T
E
L
K
O
M
NIKA
T
elec
o
mm
un
ica
t
io
n,
Co
m
pu
t
ing
,
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
1
9
,
No
.
6
,
Dec
em
b
er
202
1
,
p
p
.
18
84
~
1
891
I
SS
N:
1
6
9
3
-
6
9
3
0
,
ac
cr
ed
ited
First Gr
ad
e
b
y
Ke
m
e
n
r
is
te
k
d
i
k
ti,
Dec
r
ee
No
: 2
1
/E/KPT
/2
0
1
8
DOI
: 1
0
.
1
2
9
2
8
/
T
E
L
KOM
NI
KA
.
v
1
9
i
6
.
2
1
6
6
7
1884
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//jo
u
r
n
a
l.u
a
d
.
a
c.
id
/in
d
ex
.
p
h
p
/TELK
OM
N
I
K
A
A nov
el deep
lear
ning
architec
ture
for drug
na
m
ed
e
ntit
y
recog
nition
T
.
M
a
t
hu
,
K
u
m
ud
ha
Ra
i
m
o
nd
De
p
a
rt
m
e
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
,
Ka
ru
n
y
a
In
stit
u
te o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
S
c
ien
c
e
s,
Co
im
b
a
to
re
,
In
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
1
,
2
0
2
1
R
ev
i
s
ed
Oct
1
0
,
2
0
2
1
A
cc
ep
ted
Oct
1
8
,
2
0
2
1
Dru
g
n
a
m
e
d
e
n
ti
t
y
re
c
o
g
n
it
io
n
(
DN
ER)
b
e
c
o
m
e
s
th
e
p
re
re
q
u
isit
e
o
f
o
th
e
r
m
e
d
ica
l
re
latio
n
e
x
tra
c
ti
o
n
sy
ste
m
s.
Ex
isti
n
g
a
p
p
ro
a
c
h
e
s
to
a
u
to
m
a
ti
c
a
ll
y
re
c
o
g
n
ize
d
ru
g
n
a
m
e
s
in
c
lu
d
e
s
ru
le
-
b
a
se
d
,
m
a
c
h
in
e
lea
rn
in
g
(M
L)
a
n
d
d
e
e
p
lea
rn
in
g
(DL
)
tec
h
n
iq
u
e
s.
DL
tec
h
n
i
q
u
e
s
h
a
v
e
b
e
e
n
v
e
rif
i
e
d
to
b
e
th
e
sta
te
-
of
-
th
e
-
a
rt
a
s
it
is
in
d
e
p
e
n
d
e
n
t
o
f
h
a
n
d
c
ra
f
ted
f
e
a
tu
re
s.
T
h
e
p
re
v
io
u
s
DL
m
e
th
o
d
s
b
a
se
d
o
n
w
o
rd
e
m
b
e
d
d
in
g
in
p
u
t
re
p
re
se
n
tati
o
n
u
se
s
th
e
s
a
m
e
v
e
c
to
r
re
p
re
se
n
tatio
n
f
o
r
a
n
e
n
ti
ty
irres
p
e
c
ti
v
e
o
f
it
s
c
o
n
tex
t
in
d
if
fe
re
n
t
se
n
ten
c
e
s
a
n
d
h
e
n
c
e
c
o
u
l
d
n
o
t
c
a
p
tu
re
th
e
c
o
n
tex
t
p
ro
p
e
rly
.
A
lso
,
id
e
n
ti
f
ica
ti
o
n
o
f
th
e
n
-
g
ra
m
e
n
ti
ty
is
a
c
h
a
ll
e
n
g
e
.
In
t
h
is
p
a
p
e
r,
a
n
o
v
e
l
a
rc
h
it
e
c
tu
re
is
p
ro
p
o
se
d
th
a
t
in
c
lu
d
e
s
a
se
n
ten
c
e
e
m
b
e
d
d
in
g
la
y
e
r
th
a
t
w
o
rk
s
o
n
th
e
e
n
ti
re
se
n
ten
c
e
to
e
ff
ici
e
n
tl
y
c
a
p
tu
re
th
e
c
o
n
tex
t
o
f
a
n
e
n
ti
ty
.
A
h
y
b
rid
m
o
d
e
l
th
a
t
c
o
m
p
rise
s
a
sta
c
k
e
d
b
id
irec
ti
o
n
a
l
l
o
n
g
sh
o
rt
-
term
m
e
m
o
r
y
(Bi
-
L
S
T
M
)
w
it
h
re
sid
u
a
l
L
S
T
M
h
a
s
b
e
e
n
d
e
sig
n
e
d
to
o
v
e
rc
o
m
e
th
e
li
m
it
a
ti
o
n
s
a
n
d
to
u
p
g
ra
d
e
th
e
p
e
rf
o
r
m
a
n
c
e
o
f
th
e
m
o
d
e
l.
W
e
h
a
v
e
c
o
n
tras
ted
th
e
a
c
h
i
e
v
e
m
e
n
t
o
f
o
u
r
p
ro
p
o
se
d
a
p
p
ro
a
c
h
w
it
h
o
t
h
e
r
DN
ER
m
o
d
e
ls
a
n
d
th
e
p
e
rc
e
n
tag
e
o
f
im
p
ro
v
e
m
e
n
ts
o
f
th
e
p
ro
p
o
se
d
m
o
d
e
l
o
v
e
r
L
S
T
M
-
c
o
n
d
it
io
n
a
l
r
a
n
d
o
m
f
ield
(CRF
)
,
L
IU
a
n
d
W
BI
w
it
h
re
sp
e
c
t
to
m
icro
-
a
v
e
ra
g
e
F
1
-
sc
o
re
a
re
1
1
.
1
7
,
8
.
8
a
n
d
1
7
.
6
4
re
sp
e
c
ti
v
e
l
y
.
T
h
e
p
ro
p
o
se
d
m
o
d
e
l
h
a
s
a
lso
s
h
o
w
n
p
r
o
m
isin
g
re
su
lt
in
re
c
o
g
n
izin
g
2
-
a
n
d
3
-
g
ra
m
e
n
ti
ti
e
s.
K
ey
w
o
r
d
s
:
Dr
u
g
n
a
m
ed
en
tit
y
r
ec
o
g
n
itio
n
Natu
r
al
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
R
esid
u
al
L
ST
M
Sen
te
n
ce
lev
e
l e
m
b
ed
d
in
g
Stack
ed
B
i
-
L
ST
M
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
:
T
.
Ma
th
u
Dep
ar
t
m
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
i
n
ee
r
in
g
Kar
u
n
y
a
I
n
s
tit
u
te
o
f
T
ec
h
n
o
lo
g
y
a
n
d
Scien
ce
s
Kar
u
n
y
a
Na
g
ar
,
C
o
i
m
b
ato
r
e
6
4
1
1
1
4
,
T
am
ilNad
u
,
I
n
d
ia
E
m
ail:
m
at
h
u
@
k
ar
u
n
y
a.
ed
u
1.
I
NT
RO
D
UCT
I
O
N
Na
m
ed
en
t
it
y
r
ec
o
g
n
i
tio
n
(
N
E
R
)
is
a
n
ess
e
n
tial
tas
k
o
f
i
n
f
o
r
m
atio
n
e
x
tr
ac
tio
n
(
I
E
)
,
an
d
is
o
f
ten
u
tili
ze
d
in
n
atu
r
al
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
(
NL
P
)
.
T
h
e
NE
R
t
h
at
d
ef
i
n
es
an
d
class
if
ie
s
th
e
lab
els
o
f
d
r
u
g
s
in
to
p
r
ed
ef
in
ed
class
e
s
f
r
o
m
u
n
s
tr
u
ctu
r
ed
m
ed
ical
te
x
ts
i
s
r
ef
er
r
ed
to
as
d
r
u
g
n
a
m
ed
en
tit
y
r
ec
o
g
n
itio
n
(
DNE
R
)
[
1
]
.
