I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
4
,
A
u
g
u
s
t
20
25
,
p
p
.
4
2
0
2
~
4
2
1
2
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
4
.
pp
4
2
0
2
1
-
4
2
1
2
4202
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Enha
ncing
multi
-
cla
ss
t
ex
t
cla
ss
ific
a
tion in bio
medic
a
l lit
eratur
e
by
int
eg
ra
ting seq
uential a
nd
co
nte
x
tual l
ea
rning
wi
t
h BERT
a
nd LST
M
O
us
s
a
m
a
Nda
m
a
,
I
s
m
a
il B
e
ns
a
s
s
i,
Sa
f
a
e
Nda
m
a
,
E
l Mo
k
hta
r
E
n
-
Na
im
i
D
S
A
I
2
S
R
e
s
e
a
r
c
h
Te
a
m,
C
3
S
La
b
o
r
a
t
o
r
y
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
s
a
n
d
Te
c
h
n
o
l
o
g
i
e
s
o
f
Ta
n
g
i
e
r
,
A
b
d
e
l
ma
l
e
k
Essaâ
d
i
U
n
i
v
e
r
s
i
t
y
,
Te
t
o
u
a
n
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Oct
2
0
,
2
0
2
4
R
ev
is
ed
Ma
r
2
2
,
2
0
2
5
Acc
ep
ted
Ma
y
2
3
,
2
0
2
5
Clas
sifica
ti
o
n
o
f
se
n
ten
c
e
s
in
b
io
m
e
d
ica
l
a
b
stra
c
ts
in
to
p
re
d
e
fin
e
d
c
a
teg
o
ries
is
e
ss
e
n
ti
a
l
fo
r
e
n
h
a
n
c
in
g
re
a
d
a
b
i
li
ty
a
n
d
fa
c
il
it
a
ti
n
g
in
f
o
rm
a
ti
o
n
r
e
tri
e
v
a
l
in
sc
ien
ti
fic
li
tera
tu
re
.
We
p
r
o
p
o
se
a
n
o
v
e
l
h
y
b
r
id
m
o
d
e
l
th
a
t
in
teg
ra
tes
b
id
irec
ti
o
n
a
l
e
n
c
o
d
e
r
re
p
re
se
n
t
a
ti
o
n
s
fro
m
tran
sfo
rm
e
rs
(BE
RT)
fo
r
c
o
n
tex
t
u
a
l
lea
rn
i
n
g
,
l
o
n
g
sh
o
r
t
-
term
m
e
m
o
ry
(LS
TM
)
fo
r
se
q
u
e
n
ti
a
l
p
ro
c
e
ss
in
g
,
a
n
d
se
n
ten
c
e
o
r
d
e
r
in
fo
rm
a
ti
o
n
to
c
las
sify
se
n
ten
c
e
s
fro
m
b
io
m
e
d
ica
l
a
b
stra
c
ts.
Util
izin
g
t
h
e
P
u
b
M
e
d
2
0
0
k
ra
n
d
o
m
ize
d
c
o
n
t
ro
ll
e
d
tri
a
l
(R
CT)
d
a
tas
e
t,
o
u
r
m
o
d
e
l
a
c
h
iev
e
d
a
n
o
v
e
ra
ll
a
c
c
u
ra
c
y
o
f
8
8
.
4
2
%
,
d
e
m
o
n
stra
ti
n
g
str
o
n
g
p
e
rfo
rm
a
n
c
e
in
id
e
n
ti
fy
i
n
g
m
e
th
o
d
s
a
n
d
re
s
u
lt
s
se
c
ti
o
n
s
wh
il
e
m
a
in
tain
i
n
g
b
a
lan
c
e
d
p
re
c
isio
n
,
re
c
a
ll
,
a
n
d
F
1
-
sc
o
re
s
a
c
ro
ss
a
ll
c
a
teg
o
ries
.
Th
is
h
y
b
ri
d
a
p
p
ro
a
c
h
e
ffe
c
ti
v
e
ly
c
a
p
tu
re
s
b
o
th
c
o
n
te
x
tu
a
l
a
n
d
se
q
u
e
n
ti
a
l
p
a
tt
e
rn
s
o
f
b
io
m
e
d
i
c
a
l
tex
t,
o
ffe
rin
g
a
ro
b
u
st
so
l
u
ti
o
n
f
o
r
imp
ro
v
i
n
g
th
e
se
g
m
e
n
tati
o
n
o
f
sc
ien
ti
fic
a
b
stra
c
ts.
Th
e
m
o
d
e
l'
s
d
e
sig
n
p
ro
m
o
tes
sta
b
i
li
ty
a
n
d
g
e
n
e
ra
li
z
a
ti
o
n
,
m
a
k
i
n
g
it
a
n
e
ffe
c
ti
v
e
to
o
l
fo
r
a
u
to
m
a
ti
c
tex
t
c
las
sifica
ti
o
n
a
n
d
in
f
o
rm
a
ti
o
n
re
tri
e
v
a
l
i
n
b
io
m
e
d
ica
l
re
se
a
rc
h
.
Th
e
se
re
su
lt
s
u
n
d
e
rsc
o
re
th
e
m
o
d
e
l'
s
e
ffica
c
y
i
n
h
a
n
d
li
n
g
o
v
e
rlap
p
in
g
c
a
teg
o
r
ies
a
n
d
it
s
sig
n
if
ica
n
t
c
o
n
tri
b
u
ti
o
n
to
a
d
v
a
n
c
i
n
g
b
io
m
e
d
ica
l
tex
t
a
n
a
ly
sis.
K
ey
w
o
r
d
s
:
B
i
d
i
r
e
ct
i
o
n
a
l
e
n
c
o
d
e
r
r
e
p
r
e
s
e
n
t
a
ti
o
n
s
f
r
o
m
t
r
a
n
s
f
o
r
m
e
r
s
B
io
m
ed
ical
tex
t c
lass
if
icatio
n
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
Natu
r
al
lan
g
u
ag
e
p
r
o
ce
s
s
in
g
Sen
ten
ce
s
eg
m
en
tatio
n
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Ou
s
s
am
a
Nd
am
a
DSAI
2
S R
esear
ch
T
ea
m
,
C
3
S
L
ab
o
r
ato
r
y
,
Facu
lty
o
f
Scien
c
es a
n
d
T
ec
h
n
o
lo
g
ies
o
f
T
an
g
ie
r
,
Ab
d
elm
alek
E
s
s
aâ
d
i U
n
iv
er
s
ity
T
eto
u
an
,
M
o
r
o
cc
o
E
m
ail: o
u
s
s
am
a.
n
d
am
a@
etu
.
u
ae
.
ac
.
m
a
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
s
u
r
g
e
i
n
b
io
m
e
d
ical
liter
a
tu
r
e
h
as
m
ad
e
it
in
cr
ea
s
in
g
ly
d
if
f
icu
lt
to
e
x
tr
ac
t
v
alu
a
b
le
in
f
o
r
m
atio
n
ef
f
icien
tly
f
r
o
m
s
cien
tific
p
ap
er
s
.
As
th
e
v
o
lu
m
e
o
f
b
io
m
e
d
ical
r
esear
ch
c
o
n
tin
u
es
to
g
r
o
w,
m
an
y
ab
s
tr
ac
ts
ar
e
d
en
s
ely
wr
itten
,
m
ak
in
g
th
em
ch
allen
g
in
g
to
n
av
ig
ate
an
d
in
ter
p
r
et
[
1
]
,
[
2
]
.
T
h
is
is
s
u
e
is
co
m
p
o
u
n
d
e
d
b
y
th
e
ab
s
en
ce
o
f
s
tr
u
ctu
r
e
d
s
em
an
tic
f
r
am
ewo
r
k
s
in
th
ese
ab
s
tr
ac
ts
,
wh
ich
h
in
d
e
r
s
ef
f
ec
tiv
e
d
ata
r
etr
iev
al
a
n
d
co
m
p
r
eh
e
n
s
io
n
.
W
h
ile
a
v
a
r
iety
o
f
n
atu
r
al
lan
g
u
ag
e
p
r
o
c
ess
in
g
(
NL
P)
m
o
d
els
h
av
e
b
ee
n
d
e
v
elo
p
ed
to
ad
d
r
ess
th
ese
ch
allen
g
es
[
3
]
–
[
5
]
,
ex
is
tin
g
ap
p
r
o
ac
h
es
o
f
te
n
s
tr
u
g
g
le
to
b
alan
ce
th
e
d
e
m
an
d
s
o
f
co
n
tex
tu
al
u
n
d
er
s
tan
d
i
n
g
an
d
s
eq
u
en
tial i
n
f
o
r
m
atio
n
with
in
th
e
te
x
t.
T
o
ad
d
r
ess
th
ese
ch
allen
g
es,
th
is
r
esear
ch
in
tr
o
d
u
ce
s
a
n
o
v
el
h
y
b
r
id
m
o
d
el
th
at
i
n
teg
r
ates
th
e
s
tr
en
g
th
s
o
f
b
id
ir
ec
tio
n
al
en
co
d
er
r
ep
r
esen
tatio
n
s
f
r
o
m
tr
an
s
f
o
r
m
er
s
(
B
E
R
T
)
f
o
r
co
n
tex
tu
a
l
lear
n
in
g
an
d
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
(
L
STM
)
f
o
r
s
eq
u
en
tial
lear
n
in
g
.
T
h
i
s
co
m
b
in
atio
n
aim
s
to
ef
f
ec
tiv
ely
s
eg
m
en
t
an
d
class
if
y
b
io
m
ed
ical
r
esear
ch
p
ap
er
a
b
s
tr
ac
ts
,
o
p
tim
izin
g
th
eir
r
ea
d
ab
ilit
y
wh
ile
p
r
eser
v
in
g
ess
en
tial
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
n
h
a
n
cin
g
mu
lti
-
cla
s
s
text
cla
s
s
ifica
tio
n
in
b
io
med
ica
l litera
tu
r
e
b
y
…
(
Ou
s
s
a
ma
N
d
a
ma
)
4203
in
f
o
r
m
atio
n
.
T
h
e
m
o
d
el
le
v
er
ag
es
th
e
r
o
b
u
s
t
co
n
tex
t
u
al
u
n
d
er
s
tan
d
in
g
ca
p
a
b
ilit
ies
o
f
B
E
R
T
an
d
th
e
s
eq
u
en
tial
d
ep
en
d
en
cies
ca
p
t
u
r
ed
b
y
L
STM
,
o
f
f
er
i
n
g
a
m
o
r
e
n
u
an
ce
d
ap
p
r
o
ac
h
to
t
ex
t
s
eg
m
en
tatio
n
i
n
b
io
m
ed
ical
liter
atu
r
e.
T
h
is
s
tu
d
y
b
u
ild
s
u
p
o
n
th
e
Pu
b
Me
d
2
0
0
k
R
C
T
d
ataset,
a
wid
ely
u
s
ed
r
eso
u
r
ce
f
o
r
s
en
ten
ce
class
if
icatio
n
in
m
ed
ical
ab
s
tr
ac
ts
[
6
]
.
W
h
ile
th
e
o
r
ig
in
al
d
a
taset
co
m
p
r
is
es
2
.
3
m
illi
o
n
p
h
r
ases
f
r
o
m
2
0
0
,
0
0
0
r
an
d
o
m
ize
d
co
n
tr
o
lle
d
tr
ial
ab
s
tr
ac
ts
,
we
h
av
e
s
elec
ted
a
s
u
b
s
et
o
f
5
0
0
,
0
0
0
s
en
ten
ce
s
to
tr
ain
o
u
r
m
o
d
el.
E
ac
h
s
en
ten
ce
in
th
is
d
ataset
is
m
ap
p
ed
to
o
n
e
o
f
f
iv
e
p
r
e
d
e
f
in
ed
ca
teg
o
r
ies:
b
ac
k
g
r
o
u
n
d
,
o
b
jectiv
e,
m
eth
o
d
,
r
esu
lt,
o
r
co
n
clu
s
io
n
,
p
r
o
v
i
d
in
g
a
s
tr
u
ctu
r
ed
f
r
am
ewo
r
k
f
o
r
t
r
ain
in
g
an
d
ev
alu
atio
n
.
Ou
r
r
esear
ch
lev
er
ag
es
th
e
co
m
p
u
tatio
n
al
ca
p
ab
ilit
ies
o
f
th
e
Go
o
g
le
C
o
lab
A1
0
0
GPU
to
ef
f
icien
tly
tr
ain
an
d
f
in
e
-
tu
n
e
th
e
h
y
b
r
i
d
m
o
d
el
with
in
r
ea
l
-
wo
r
ld
r
eso
u
r
ce
co
n
s
tr
ain
ts
.
B
y
h
ar
n
ess
in
g
b
o
t
h
co
n
te
x
tu
al
an
d
s
eq
u
en
tial
f
ea
t
u
r
es,
th
is
ap
p
r
o
ac
h
ef
f
ec
tiv
ely
en
h
a
n
ce
s
th
e
r
ea
d
ab
ilit
y
an
d
ac
ce
s
s
ib
ilit
y
o
f
s
cien
tific
ab
s
tr
ac
ts
.
Ad
d
itio
n
ally
,
th
e
m
eth
o
d
m
ain
tain
s
co
m
p
u
tati
o
n
al
ef
f
icien
cy
,
en
s
u
r
in
g
th
a
t
th
e
h
y
b
r
i
d
m
o
d
el
r
em
ain
s
v
iab
le
f
o
r
p
r
ac
tical
a
p
p
licatio
n
s
.
T
h
e
p
r
i
m
ar
y
o
b
jectiv
e
o
f
th
is
wo
r
k
is
to
b
r
id
g
e
g
ap
s
in
cu
r
r
en
t
NL
P
ap
p
r
o
ac
h
es
b
y
o
f
f
er
i
n
g
a
h
y
b
r
id
m
o
d
el
th
at
ex
ce
ls
in
u
n
d
er
s
ta
n
d
in
g
co
n
tex
t
wh
ile
p
r
eser
v
in
g
th
e
lo
g
ical
f
lo
w
o
f
b
io
m
e
d
ical
tex
t.
