g
,
a
n
d
Y
.
P
a
n
,
“
G
r
a
d
i
e
n
t
a
mp
l
i
f
i
c
a
t
i
o
n
:
A
n
e
f
f
i
c
i
e
n
t
w
a
y
t
o
t
r
a
i
n
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
B
i
g
D
a
t
a
Mi
n
i
n
g
a
n
d
A
n
a
l
y
t
i
c
s
,
v
o
l
.
3
,
n
o
.
3
,
p
p
.
1
9
6
–
2
0
7
,
S
e
p
.
2
0
2
0
,
d
o
i
:
1
0
.
2
6
5
9
9
/
b
d
m
a
.
2
0
2
0
.
9
0
2
0
0
0
4
.
[
1
6
]
A
.
A
.
G
.
C
a
d
e
n
a
,
O
.
A
.
M
.
G
r
a
n
i
z
o
,
E.
P
.
C
ó
r
d
o
v
a
,
a
n
d
A
.
J
.
P
.
R
o
m
o
,
“
A
p
p
l
e
q
u
a
l
i
t
y
g
r
a
d
i
n
g
f
o
r
h
a
r
v
e
st
a
b
i
l
i
t
y
m
o
n
i
t
o
r
i
n
g
u
si
ng
c
o
m
p
u
t
e
r
v
i
s
i
ó
n
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
ms
,
”
I
n
g
e
n
i
a
re
R
e
v
i
st
a
C
h
i
l
e
n
a
d
e
I
n
g
e
n
i
e
r
í
a
,
v
o
l
.
3
1
,
p
p
.
1
–
1
8
,
2
0
2
3
,
d
o
i
:
1
0
.
4
0
6
7
/
s
0
7
1
8
-
3
3
0
5
2
0
2
3
0
0
0
1
0
0
2
1
5
.
[
1
7
]
M
.
C
.
S
.
Ta
n
g
a
n
d
S
.
S
.
T
e
o
h
,
“
B
r
a
i
n
t
u
m
o
r
d
e
t
e
c
t
i
o
n
f
r
o
m
M
R
I
i
ma
g
e
s
b
a
se
d
o
n
R
e
sN
e
t
1
8
,
”
i
n
2
0
2
3
6
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
n
f
o
rm
a
t
i
o
n
S
y
s
t
e
m
s
a
n
d
C
o
m
p
u
t
e
r
N
e
t
w
o
rks
(
I
S
C
O
N
)
,
M
a
r
.
2
0
2
3
,
p
p
.
1
–
5
,
d
o
i
:
1
0
.
1
1
0
9
/
i
sc
o
n
5
7
2
9
4
.
2
0
2
3
.
1
0
1
1
2
0
2
5
.
Evaluation Warning : The document was created with Spire.PDF for Python.