T
h
e
r
esear
ch
o
n
th
e
D
NE
R
b
ec
o
m
e
s
p
r
o
m
i
n
en
t
e
v
er
s
i
n
ce
t
h
e
id
en
ti
f
ic
atio
n
o
f
d
r
u
g
-
d
r
u
g
in
ter
ac
tio
n
s
(
DDI
)
an
d
ad
v
er
s
e
d
r
u
g
r
ea
ctio
n
(
ADR)
h
av
e
b
ec
o
m
e
i
m
p
o
r
tan
t
i
n
th
e
b
r
an
ch
o
f
p
h
ar
m
ac
o
d
y
n
a
m
ics
an
d
p
h
ar
m
ac
o
k
i
n
etic
s
.
Ho
w
e
v
er
,
m
an
y
s
tu
d
ies
[
2
]
-
[
5
]
h
av
e
s
h
o
w
n
th
at
th
er
e
is
n
o
t
m
u
c
h
s
p
ec
if
ic
w
o
r
k
f
o
r
DNE
R
in
r
ec
en
t
y
ea
r
s
.
T
ec
h
n
iq
u
e
s
lik
e
r
u
le
-
b
ased
f
r
a
m
e
w
o
r
k
,
m
ac
h
in
e
lear
n
i
n
g
(
ML
)
m
et
h
o
d
s
an
d
d
ee
p
lear
n
i
n
g
(
DL
)
tec
h
n
iq
u
es
w
er
e
e
m
p
l
o
y
ed
f
o
r
th
e
DNE
R
.
T
h
e
r
u
le
-
b
ased
an
d
M
L
tech
n
iq
u
es
h
ea
v
il
y
d
ep
en
d
o
n
th
e
f
ie
ld
/s
u
b
j
ec
t
k
n
o
w
led
g
e
o
f
th
e
h
u
m
a
n
p
r
o
f
ess
io
n
als
to
d
ev
is
e
th
e
f
ea
t
u
r
es
f
o
r
d
esig
n
i
n
g
th
e
r
ec
o
g
n
itio
n
m
o
d
el.
DL
u
s
es
m
o
r
e
th
an
o
n
e
la
y
er
o
f
ar
tif
icia
l
n
eu
r
al
n
et
w
o
r
k
s
th
at
r
ec
o
g
n
izes
t
h
e
n
a
m
ed
en
tit
i
es
[
2
]
,
[
6
]
–
[
8
]
.
W
h
en
co
m
p
ar
ed
to
tr
ad
itio
n
al
ap
p
r
o
ac
h
es,
D
L
is
m
o
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
n
o
ve
l d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
d
r
u
g
n
a
med
en
tity reco
g
n
itio
n
(
T.
Ma
th
u
)
1885
ad
v
an
ta
g
eo
u
s
i
n
a
u
to
m
at
icall
y
r
ec
o
g
n
izi
n
g
h
id
d
en
f
ea
tu
r
e
s
.
T
h
e
latest
D
L
tec
h
n
iq
u
es
d
o
n
o
t
r
eq
u
ir
e
th
e
in
ter
v
e
n
tio
n
o
f
t
h
e
h
u
m
a
n
ex
p
er
ts
in
co
n
s
tr
u
c
tin
g
t
h
e
f
ea
tu
r
es
f
r
o
m
u
n
s
tr
u
ct
u
r
ed
tex
t.
T
h
e
m
o
s
t
co
m
m
o
n
D
L
m
o
d
el
u
s
ed
is
lo
n
g
s
h
o
r
t
-
te
r
m
m
e
m
o
r
y
(
L
ST
M)
w
h
ic
h
h
elp
s
to
p
r
eser
v
e
th
e
lo
n
g
-
r
an
g
e
d
ep
en
d
en
c
y
esp
ec
iall
y
w
h
ile
d
ea
lin
g
w
ith
s
eq
u
en
tial
te
x
t
.
B
id
ir
ec
tio
n
al
L
ST
M
m
o
d
el
(
B
i
-
L
ST
M)
w
h
ich
r
ea
d
s
th
e
te
x
t
b
o
th
in
f
o
r
w
ar
d
an
d
r
ev
er
s
e
d
ir
ec
tio
n
s
i
s
u
s
ed
to
ca
p
tu
r
e
t
h
e
co
n
tex
t
o
f
t
h
e
w
o
r
d
f
o
r
b
ett
er
p
r
ed
ictio
n
.
W
o
r
d
e
m
b
ed
d
in
g
m
o
d
el
s
lik
e
W
o
r
d
2
Vec
,
Glo
Ve
an
d
Fas
tT
ex
t
ar
e
u
s
u
all
y
u
s
ed
f
o
r
w
o
r
d
e
m
b
ed
d
in
g
i
n
DL
alg
o
r
ith
m
s
.
W
e
h
a
v
e
o
b
s
er
v
ed
th
at
th
e
m
aj
o
r
lim
itat
io
n
o
f
wo
r
d
em
b
ed
d
in
g
m
o
d
els
i
s
th
at
th
e
y
w
o
r
k
w
i
th
t
h
e
s
a
m
e
v
ec
to
r
f
o
r
all
th
e
m
en
tio
n
s
o
f
a
n
en
tit
y
i
n
t
h
e
ar
ticle
an
d
h
en
ce
co
u
ld
n
o
t c
ap
tu
r
e
t
h
e
co
n
tex
t p
r
o
p
er
ly
.
T
h
e
p
r
ev
io
u
s
r
esear
ch
w
o
r
k
s
f
o
r
DNE
R
m
o
d
els
b
a
s
ed
o
n
DL
h
a
v
e
u
s
ed
w
o
r
d
2
v
ec
[
9
]
an
d
Glo
v
e
[
1
0
]
w
o
r
d
e
m
b
ed
d
in
g
an
d
ch
ar
ac
ter
e
m
b
ed
d
in
g
m
o
d
els
to
r
ep
r
esen
t
th
e
in
p
u
t.
I
t
w
as
f
o
u
n
d
th
at
t
h
e
w
o
r
d
e
m
b
ed
d
in
g
m
o
d
el
co
u
l
d
n
o
t
ca
p
tu
r
e
th
e
s
e
m
an
tic
f
ea
tu
r
e
o
f
t
h
e
w
o
r
d
s
i
n
t
h
e
s
en
ten
ce
co
m
p
letel
y
.
B
ec
au
s
e
in
w
o
r
d
2
v
ec
,
ev
er
y
u
n
iq
u
e
w
o
r
d
th
r
o
u
g
h
o
u
t
t
h
e
co
r
p
u
s
w
ill
h
av
e
t
h
e
s
a
m
e
v
ec
to
r
in
th
e
v
ec
to
r
s
p
ac
e.
C
o
n
s
id
er
th
e
f
o
llo
w
i
n
g
s
en
te
n
ce
s
:
Sen
te
n
ce
1
:
M
A
O
i
n
h
ib
ito
r
s
p
r
o
lo
n
g
an
d
in
ten
s
i
f
y
t
h
e
a
n
ti
ch
o
lin
er
g
ic
(
d
r
y
i
n
g
)
e
f
f
ec
ts
o
f
an
ti
h
is
ta
m
i
n
es.
(
MA
O
–
B
-
g
r
o
u
p
,
in
h
ib
ito
r
s
–
I
-
g
r
o
u
p
)
.