T
h
is
m
o
d
el
im
p
r
o
v
es
th
e
s
eg
m
e
n
tatio
n
o
f
b
io
m
ed
ical
ab
s
tr
ac
ts
b
y
ac
cu
r
ately
d
is
s
ec
tin
g
th
eir
s
tr
u
ctu
r
al
an
d
s
em
an
tic
co
m
p
o
n
en
ts
.
Fu
r
th
er
m
o
r
e,
it
a
im
s
to
s
et
a
n
ew
s
tan
d
a
r
d
f
o
r
r
ea
d
ab
ilit
y
an
d
co
m
p
r
eh
e
n
s
io
n
with
in
th
e
f
ield
o
f
b
io
m
ed
ical
r
esear
ch
.
2.
L
I
T
E
R
AT
U
RE
R
E
VI
E
W
B
io
m
ed
ical
tex
t
clas
s
if
icatio
n
h
as
s
ee
n
s
ig
n
if
ican
t
ad
v
an
ce
m
en
ts
,
p
ar
ticu
lar
ly
with
th
e
in
teg
r
atio
n
o
f
m
ac
h
in
e
lear
n
in
g
an
d
d
ee
p
le
ar
n
in
g
m
o
d
els.
R
io
s
an
d
Kav
u
lu
r
u
[
7
]
(
co
n
v
o
l
u
tio
n
al
n
e
u
r
al
n
etwo
r
k
s
(
C
NNs)
f
o
r
b
io
m
e
d
ical
tex
t
class
if
ica
tio
n
)
d
em
o
n
s
tr
ated
th
e
ef
f
ec
tiv
en
ess
o
f
C
NN
s
in
as
s
ig
n
in
g
m
ed
ical
s
u
b
ject
h
ea
d
in
g
s
(
Me
SH)
to
b
io
m
ed
ical
ar
ticles,
o
u
tp
er
f
o
r
m
i
n
g
t
r
ad
itio
n
al
m
et
h
o
d
s
lik
e
lo
g
is
tic
r
eg
r
ess
io
n
a
n
d
s
u
p
p
o
r
t
v
ec
t
o
r
m
ac
h
i
n
es
b
y
i
m
p
r
o
v
i
n
g
m
ac
r
o
F
-
s
co
r
es.
T
h
is
wo
r
k
em
p
h
asized
th
e
ad
v
a
n
tag
es
o
f
C
NNs
in
h
an
d
lin
g
lar
g
e
f
ea
t
u
r
e
s
p
ac
es
an
d
co
m
p
lex
b
i
o
m
ed
ical
tex
t s
tr
u
ctu
r
es.
On
th
e
o
th
er
h
an
d
,
Dr
am
é
e
t
a
l.
[
8
]
e
x
p
lo
r
e
d
a
k
-
n
ea
r
es
t
n
eig
h
b
o
r
s
(
k
NN)
b
ased
a
n
d
ex
p
licit
s
em
an
tic
an
aly
s
is
(
E
SA)
b
ase
d
ap
p
r
o
ac
h
f
o
r
lar
g
e
-
s
ca
le
b
io
m
ed
ical
tex
t
class
if
icatio
n
.
T
h
eir
k
NN
ap
p
r
o
ac
h
,
co
m
b
in
ed
with
r
an
d
o
m
f
o
r
est (
R
F),
ac
h
iev
ed
co
m
p
etitiv
e
p
e
r
f
o
r
m
a
n
ce
with
an
F
-
m
ea
s
u
r
e
o
f
0
.
5
5
,
wh
ile
th
ei
r
E
SA
m
eth
o
d
u
n
d
er
p
er
f
o
r
m
ed
.
T
h
eir
s
tu
d
y
h
ig
h
lig
h
te
d
th
e
o
n
g
o
in
g
c
h
allen
g
e
o
f
u
s
in
g
p
ar
tial
in
f
o
r
m
atio
n
to
class
if
y
d
o
cu
m
en
ts
in
th
e
b
io
m
ed
ical
d
o
m
ain
.
I
n
a
b
r
o
ad
er
r
e
v
iew
o
f
b
io
m
ed
ical
tex
t
m
in
in
g
,
C
o
h
en
[
9
]
s
u
m
m
ar
ized
th
e
cu
r
r
en
t
p
r
o
g
r
ess
in
ap
p
ly
in
g
tex
t
m
in
in
g
tech
n
i
q
u
es
to
task
s
lik
e
n
am
e
d
en
tit
y
r
ec
o
g
n
itio
n
,
tex
t
class
if
icati
o
n
,
a
n
d
h
y
p
o
th
esis
g
en
er
atio
n
.
T
h
ey
h
ig
h
lig
h
ted
s
u
b
s
tan
tial
ad
v
an
ce
m
en
ts
in
co
m
p
u
tatio
n
al
m
eth
o
d
o
lo
g
ies
an
d
alg
o
r
ith
m
s
,
en
ab
lin
g
m
o
r
e
ef
f
ec
ti
v
e
ex
tr
a
ctio
n
o
f
m
ea
n
in
g
f
u
l
p
atter
n
s
f
r
o
m
b
io
m
ed
ical
tex
ts
.
Ho
wev
er
,
th
ey
n
o
ted
th
at
d
esp
ite
th
ese
ad
v
an
ce
m
e
n
ts
,
co
n
s
id
er
ab
le
ch
allen
g
es
p
er
s
i
s
t,
p
ar
ticu
lar
ly
in
im
p
r
o
v
in
g
s
y
s
tem
u
s
ab
ilit
y
f
o
r
b
io
m
ed
ical
r
esear
ch
er
s
an
d
en
h
an
cin
g
ac
ce
s
s
to
f
u
ll
-
tex
t
ar
ticles,
wh
ich
ar
e
cr
itical
b
ar
r
ier
s
lim
itin
g
wid
esp
r
ea
d
ad
o
p
tio
n
a
n
d
p
r
ac
tical
u
tili
ty
o
f
b
io
m
e
d
ical
tex
t
m
in
in
g
to
o
ls
.
Mo
n
d
al
in
tr
o
d
u
ce
d
b
io
m
ed
ica
l
B
E
R
T
-
b
ased
ad
v
er
s
ar
i
al
ex
am
p
le
g
en
er
atio
n
(
B
B
AE
G)
[
1
0
]
,
a
n
o
v
el
ad
v
er
s
ar
ial
ex
am
p
le
g
en
e
r
atio
n
tech
n
iq
u
e
s
p
ec
if
ically
f
o
r
b
io
m
ed
ical
tex
t
class
if
icati
o
n
.
B
y
lev
er
ag
in
g
B
E
R
T
-
m
ask
ed
lan
g
u
ag
e
m
o
d
el
(
ML
M)
p
r
ed
ictio
n
s
an
d
s
y
n
o
n
y
m
r
ep
lace
m
en
t
f
o
r
b
io
m
ed
ical
en
titi
es,
B
B
AE
G
d
em
o
n
s
tr
ated
th
e
p
o
t
en
tial
o
f
g
en
e
r
atin
g
r
o
b
u
s
t
ad
v
er
s
ar
ial
attac
k
s
th
at
co
u
l
d
ex
p
o
s
e
v
u
ln
er
ab
ilit
ies
in
cu
r
r
en
t
b
io
m
e
d
ical
NL
P
m
o
d
els,
h
ig
h
lig
h
tin
g
th
e
n
ee
d
f
o
r
m
o
r
e
r
esil
ien
t
p
r
ed
icti
v
e
s
y
s
tem
s
.
Fu
r
th
er
ad
v
an
ce
m
e
n
ts
in
b
i
o
m
ed
ical
m
u
lti
-
lab
el
class
if
icatio
n
wer
e
ex
p
lo
r
ed
b
y
Z
h
an
g
et
a
l.
[
1
1
]
,
wh
o
in
tr
o
d
u
ce
d
a
m
u
lti
-
lay
er
s
elf
-
atten
tio
n
m
ec
h
an
is
m
co
m
b
in
ed
with
B
E
R
T
to
en
h
an
ce
class
if
icatio
n
ac
cu
r
ac
y
.
T
h
eir
m
o
d
el
o
u
tp
er
f
o
r
m
ed
b
ase
lin
es
in
asp
ec
t
ca
teg
o
r
y
d
etec
tio
n
an
d
b
i
o
m
ed
ical
d
o
cu
m
en
t
class
if
icatio
n
,
s
h
o
wca
s
in
g
th
e
u
tili
ty
o
f
s
elf
-
atten
tio
n
f
o
r
ca
p
tu
r
in
g
co
m
p
lex
d
ep
en
d
en
cies in
b
io
m
e
d
ical
tex
ts
.
Neu
m
an
n
et
a
l.
[
1
2
]
co
n
tr
i
b
u
t
ed
s
ig
n
if
ican
tly
to
th
e
f
ield
with
Scis
p
aCy
,
a
s
p
ec
ialized
Py
t
h
o
n
lib
r
a
r
y
b
u
ilt
u
p
o
n
s
p
aCy
,
o
p
tim
ized
s
p
ec
if
ically
f
o
r
p
r
o
ce
s
s
in
g
b
i
o
m
ed
ical
tex
ts
.
Scis
p
aCy
p
r
o
v
id
es
f
ast,
s
ca
lab
le
m
o
d
els
ac
h
iev
in
g
n
ea
r
-
s
tate
-
of
-
th
e
-
a
r
t
p
er
f
o
r
m
an
ce
ac
r
o
s
s
m
u
ltip
le
b
io
m
ed
ical
NL
P
task
s
,
s
u
ch
as
n
am
ed
en
tity
r
ec
o
g
n
itio
n
an
d
p
ar
s
in
g
.
C
o
n
s
eq
u
en
tly
,
it
s
er
v
es
a
s
a
r
o
b
u
s
t
an
d
h
ig
h
l
y
ac
ce
s
s
ib
le
to
o
l,
f
ac
ilit
atin
g
wid
er
ad
o
p
tio
n
am
o
n
g
b
io
m
ed
ical
r
esear
ch
er
s
an
d
p
r
ac
titi
o
n
er
s
.
Do
cu
m
en
t
-
lev
el
b
io
m
ed
ical
r
elatio
n
ex
tr
ac
tio
n
was
s
y
s
te
m
atica
lly
ad
d
r
ess
ed
b
y
Yu
an
et
a
l.
[
1
3
]
th
r
o
u
g
h
th
e
in
tr
o
d
u
ctio
n
o
f
t
h
e
HT
GR
S
f
r
am
ewo
r
k
,
wh
ic
h
em
p
lo
y
s
h
ier
a
r
ch
ical
tr
ee
g
r
a
p
h
s
an
d
a
d
e
d
icate
d
r
elatio
n
s
eg
m
en
tatio
n
m
o
d
u
l
e.
T
h
eir
f
r
a
m
ew
o
r
k
s
tr
ateg
i
ca
lly
m
o
d
els
in
ter
ac
tio
n
s
b
e
twee
n
en
tity
p
air
s
,
s
ig
n
if
ican
tly
en
h
an
cin
g
th
e
ac
cu
r
ac
y
o
f
p
r
ed
ictin
g
r
elatio
n
s
ac
r
o
s
s
m
u
ltip
le
b
io
m
ed
ical
en
titi
es.
E
x
p
er
im
en
tal
ev
alu
atio
n
s
d
em
o
n
s
tr
ated
th
a
t
th
eir
m
eth
o
d
c
o
n
s
is
ten
tly
o
u
tp
er
f
o
r
m
ed
p
r
e
v
io
u
s
s
tate
-
o
f
-
th
e
-
a
r
t
m
o
d
els,
u
n
d
er
s
co
r
i
n
g
th
e
v
alu
e
o
f
s
tr
u
ctu
r
al
m
o
d
elin
g
in
b
i
o
m
ed
ical
r
elatio
n
ex
tr
ac
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
25
:
4
2
0
2
-
4212
4204
Du
an
et
a
l.
[
1
4
]
tack
led
th
e
c
h
allen
g
e
o
f
s
eq
u
en
tial
s
en
ten
ce
class
if
icatio
n
in
b
i
o
m
ed
ica
l
liter
atu
r
e
b
y
p
r
o
p
o
s
in
g
th
e
b
o
u
n
d
ar
y
-
a
war
e
d
u
al
b
iaf
f
in
e
m
o
d
el.
T
h
eir
in
n
o
v
ativ
e
ap
p
r
o
ac
h
ef
f
ec
tiv
ely
lev
er
ag
ed
d
o
cu
m
e
n
t
s
tr
u
ctu
r
al
in
f
o
r
m
ati
o
n
,
en
a
b
lin
g
p
r
ec
is
e
d
etec
tio
n
o
f
s
en
ten
ce
b
o
u
n
d
a
r
ies
an
d
r
elatio
n
s
h
ip
s
.
T
h
is
m
eth
o
d
n
o
tab
ly
r
ed
u
ce
d
class
if
icati
o
n
er
r
o
r
s
,
p
a
r
ticu
lar
ly
i
n
co
m
p
lex
b
io
m
ed
ical
d
o
c
u
m
en
ts
ch
ar
ac
ter
ized
b
y
in
tr
icate
s
en
ten
ce
s
eq
u
en
ce
s
a
n
d
r
elatio
n
s
h
ip
s
.
Fin
al
ly
,
W
an
g
e
t
a
l.
[
1
5
]
p
r
o
v
id
ed
a
co
m
p
r
eh
en
s
iv
e
s
u
r
v
ey
o
n
th
e
u
s
e
o
f
p
r
e
-
tr
a
in
e
d
lan
g
u
ag
e
m
o
d
el
s
(
PL
Ms
)
in
b
io
m
ed
ica
l
ap
p
l
ic
at
io
n
s
.
T
h
ey
ca
teg
o
r
iz
ed
th
e
ex
i
s
t
in
g
b
io
m
ed
ica
l
P
L
M
s
an
d
d
i
s
c
u
s
s
e
d
th
e
ir
ap
p
li
ca
tio
n
s
in
v
ar
io
u
s
t
ask
s
,
n
o
t
in
g
b
o
t
h
th
e
ad
v
an
c
em
en
t
s
an
d
l
im
it
at
io
n
s
i
n
th
e
f
ie
ld
.
T
h
i
s
s
u
r
v
e
y
em
p
h
a
s
ize
d
th
e
i
m
p
o
r
tan
ce
o
f
cr
o
s
s
-
d
is
ci
p
l
in
ar
y
co
ll
ab
o
r
at
io
n
t
o
d
r
i
v
e
f
u
r
th
er
i
n
n
o
v
a
tio
n
in
b
io
m
ed
i
ca
l
NL
P.
3.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
s
ec
tio
n
o
u
tlin
es
th
e
m
eth
o
d
o
lo
g
y
em
p
lo
y
ed
to
d
ev
el
o
p
an
d
ev
alu
ate
th
e
h
y
b
r
id
m
o
d
el
u
s
ed
f
o
r
s
eg
m
en
tin
g
an
d
class
if
y
in
g
b
i
o
m
ed
ical
r
esear
ch
p
ap
e
r
ab
s
tr
ac
ts
.
T
h
e
m
o
d
el
in
teg
r
ates
B
E
R
T
f
o
r
co
n
tex
tu
al
lear
n
in
g
an
d
L
STM
f
o
r
s
eq
u
en
tial
lear
n
in
g
to
en
h
an
ce
th
e
r
ea
d
ab
ilit
y
an
d
s
eg
m
en
tatio
n
o
f
th
ese
ab
s
tr
ac
ts
.
W
e
u
tili
ze
d
th
e
Pu
b
Me
d
2
0
0
k
R
C
T
d
ata
s
et
as
a
b
en
ch
m
ar
k
,
f
o
cu
s
in
g
o
n
a
s
u
b
s
et
o
f
5
0
0
,
0
0
0
s
en
ten
ce
s
to
en
s
u
r
e
co
m
p
u
tatio
n
al
ef
f
icien
cy
wh
ile
m
ain
tain
in
g
r
o
b
u
s
t
m
o
d
el
p
er
f
o
r
m
a
n
ce
.
T
h
e
m
o
d
el
was
tr
ain
ed
u
s
in
g
Go
o
g
le
C
o
lab
’
s
A1
0
0
GPU,
a
d
h
er
in
g
to
c
o
n
s
tr
ain
ts
o
f
r
eso
u
r
ce
av
ailab
ilit
y
a
n
d
co
m
p
u
tatio
n
al
ef
f
icien
c
y
.
I
n
th
is
s
ec
tio
n
,
we
d
etail
th
e
d
a
taset,
m
o
d
el
ar
c
h
itectu
r
e,
tr
ai
n
in
g
p
r
o
ce
d
u
r
e,
an
d
ev
al
u
atio
n
m
etr
ics
u
s
ed
to
ass
es
s
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
h
y
b
r
id
m
o
d
el.
3
.
1
.
Da
t
a
s
et
T
h
e
d
ataset
u
s
ed
in
th
is
s
tu
d
y
is
th
e
Pu
b
Me
d
2
0
0
k
R
C
T
d
ataset,
a
lar
g
e
-
s
ca
le
r
eso
u
r
ce
d
esig
n
ed
f
o
r
s
eq
u
en
tial
s
en
ten
ce
class
if
icatio
n
in
b
io
m
e
d
ical
ab
s
tr
ac
ts
.
I
t
co
m
p
r
is
es
ap
p
r
o
x
im
ately
2
0
0
,
0
0
0
r
a
n
d
o
m
ize
d
co
n
tr
o
lled
tr
ial
ab
s
tr
ac
ts
,
to
talin
g
2
.
3
m
illi
o
n
s
en
ten
ce
s
.
E
ac
h
s
en
ten
ce
is
lab
eled
with
o
n
e
o
f
f
iv
e
p
r
ed
ef
i
n
ed
ca
teg
o
r
ies
:
b
ac
k
g
r
o
u
n
d
,
o
b
jec
tiv
e,
m
et
h
o
d
,
r
esu
lt,
o
r
co
n
cl
u
s
io
n
.
T
h
is
d
ataset
was
r
elea
s
ed
to
ad
d
r
ess
two
k
ey
ch
allen
g
es:
th
e
lack
o
f
lar
g
e
-
s
ca
le
d
atasets
f
o
r
s
eq
u
en
tia
l
s
h
o
r
t
-
tex
t
class
if
icatio
n
an
d
th
e
n
ee
d
f
o
r
b
etter
to
o
ls
to
h
elp
r
esear
c
h
er
s
ef
f
ici
en
tly
n
av
ig
ate
le
n
g
th
y
b
io
m
e
d
ical
ab
s
tr
ac
ts
.
Fo
r
th
e
p
u
r
p
o
s
es
o
f
th
is
r
ese
ar
ch
,
a
s
u
b
s
et
o
f
5
0
0
,
0
0
0
s
en
ten
ce
s
was
s
am
p
led
f
r
o
m
th
e
d
ataset
to
b
alan
ce
co
m
p
u
tatio
n
al
ef
f
icie
n
cy
an
d
m
o
d
el
p
e
r
f
o
r
m
an
ce
.
Sp
ec
if
ically
,
2
2
.
6
1
%
o
f
th
e
o
r
ig
in
al
d
ataset
was
s
elec
ted
f
o
r
tr
ain
in
g
,
wh
ich
r
e
s
u
lted
in
5
0
0
,
1
0
2
s
am
p
les
in
th
e
tr
ain
in
g
s
et
an
d
2
9
,
4
9
3
s
a
m
p
les
in
th
e
test
s
et.
E
ac
h
s
am
p
le
in
clu
d
es th
e
f
o
llo
win
g
f
ield
s
:
−
T
ex
t: T
h
e
s
en
ten
ce
f
r
o
m
t
h
e
a
b
s
tr
ac
t.
−
C
h
ar
s
: A
ch
ar
ac
ter
-
lev
el
r
ep
r
e
s
en
tatio
n
o
f
th
e
s
en
ten
ce
.
−
Or
d
er
: T
h
e
s
eq
u
en
tial p
o
s
itio
n
o
f
th
e
s
en
ten
ce
with
in
th
e
ab
s
tr
ac
t.
−
L
ab
el:
T
h
e
s
en
ten
ce
'
s
ca
teg
o
r
y
(
o
n
e
o
f
th
e
f
iv
e
p
r
ed
ef
i
n
ed
c
lass
es).
T
o
p
r
e
p
r
o
ce
s
s
th
e
d
ata,
we
e
m
p
lo
y
ed
a
d
u
al
-
lev
el
t
o
k
en
iza
tio
n
s
tr
ateg
y
,
t
r
an
s
f
o
r
m
i
n
g
t
h
e
s
en
ten
ce
s
at
b
o
th
th
e
wo
r
d
a
n
d
ch
a
r
ac
te
r
lev
els.
C
h
ar
ac
ter
-
lev
el
to
k
e
n
izatio
n
was
ac
h
iev
ed
u
s
in
g
a
T
ex
tVec
to
r
izatio
n
lay
er
,
co
n
f
ig
u
r
e
d
with
a
cu
s
to
m
v
o
ca
b
u
lar
y
c
o
n
s
is
t
in
g
o
f
d
ig
its
,
p
u
n
ctu
atio
n
m
ar
k
s
,
an
d
ASC
I
I
ch
ar
ac
ter
s
.
T
h
is
ap
p
r
o
ac
h
ca
p
tu
r
es
th
e
f
i
n
er
g
r
an
u
lar
ity
o
f
s
en
ten
ce
s
tr
u
ctu
r
e,
en
s
u
r
in
g
th
at
ev
er
y
i
n
d
iv
id
u
al
ch
ar
ac
ter
co
n
tr
ib
u
tes to
th
e
m
o
d
el'
s
u
n
d
er
s
tan
d
in
g
o
f
th
e
in
p
u
t.
Fo
r
wo
r
d
-
le
v
el
to
k
en
izatio
n
,
th
e
v
o
c
ab
u
lar
y
was
d
er
i
v
ed
f
r
o
m
a
clea
n
ed
v
e
r
s
io
n
o
f
th
e
d
ataset,
wh
er
e
p
u
n
ct
u
atio
n
an
d
u
n
n
ec
ess
ar
y
s
y
m
b
o
ls
wer
e
s
y
s
tem
at
ically
r
em
o
v
ed
to
s
tan
d
a
r
d
ize
th
e
tex
t
[
1
6
]
,
[
1
7
]
.
T
h
e
r
esu
ltin
g
s
eq
u
e
n
ce
s
o
f
wo
r
d
s
wer
e
p
ad
d
ed
to
alig
n
with
th
e
9
5
th
p
er
ce
n
tile
o
f
s
en
ten
ce
len
g
th
s
,
o
p
tim
izin
g
co
m
p
u
tatio
n
al
e
f
f
icien
cy
b
y
s
ettin
g
a
p
r
ac
tica
l
in
p
u
t
len
g
t
h
th
r
esh
o
l
d
.
T
h
i
s
ca
r
ef
u
l
ap
p
r
o
ac
h
en
s
u
r
ed
lo
n
g
er
s
en
ten
ce
s
wer
e
ef
f
ec
tiv
ely
ac
co
m
m
o
d
ate
d
with
o
u
t lo
s
in
g
ess
en
tial sem
an
tic
in
f
o
r
m
atio
n
.
L
ab
el
p
r
e
p
r
o
ce
s
s
in
g
in
v
o
lv
ed
co
n
v
er
tin
g
ca
teg
o
r
ical
lab
els
i
n
to
n
u
m
er
ical
v
alu
es
t
h
r
o
u
g
h
th
e
u
s
e
o
f
a
L
ab
elE
n
co
d
e
r
,
f
ac
ilit
atin
g
ef
f
icien
t
co
m
p
u
tatio
n
al
h
an
d
l
in
g
.
Su
b
s
eq
u
e
n
tly
,
th
ese
n
u
m
er
ical
v
alu
es
wer
e
tr
an
s
f
o
r
m
ed
v
ia
o
n
e
-
h
o
t
en
co
d
in
g
,
m
a
k
in
g
th
em
s
u
itab
le
f
o
r
th
e
m
u
lti
-
class
clas
s
i
f
icatio
n
task
[
1
8
]
.
C
o
n
s
eq
u
en
tly
,
th
is
s
tr
u
ctu
r
ed
lab
elin
g
ap
p
r
o
ac
h
en
ab
led
th
e
m
o
d
el
to
class
if
y
ea
ch
s
en
te
n
ce
ac
cu
r
ately
in
to
o
n
e
o
f
th
e
f
iv
e
p
r
ed
e
f
in
ed
ca
te
g
o
r
ies:
b
ac
k
g
r
o
u
n
d
,
o
b
jectiv
e,
m
eth
o
d
,
r
esu
lt,
o
r
c
o
n
clu
s
io
n
.
T
h
e
d
ataset
was
th
en
s
p
lit
in
t
o
tr
ain
in
g
,
v
alid
atio
n
,
an
d
te
s
t
s
ets,
with
an
8
0
/2
0
d
iv
is
io
n
b
etwe
en
tr
ain
in
g
an
d
v
alid
atio
n
.
T
h
is
b
alan
ce
d
s
p
lit
f
ac
ilit
ated
ef
f
ec
tiv
e
tr
ain
in
g
an
d
m
o
d
el
tu
n
i
n
g
wh
ile
p
r
eser
v
in
g
a
p
o
r
tio
n
o
f
th
e
d
ata
f
o
r
u
n
b
iased
ev
alu
atio
n
.
T
h
is
p
r
ep
r
o
ce
s
s
in
g
f
r
am
ewo
r
k
laid
th
e
f
o
u
n
d
atio
n
f
o
r
th
e
ef
f
icien
t tr
ain
in
g
o
f
th
e
h
y
b
r
id
B
E
R
T
-
L
STM
m
o
d
el,
wh
ich
i
s
d
etailed
in
th
e
f
o
llo
win
g
s
ec
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
n
h
a
n
cin
g
mu
lti
-
cla
s
s
text
cla
s
s
ifica
tio
n
in
b
io
med
ica
l litera
tu
r
e
b
y
…
(
Ou
s
s
a
ma
N
d
a
ma
)
4205
3
.
2
.
M
o
del a
rc
hite
ct
ure
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
in
te
g
r
a
tes
b
o
th
co
n
te
x
tu
al
an
d
s
eq
u
en
tial
lear
n
in
g
to
ef
f
ec
tiv
ely
class
if
y
s
en
ten
ce
s
f
r
o
m
b
io
m
ed
ical
ab
s
tr
ac
ts
.
T
h
is
h
y
b
r
id
ar
ch
itectu
r
e
co
m
b
in
es
th
e
s
tr
en
g
th
s
o
f
B
E
R
T
f
o
r
co
n
tex
tu
al
em
b
ed
d
in
g
a
n
d
L
STM
f
o
r
s
eq
u
en
tial
u
n
d
er
s
tan
d
in
g
.
I
n
ad
d
it
io
n
,
it
in
co
r
p
o
r
ates
s
en
ten
ce
o
r
d
er
i
n
f
o
r
m
atio
n
to
f
u
r
th
er
im
p
r
o
v
e
class
if
icatio
n
p
er
f
o
r
m
an
ce
.
3
.
2
.
1
.
B
E
RT
enco
der
f
o
r
co
n
t
ex
t
ua
l le
a
rning
T
h
e
co
r
e
o
f
th
e
m
o
d
el
is
th
e
B
E
R
T
en
co
d
er
,
a
s
tate
-
of
-
th
e
-
ar
t
m
o
d
el
k
n
o
wn
f
o
r
its
ab
ilit
y
to
ca
p
tu
r
e
d
ee
p
co
n
tex
t
u
al
r
elatio
n
s
h
ip
s
in
tex
t
[
1
9
]
.
Fo
r
th
is
r
esear
ch
,
we
u
s
e
th
e
p
r
e
-
tr
ain
ed
"
b
ert_
b
a
s
e_
en
_
u
n
ca
s
ed
"
m
o
d
el,
wh
ich
is
f
in
e
-
tu
n
ed
o
n
th
e
b
io
m
ed
ical
tex
t
d
ataset
t
o
ad
ap
t
it
to
d
o
m
ain
-
s
p
ec
if
ic
lan
g
u
ag
e
[
2
0
]
.
T
h
e
B
E
R
T
to
k
en
izer
an
d
p
r
ep
r
o
ce
s
s
o
r
ar
e
in
itialized
with
a
s
eq
u
en
ce
len
g
t
h
o
f
2
5
6
to
k
en
s
,
en
s
u
r
in
g
th
at
s
en
ten
ce
s
ar
e
tr
u
n
ca
te
d
o
r
p
a
d
d
ed
t
o
a
c
o
n
s
is
ten
t
len
g
th
[
2
1
]
,
[
2
2
]
.