Sen
te
n
ce
2
:
I
n
t
h
e
ab
s
e
n
ce
o
f
f
o
r
m
al
cli
n
ical
d
r
u
g
i
n
ter
ac
ti
o
n
s
t
u
d
ies,
ca
u
tio
n
s
h
o
u
ld
b
e
ex
er
cised
w
h
e
n
ad
m
in
i
s
ter
i
n
g
T
A
XO
L
co
n
c
o
m
itan
tl
y
w
it
h
k
n
o
w
n
s
u
b
s
t
r
ates
o
r
in
h
ib
ito
r
s
o
f
th
e
c
y
to
ch
r
o
m
e
P
4
5
0
is
o
en
z
y
m
e
s
C
YP
2
C
8
an
d
C
Y
P
3
A
4
.
I
n
s
e
n
te
n
ce
1
,
t
h
e
w
o
r
d
“
M
A
O
i
n
h
ib
ito
r
s
”
is
an
n
o
tated
as
a
cla
s
s
o
f
d
r
u
g
‘
g
r
o
u
p
’
a
n
d
h
e
n
c
e
“
i
n
h
ib
ito
r
s
”
is
an
n
o
tated
w
i
th
“
I
-
g
r
o
u
p
”.
B
u
t
i
n
s
en
te
n
ce
2
,
th
e
w
o
r
d
“
i
n
h
ib
ito
r
s
”
d
o
es
n
o
t
r
ef
er
to
a
n
y
cla
s
s
o
f
d
r
u
g
a
n
d
h
e
n
ce
s
h
o
u
ld
b
e
i
d
en
tifie
d
as
“O”
.
I
f
w
o
r
d
e
m
b
ed
d
in
g
m
o
d
els
lik
e
W
o
r
d
2
Ve
c
o
r
Glo
v
e
ar
e
u
s
ed
,
th
e
s
a
m
e
w
o
r
d
v
ec
to
r
is
u
s
ed
f
o
r
b
o
th
th
e
m
e
n
tio
n
s
a
n
d
h
e
n
c
e
w
o
u
ld
n
o
t
b
e
r
ec
o
g
n
ized
co
r
r
ec
tl
y
b
as
ed
o
n
th
e
co
n
tex
t.
T
h
o
u
g
h
r
ec
u
r
r
en
t
n
e
u
r
al
n
et
w
o
r
k
s
(
R
NN)
h
a
v
e
b
ee
n
u
s
ed
f
o
r
v
ar
io
u
s
NE
R
m
o
d
els,
th
e
f
u
ll
p
o
ten
tia
l
is
n
o
t
r
ea
lized
w
h
e
n
a
n
L
ST
M
o
r
B
i
-
L
ST
M
m
o
d
el
alo
n
e
is
u
s
ed
.
A
ls
o
,
i
n
D
NE
R
,
s
ev
e
r
al
d
r
u
g
n
a
m
es
ar
e
n
-
g
r
a
m
e
n
titi
e
s
.
Fo
r
in
s
ta
n
ce
,
“
alb
en
d
az
o
le
s
u
lf
o
x
id
e
”
(
2
-
g
r
a
m
en
tit
y
)
,
“
ce
n
tr
al
n
er
v
o
u
s
s
y
s
te
m
d
ep
r
ess
an
t
s
”
(4
-
g
r
a
m
e
n
tit
y
).
I
n
th
is
p
ap
er
,
to
o
v
er
co
m
e
th
e
ch
allen
g
es
m
en
tio
n
ed
a
b
o
v
e,
a
n
o
v
el
ar
ch
itectu
r
e
h
a
s
b
ee
n
p
r
o
p
o
s
ed
b
y
i
n
co
r
p
o
r
atin
g
a
s
e
n
te
n
ce
em
b
ed
d
i
n
g
m
o
d
el,
s
tack
ed
B
i
-
L
ST
M
an
d
r
esid
u
al
L
ST
M.
A
s
en
te
n
ce
e
m
b
ed
d
in
g
m
o
d
el
ca
lled
E
L
Mo
[
1
1
]
,
is
u
s
ed
to
d
ea
l
w
it
h
e
n
tire
s
en
te
n
ce
to
ca
p
tu
r
e
t
h
e
co
n
te
x
t p
r
o
p
er
ly
.
Si
n
ce
th
e
m
o
d
el
is
ex
tr
e
m
el
y
co
n
tex
t
u
alize
d
,
b
o
th
s
y
n
ta
x
a
n
d
s
e
m
an
tic
c
h
a
r
ac
ter
is
tics
o
f
th
e
w
o
r
d
ar
e
m
o
d
elled
.
T
h
e
m
ain
ad
v
an
ta
g
e
is
th
at
it
w
o
u
ld
b
e
ab
le
to
g
en
er
ate
v
ec
to
r
s
f
o
r
w
o
r
d
s
th
at
ar
e
n
o
t
s
ee
n
d
u
r
i
n
g
tr
ain
i
n
g
.
A
l
s
o
,
to
en
h
a
n
ce
t
h
e
D
NE
R
m
o
d
el,
t
h
e
d
esig
n
ed
ar
ch
itec
tu
r
e
u
til
ize
s
t
h
e
p
o
w
er
o
f
t
h
e
R
NN
i
n
d
r
u
g
p
r
ed
ictio
n
.
T
h
e
ar
ch
itect
u
r
e
co
n
s
is
ts
o
f
s
tack
e
d
B
i
-
L
ST
M
la
y
er
s
[
1
2
]
an
d
a
r
esid
u
al
L
ST
M
la
y
er
[
1
3
]
.
I
n
itiall
y
,
th
e
s
e
n
te
n
ce
e
m
b
ed
d
in
g
m
o
d
el
E
L
Mo
u
s
ed
in
t
h
is
ar
ch
itect
u
r
e
cr
ea
tes
w
o
r
d
v
ec
to
r
s
b
y
f
u
n
ctio
n
i
n
g
o
n
a
n
en
tire
s
e
n
te
n
ce
to
ef
f
icien
tl
y
ca
p
t
u
r
e
th
e
co
n
tex
t
o
f
t
h
e
w
o
r
d
.
T
h
e
s
p
ec
if
ic
co
n
te
x
t
an
d
t
h
e
v
ar
iatio
n
s
i
n
t
h
e
co
n
ten
t
is
id
en
ti
f
ied
a
n
d
h
elp
s
t
h
e
m
ac
h
i
n
e
to
u
n
d
er
s
ta
n
d
b
etter
,
u
n
l
ik
e
h
a
v
in
g
th
e
s
a
m
e
w
o
r
d
v
ec
to
r
f
o
r
ev
er
y
m
e
n
tio
n
o
f
th
e
w
o
r
d
in
o
th
er
w
o
r
d
e
m
b
ed
d
in
g
m
o
d
els.
T
h
o
u
g
h
a
s
i
n
g
le
B
i
-
L
ST
M
la
y
er
its
el
f
co
u
ld
p
o
s
s
ib
l
y
r
ec
o
g
n
iz
e
th
e
en
titi
e
s
,
th
e
p
o
w
er
o
f
th
e
R
NN
s
h
all
b
e
en
h
an
ce
d
b
y
ar
r
an
g
i
n
g
m
u
ltip
le
Bi
-
L
ST
M
la
y
er
s
o
n
to
p
o
f
ea
ch
o
t
h
er
.
W
e
h
a
v
e
e
x
p
er
i
m
e
n
ted
w
it
h
la
y
er
s
o
f
B
i
-
L
ST
M
in
t
h
is
m
o
d
el.
T
o
av
o
id
s
tack
ed
B
i
-
L
ST
M
s
u
f
f
er
in
g
f
r
o
m
th
e
v
a
n
is
h
i
n
g
g
r
a
d
ien
t
p
r
o
b
lem
,
th
e
r
es
id
u
al
L
ST
M
is
u
s
ed
.
T
h
e
r
esid
u
al
co
n
n
ec
tio
n
i
s
u
s
ed
b
et
w
ee
n
t
h
e
t
w
o
B
i
-
L
ST
M
lay
er
s
.