T
h
e
B
E
R
T
en
co
d
e
r
tak
es
as
in
p
u
t
th
e
to
k
en
ized
s
en
ten
ce
=
{
1
,
2
,
…
,
}
,
wh
er
e
r
ep
r
esen
ts
th
e
-
th
to
k
en
in
t
h
e
-
th
s
en
ten
ce
.
T
h
e
B
E
R
T
m
o
d
el
g
e
n
er
ates a
d
ee
p
c
o
n
tex
tu
al
em
b
ed
d
i
n
g
:
ℎ
be
r
t
=
B
E
R
T
(
)
∈
br
(
1
)
wh
er
e
be
r
t
is
th
e
d
im
en
s
io
n
ality
o
f
th
e
B
E
R
T
em
b
ed
d
in
g
s
p
ac
e
(
7
6
8
in
th
e
b
ase
m
o
d
el)
.
T
h
e
p
o
o
led
o
u
tp
u
t
fr
o
m
th
e
B
E
R
T
en
co
d
er
,
ℎ
be
r
t
,
is
p
ass
ed
th
r
o
u
g
h
a
f
u
lly
co
n
n
ec
ted
d
en
s
e
lay
er
:
ℎ
de
ns
e
=
R
eL
U
(
1
ℎ
be
r
t
+
1
)
(
2
)
wh
er
e
1
∈
dne
×
br
an
d
1
∈
dne
ar
e
th
e
lear
n
ab
le
weig
h
ts
an
d
b
iases
o
f
th
e
d
en
s
e
lay
er
.
W
e
ap
p
ly
L
2
r
eg
u
lar
izatio
n
to
p
r
ev
e
n
t o
v
er
f
itti
n
g
,
f
o
llo
wed
b
y
a
d
r
o
p
o
u
t la
y
er
to
f
u
r
t
h
er
im
p
r
o
v
e
g
en
er
aliza
tio
n
.
3
.
2
.
2
.
L
ST
M
f
o
r
s
equentia
l l
ea
rning
I
n
p
ar
allel
with
B
E
R
T
’
s
co
n
tex
tu
al
em
b
ed
d
in
g
,
th
e
m
o
d
el
in
co
r
p
o
r
ates a
n
L
STM
n
etwo
r
k
to
ca
p
tu
r
e
wo
r
d
-
lev
el
s
eq
u
en
tial
d
e
p
en
d
en
cies
[
2
3
]
,
[
2
4
]
.
T
h
e
s
en
ten
c
e
is
f
ir
s
t
to
k
en
ized
in
to
wo
r
d
s
,
wh
ich
ar
e
th
en
em
b
ed
d
e
d
in
to
a
1
2
8
-
d
im
en
s
i
o
n
al
v
ec
to
r
s
p
ac
e
u
s
in
g
a
n
em
b
ed
d
in
g
lay
er
:
wor
d
=
E
m
b
ed
d
in
g
(
)
∈
×
eb
d
(
3
)
wh
er
e
is
th
e
n
u
m
b
er
o
f
w
o
r
d
s
in
th
e
s
en
ten
ce
an
d
e
mbed
=
128
is
th
e
d
im
en
s
io
n
ality
o
f
th
e
wo
r
d
em
b
ed
d
in
g
s
p
ac
e.
T
h
e
em
b
ed
d
ed
s
eq
u
en
ce
is
p
r
o
ce
s
s
ed
b
y
th
e
L
STM
lay
er
to
ca
p
tu
r
e
tem
p
o
r
a
l
d
ep
en
d
e
n
cies:
ℎ
ls
tm
=
L
STM
(
wor
d
)
∈
lt
(
4
)
wh
er
e
ls
tm
=
32
r
ep
r
esen
ts
th
e
h
id
d
e
n
s
tate
s
ize
o
f
th
e
L
STM
.
T
h
is
c
ap
tu
r
es
s
eq
u
en
tial
p
atter
n
s
th
a
t
ar
e
n
o
t
ex
p
licitly
m
o
d
eled
b
y
th
e
B
E
R
T
en
co
d
er
.
T
h
e
L
STM
o
u
tp
u
t
is
p
ass
ed
th
r
o
u
g
h
a
f
u
lly
co
n
n
ec
ted
d
en
s
e
lay
er
with
1
6
u
n
its
:
ℎ
ls
tm
_
de
ns
e
=
R
eL
U
(
2
ℎ
ls
tm
+
2
)
(
5
)
A
d
r
o
p
o
u
t la
y
er
with
a
r
ate
o
f
0
.
5
is
ap
p
lied
t
o
p
r
e
v
en
t o
v
er
f
i
ttin
g
d
u
r
in
g
tr
ain
in
g
.
T
h
is
co
m
p
o
n
e
n
t f
o
cu
s
es o
n
ca
p
tu
r
in
g
s
eq
u
en
tial
r
elatio
n
s
h
ip
s
b
etwe
en
wo
r
d
s
[
2
5
]
,
[
2
6
]
,
p
r
o
v
id
i
n
g
a
co
m
p
lem
en
ta
r
y
v
iew
to
B
E
R
T
's
co
n
tex
tu
al
em
b
ed
d
in
g
s
.
3
.
2
.
3
.
I
nco
rpo
ra
t
ing
s
ent
ence
o
rder
info
rm
a
t
io
n
T
h
e
o
r
d
er
in
wh
ich
s
en
te
n
ce
s
ap
p
ea
r
with
in
an
ab
s
tr
ac
t
is
c
r
itical
f
o
r
u
n
d
e
r
s
tan
d
in
g
th
eir
r
o
le
in
th
e
o
v
er
all
n
ar
r
ativ
e.
T
o
lev
er
ag
e
th
is
,
th
e
s
en
ten
ce
’
s
p
o
s
itio
n
is
f
ed
in
to
a
s
im
p
le
d
en
s
e
n
etwo
r
k
:
ℎ
or
de
r
=
R
eL
U
(
3
+
3
)
(
6
)
wh
er
e
is
a
s
ca
lar
r
ep
r
esen
tin
g
th
e
s
en
ten
ce
o
r
d
er
,
an
d
3
∈
odr
×
,
3
∈
odr
ar
e
th
e
weig
h
ts
an
d
b
iases
o
f
th
e
f
u
lly
co
n
n
ec
ted
l
ay
er
.
T
h
is
en
co
d
in
g
p
r
o
v
id
es
ad
d
itio
n
al
in
s
ig
h
t in
to
th
e
f
u
n
ctio
n
o
f
th
e
s
en
ten
ce
b
ased
o
n
its
p
o
s
itio
n
in
th
e
ab
s
tr
ac
t.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
25
:
4
2
0
2
-
4212
4206
3
.
2
.
4
.
F
us
io
n o
f
f
ea
t
ures
T
h
e
o
u
t
p
u
ts
f
r
o
m
th
e
B
E
R
T
en
co
d
er
,
L
STM
,
an
d
o
r
d
er
p
r
o
ce
s
s
in
g
lay
er
s
ar
e
co
n
ca
ten
ate
d
to
c
r
ea
te
a
u
n
if
ied
r
ep
r
esen
tatio
n
o
f
th
e
s
en
ten
ce
:
ℎ
c
onc
a
t
=
[
ℎ
de
ns
e
,
ℎ
ls
tm
_
de
ns
e
,
ℎ
or
de
r
]
(
7
)
T
h
is
co
m
b
in
ed
v
ec
to
r
ℎ
c
onc
a
t
∈
c
na
is
th
en
p
ass
ed
th
r
o
u
g
h
a
f
u
lly
co
n
n
e
cted
d
en
s
e
lay
er
with
8
u
n
its
an
d
a
R
eL
U
ac
tiv
atio
n
f
u
n
ctio
n
[
2
7
]
:
ℎ
f
inal
=
R
eL
U
(
4
ℎ
c
onc
a
t
+
4
)
(
8
)
T
o
r
ed
u
ce
o
v
e
r
f
itti
n
g
,
a
d
r
o
p
o
u
t
lay
er
with
a
r
ate
o
f
0
.
2
is
a
d
d
ed
.
Fin
ally
,
a
So
f
tMa
x
lay
e
r
[
2
8
]
is
ap
p
lied
to
p
r
o
d
u
ce
th
e
p
r
o
b
a
b
ilit
y
d
is
tr
ib
u
tio
n
∈
o
v
er
th
e
f
iv
e
s
en
ten
ce
c
ateg
o
r
ies:
=
So
f
tm
ax
(
5
ℎ
f
inal
+
5
)
(
9
)
wh
er
e
r
ep
r
esen
ts
th
e
p
r
o
b
ab
il
ity
th
at
th
e
s
en
ten
ce
b
el
o
n
g
s
t
o
ca
teg
o
r
y
.
Fig
u
r
e
1
p
r
o
v
id
es
a
d
etailed
b
r
ea
k
d
o
wn
o
f
th
e
m
o
d
el’
s
ar
ch
itectu
r
e,
o
u
tlin
in
g
th
e
v
ar
io
u
s
lay
er
s
,
th
eir
r
esp
ec
tiv
e
o
u
tp
u
t
s
h
ap
es
,
an
d
th
e
n
u
m
b
er
o
f
p
ar
am
et
er
s
ass
o
ciate
d
with
ea
ch
.
Dir
ec
tly
f
o
llo
win
g
th
e
tab
le,
Fig
u
r
e
2
o
f
f
er
s
a
v
is
u
al
r
ep
r
esen
tatio
n
o
f
th
e
m
o
d
el'
s
s
tr
u
ctu
r
e.
T
h
e
ar
ch
itectu
r
e
co
n
s
is
ts
o
f
th
r
ee
m
ain
co
m
p
o
n
en
ts
:
th
e
B
E
R
T
en
c
o
d
er
f
o
r
ca
p
tu
r
i
n
g
c
o
n
tex
tu
al
i
n
f
o
r
m
atio
n
,
t
h
e
L
STM
la
y
er
f
o
r
m
o
d
elin
g
wo
r
d
-
lev
el
s
eq
u
en
ce
s
,
a
n
d
th
e
d
e
n
s
e
lay
er
s
p
r
o
ce
s
s
in
g
s
en
te
n
ce
o
r
d
er
in
f
o
r
m
atio
n
.
T
h
es
e
co
m
p
o
n
en
ts
a
r
e
co
m
b
in
ed
to
f
o
r
m
a
co
m
p
r
e
h
en
s
iv
e
f
ea
tu
r
e
r
e
p
r
esen
tatio
n
,
wh
ich
is
th
en
p
ass
ed
th
r
o
u
g
h
f
u
lly
co
n
n
ec
ted
lay
er
s
f
o
r
f
i
n
al
class
if
icatio
n
.
T
h
e
m
o
d
el
c
o
n
tain
s
o
v
er
1
0
9
m
illi
o
n
tr
ain
ab
le
p
a
r
am
eter
s
,
en
s
u
r
in
g
f
lex
ib
ilit
y
an
d
ca
p
ac
ity
f
o
r
lear
n
in
g
co
m
p
lex
p
atter
n
s
in
b
io
m
ed
ical
te
x
t.
Fig
u
r
e
1
.
Mo
d
el
s
u
m
m
ar
y
an
d
d
etailed
b
r
ea
k
d
o
wn
o
f
ar
c
h
itectu
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
n
h
a
n
cin
g
mu
lti
-
cla
s
s
text
cla
s
s
ifica
tio
n
in
b
io
med
ica
l litera
tu
r
e
b
y
…
(
Ou
s
s
a
ma
N
d
a
ma
)
4207
Fig
u
r
e
2
.
Ar
c
h
itectu
r
e
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
id
m
o
d
el
3
.
3
.
T
ra
ini
ng
p
ro
ce
du
re
T
o
tr
ain
th
e
h
y
b
r
id
m
o
d
el,
th
e
d
ataset
was
s
p
lit
in
to
tr
ain
in
g
an
d
v
alid
atio
n
s
ets,
with
8
0
%
u
s
ed
f
o
r
tr
ain
in
g
a
n
d
2
0
%
f
o
r
v
alid
a
tio
n
.
T
h
e
in
p
u
ts
in
clu
d
e
d
to
k
en
ized
s
en
ten
ce
tex
t
f
o
r
b
o
th
wo
r
d
-
lev
el
an
d
co
n
tex
tu
al
-
lev
el
em
b
ed
d
i
n
g
(
u
s
in
g
B
E
R
T
)
as
well
as
s
en
te
n
ce
o
r
d
e
r
in
f
o
r
m
atio
n
.
T
h
es
e
co
m
p
o
n
en
ts
wer
e
ess
en
tial f
o
r
f
ee
d
in
g
i
n
to
th
e
B
E
R
T
en
co
d
er
,
L
STM
,
an
d
o
r
d
er
m
o
d
el
lay
er
s
,
r
esp
ec
tiv
ely
.
3
.
3
.
1
.
O
ptim
izer
a
nd
lo
s
s
f
un
ct
io
n
T
h
e
m
o
d
el
was
co
m
p
iled
u
s
in
g
th
e
Ad
am
o
p
tim
izer
[
2
9
]
,
with
a
lear
n
in
g
r
ate
o
f
2
×
10
−
5
.
T
h
i
s
s
m
all
lear
n
in
g
r
ate
was
s
elec
t
ed
to
b
alan
ce
f
ast
c
o
n
v
e
r
g
en
c
e
an
d
s
tab
le
tr
ai
n
in
g
.
T
h
e
m
o
d
el’
s
p
er
f
o
r
m
a
n
ce
was
o
p
tim
ized
u
s
in
g
ca
teg
o
r
i
ca
l
cr
o
s
s
en
tr
o
p
y
,
th
e
s
tan
d
ar
d
lo
s
s
f
u
n
ctio
n
f
o
r
m
u
lti
-
class
class
if
ica
tio
n
task
s
[
3
0
]
.
T
h
is
f
u
n
ctio
n
is
d
ef
in
e
d
as:
=
−
∑
5
=
1
l
og
(
)
(
1
0
)
wh
er
e
is
th
e
tr
u
e
lab
el
(
o
n
e
-
h
o
t e
n
co
d
e
d
)
,
a
n
d
is
th
e
p
r
ed
ic
ted
p
r
o
b
ab
ilit
y
f
o
r
ea
ch
class
.
3
.
3
.
2
.
L
ea
rning
ra
t
e
s
chedul
er
T
o
f
u
r
th
er
o
p
tim
ize
th
e
lear
n
in
g
p
r
o
ce
s
s
,
a
cu
s
to
m
lear
n
in
g
r
ate
s
ch
ed
u
ler
was
u
s
ed
.
F
o
r
th
e
f
ir
s
t
th
r
ee
ep
o
ch
s
,
th
e
lear
n
i
n
g
r
ate
was
k
ep
t
co
n
s
tan
t,
b
u
t
af
ter
th
e
th
ir
d
ep
o
c
h
,
th
e
lear
n
in
g
r
a
te
was
r
ed
u
ce
d
b
y
a
f
ac
to
r
o
f
−
0
.
1
at
ea
ch
ep
o
ch
.
T
h
is
d
y
n
am
ic
ad
ju
s
tm
en
t
h
el
p
ed
f
in
e
-
tu
n
e
th
e
m
o
d
el
as
it
ap
p
r
o
ac
h
e
d
co
n
v
er
g
en
ce
,
s
lo
win
g
d
o
wn
l
ea
r
n
in
g
t
o
av
o
i
d
o
v
e
r
s
h
o
o
tin
g
o
p
tim
al
weig
h
ts
[
3
1
]
,
[
3
2
]
.
T
h
e
lear
n
i
n
g
r
ate
s
ch
ed
u
ler
is
d
ef
in
e
d
as:
lr
_
s
c
he
dul
e
r
(
e
p
oc
h
,
lr
)
=
if
e
p
oc
h
<
3
the
n
l
r
e
l
s
e
l
r
⋅
−
0
.
1
(
1
1
)
T
h
is
ap
p
r
o
ac
h
en
s
u
r
e
d
th
at
th
e
m
o
d
el
lear
n
e
d
m
o
r
e
ag
g
r
ess
iv
ely
in
th
e
in
itial e
p
o
c
h
s
wh
ile
g
r
ad
u
ally
r
ef
in
in
g
th
e
weig
h
ts
as tr
ain
in
g
p
r
o
g
r
es
s
ed
.
3
.
3
.
3
.
Ca
llb
a
ck
s
T
o
en
s
u
r
e
th
e
m
o
d
el
d
id
n
o
t
o
v
er
f
it a
n
d
to
s
p
ee
d
u
p
co
n
v
er
g
en
ce
,
two
k
e
y
ca
llb
ac
k
s
wer
e
em
p
lo
y
ed
:
−
E
ar
ly
s
to
p
p
i
n
g
:
T
h
is
ca
llb
ac
k
m
o
n
ito
r
ed
th
e
v
alid
atio
n
ac
cu
r
ac
y
a
n
d
s
to
p
p
ed
tr
ain
in
g
i
f
th
er
e
was
n
o
im
p
r
o
v
em
e
n
t
f
o
r
3
co
n
s
ec
u
tiv
e
ep
o
c
h
s
.
T
h
e
b
est
weig
h
ts
wer
e
r
esto
r
ed
a
f
ter
tr
ain
i
n
g
,
e
n
s
u
r
in
g
t
h
at
th
e
m
o
d
el
u
s
ed
th
e
p
ar
am
eter
s
f
r
o
m
th
e
ep
o
c
h
th
at
y
ield
e
d
th
e
h
ig
h
est v
alid
atio
n
ac
cu
r
ac
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
25
:
4
2
0
2
-
4212
4208
−
L
ea
r
n
in
g
r
ate
s
ch
e
d
u
ler
:
T
h
e
cu
s
to
m
lear
n
in
g
r
ate
s
ch
e
d
u
l
er
d
escr
ib
ed
ab
o
v
e
d
y
n
a
m
ically
ad
ju
s
ted
th
e
lear
n
in
g
r
ate
b
ased
o
n
th
e
tr
ai
n
in
g
ep
o
ch
.
3
.
3
.
4
.
T
ra
ini
ng
co
nfig
ura
t
io
n
T
h
e
m
o
d
el
was
tr
ain
ed
f
o
r
1
0
ep
o
c
h
s
,
with
a
b
atch
s
ize
o
f
1
6
.
T
h
is
r
elativ
ely
s
m
all
b
atch
s
ize
allo
wed
th
e
m
o
d
el
to
ef
f
ec
t
iv
ely
ca
p
tu
r
e
th
e
co
m
p
lex
r
elatio
n
s
h
ip
s
in
th
e
d
ata
wit
h
o
u
t
o
v
er
wh
elm
i
n
g
m
em
o
r
y
.
T
h
e
tr
ain
i
n
g
in
p
u
t
co
n
s
is
ted
o
f
:
−
Sen
ten
ce
tex
t: Pass
ed
to
b
o
th
t
h
e
to
k
en
an
d
s
en
ten
ce
m
o
d
els.
−
Sen
ten
ce
o
r
d
e
r
: Pass
ed
to
th
e
o
r
d
er
m
o
d
el.
T
r
ain
in
g
was
p
er
f
o
r
m
e
d
o
n
a
Go
o
g
le
C
o
lab
A1
0
0
GPU,
le
v
er
ag
in
g
th
e
GPU'
s
h
ig
h
co
m
p
u
tatio
n
al
p
o
wer
t
o
s
p
ee
d
u
p
tr
ai
n
in
g
a
n
d
m
a
n
ag
e
th
e
r
eso
u
r
ce
-
i
n
ten
s
iv
e
n
atu
r
e
o
f
B
E
R
T
f
in
e
-
tu
n
in
g
.
3
.
4
.
E
v
a
lua
t
io
n
m
et
rics
Af
ter
tr
ain
in
g
,
th
e
m
o
d
el
wa
s
ev
alu
ated
o
n
t
h
e
test
s
et
to
ass
ess
its
p
er
f
o
r
m
an
ce
.
T
h
e
ev
alu
atio
n
f
o
cu
s
ed
o
n
m
ea
s
u
r
in
g
k
e
y
m
etr
ics
th
at
d
em
o
n
s
tr
ate
th
e
m
o
d
el’
s
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
e
ca
ll,
an
d
F1
-
s
co
r
e,
p
r
o
v
id
i
n
g
a
co
m
p
r
eh
e
n
s
iv
e
v
iew
o
f
its
class
if
icatio
n
p
er
f
o
r
m
an
ce
[
3
3
]
,
[
3
4
]
.
T
h
is
th
o
r
o
u
g
h
ass
ess
m
en
t
en
s
u
r
es th
e
r
eliab
ilit
y
an
d
g
en
er
aliza
b
ilit
y
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
in
h
a
n
d
lin
g
v
ar
io
u
s
b
io
m
ed
ical
ab
s
tr
ac
ts
.
3.
4
.
1
.
T
est
i
np
uts
T
h
e
test
s
et
co
n
s
is
ted
o
f
2
9
,
4
9
3
s
en
ten
ce
s
,
wh
ic
h
wer
e
p
r
o
ce
s
s
ed
in
th
e
s
am
e
way
as
th
e
tr
ain
in
g
an
d
v
alid
atio
n
d
ata.
T
h
e
m
o
d
e
l to
o
k
as in
p
u
t:
−
Sen
ten
ce
tex
t: T
o
k
en
ize
d
an
d
p
ass
ed
to
b
o
th
th
e
B
E
R
T
e
n
co
d
er
an
d
L
STM
lay
er
s
.
−
Sen
ten
ce
o
r
d
e
r
: A
f
lo
at
v
alu
e
in
d
icatin
g
th
e
o
r
d
e
r
o
f
ea
c
h
s
e
n
ten
ce
in
th
e
a
b
s
tr
ac
t.
3.
4
.
2
.
E
v
a
lua
t
io
n
pro
ce
s
s
a
nd
perf
o
rm
a
nce
m
e
t
rics
T
o
ev
alu
ate
th
e
m
o
d
el,
we
b
u
ild
a
cu
s
to
m
f
u
n
ctio
n
ad
ap
t
ed
to
o
u
r
h
y
b
r
id
m
o
d
el.
T
h
i
s
f
u
n
ctio
n
co
m
p
u
tes
th
e
m
o
d
el'
s
p
r
ed
ictio
n
s
b
y
ap
p
l
y
in
g
th
e
So
f
tMa
x
o
u
tp
u
t
to
d
er
iv
e
class
p
r
o
b
ab
ilit
ies
an
d
th
en
s
elec
tin
g
th
e
class
with
t
h
e
h
ig
h
est
p
r
o
b
ab
ilit
y
u
s
in
g
ar
g
m
a
x
.
T
h
ese
p
r
e
d
ictio
n
s
wer
e
th
e
n
co
m
p
ar
ed
ag
ai
n
s
t
th
e
tr
u
e
lab
els f
r
o
m
th
e
test
s
et
to
co
m
p
u
te
th
e
f
o
llo
win
g
m
et
r
ics:
−
Acc
u
r
ac
y
:
T
h
e
p
r
o
p
o
r
tio
n
o
f
c
o
r
r
ec
t
p
r
ed
ictio
n
s
o
u
t
o
f
all
p
r
ed
ictio
n
s
.
I
t
p
r
o
v
i
d
es
a
h
i
g
h
-
le
v
el
m
ea
s
u
r
e
o
f
h
o
w
well
th
e
m
o
d
el
class
if
ied
t
h
e
s
en
ten
ce
s
:
A
c
c
ura
c
y
=
C
o
rre
c
t
Pre
d
ic
t
ion
s
T
o
t
a
l
Pre
d
ic
t
ion
s
(
1
2
)
−
Pre
cisi
o
n
(
Mic
r
o
-
Av
er
a
g
ed
)
:
Pre
cisi
o
n
m
ea
s
u
r
es
h
o
w
m
an
y
o
f
th
e
p
r
ed
icted
p
o
s
itiv
e
in
s
tan
ce
s
wer
e
co
r
r
ec
t,
an
d
is
ca
lcu
lated
as:
Pr
e
c
isio
n
=
T
rue
Po
sit
iv
e
s
T
rue
Po
sit
iv
e
s
+
Fa
l
se
Po
sit
iv
e
s
(
1
3
)
Mic
r
o
-
av
er
a
g
in
g
c
o
m
p
u
tes th
i
s
m
etr
ic
b
y
ag
g
r
eg
atin
g
co
n
tr
i
b
u
tio
n
s
f
r
o
m
all
class
es.
−
R
ec
all
(
Mic
r
o
-
Av
er
ag
ed
)
:
R
ec
all
m
ea
s
u
r
es
h
o
w
m
an
y
o
f
th
e
ac
tu
al
p
o
s
itiv
e
in
s
tan
ce
s
wer
e
co
r
r
ec
tly
id
en
tifie
d
b
y
th
e
m
o
d
el.
I
t is d
ef
in
ed
as:
R
e
c
a
l
l
=
T
rue
Po
si
t
iv
e
s
T
rue
Po
sit
iv
e
s
+
Fa
l
se
N
e
g
a
t
iv
e
s
(
1
4
)
−
F1
-
s
co
r
e
(
Mic
r
o
-
Av
e
r
ag
ed
)
:
T
h
e
F1
-
s
co
r
e
is
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
p
r
ec
is
io
n
a
n
d
r
ec
all,
p
r
o
v
id
in
g
a
b
alan
ce
d
m
ea
s
u
r
e
th
at
co
n
s
id
e
r
s
b
o
th
f
alse p
o
s
itiv
es a
n
d
f
alse n
eg
ativ
es:
F1
=
2
⋅
Pre
c
isi
o
n
⋅
Re
c
a
l
l
Pre
c
isi
o
n
+
Re
c
a
l
l
(
1
5
)
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
p
e
r
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
h
y
b
r
i
d
m
o
d
el
was
ev
alu
ated
o
n
th
e
test
s
et,
a
n
d
th
e
r
esu
lts
d
em
o
n
s
tr
ate
a
s
tr
o
n
g
ab
ilit
y
to
class
if
y
s
en
ten
ce
s
f
r
o
m
b
io
m
ed
ical
ab
s
tr
ac
ts
in
to
th
eir
r
e
s
p
ec
tiv
e
ca
teg
o
r
ies.
T
h
e
ev
alu
atio
n
m
etr
ics
in
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
in
d
icate
t
h
at
th
e
m
o
d
el
ef
f
ec
tiv
ely
ca
p
tu
r
es
b
o
th
co
n
te
x
tu
al
an
d
s
eq
u
en
tial
in
f
o
r
m
atio
n
in
th
e
tex
t.
T
h
e
m
o
d
el
ac
h
iev
ed
a
n
o
v
er
all
ac
cu
r
ac
y
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
n
h
a
n
cin
g
mu
lti
-
cla
s
s
text
cla
s
s
ifica
tio
n
in
b
io
med
ica
l litera
tu
r
e
b
y
…
(
Ou
s
s
a
ma
N
d
a
ma
)
4209
8
8
.
4
2
%,
with
a
p
r
ec
is
io
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
o
f
8
8
.
4
2
%
ac
r
o
s
s
all
ca
teg
o
r
ies,
r
e
f
lectin
g
a
b
ala
n
ce
d
p
er
f
o
r
m
an
ce
ac
r
o
s
s
d
if
f
er
en
t
ev
alu
atio
n
d
im
en
s
io
n
s
a
n
d
h
i
g
h
lig
h
tin
g
th
e
m
o
d
el'
s
ab
ilit
y
to
co
r
r
ec
tly
class
if
y
th
e
s
en
ten
ce
s
wh
ile
m
ain
tain
in
g
a
h
i
g
h
d
e
g
r
ee
o
f
p
r
ec
is
io
n
an
d
r
ec
all.
T
h
e
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
o
f
th
e
m
o
d
el
s
h
o
wed
c
o
n
s
is
ten
t
im
p
r
o
v
em
en
t
o
v
er
s
ev
e
n
ep
o
ch
s
,
as
illu
s
tr
ated
in
Fig
u
r
e
3
.
T
h
e
tr
ain
in
g
lo
s
s
d
ec
r
ea
s
ed
f
r
o
m
1
.
1
4
to
0
.
4
1
,
wh
ile
th
e
v
alid
atio
n
lo
s
s
f
o
llo
wed
a
s
im
ilar
p
atter
n
,
r
e
d
u
cin
g
f
r
o
m
0
.
5
0
to
0
.