I
t
allo
w
s
t
h
e
g
r
ad
ien
ts
t
o
p
ass
th
r
o
u
g
h
t
h
e
n
et
w
o
r
k
d
ir
ec
tl
y
an
d
also
h
e
lp
s
to
p
r
eser
v
e
th
e
lo
n
g
r
an
g
e
d
ep
en
d
en
cies
[
1
4
]
,
[
1
5
]
.
W
e
h
av
e
tes
ted
o
u
r
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
w
it
h
a
test
d
ata
s
et
an
d
th
e
p
er
f
o
r
m
an
ce
is
co
m
p
ar
ed
w
it
h
o
t
h
er
DNE
R
m
o
d
els
an
d
f
o
u
n
d
to
i
m
p
r
o
v
e
t
h
e
r
ec
o
g
n
itio
n
r
ate
o
v
er
t
h
e
p
r
e
ce
d
in
g
s
tate
-
of
-
t
h
e
-
ar
t
m
o
d
el
s
.
W
e
h
a
v
e
also
ev
al
u
ated
t
h
e
p
er
f
o
r
m
a
n
ce
in
id
en
ti
f
y
i
n
g
th
e
2
-
,
3
-
,
an
d
4
-
g
r
a
m
en
t
ities
w
h
ic
h
i
s
a
m
aj
o
r
ch
alle
n
g
e
in
DNE
R
.
T
h
e
c
o
n
t
en
t
o
f
th
is
p
a
p
e
r
i
s
c
at
eg
o
r
i
z
e
d
a
s
f
o
l
l
o
w
s
:
s
e
ct
i
o
n
2
e
x
te
n
d
s
th
e
r
e
s
e
a
r
ch
m
e
th
o
d
.
R
e
s
u
lt
s
a
r
e
d
is
cu
s
s
e
d
in
s
ec
t
i
o
n
3
an
d
co
n
clu
s
io
n
i
s
g
i
v
e
n
in
s
ec
t
io
n
4.
2.
RE
S
E
ARCH
M
E
T
H
O
D
I
n
o
u
r
m
o
d
el,
w
e
h
a
v
e
u
s
ed
B
i
-
L
ST
M
n
et
w
o
r
k
w
h
er
e
th
e
s
en
te
n
ce
is
r
ea
d
in
b
o
th
f
o
r
w
ar
d
an
d
r
ev
er
s
e
d
ir
ec
tio
n
.
T
h
e
w
o
r
k
s
r
ep
r
esen
ted
in
th
e
f
o
llo
w
in
g
p
a
p
er
s
[
1
6
]
-
[
1
9
]
h
av
e
s
h
o
w
n
t
h
a
t
th
e
class
if
ica
tio
n
p
er
f
o
r
m
a
n
ce
co
u
ld
b
e
en
h
an
c
ed
f
u
r
t
h
er
b
y
s
tac
k
in
g
m
an
y
B
i
-
L
ST
M
lay
er
s
.
Hen
ce
,
i
n
o
u
r
m
o
d
el,
w
e
h
a
v
e
u
s
ed
t
w
o
B
i
-
L
ST
M
la
y
er
s
s
ta
ck
ed
ab
o
v
e
ea
c
h
o
t
h
er
as
s
h
o
w
n
i
n
Fi
g
u
r
e
1
.
W
h
e
n
t
h
e
s
e
n
ten
ce
s
ar
e
lo
n
g
a
s
g
iv
e
n
in
th
e
ex
a
m
p
le
i
n
s
ec
tio
n
1
,
it
i
s
n
ec
es
s
ar
y
to
r
e
m
e
m
b
e
r
th
e
lo
n
g
r
an
g
e
d
ep
en
d
en
c
y
o
f
t
h
e
e
n
tit
y
f
r
o
m
th
e
f
ir
s
t
i
n
s
ta
n
ce
a
n
d
t
h
e
n
e
x
t
i
n
s
tan
ce
.
W
h
e
n
t
h
e
d
ep
th
o
f
t
h
e
n
e
u
r
al
n
et
w
o
r
k
i
n
cr
ea
s
es,
ac
c
u
r
ac
y
o
f
p
r
ed
ictio
n
also
i
n
cr
ea
s
es.
R
es
id
u
al
L
ST
M
is
u
s
ed
to
av
o
id
s
tac
k
ed
B
i
-
L
ST
M
s
u
f
f
er
i
n
g
f
r
o
m
t
h
e
v
a
n
is
h
i
n
g
g
r
ad
ien
t p
r
o
b
le
m
an
d
also
it is
s
u
itab
le
f
o
r
h
an
d
li
n
g
s
u
c
h
lo
n
g
r
an
g
e
d
ep
en
d
en
cies.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
1
9
,
No
.
6
,
Decem
b
er
2021
:
18
84
-
1
891
1886
2
.
1
.
Da
t
a
s
o
urce
a
nd
prepro
ce
s
s
ing
I
n
th
i
s
p
ap
er
,
w
e
h
av
e
ad
o
p
ted
th
e
tag
g
ed
co
r
p
u
s
n
a
m
el
y
,
t
h
e
DDI
2
0
1
3
d
r
u
g
b
an
k
d
ataset
[
2
0
]
,
th
e
b
en
ch
m
ar
k
d
ataset
to
tr
ain
th
e
d
ee
p
lear
n
in
g
m
o
d
el.
T
h
e
d
at
a
is
ca
teg
o
r
ized
w
i
th
t
h
e
f
o
llo
w
i
n
g
lab
els:
d
r
u
g
,
b
r
an
d
,
g
r
o
u
p
an
d
d
r
u
g
_
n
[
2
0
]
.
W
e
h
av
e
p
r
ep
r
o
ce
s
s
ed
th
e
DDI
2
0
1
3
t
r
ain
in
g
d
ataset
in
s
u
ch
a
w
a
y
th
at
t
h
e
s
en
te
n
ce
is
s
p
lit
in
to
to
k
e
n
s
a
n
d
ev
er
y
to
k
e
n
i
s
lab
eled
w
i
t
h
t
h
e
co
r
r
esp
o
n
d
i
n
g
clas
s
lab
els.
Si
n
ce
th
er
e
ar
e
s
ev
er
al
n
-
g
r
a
m
w
o
r
d
s
av
ai
la
b
le
as
d
r
u
g
n
a
m
es
i
n
t
h
e
b
i
o
m
ed
ical
ar
ticles,
it
is
n
ec
es
s
ar
y
to
ca
p
tu
r
e
t
h
e
b
eg
in
n
i
n
g
a
n
d
en
d
o
f
t
h
e
en
tit
y
.
T
h
e
m
o
s
t
co
m
m
o
n
tag
g
in
g
s
ch
e
m
e
k
n
o
w
n
as
B
I
O
tag
g
i
n
g
is
u
s
e
f
u
l
to
ca
p
tu
r
e
t
h
ese
d
etail
s
.
T
ab
le
1
g
iv
e
s
th
e
co
u
n
t
o
f
t
h
e
v
ar
io
u
s
B
I
O
tag
s
av
ai
lab
le
i
n
t
h
e
tr
ain
in
g
d
ataset.
I
n
B
I
O
tag
g
in
g
,
B
,
I
a
n
d
O
co
r
r
esp
o
n
d
to
t
h
e
b
e
g
in
n
i
n
g
,
i
n
s
id
e
a
n
d
o
u
t
s
id
e
o
r
n
o
n
-
en
t
it
y
to
k
e
n
r
esp
ec
ti
v
el
y
.
Fo
r
ex
a
m
p
le,
t
h
e
lab
el
B
-
g
r
o
u
p
,
I
-
g
r
o
u
p
r
ep
r
esen
t
t
h
e
b
eg
i
n
n
i
n
g
a
n
d
in
s
id
e
o
f
th
e
g
r
o
u
p
r
esp
ec
tiv
el
y
an
d
O
r
ep
r
esen
ts
n
o
n
-
e
n
ti
t
y
to
k
en
s
.