3
9
,
wit
h
m
in
o
r
f
lu
ct
u
atio
n
s
to
war
d
th
e
e
n
d
.
T
h
is
c
o
n
v
er
g
en
ce
o
f
b
o
th
lo
s
s
es
in
d
icate
s
th
at
th
e
m
o
d
el
was
ef
f
ec
tiv
ely
lear
n
in
g
th
e
p
atter
n
s
f
r
o
m
th
e
d
ata
with
o
u
t
o
v
er
f
itti
n
g
,
m
ain
tain
in
g
s
tab
ilit
y
th
r
o
u
g
h
o
u
t
tr
ain
in
g
.
T
h
e
to
tal
tr
ai
n
in
g
tim
e
f
o
r
th
is
p
r
o
ce
s
s
was
a
p
p
r
o
x
im
ately
5
3
4
.
4
7
m
in
u
tes
(
~8
.
9
h
o
u
r
s
)
,
s
ig
n
if
ican
tly
r
ed
u
ce
d
u
s
in
g
a
Go
o
g
le
C
o
lab
A1
0
0
GPU,
m
ak
in
g
th
e
t
r
ain
in
g
ef
f
icien
t g
iv
en
th
e
m
o
d
el'
s
co
m
p
lex
ity
an
d
th
e
s
ize
o
f
t
h
e
d
a
taset.
As s
h
o
wn
in
T
ab
le
1
,
th
e
m
o
d
el
d
em
o
n
s
tr
ated
s
tr
o
n
g
p
er
f
o
r
m
an
ce
in
th
e
'
M
eth
o
d
s
'
an
d
'
C
o
n
clu
s
io
n
s
'
ca
teg
o
r
ies,
with
F1
-
s
co
r
es
o
f
9
3
.
5
6
%
an
d
8
8
.
7
9
%,
r
esp
ec
tiv
ely
,
h
ig
h
lig
h
tin
g
its
ab
i
lity
to
ef
f
ec
tiv
ely
d
if
f
er
en
tiate
th
ese
well
-
d
ef
i
n
ed
s
ec
tio
n
s
o
f
b
io
m
e
d
ical
ab
s
tr
ac
ts
.
Ho
wev
er
,
its
p
er
f
o
r
m
a
n
ce
in
th
e
'B
ac
k
g
r
o
u
n
d
'
an
d
'
Ob
ject
iv
e'
ca
teg
o
r
ies
was
co
m
p
ar
ativ
ely
lo
wer
,
with
F1
-
s
co
r
es
o
f
7
4
.
0
3
%
an
d
7
0
.
4
7
%,
r
ef
lectin
g
ch
allen
g
es
in
d
is
tin
g
u
is
h
in
g
th
ese
s
ec
tio
n
s
.
T
h
e
weig
h
ted
av
er
ag
es
f
o
r
p
r
e
cisi
o
n
,
r
ec
all,
an
d
F1
-
s
co
r
e
8
8
.
5
3
%,
8
8
.
4
2
%,
an
d
8
8
.
4
5
%,
r
esp
ec
tiv
ely
u
n
d
er
s
co
r
e
t
h
e
m
o
d
el’
s
o
v
e
r
all
ef
f
ec
tiv
en
ess
in
m
an
ag
in
g
class
im
b
alan
ce
with
in
th
e
d
ataset.
Fig
u
r
e
3
.
T
r
ain
in
g
a
n
d
v
alid
atio
n
lo
s
s
o
v
er
e
p
o
ch
s
T
ab
le
1
.
C
lass
if
icatio
n
p
er
f
o
r
m
an
ce
b
y
ca
teg
o
r
y
A
c
c
u
r
a
c
y
R
e
c
a
l
l
F
1
S
c
o
r
e
S
u
p
p
o
r
t
B
a
c
k
g
r
o
u
n
d
0
.
7
2
2
1
0
.
7
5
9
3
0
.
7
4
0
2
2
6
6
3
C
o
n
c
l
u
s
i
o
n
s
0
.
8
6
6
1
0
.
9
1
0
8
0
.
8
8
7
9
4
4
2
6
M
e
t
h
o
d
s
0
.
9
3
6
1
0
.
9
3
5
0
0
.
9
3
5
6
9
7
5
1
O
b
j
e
c
t
i
v
e
0
.
7
1
3
4
0
.
6
9
6
3
0
.
7
0
4
7
2
3
7
7
R
e
s
u
l
t
s
0
.
9
2
7
3
0
.
9
0
0
5
0
.
9
1
3
7
1
0
2
7
6
A
c
c
u
r
a
c
y
-
-
0
.
8
8
4
2
2
9
4
9
3
W
e
i
g
h
t
e
d
A
v
g
0
.
8
8
5
3
0
.
8
8
4
2
0
.
8
8
4
5
2
9
4
9
3
T
h
e
r
esu
lts
in
d
icate
th
at
th
e
h
y
b
r
id
m
o
d
el,
wh
ich
c
o
m
b
in
es
B
E
R
T
'
s
co
n
tex
tu
al
em
b
e
d
d
in
g
s
with
L
STM
's
ab
ilit
y
to
ca
p
tu
r
e
s
eq
u
en
tial
d
ep
en
d
en
cies,
is
h
ig
h
l
y
ef
f
ec
tiv
e
f
o
r
m
u
lti
-
class
s
en
ten
ce
class
if
icatio
n
in
b
io
m
e
d
ical
ab
s
tr
ac
ts
.
T
h
e
h
ig
h
ac
cu
r
ac
y
an
d
b
alan
ce
d
p
r
ec
is
io
n
an
d
r
ec
all
ac
r
o
s
s
m
o
s
t
ca
teg
o
r
ies
d
em
o
n
s
tr
ate
th
e
m
o
d
el'
s
r
o
b
u
s
tn
ess
an
d
g
en
e
r
aliza
tio
n
c
ap
ab
ilit
y
.
T
h
e
ex
ce
p
tio
n
al
p
er
f
o
r
m
a
n
ce
in
th
e
'
Me
th
o
d
s
'
an
d
'
R
e
s
u
lts
'
ca
teg
o
r
ies
ca
n
b
e
attr
ib
u
te
d
to
th
e
d
is
tin
ct
lan
g
u
ag
e
an
d
s
tr
u
ctu
r
e
ty
p
ically
f
o
u
n
d
i
n
th
ese
s
ec
tio
n
s
,
wh
ich
th
e
m
o
d
el
co
u
ld
lear
n
ef
f
ec
tiv
ely
.
T
h
e
lo
wer
p
er
f
o
r
m
an
ce
in
th
e
'B
ac
k
g
r
o
u
n
d
'
an
d
'
Ob
jectiv
e'
ca
teg
o
r
ies
s
u
g
g
ests
th
at
th
e
s
e
s
ec
tio
n
s
m
ay
co
n
tain
m
o
r
e
n
u
an
ce
d
la
n
g
u
ag
e
o
r
s
h
ar
e
s
im
ilar
ities
with
o
th
er
s
ec
tio
n
s
,
m
ak
in
g
th
e
m
h
ar
d
er
t
o
class
if
y
.
T
h
is
o
v
er
lap
c
o
u
ld
b
e
d
u
e
to
th
e
in
tr
o
d
u
cto
r
y
n
atu
r
e
o
f
th
e
s
e
s
ec
tio
n
s
,
wh
er
e
b
ac
k
g
r
o
u
n
d
in
f
o
r
m
atio
n
o
f
ten
s
ets
th
e
s
tag
e
f
o
r
th
e
o
b
jectiv
es
o
f
th
e
s
tu
d
y
.
Fu
tu
r
e
wo
r
k
co
u
ld
f
o
cu
s
o
n
e
n
h
an
cin
g
th
e
m
o
d
el'
s
ab
ilit
y
to
d
is
tin
g
u
is
h
b
etwe
en
th
ese
o
v
e
r
lap
p
in
g
ca
te
g
o
r
ies
b
y
in
co
r
p
o
r
atin
g
ad
d
itio
n
al
lin
g
u
is
tic
f
ea
tu
r
es
o
r
lev
er
a
g
in
g
d
o
m
ain
-
s
p
ec
if
ic
k
n
o
wled
g
e.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
4
,
Au
g
u
s
t
20
25
:
4
2
0
2
-
4212
4210
Mo
r
eo
v
er
,
th
e
co
n
s
is
ten
t
d
ec
r
ea
s
e
in
tr
ain
in
g
an
d
v
alid
atio
n
lo
s
s
o
v
er
th
e
ep
o
ch
s
with
o
u
t
s
ig
n
if
ican
t
f
lu
ctu
atio
n
s
in
d
icate
s
th
at
th
e
m
o
d
el
d
i
d
n
o
t
o
v
er
f
it
an
d
ca
n
g
en
e
r
alize
we
ll
to
u
n
s
e
en
d
ata.
T
h
is
tr
en
d
u
n
d
er
s
co
r
es
th
e
r
o
b
u
s
tn
ess
o
f
th
e
tr
ain
in
g
p
r
o
ce
d
u
r
e
a
n
d
th
e
s
tab
ilit
y
o
f
th
e
m
o
d
el’
s
ar
ch
itectu
r
e.
T
h
e
s
u
b
s
tan
tial
tr
ain
in
g
tim
e
f
u
r
th
er
h
ig
h
lig
h
ts
th
e
co
m
p
u
tatio
n
a
l
in
ten
s
ity
o
f
tr
ain
in
g
s
u
ch
d
ee
p
lear
n
in
g
m
o
d
els;
h
o
wev
er
,
t
h
e
u
s
e
o
f
h
ig
h
-
p
er
f
o
r
m
an
ce
c
o
m
p
u
tin
g
r
eso
u
r
ce
s
lik
e
GPUs
s
ig
n
if
ican
tly
allev
iates th
is
ch
allen
g
e.
I
n
c
o
n
clu
s
io
n
,
th
e
h
y
b
r
id
m
o
d
el
d
em
o
n
s
tr
ates
s
tr
o
n
g
p
o
ten
tial
f
o
r
au
to
m
atin
g
th
e
class
if
icatio
n
o
f
s
en
ten
ce
s
in
b
io
m
e
d
ical
liter
atu
r
e,
th
er
e
b
y
f
ac
ilit
atin
g
m
o
r
e
ef
f
icien
t
in
f
o
r
m
atio
n
r
etr
ie
v
al
an
d
k
n
o
wled
g
e
ex
tr
ac
tio
n
.
Fu
tu
r
e
en
h
a
n
ce
m
e
n
ts
co
u
ld
f
u
r
th
er
im
p
r
o
v
e
th
e
m
o
d
el’
s
p
er
f
o
r
m
an
ce
,
p
a
r
ticu
lar
ly
in
ch
allen
g
in
g
ca
teg
o
r
ies,
m
ak
in
g
it
an
ev
e
n
m
o
r
e
v
alu
a
b
le
to
o
l
f
o
r
b
i
o
m
ed
ical
r
esear
ch
er
s
an
d
p
r
ac
titi
o
n
er
s
.
On
g
o
i
n
g
r
ef
in
em
en
ts
to
th
e
m
o
d
el’
s
ar
ch
itectu
r
e
an
d
o
p
tim
izatio
n
s
t
r
ateg
ies
m
ay
y
ield
ev
e
n
m
o
r
e
r
o
b
u
s
t
an
d
s
ca
lab
le
s
o
lu
tio
n
s
in
th
e
lo
n
g
ter
m
.
5.
CO
NCLU
SI
O
N
I
n
th
is
s
tu
d
y
,
we
in
tr
o
d
u
ce
d
a
h
y
b
r
id
m
o
d
el
th
at
c
o
m
b
in
es
B
E
R
T
f
o
r
c
o
n
tex
tu
al
lea
r
n
in
g
,
L
STM
f
o
r
s
eq
u
en
t
ial
p
r
o
ce
s
s
in
g
,
an
d
s
en
ten
ce
o
r
d
er
in
f
o
r
m
atio
n
to
class
if
y
s
en
ten
ce
s
in
b
io
m
ed
ical
ab
s
tr
ac
ts
.
T
h
e
m
o
d
el
ac
h
ie
v
ed
s
tr
o
n
g
p
er
f
o
r
m
an
ce
o
n
th
e
Pu
b
Me
d
2
0
0
k
R
C
T
d
ataset,
with
an
o
v
er
all
ac
cu
r
ac
y
o
f
8
8
.
4
2
%
an
d
b
alan
ce
d
p
r
ec
is
io
n
,
r
ec
all
,
an
d
F1
-
s
co
r
e
ac
r
o
s
s
ca
teg
o
r
i
es.
I
t
ex
ce
lled
in
class
if
y
in
g
m
eth
o
d
s
an
d
r
esu
lts
s
ec
tio
n
s
,
th
o
u
g
h
f
u
r
th
er
im
p
r
o
v
em
en
ts
co
u
ld
b
e
m
a
d
e
in
d
is
tin
g
u
is
h
in
g
b
ac
k
g
r
o
u
n
d
an
d
o
b
jectiv
e
s
en
ten
ce
s
.
B
y
ef
f
ec
tiv
ely
in
teg
r
atin
g
b
o
th
co
n
tex
tu
al
an
d
s
eq
u
en
tial
in
f
o
r
m
atio
n
,
o
u
r
m
o
d
el
d
em
o
n
s
tr
ates
it
s
p
o
ten
tial
f
o
r
im
p
r
o
v
in
g
th
e
r
ea
d
ab
ilit
y
an
d
s
eg
m
e
n
tatio
n
o
f
c
o
m
p
lex
b
io
m
ed
ical
tex
ts
.
T
h
e
u
s
e
o
f
f
i
n
e
-
tu
n
in
g
,
lear
n
in
g
r
ate
s
ch
ed
u
lin
g
,
an
d
ea
r
ly
s
to
p
p
in
g
co
n
tr
i
b
u
ted
to
th
e
m
o
d
el'
s
co
n
v
er
g
en
ce
a
n
d
s
tab
il
ity
,
en
s
u
r
in
g
r
o
b
u
s
t
g
en
er
aliza
tio
n
to
u
n
s
ee
n
d
at
a
.
T
h
is
h
y
b
r
id
ap
p
r
o
ac
h
o
f
f
e
r
s
a
v
alu
ab
le
to
o
l
f
o
r
en
h
an
cin
g
th
e
au
to
m
atic
p
r
o
ce
s
s
in
g
o
f
b
io
m
e
d
ical
liter
atu
r
e,
en
ab
lin
g
m
o
r
e
ef
f
icien
t
in
f
o
r
m
atio
n
r
etr
iev
al
f
o
r
r
esea
r
ch
er
s
.
Fu
tu
r
e
wo
r
k
co
u
ld
ex
p
lo
r
e
f
u
r
th
e
r
r
ef
in
e
m
en
ts
in
class
if
icatio
n
ac
cu
r
ac
y
,
p
ar
ticu
l
ar
l
y
in
o
v
er
lap
p
in
g
ca
teg
o
r
ies,
an
d
th
e
in
teg
r
atio
n
o
f
ad
d
itio
n
al
tech
n
iq
u
es su
ch
as a
tten
tio
n
m
ec
h
a
n
is
m
s
to
en
h
an
ce
p
er
f
o
r
m
an
ce
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
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
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ou
s
s
am
a
Nd
am
a
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
I
s
m
ail
B
en
s
as
s
i
✓
✓
Saf
ae
Nd
am
a
✓
✓
E
l M
o
k
h
tar
E
n
-
Naim
i
✓
✓
✓
✓
✓
✓
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
ma
l
a
n
a
l
y
si
s
I
:
I
n
v
e
st
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
s
u
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
si
o
n
P
:
P
r
o
j
e
c
t
a
d
m
i
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
s
i
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
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
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
i
n
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
o
p
e
n
ly
av
ailab
le
in
th
e
AC
L
A
n
th
o
lo
g
y
at
h
ttp
s
:
//w
w
w
.
a
clw
eb
.
o
r
g
/a
n
th
o
lo
g
y/I
1
7
-
2
0
5
2
.
p
d
f
,
r
e
f
er
en
ce
:
Der
n
o
n
co
u
r
t
a
n
d
L
ee
(
2
0
1
7
)
[
6
]
.
Pu
b
Me
d
2
0
0
k
R
C
T
: a
Data
s
et
f
o
r
s
eq
u
en
tial
s
en
ten
ce
class
if
icatio
n
in
m
ed
ical
ab
s
tr
ac
ts
.
I
J
C
NL
P.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
E
n
h
a
n
cin
g
mu
lti
-
cla
s
s
text
cla
s
s
ifica
tio
n
in
b
io
med
ica
l litera
tu
r
e
b
y
…
(
Ou
s
s
a
ma
N
d
a
ma
)
4211
RE
F
E
R
E
NC
E
S
[
1
]
C
.
J
.
C
r
e
m
i
n
,
S
.
D
a
s
h
,
a
n
d
X
.
H
u
a
n
g
,
“
B
i
g
d
a
t
a
:
H
i
s
t
o
r
i
c
a
d
v
a
n
c
e
s
a
n
d
e
mer
g
i
n
g
t
r
e
n
d
s
i
n
b
i
o
m
e
d
i
c
a
l
r
e
sea
r
c
h
,
”
C
u
rr
e
n
t
Re
se
a
rc
h
i
n
Bi
o
t
e
c
h
n
o
l
o
g
y
,
v
o
l
.
4
,
p
p
.
1
3
8
–
1
5
1
,
2
0
2
2
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
r
b
i
o
t
.
2
0
2
2
.
0
2
.
0
0
4
.
[
2
]
Q
.
Ji
n
e
t
a
l
.
,
“
B
i
o
me
d
i
c
a
l
q
u
e
st
i
o
n
a
n
sw
e
r
i
n
g
:
A
su
r
v
e
y
o
f
a
p
p
r
o
a
c
h
e
s
a
n
d
c
h
a
l
l
e
n
g
e
s,”
A
C
M
C
o
m
p
u
t
i
n
g
S
u
rve
y
s
,
v
o
l
.
5
5
,
n
o
.
2
,
p
p
.
1
–
3
6
,
F
e
b
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
4
9
0
2
3
8
.
[
3
]
T.
A
u
g
u
st
,
L
.
L.
W
a
n
g
,
J
.
B
r
a
g
g
,
M
.
A
.
H
e
a
r
st
,
A
.
H
e
a
d
,
a
n
d
K
.
L
o
,
“
P
a
p
e
r
P
l
a
i
n
:
M
a
k
i
n
g
me
d
i
c
a
l
r
e
s
e
a
r
c
h
p
a
p
e
r
s
a
p
p
r
o
a
c
h
a
b
l
e
t
o
h
e
a
l
t
h
c
a
r
e
c
o
n
su
m
e
r
s
w
i
t
h
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ssi
n
g
,
”
AC
M
T
r
a
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
e
r
-
H
u
m
a
n
I
n
t
e
r
a
c
t
i
o
n
,
v
o
l
.
3
0
,
n
o
.
5
,
p
p
.
1
–
3
8
,
O
c
t
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
8
9
9
5
5
.
[
4
]
G
.
F
r
i
s
o
n
i
,
G
.
M
o
r
o
,
a
n
d
A
.
C
a
r
b
o
n
a
r
o
,
“
A
s
u
r
v
e
y
o
n
e
v
e
n
t
e
x
t
r
a
c
t
i
o
n
f
o
r
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
u
n
d
e
r
st
a
n
d
i
n
g
:
R
i
d
i
n
g
t
h
e
b
i
o
me
d
i
c
a
l
l
i
t
e
r
a
t
u
r
e
W
a
v
e
,
”
I
E
EE
Ac
c
e
ss
,
v
o
l
.
9
,
p
p
.
1
6
0
7
2
1
–
1
6
0
7
5
7
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
1
.
3
1
3
0
9
5
6
.
[
5
]
D
.
D
e
m
n
er
-
F
u
sh
m
a
n
,
N
.
El
h
a
d
a
d
,
a
n
d
C
.
F
r
i
e
d
ma
n
,
“
N
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ss
i
n
g
f
o
r
h
e
a
l
t
h
-
r
e
l
a
t
e
d
t
e
x
t
s
,
”
i
n
Bi
o
m
e
d
i
c
a
l
I
n
f
o
rm
a
t
i
c
s
,
C
h
a
m:
S
p
r
i
n
g
e
r
I
n
t
e
r
n
a
t
i
o
n
a
l
P
u
b
l
i
sh
i
n
g
,
2
0
2
1
,
p
p
.
2
4
1
–
2
7
2
.
[
6
]
F
.
D
e
r
n
o
n
c
o
u
r
t
a
n
d
J
.
Y
.
L
e
e
,
“
P
u
b
M
e
d
2
0
0
k
R
C
T
:
A
d
a
t
a
se
t
f
o
r
se
q
u
e
n
t
i
a
l
s
e
n
t
e
n
c
e
c
l
a
ssi
f
i
c
a
t
i
o
n
i
n
m
e
d
i
c
a
l
a
b
st
r
a
c
t
s
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
rXi
v
:
1
7
1
0
.
0
6
0
7
1
,
2
0
1
7
.
[
7
]
A
.
R
i
o
s
a
n
d
R
.
K
a
v
u
l
u
r
u
,
“
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
f
o
r
b
i
o
me
d
i
c
a
l
t
e
x
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
6
t
h
A
C
M
C
o
n
f
e
re
n
c
e
o
n
B
i
o
i
n
f
o
rm
a
t
i
c
s,
C
o
m
p
u
t
a
t
i
o
n
a
l
B
i
o
l
o
g
y
a
n
d
H
e
a
l
t
h
I
n
f
o
rm
a
t
i
c
s
,
S
e
p
.
2
0
1
5
,
p
p
.
2
5
8
–
2
6
7
,
d
o
i
:
1
0
.
1
1
4
5
/
2
8
0
8
7
1
9
.
2
8
0
8
7
4
6
.
[
8
]
K
.
D
r
a
mé
,
F
.
M
o
u
g
i
n
,
a
n
d
G
.
D
i
a
l
l
o
,
“
L
a
r
g
e
sc
a
l
e
b
i
o
me
d
i
c
a
l
t
e
x
t
s
c
l
a
ssi
f
i
c
a
t
i
o
n
:
A
k
N
N
a
n
d
a
n
ESA
-
b
a
s
e
d
a
p
p
r
o
a
c
h
e
s,”
J
o
u
rn
a
l
o
f
Bi
o
m
e
d
i
c
a
l
S
e
m
a
n
t
i
c
s
,
v
o
l
.
7
,
n
o
.
1
,
p
.
4
0
,
D
e
c
.
2
0
1
6
,
d
o
i
:
1
0
.
1
1
8
6
/
s
1
3
3
2
6
-
0
1
6
-
0
0
7
3
-
1.
[
9
]
A
.
M
.
C
o
h
e
n
,
“
A
s
u
r
v
e
y
o
f
c
u
r
r
e
n
t
w
o
r
k
i
n
b
i
o
me
d
i
c
a
l
t
e
x
t
m
i
n
i
n
g
,
”
Br
i
e
f
i
n
g
s
i
n
Bi
o
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
6
,
n
o
.
1
,
p
p
.
5
7
–
7
1
,
Ja
n
.
2
0
0
5
,
d
o
i
:
1
0
.
1
0
9
3
/
b
i
b
/
6
.
1
.
5
7
.
[
1
0
]
I
.
M
o
n
d
a
l
,
“
B
B
A
EG
:
T
o
w
a
r
d
s
B
ER
T
-
b
a
se
d
b
i
o
m
e
d
i
c
a
l
a
d
v
e
r
sari
a
l
e
x
a
m
p
l
e
g
e
n
e
r
a
t
i
o
n
f
o
r
t
e
x
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
2
0
2
1
C
o
n
f
e
re
n
c
e
o
f
t
h
e
N
o
rt
h
Am
e
ri
c
a
n
C
h
a
p
t
e
r
o
f
t
h
e
Ass
o
c
i
a
t
i
o
n
f
o
r
C
o
m
p
u
t
a
t
i
o
n
a
l
L
i
n
g
u
i
st
i
c
s:
H
u
m
a
n
L
a
n
g
u
a
g
e
T
e
c
h
n
o
l
o
g
i
e
s
,
2
0
2
1
,
p
p
.
5
3
7
8
–
5
3
8
4
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
n
a
a
c
l
-
mai
n
.
4
2
3
.
[
1
1
]
X
.
Z
h
a
n
g
,
X
.
S
o
n
g
,
A
.
F
e
n
g
,
a
n
d
Z
.
G
a
o
,
“
M
u
l
t
i
-
sel
f
-
a
t
t
e
n
t
i
o
n
f
o
r
a
s
p
e
c
t
c
a
t
e
g
o
r
y
d
e
t
e
c
t
i
o
n
a
n
d
b
i
o
me
d
i
c
a
l
m
u
l
t
i
l
a
b
e
l
t
e
x
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
w
i
t
h
B
E
R
T,
”
Ma
t
h
e
m
a
t
i
c
a
l
Pro
b
l
e
m
s
i
n
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
2
0
2
1
,
p
p
.
1
–
6
,
N
o
v
.
2
0
2
1
,
d
o
i
:
1
0
.
1
1
5
5
/
2
0
2
1
/
6
6
5
8
5
2
0
.
[
1
2
]
M
.
N
e
u
m
a
n
n
,
D
.
K
i
n
g
,
I
.
B
e
l
t
a
g
y
,
a
n
d
W
.
A
mm
a
r
,
“
S
c
i
sp
a
C
y
:
F
a
st
a
n
d
r
o
b
u
st
mo
d
e
l
s
f
o
r
b
i
o
m
e
d
i
c
a
l
n
a
t
u
r
a
l
l
a
n
g
u
a
g
e
p
r
o
c
e
ss
i
n
g
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
1
8
t
h
B
i
o
N
L
P
W
o
rks
h
o
p
a
n
d
S
h
a
re
d
T
a
sk
,
2
0
1
9
,
p
p
.
3
1
9
–
3
2
7
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
W
1
9
-
5
0
3
4
.
[
1
3
]
J.
Y
u
a
n
,
F
.
Z
h
a
n
g
,
Y
.
Q
i
u
,
H
.
L
i
n
,
a
n
d
Y
.
Zh
a
n
g
,
“
D
o
c
u
me
n
t
-
l
e
v
e
l
b
i
o
m
e
d
i
c
a
l
r
e
l
a
t
i
o
n
e
x
t
r
a
c
t
i
o
n
v
i
a
h
i
e
r
a
r
c
h
i
c
a
l
t
r
e
e
g
r
a
p
h
a
n
d
r
e
l
a
t
i
o
n
se
g
me
n
t
a
t
i
o
n
mo
d
u
l
e
,
”
Bi
o
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
4
0
,
n
o
.
7
,
J
u
l
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
9
3
/
b
i
o
i
n
f
o
r
ma
t
i
c
s/
b
t
a
e
4
1
8
.
[
1
4
]
J.
D
u
a
n
,
H
.
G
u
o
,
H
.
J
i
a
n
g
,
F
.
G
u
o
,
a
n
d
J
.
W
a
n
g
,
“
B
o
u
n
d
a
r
y
-
a
w
a
r
e
d
u
a
l
b
i
a
f
f
i
n
e
m
o
d
e
l
f
o
r
se
q
u
e
n
t
i
a
l
s
e
n
t
e
n
c
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
i
n
b
i
o
me
d
i
c
a
l
d
o
c
u
m
e
n
t
s,
”
I
EEE
/
AC
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
Bi
o
l
o
g
y
a
n
d
B
i
o
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
1
,
n
o
.
5
,
p
p
.
1
2
0
2
–
1
2
1
0
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
TC
B
B
.
2
0
2
4
.
3
3
7
6
5
6
6
.
[
1
5
]
B
.
W
a
n
g
e
t
a
l
.
,
“
P
r
e
-
t
r
a
i
n
e
d
l
a
n
g
u
a
g
e
mo
d
e
l
s
i
n
b
i
o
m
e
d
i
c
a
l
d
o
ma
i
n
:
A
s
y
st
e
ma
t
i
c
su
r
v
e
y
,
”
A
C
M
C
o
m
p
u
t
i
n
g
S
u
r
v
e
y
s
,
v
o
l
.
5
6
,
n
o
.
3
,
p
p
.
1
–
5
2
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
4
5
/
3
6
1
1
6
5
1
.
[
1
6
]
S
.
J.
M
i
e
l
k
e
e
t
a
l
.