T
ab
le
1
.
T
ag
s
an
d
its
co
u
n
ts
T
a
g
B
-
b
r
a
n
d
I
-
b
r
a
n
d
B
-
d
r
u
g
I
-
d
r
u
g
B
-
g
r
o
u
p
I
-
g
r
o
u
p
B
-
d
r
u
g
_
n
I
-
d
r
u
g
_
n
C
o
u
n
t
1
4
2
5
48
8
3
3
3
5
4
9
3
0
2
7
1
9
3
7
96
29
2
.
2.
Co
m
po
nent
s
o
f
t
he
m
o
del
2
.
2
.1
.
Sente
nce
e
m
bedd
ing
la
y
er
Sen
te
n
ce
e
m
b
ed
d
in
g
tec
h
n
iq
u
es
ad
d
r
ess
es
w
h
o
le
s
e
n
te
n
ce
s
an
d
t
h
eir
s
e
m
a
n
tic
d
ata
as
v
ec
to
r
s
.
T
h
is
aid
s
th
e
m
ac
h
i
n
e
i
n
u
n
d
er
s
tan
d
in
g
th
e
s
p
ec
i
f
ic
co
n
te
x
t
an
d
d
if
f
er
e
n
t
s
u
b
tlet
ies
i
n
th
e
w
h
o
le
co
n
ten
t
.
E
L
M
o
(
e
m
b
ed
d
in
g
f
r
o
m
lan
g
u
ag
e
m
o
d
els
)
is
an
e
m
b
ed
d
in
g
m
o
d
e
l
[
1
1
]
w
h
ic
h
f
u
n
c
tio
n
s
f
o
r
an
e
n
tire
s
e
n
te
n
ce
.
T
h
e
w
o
r
d
r
ep
r
esen
tatio
n
u
s
ed
i
n
th
is
m
o
d
el
is
d
ee
p
l
y
co
n
tex
t
u
ali
ze
d
th
at
ca
n
m
o
d
el
c
h
ar
ac
ter
is
tics
s
u
ch
a
s
s
y
n
ta
x
an
d
s
e
m
a
n
tic
s
o
f
th
e
w
o
r
d
an
d
also
f
in
d
s
h
o
w
t
h
ese
ch
a
r
ac
ter
is
tics
ca
n
b
e
u
s
e
d
f
o
r
d
if
f
er
e
n
t
lin
g
u
is
t
ic
co
n
tex
t
s
[
2
1
]
.
Sin
ce
th
i
s
is
a
l
s
o
a
ch
ar
ac
ter
b
ased
r
ep
r
esen
tatio
n
,
in
s
tead
o
f
s
i
m
p
l
y
lo
o
k
in
g
in
to
w
o
r
d
s
an
d
th
eir
v
ec
to
r
s
,
it
g
en
er
ate
s
v
ec
t
o
r
s
th
at
f
o
r
m
r
ep
r
ese
n
tatio
n
s
o
f
to
k
e
n
s
t
h
at
ar
e
n
o
t see
n
d
u
r
i
n
g
tr
ai
n
i
n
g
.
2
.
2
.
2.
Sta
ck
ed
B
i
-
L
ST
M
la
y
er
s
Bi
-
L
ST
M,
w
h
ich
to
o
k
its
id
e
a
f
r
o
m
b
id
ir
ec
tio
n
al
R
NN
t
h
a
t
p
r
o
ce
ed
s
in
b
o
th
d
ir
ec
tio
n
s
–
f
o
r
w
ar
d
an
d
r
ev
er
s
e
–
h
a
v
in
g
in
d
ep
e
n
d
en
t
h
id
d
en
la
y
er
s
f
o
r
ea
c
h
d
ir
ec
tio
n
.
T
h
ese
h
id
d
en
la
y
er
s
ar
e
lin
k
ed
to
a
co
m
m
o
n
o
u
tp
u
t
la
y
er
.
B
i
-
L
S
T
M
n
et
w
o
r
k
s
ar
e
f
o
u
n
d
to
b
e
b
etter
in
m
a
n
y
r
esear
c
h
ar
ea
s
s
u
c
h
as
tr
af
f
ic
p
r
ed
ictio
n
[
1
2
]
,
s
p
ee
ch
r
ec
o
g
n
itio
n
[
2
2
]
an
d
p
h
o
n
e
m
e
clas
s
i
f
icatio
n
[
2
3
]
.
T
h
e
p
r
ev
io
u
s
s
tu
d
ies
[
2
2
]
,
[
2
3
]
h
av
e
p
r
o
v
ed
th
at
d
ee
p
L
STM
m
o
d
els
ie.
s
tack
ed
L
ST
M
m
o
d
el
s
w
it
h
m
an
y
h
id
d
en
la
y
er
s
ca
n
d
ev
el
o
p
a
s
u
cc
ess
i
v
el
y
m
o
r
e
s
i
g
n
if
ican
t
le
v
el
o
f
d
escr
ip
tio
n
f
o
r
th
e
s
eq
u
e
n
tial
d
at
a
an
d
h
e
n
ce
co
u
ld
p
er
f
o
r
m
m
o
r
e
ef
f
ec
ti
v
el
y
a
s
ill
u
s
tr
ated
in
[
1
2
]
u
s
in
g
a
t
w
o
-
la
y
er
Bi
-
L
ST
M
m
o
d
el
f
o
r
tr
af
f
ic
p
r
ed
ictio
n
.
T
h
e
ef
f
ec
ti
v
en
e
s
s
o
f
s
tac
k
ed
B
i
-
LS
T
M
n
et
w
o
r
k
s
f
o
r
b
etter
class
i
f
icatio
n
an
d
r
eg
r
ess
io
n
tas
k
s
w
as
also
d
e
m
o
n
s
tr
ated
i
n
[
1
7
]
,
[
1
8
]
,
an
d
[
2
4
]
.
I
n
o
u
r
w
o
r
k
,
b
ased
o
n
t
h
e
p
r
ev
io
u
s
r
esear
c
h
w
o
r
k
s
,
w
e
h
a
v
e
also
ad
o
p
ted
th
e
t
w
o
la
y
er
s
o
f
s
tac
k
e
d
Bi
-
L
ST
M.
T
h
e
lo
w
er
la
y
er
o
f
B
i
-
L
ST
M
is
m
o
r
e
ap
p
r
o
p
r
iat
e
f
o
r
ex
tr
ac
ti
n
g
u
s
e
f
u
l
i
n
f
o
r
m
atio
n
f
r
o
m
th
e
i
n
p
u
t
v
ec
to
r
s
.
T
h
e
u
n
iq
u
e
v
ec
to
r
s
o
b
tain
ed
f
o
r
ea
c
h
w
o
r
d
i
n
t
h
e
s
en
te
n
ce
u
s
in
g
t
h
e
s
e
n
ten
ce
e
m
b
ed
d
i
n
g
m
o
d
el
is
g
iv
e
n
a
s
i
n
p
u
t
to
t
h
e
B
i
-
L
S
T
M
lay
er
1
w
h
ic
h
h
elp
s
i
n
ca
p
tu
r
in
g
t
h
e
f
ea
t
u
r
es
f
o
r
p
r
ed
ictin
g
t
h
e
d
r
u
g
ca
teg
o
r
ies.
As
w
e
h
av
e
u
s
ed
t
w
o
s
tac
k
ed
la
y
er
s
,
th
e
s
ec
o
n
d
lay
er
o
r
th
e
to
p
lay
er
o
f
th
e
s
tack
u
til
iz
es
th
e
f
ea
t
u
r
es
lear
n
ed
f
r
o
m
t
h
e
o
u
t
p
u
t
o
f
t
h
e
lo
w
er
la
y
er
.
I
t
al
s
o
lear
n
s
m
a
n
y
co
m
p
le
x
f
ea
t
u
r
es
to
e
n
h
a
n
ce
t
h
e
ac
h
iev
e
m
e
n
t o
f
t
h
e
m
o
d
el.
2
.
2
.
3
.
Resid
ua
l LS
T
M
co
nn
ec
t
io
n
T
o
o
v
er
co
m
e
th
e
is
s
u
e
o
f
v
an
i
s
h
i
n
g
g
r
ad
ie
n
ts
,
a
r
e
s
id
u
al
L
S
T
M
co
n
n
ec
tio
n
i
s
u
s
ed
w
h
ich
p
r
o
v
id
es
a
b
y
p
as
s
li
n
k
b
et
w
ee
n
th
e
la
y
er
s
[
1
3
]
.
T
h
e
s
h
o
r
tcu
t
p
at
h
co
u
ld
b
e
f
r
o
m
a
n
y
lo
w
er
la
y
er
s
.
I
n
t
h
is
p
ap
er
,
w
e
h
a
v
e
u
s
ed
r
esid
u
al
L
ST
M
as a
s
h
o
r
tc
u
t b
et
w
ee
n
t
h
e
s
tac
k
ed
B
i
-
L
S
T
M
lay
er
s
.
Si
n
ce
w
e
h
a
v
e
u
s
e
d
o
n
l
y
t
w
o
s
tac
k
ed
Bi
-
L
ST
M
la
y
er
s
,
t
h
e
s
h
o
r
tcu
t
is
ta
k
en
f
r
o
m
t
h
e
o
u
tp
u
t
o
f
la
y
er
1
an
d
ad
d
ed
w
ith
th
e
o
u
tp
u
t
o
f
la
y
er
2
a
s
s
h
o
w
n
in
F
ig
u
r
e
1
.
2
.
3
.
Arc
hite
ct
ure
o
f
t
he
pro
po
s
ed
s
y
s
t
e
m
T
h
e
s
y
s
te
m
ar
c
h
itect
u
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
DNE
R
s
y
s
te
m
i
s
s
h
o
w
n
in
Fi
g
u
r
e
2
.
T
h
e
m
ai
n
n
o
v
elt
y
o
f
th
i
s
ar
c
h
itect
u
r
e
i
s
th
e
i
n
cl
u
s
io
n
o
f
s
en
ten
ce
e
m
b
ed
d
in
g
la
y
er
w
h
ic
h
en
ab
les
t
h
e
s
y
s
te
m
to
ca
p
tu
r
e
th
e
s
e
m
an
t
ic
in
f
o
r
m
atio
n
b
etter
th
an
w
o
r
d
o
r
ch
ar
ac
ter
e
m
b
ed
d
i
n
g
m
o
d
el
s
.
I
n
ad
d
itio
n
,
th
e
ar
ch
itect
u
r
e
co
m
p
r
is
e
s
o
f
s
tac
k
ed
B
i
-
L
STM
(
w
it
h
t
w
o
la
y
er
s
)
an
d
r
esid
u
al
L
ST
M
to
ca
p
tu
r
e
th
e
c
o
m
p
lex
f
ea
t
u
r
es
o
f
th
e
s
en
te
n
ce
s
a
n
d
to
o
v
er
co
m
e
th
e
p
r
o
b
lem
o
f
v
a
n
i
s
h
i
n
g
g
r
ad
ien
ts
r
esp
ec
tiv
el
y
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
n
o
ve
l d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
d
r
u
g
n
a
med
en
tity reco
g
n
itio
n
(
T.
Ma
th
u
)
1887
C
o
n
s
id
er
an
in
p
u
t
s
eq
u
en
ce
w
=(
w
1
,
w
2
,
…
w
n
)
w
h
er
e
w
1
,
.
.
w
n
r
ep
r
esen
t
s
th
e
w
o
r
d
s
in
th
e
s
eq
u
en
c
e
p
ad
d
ed
w
it
h
a
f
ix
ed
le
n
g
t
h
f
o
r
ea
ch
s
en
te
n
ce
a
n
d
s
er
ie
s
o
f
o
u
tp
u
t
tag
s
e=
(
e
1
,
e
2
…e
n
)
w
h
er
e
e
1
,
e
2
,..e
n
r
ef
er
s
to
en
titi
e
s
.
T
h
e
s
en
te
n
ce
e
m
b
ed
d
in
g
la
y
er
E
L
Mo
cr
ea
tes
t
h
e
v
e
cto
r
to
ev
er
y
w
o
r
d
f
o
r
s
en
te
n
c
e
w
.
T
h
e
in
p
u
t
f
o
r
Bi
-
L
ST
M
la
y
er
1
is
th
e
s
eq
u
e
n
ce
o
f
w
o
r
d
v
ec
to
r
s
f
o
u
n
d
f
r
o
m
t
h
e
s
e
n
ten
ce
e
m
b
ed
d
in
g
la
y
er
.
T
h
e
s
eq
u
e
n
ce
o
f
h
id
d
en
s
tate
s
f
o
r
m
s
th
e
o
u
tp
u
t
o
f
th
e
la
y
er
1
an
d
th
at
i
n
tu
r
n
b
ec
o
m
es
t
h
e
in
p
u
t
to
th
e
Bi
-
L
ST
M
la
y
er
2
.
T
h
e
Bi
-
L
ST
M
lay
er
s
h
av
e
t
w
o
p
a
s
s
es
i
n
ea
ch
la
y
er
,
n
a
m
e
l
y
f
o
r
w
ar
d
p
ass
/
f
o
r
w
ar
d
la
y
er
an
d
r
ev
er
s
e
p
ass
/r
e
v
er
s
e
la
y
er
.
Fig
u
r
e
1
.
Stack
ed
B
i
-
L
ST
M
w
ith
r
esid
u
a
l
L
ST
M
co
n
n
ec
tio
n
Fig
u
r
e
2
.
S
y
s
te
m
ar
ch
itect
u
r
e
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
I
n
th
e
f
o
r
w
ar
d
la
y
er
,
t
h
e
i
n
p
u
t
s
eq
u
e
n
ce
is
f
ed
f
r
o
m
t
i
m
e
t=
1
to
T
n
an
d
f
r
o
m
t=T
n
to
1
in
th
e
r
ev
er
s
e
la
y
er
.
T
h
e
h
id
d
en
v
ec
to
r
s
eq
u
en
ce
an
d
th
e
o
u
tp
u
t
s
eq
u
en
ce
ar
e
co
m
p
u
ted
f
r
o
m
t
h
e
B
i
-
L
ST
M
lay
er
.
T
h
e
h
id
d
en
v
ec
to
r
s
eq
u
en
ce
ca
n
b
e
f
o
r
w
ar
d
s
eq
u
en
ce
a
n
d
r
ev
er
s
e
s
eq
u
en
ce
r
ep
r
esen
ted
b
y
⃗
⃖
r
esp
ec
tiv
el
y
.
T
h
e
f
o
r
w
ar
d
la
y
er
is
iter
ated
f
r
o
m
t=1
to
T
n
a
n
d
th
e
r
ev
er
s
e
la
y
er
is
iter
ated
f
r
o
m
t=T
n
to
1
.
T
h
e
ca
lcu
latio
n
o
f
f
o
r
w
ar
d
h
id
d
en
v
ec
to
r
,
r
ev
er
s
e
h
id
d
en
v
ec
to
r
an
d
o
u
tp
u
t
s
eq
u
e
n
ce
r
esp
ec
ti
v
el
y
ar
e
s
h
o
w
n
as
i
n
(
1
)
,
(
2
)
an
d
(
3
)
.