,
“
B
e
t
w
e
e
n
w
o
r
d
s a
n
d
c
h
a
r
a
c
t
e
r
s
:
A
b
r
i
e
f
h
i
st
o
r
y
o
f
o
p
e
n
-
v
o
c
a
b
u
l
a
r
y
mo
d
e
l
i
n
g
a
n
d
t
o
k
e
n
i
z
a
t
i
o
n
i
n
N
LP,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
2
1
1
2
.
1
0
5
0
8
,
2
0
2
1
.
[
1
7
]
B
.
K
i
m
a
n
d
H
.
J.
J
a
n
g
,
“
W
o
r
d
-
l
e
v
e
l
e
mb
e
d
d
i
n
g
t
o
i
m
p
r
o
v
e
p
e
r
f
o
r
man
c
e
o
f
r
e
p
r
e
se
n
t
a
t
i
v
e
S
p
a
t
i
o
-
t
e
mp
o
r
a
l
d
o
c
u
me
n
t
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
J
o
u
r
n
a
l
o
f
I
n
f
o
rm
a
t
i
o
n
Pr
o
c
e
ssi
n
g
S
y
s
t
e
m
s
,
v
o
l
.
1
9
,
n
o
.
6
,
p
p
.
8
3
0
–
8
4
1
,
2
0
2
3
,
d
o
i
:
1
0
.
3
7
4
5
/
JI
P
S
.
0
4
.
0
2
9
6
.
[
1
8
]
N
.
K
o
sar
a
j
u
,
S
.
R
.
S
a
n
k
e
p
a
l
l
y
,
a
n
d
K
.
M
a
l
l
i
k
h
a
r
j
u
n
a
R
a
o
,
“
C
a
t
e
g
o
r
i
c
a
l
d
a
t
a
:
N
e
e
d
,
e
n
c
o
d
i
n
g
,
s
e
l
e
c
t
i
o
n
o
f
e
n
c
o
d
i
n
g
m
e
t
h
o
d
a
n
d
i
t
s
e
m
e
r
g
e
n
c
e
i
n
ma
c
h
i
n
e
l
e
a
r
n
i
n
g
m
o
d
e
l
s
—
a
p
r
a
c
t
i
c
a
l
r
e
v
i
e
w
s
t
u
d
y
o
n
h
e
a
r
t
d
i
s
e
a
s
e
p
r
e
d
i
c
t
i
o
n
d
a
t
a
s
e
t
u
s
i
n
g
P
e
a
r
so
n
c
o
r
r
e
l
a
t
i
o
n
,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
D
a
t
a
S
c
i
e
n
c
e
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s:
I
C
D
S
A
2
0
2
2
,
V
o
l
u
m
e
1
,
2
0
2
3
,
p
p
.
3
6
9
–
382
.
[
1
9
]
J.
D
e
v
l
i
n
,
M
.
-
W
.
C
h
a
n
g
,
K
.
Le
e
,
K
.
T.
G
o
o
g
l
e
,
a
n
d
A
.
I
.
L
a
n
g
u
a
g
e
,
“
B
E
R
T:
P
r
e
-
t
r
a
i
n
i
n
g
o
f
d
e
e
p
b
i
d
i
r
e
c
t
i
o
n
a
l
t
r
a
n
sf
o
r
mers
f
o
r
l
a
n
g
u
a
g
e
u
n
d
e
r
st
a
n
d
i
n
g
,
”
N
a
a
c
l
-
H
l
t
2
0
1
9
,
n
o
.
M
l
m,
p
p
.
4
1
7
1
–
4
1
8
6
,
2
0
1
8
,
d
o
i
:
1
0
.
4
8
5
5
0
/
a
r
x
i
v
.
1
8
1
0
.
0
4
8
0
5
.
[
2
0
]
“
B
e
r
t
Te
x
t
C
l
a
ssi
f
i
e
r
m
o
d
e
l
,
”
K
e
r
a
s
,
h
t
t
p
s
:
/
/
k
e
r
a
s.
i
o
/
a
p
i
/
k
e
r
a
s
_
n
l
p
/
m
o
d
e
l
s/
b
e
r
t
/
b
e
r
t
_
c
l
a
ss
i
f
i
e
r
/
(a
c
c
e
ss
e
d
A
u
g
.
0
7
,
2
0
2
4
)
.
[
2
1
]
A
.
M
e
r
c
h
a
n
t
,
E.
R
a
h
i
m
t
o
r
o
g
h
i
,
E
.
P
a
v
l
i
c
k
,
a
n
d
I
.
Te
n
n
e
y
,
“
W
h
a
t
h
a
p
p
e
n
s
t
o
B
ER
T
e
m
b
e
d
d
i
n
g
s
d
u
r
i
n
g
f
i
n
e
-
t
u
n
i
n
g
?
,
”
a
r
Xi
v
p
re
p
ri
n
t
a
rXi
v
:
2
0
0
4
.
1
4
4
4
8
,
p
p
.
3
3
–
4
4
,
2
0
2
0
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
0
.
b
l
a
c
k
b
o
x
n
l
p
-
1
.
4
.
[
2
2
]
G
.
P
u
c
c
e
t
t
i
,
A
.
M
i
a
s
c
h
i
,
a
n
d
F
.
D
e
l
l
’
O
r
l
e
t
t
a
,
“
H
o
w
d
o
B
ER
T
e
m
b
e
d
d
i
n
g
s
o
r
g
a
n
i
z
e
l
i
n
g
u
i
st
i
c
k
n
o
w
l
e
d
g
e
?
,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
o
f
D
e
e
p
L
e
a
r
n
i
n
g
I
n
si
d
e
O
u
t
(
D
e
e
L
I
O
)
:
T
h
e
2
n
d
W
o
rks
h
o
p
o
n
K
n
o
w
l
e
d
g
e
E
x
t
r
a
c
t
i
o
n
a
n
d
I
n
t
e
g
r
a
t
i
o
n
f
o
r
D
e
e
p
L
e
a
r
n
i
n
g
Arc
h
i
t
e
c
t
u
r
e
s
,
2
0
2
1
,
p
p
.
4
8
–
5
7
,
d
o
i
:
1
0
.
1
8
6
5
3
/
v
1
/
2
0
2
1
.
d
e
e
l
i
o
-
1
.
6
.
[
2
3
]
D
.
N
i
u
,
Z
.
X
i
a
,
Y
.
L
i
u
,
T.
C
a
i
,
T.
L
i
u
,
a
n
d
Y
.
Zh
a
n
,
“
A
LST
M
:
A
d
a
p
t
i
v
e
L
S
TM
f
o
r
d
u
r
a
t
i
v
e
se
q
u
e
n
t
i
a
l
d
a
t
a
,
”
i
n
2
0
1
8
I
E
EE
3
0
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
T
o
o
l
s
w
i
t
h
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
I
C
T
AI
)
,
N
o
v
.
2
0
1
8
,
p
p
.
1
5
1
–
1
5
7
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
T
A
I
.
2
0
1
8
.
0
0
0
3
2
.
[
2
4
]
Y
u
W
a
n
g
,
“
A
n
e
w
c
o
n
c
e
p
t
u
si
n
g
LSTM
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
d
y
n
a
m
i
c
s
y
st
e
m
i
d
e
n
t
i
f
i
c
a
t
i
o
n
,
”
i
n
2
0
1
7
A
m
e
ri
c
a
n
C
o
n
t
r
o
l
C
o
n
f
e
re
n
c
e
(
A
C
C
)
,
M
a
y
2
0
1
7
,
p
p
.
5
3
2
4
–
5
3
2
9
,
d
o
i
:
1
0
.
2
3
9
1
9
/
A
C
C
.
2
0
1
7
.
7
9
6
3
7
8
2
.
[
2
5
]
Y
.
Y
u
,
X
.
S
i
,
C
.
H
u
,
a
n
d
J.
Z
h
a
n
g
,
“
A
r
e
v
i
e
w
o
f
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s:
LST
M
c
e
l
l
s
a
n
d
n
e
t
w
o
r
k
a
r
c
h
i
t
e
c
t
u
r
e
s
,
”
N
e
u
r
a
l
C
o
m
p
u
t
a
t
i
o
n
,
v
o
l
.
3
1
,
n
o
.
7
,
p
p
.
1
2
3
5
–
1
2
7
0
,
J
u
l
.
2
0
1
9
,
d
o
i
:
1
0
.
1
1
6
2
/
n
e
c
o
_
a
_
0
1
1
9
9
.
[
2
6
]
G
.
V
a
n
H
o
u
d
t
,
C
.
M
o
s
q
u
e
r
a
,
a
n
d
G
.
N
á
p
o
l
e
s,
“
A
r
e
v
i
e
w
o
n
t
h
e
l
o
n
g
s
h
o
r
t
-
t
e
r
m
mem
o
r
y
m
o
d
e
l
,
”
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
Re
v
i
e
w
,
v
o
l
.
5
3
,
n
o
.
8
,
p
p
.
5
9
2
9
–
5
9
5
5
,
D
e
c
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
4
6
2
-
0
2
0
-
0
9
8
3
8
-
1.
[
2
7
]
A
.
M
.
J
a
v
i
d
,
S
.
D
a
s
,
M
.
S
k
o
g
l
u
n
d
,
a
n
d
S
.
C
h
a
t
t
e
r
j
e
e
,
“
A
R
e
L
U
d
e
n
s
e
l
a
y
e
r
t
o
i
m
p
r
o
v
e
t
h
e
p
e
r
f
o
r
ma
n
c
e
o
f
n
e
u
r
a
l
n
e
t
w
o
r
k
s
,
”
i
n
I
C
AS
S
P
2
0
2
1
-
2
0
2
1
I
E
EE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
Ac
o
u
st
i
c
s,
S
p
e
e
c
h
a
n
d
S
i
g
n
a
l
Pr
o
c
e
s
si
n
g
(
I
C
A
S
S
P)
,
J
u
n
.
2
0
2
1
,
p
p
.
2
8
1
0
–
2
8
1
4
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
S
S
P
3
9
7
2
8
.
2
0
2
1
.
9
4
1
4
2
6
9
.
[
2
8
]
M
.
W
a
n
g
,
S
.
L
u
,
D
.
Z
h
u
,
J.
L
i
n
,
a
n
d
Z.
W
a
n
g
,
“
A
h
i
g
h
-
s
p
e
e
d
a
n
d
l
o
w
-
c
o
m
p
l
e
x
i
t
y
a
r
c
h
i
t
e
c
t
u
r
e
f
o
r
S
o
f
t
M
a
x
f
u
n
c
t
i
o
n
i
n
d
e
e
p
l
e
a
r
n
i
n
g
,
”
i
n
2
0
1
8
I
EE
E
Asi
a
Pa
c
i
f
i
c
C
o
n
f
e
re
n
c
e
o
n
C
i
r
c
u
i
t
s
a
n
d
S
y
s
t
e
m
s
(
APC
C
A
S
)
,
O
c
t
.
2
0
1
8
,
p
p
.
2
2
3
–
2
2
6
,
d
o
i
:
1
0
.
1
1
0
9
/
A
P
C
C
A
S
.
2
0
1
8
.
8
6
0
5
6
5
4
.
[
2
9
]
S
.
B
o
c
k
,
J
.
G
o
p
p
o
l
d
,
a
n
d
M
.
W
e
i
ß
,
“
A
n
i
m
p
r
o
v
e
me
n
t
o
f
t
h
e
c
o
n
v
e
r
g
e
n
c
e
p
r
o
o
f
o
f
t
h
e
A
D
A
M
-
o
p
t
i
mi
z
e
r
,
”
a
rX
i
v
p
re
p
ri
n
t
a
rXi
v
:
1
8
0
4
.
1
0
5
8
7
,
2
0
1
8
.
[
3
0
]
J.
G
h
o
sh
a
n
d
S
.
G
u
p
t
a
,
“
A
D
A
M
o
p
t
i
mi
z
e
r
a
n
d
c
a
t
e
g
o
r
i
c
a
l
C
r
o
ss
e
n
t
r
o
p
y
l
o
ss
f
u
n
c
t
i
o
n
-
b
a
s
e
d
C
N
N
me
t
h
o
d
f
o
r
d
i
a
g
n
o
si
n
g
c
o
l
o
r
e
c
t
a
l
c
a
n
c
e
r
,
”
i
n
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
a
n
d
S
u
st
a
i
n
a
b
l
e
En
g
i
n
e
e
ri
n
g
S
o
l
u
t
i
o
n
s
(
C
I
S
E
S
)
,
A
p
r
.
2
0
2
3
,
p
p
.
4
7
0
–
4
7
4
,
d
o
i
:
1
0
.
1
1
0
9
/
C
I
S
ES5
8
7
2
0
.
2
0
2
3
.
1
0
1
8
3
4
9
1
.
[
3
1
]
C
.
K
i
m,
S
.
K
i
m
,
J
.
K
i
m,
D
.
Le
e
,
a
n
d
S
.
K
i
m
,
“
A
u
t
o
mat
e
d
l
e
a
r
n
i
n
g
r
a
t
e
s
c
h
e
d
u
l
e
r
f
o
r
l
a
r
g
e
-
b
a
t
c
h
t
r
a
i
n
i
n
g
,
”
a
rX
i
v
p
r
e
p
r
i
n
t
a
rXi
v
:
2
1
0
7
.
0
5
8
5
5
,
2
0
2
1
.
[
3
2
]
G
.
I
o
a
n
n
o
u
,
T.
T
a
g
a
r
i
s,
a
n
d
A
.
S
t
a
f
y
l
o
p
a
t
i
s
,
“
A
d
a
Li
p
:
A
n
a
d
a
p
t
i
v
e
l
e
a
r
n
i
n
g
r
a
t
e
m
e
t
h
o
d
p
e
r
l
a
y
e
r
f
o
r
s
t
o
c
h
a
st
i
c
o
p
t
i
m
i
z
a
t
i
o
n
,
”
N
e
u
ra
l
Pr
o
c
e
ss
i
n
g
L
e
t
t
e
rs
,
v
o
l
.
5
5
,
n
o
.
5
,
p
p
.
6
3
1
1
–
6
3
3
8
,
O
c
t
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
1
0
6
3
-
022
-
1
1
1
4
0
-
w.
Evaluation Warning : The document was created with Spire.PDF for Python.