⃗
=
(
⃗
+
⃗
⃗
⃗
−
1
+
⃗
)
(1
)
⃖
=
(
⃖
+
⃖
⃖
⃖
+
1
+
⃖
)
(2
)
=
⃗
⃗
+
⃖
⃖
+
(3
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
1
9
,
No
.
6
,
Decem
b
er
2021
:
18
84
-
1
891
1888
W
h
er
e
s
y
m
b
o
lize
s
t
h
e
h
id
d
en
la
y
er
f
u
n
ct
io
n
,
s
y
m
b
o
lizes
t
h
e
v
ar
io
u
s
w
ei
g
h
t
m
atr
ices,
s
y
m
b
o
lize
s
t
h
e
b
ias
v
ec
to
r
s
a
n
d
d
en
o
tes
t
h
e
o
u
tp
u
t
la
y
er
v
ar
iab
le.
T
h
e
w
e
i
g
h
t
s
⃗
,
⃗
⃗
,
⃖
,
⃖
⃖
an
d
t
h
e
b
iases
⃗
,
⃖
r
ep
r
esen
ts
t
h
e
m
o
d
el
p
ar
a
m
e
t
er
s
in
(
1
)
an
d
(
2
)
.
T
h
en
th
e
b
i
as
p
ar
a
m
eter
is
co
n
ca
te
n
ated
w
it
h
f
o
r
w
ar
d
h
id
d
en
la
y
er
⃗
an
d
r
ev
er
s
e
h
id
d
en
la
y
er
⃖
to
g
et
th
e
o
u
tp
u
t
la
y
er
as
in
(
3
)
.
I
n
g
en
er
al,
w
h
e
n
m
o
r
e
th
a
n
on
e
B
i
-
L
ST
M
lay
er
is
u
s
ed
,
t
h
e
f
o
r
w
ar
d
an
d
r
ev
er
s
e
h
id
d
en
s
eq
u
e
n
ce
ca
n
b
e
co
m
p
u
ted
f
o
r
n
=1
to
N
an
d
t=1
to
T
n
as g
iv
e
n
i
n
(
4
)
an
d
(
5
).
⃗
=
(
⃗
−
1
⃗
−
1
+
⃗
⃗
+
)
(4
)
⃖
=
(
⃖
−
1
⃖
−
1
+
⃖
⃖
+
)
(5
)
T
h
e
o
u
tp
u
t seq
u
e
n
ce
is
ca
lc
u
la
ted
as g
iv
e
n
i
n
(
6
)
.
=
+
(6
)
No
w
,
t
h
e
r
esid
u
al
L
ST
M
co
n
n
ec
tio
n
is
ap
p
lied
b
y
ad
d
in
g
t
h
e
o
u
tp
u
t seq
u
e
n
ce
s
o
f
B
i
-
L
ST
M
la
y
er
2
w
it
h
w
.
I
t
is
r
ef
er
r
ed
b
y
H(
w
)
an
d
i
s
s
h
o
w
n
i
n
(
7
)
.
(
)
=
+
(7
)
T
h
e
v
an
i
s
h
i
n
g
g
r
ad
ien
t
p
r
o
b
le
m
co
u
ld
b
e
r
eso
l
v
ed
b
y
t
h
e
ap
p
licatio
n
o
f
r
e
s
id
u
al
L
ST
M
s
in
ce
th
e
g
r
ad
ien
t
s
co
u
ld
p
ass
th
r
o
u
g
h
t
h
e
la
y
er
s
d
ir
ec
tl
y
b
y
u
s
i
n
g
t
h
e
ad
d
it
io
n
o
p
er
ato
r
.
T
h
e
r
esid
u
al
L
ST
M
[
1
3
]
p
er
m
its
d
if
f
er
e
n
t
la
y
er
s
o
f
L
ST
M
to
ad
eq
u
atel
y
tr
ai
n
co
m
p
lex
n
et
w
o
r
k
s
w
it
h
an
o
p
tio
n
al
te
m
p
o
r
al
s
h
o
r
tcu
t
p
ath
f
r
o
m
d
ee
p
er
lev
els.
Fin
a
ll
y
,
t
h
e
s
co
r
es g
i
v
e
n
f
o
r
ea
ch
lab
el
b
y
th
e
B
i
-
L
ST
M
la
y
er
s
ar
e
p
r
o
v
id
ed
as a
n
i
n
p
u
t
in
to
t
h
e
s
o
f
t
m
ax
cla
s
s
i
f
ier
o
u
tp
u
t la
y
er
,
as g
iv
e
n
i
n
(
8
)
.
=
(
(
)
)
(
8
)
T
h
is
la
y
er
p
r
o
d
u
ce
s
th
e
p
r
ed
i
cted
p
r
o
b
ab
ilit
ies
f
o
r
all
th
e
lab
els
to
ea
ch
w
o
r
d
u
s
ed
f
o
r
class
i
f
icatio
n
w
h
ic
h
in
cl
u
d
es
B
-
d
r
u
g
,
I
-
d
r
u
g
,
B
-
b
r
an
d
,
I
-
b
r
an
d
,
B
-
g
r
o
u
p
,
I
-
g
r
o
u
p
an
d
O.
T
h
e
lab
el
w
h
ic
h
h
as
g
o
t
th
e
h
i
g
h
e
s
t
p
r
ed
ictio
n
in
th
e
s
eq
u
en
ce
w
o
u
ld
b
e
co
n
s
id
e
r
ed
as th
e
lab
el
f
o
r
th
e
w
o
r
d
.
2
.
4
.
P
s
eudo
co
de
o
f
t
he
pro
po
s
ed
DNER M
o
del
T
h
e
g
en
er
al
s
tep
s
o
f
t
h
e
p
r
o
p
o
s
ed
DNE
R
s
y
s
te
m
b
ased
o
n
s
tack
ed
B
i
-
L
ST
M
an
d
r
esid
u
a
l
L
ST
M
is
s
h
o
w
n
in
A
l
g
o
r
it
h
m
1
.
A
l
g
o
r
i
t
h
m
1
:
1
Input
Sentences
of
various
l
engths
from
DDI2013
Dr
ugbank
training
dataset.
2
Preprocessing
Tokenize the sentences from the input dataset.
3
Fo
r
ea
ch
to
ke
n,
in
cl
ud
e
Pa
rt
-
of
-
Sp
ee
ch
(P
OS
)
t
ag
s
an
d
BI
O
dr
ug
labels.
4
Ea
ch
to
ke
ni
ze
d
se
nt
en
ce
is
pa
dd
ed
wi
th
_P
AD
_
to
ke
ns
to
br
in
g
it
to
a fixed length.
5
Model
Construction
Construct Sentence embedd
ing using ELMo for
the pr
e
-
processed input
dataset.
6
Implement
Stacked
Bi
-
LS
TM
la
ye
rs
(t
wo
la
ye
rs
)
to
ob
ta
in
th
e
previous
and
future
co
ntextual
information
fo
r
more
accurate
prediction ie. the sequen
ce of hidden vectors
obta
ined from Bi
-
LS
TM
la
ye
r
1
is
gi
ve
n
to
Bi
-
L
ST
M
la
ye
r
2
us
in
g
(2
),
(3
),(
4),(5,(6)
,
and
(7).
7
Establish a Residual conn
ection using a vector
add
ition between the
Bi
-
LS
TM
la
ye
r
1
ou
tp
ut
a
nd
Bi
-
LS
TM
la
ye
r
2
ou
tp
u
t
to
pr
ev
en
t
Bi
-
LSTM
suffering from the vanishing gradient problem as in (8)
8
Finally,
apply
softmax
f
unction
in
the
output
l
ay
er
to
cl
as
si
fy
drug names into multiple categories of drugs as in (9)
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
P
er
f
o
rm
a
nce
m
et
ric
s
T
h
e
DNE
R
m
o
d
el
n
ee
d
s
to
b
e
ev
alu
ated
b
y
ap
p
r
o
p
r
iate
an
d
u
n
a
m
b
i
g
u
o
u
s
m
etr
ics
to
r
ig
h
tl
y
j
u
d
g
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
m
o
d
el
.
P
r
ec
is
io
n
,
r
ec
all
an
d
F1
-
s
co
r
e
ar
e
u
s
ed
as
m
ea
s
u
r
e
m
e
n
ts
t
o
ass
ess
th
e
m
o
d
el.
As
f
o
u
r
d
if
f
er
en
t
en
titi
e
s
ar
e
av
ailab
le
in
DDI
2
0
1
3
co
r
p
u
s
,
it
is
n
ec
ess
ar
y
to
co
m
p
u
te
th
e
o
v
er
all
p
er
f
o
r
m
an
c
e
o
f
all
th
e
e
n
tit
y
cla
s
s
es.
I
n
th
is
r
eg
ar
d
,
w
e
ta
k
e
t
h
e
m
icr
o
-
av
er
ag
e
F1
-
s
co
r
e
[
2
5
]
m
etr
ic
f
o
r
th
e
co
m
p
ar
is
o
n
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
n
o
ve
l d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
d
r
u
g
n
a
med
en
tity reco
g
n
itio
n
(
T.
Ma
th
u
)
1889
w
it
h
o
th
er
s
y
s
te
m
s
.
M
icr
o
-
av
er
ag
e
F1
-
s
co
r
e,
as
in
(
9
)
is
d
ef
i
n
ed
as
th
e
h
ar
m
o
n
ic
m
ea
n
o
f
m
icr
o
-
a
v
er
ag
e
p
r
ec
is
io
n
(
m
P
)
an
d
m
icr
o
-
a
v
er
ag
e
r
ec
all
(
m
R
)
an
d
it is
g
iv
e
n
in
(
1
0
)
an
d
(
1
1
)
r
esp
ec
tiv
el
y
.
Mic
r
o
-
av
er
ag
e
F1
-
s
co
r
e
=
2
*
(
m
P
x
m
R
)
/(
m
P
+
m
R
)
,
w
h
er
e
(
9
)
m
P
=tp
1
+
tp
2
+…+
tp
n
/(
tp
1
+
tp
2
+…+
tp
n
+
f
p
1
+
f
p
2
+…+
f
p
n
)
(
1
0
)
m
R
=
tp
1
+
tp
2
+…+
tp
n
/(
tp
1
+
tp
2
+…+
tp
n
+
f
n
1
+
f
n
2
+
…+f
n
n
)
(
1
1
)
tp
,
f
p
an
d
f
n
r
ep
r
esen
ts
t
h
e
tr
u
e
p
o
s
itiv
e,
f
al
s
e
p
o
s
itiv
e
a
n
d
f
alse n
e
g
ati
v
e
r
esp
ec
tiv
el
y
.
3
.
2
.
E
x
peri
m
ent
a
l
s
et
up
T
h
e
d
ataso
u
r
ce
u
s
ed
is
d
escr
ib
ed
in
s
ec
tio
n
2
.
2
.
T
h
e
tr
ai
n
in
g
d
ata
g
i
v
e
n
in
D
DI
2
0
1
3
d
r
u
g
b
an
k
co
r
p
u
s
is
p
r
ep
r
o
ce
s
s
ed
an
d
we
h
a
v
e
u
s
ed
4
9
9
0
s
en
te
n
ce
s
with
8
0
0
6
u
n
iq
u
e
w
o
r
d
s
a
n
d
3
0
%
o
f
t
h
e
d
atase
t
i
s
co
n
s
id
er
ed
as
test
d
ata
s
et.
W
e
u
s
ed
‘
ad
a
m
’
o
p
ti
m
izer
w
it
h
lo
s
s
as
‘
s
p
ar
s
e_
ca
teg
o
r
i
ca
l_
cr
o
s
s
en
tr
o
p
y
’
f
o
r
co
m
p
il
in
g
t
h
e
m
o
d
el.
T
h
e
b
atch
s
ize
is
3
2
a
n
d
t
h
e
n
u
m
b
er
o
f
ep
o
ch
s
is
m
ad
e
as
8
.
T
h
e
r
ec
u
r
r
en
t
d
r
o
p
o
u
t
i
s
tak
en
a
s
0
.
2
.
T
h
e
s
o
f
t
m
a
x
f
u
n
ctio
n
is
f
u
ll
y
u
s
ed
in
t
h
e
class
if
ier
'
s
o
u
tp
u
t
la
y
er
as
t
h
e
ac
ti
v
atio
n
la
y
er
w
h
er
e
th
e
p
r
o
b
ab
ilit
ies o
f
id
en
ti
f
y
in
g
th
e
in
p
u
t c
las
s
ar
e
ef
f
ec
tiv
e
l
y
ac
h
iev
ed
.
3
.
3
.
Resul
t
a
na
ly
s
is
Fig
u
r
e
3
s
h
o
w
s
t
h
e
d
etailed
r
esu
lt
s
o
b
tai
n
ed
i
n
t
h
e
f
o
r
m
o
f
p
r
ec
is
io
n
,
r
ec
all
an
d
f
1
-
s
co
r
e
f
o
r
e
v
er
y
class
lab
el
o
f
d
r
u
g
e
n
tit
y
u
s
i
n
g
t
h
e
p
r
o
p
o
s
ed
m
o
d
el.
T
h
e
m
o
d
el
h
as
p
er
f
o
r
m
ed
w
e
ll
w
i
th
F1
-
s
co
r
e
v
al
u
e
o
f
m
o
r
e
th
a
n
8
5
%
i
n
ca
te
g
o
r
izin
g
t
h
e
d
r
u
g
en
t
ities
ex
ce
p
t
th
e
d
r
u
g
_
n
cla
s
s
lab
el.
T
h
is
m
a
y
b
e
d
u
e
to
th
e
f
ac
t
th
at
les
s
n
u
m
b
er
o
f
d
ata
is
av
a
ilab
le
in
th
e
tr
ai
n
in
g
d
ataset
to
lear
n
d
r
u
g
_
n
clas
s
lab
el.
Ho
w
ev
er
,
th
is
co
u
ld
b
e
ig
n
o
r
ed
as
lar
g
er
p
ar
t
o
f
th
e
co
r
p
u
s
co
n
s
is
tin
g
o
f
d
r
u
g
,
g
r
o
u
p
a
n
d
b
r
an
d
lab
els
h
a
v
e
b
ee
n
clas
s
i
f
ied
ef
f
icien
tl
y
.
Sin
ce
i
t
is
n
ec
es
s
ar
y
to
co
m
p
u
te
th
e
o
v
er
all
p
er
f
o
r
m
an
ce
o
f
all
th
e
e
n
tit
y
clas
s
es,
an
al
y
s
i
s
h
as
b
ee
n
ca
r
r
ied
o
u
t
b
ased
o
n
th
e
p
er
f
o
r
m
a
n
ce
m
e
tr
ics
s
h
o
w
n
i
n
(
9
)
-
(
1
1
)
to
c
o
n
tr
ast
th
e
p
r
o
p
o
s
ed
m
o
d
el
w
it
h
t
h
e
ex
is
t
in
g
D
L
b
ased
D
NE
R
m
o
d
el
(
L
ST
M
-
co
n
d
itio
n
al
r
an
d
o
m
f
ield
(
C
R
F)
)
[
9
]
as
w
ell
as
o
th
er
m
o
d
els
f
r
o
m
th
e
D
DI
2
0
1
3
ch
allen
g
e
[
2
6
]
,
[
2
7
]
.
T
h
e
r
esu
lts
ar
e
s
h
o
w
n
g
r
a
p
h
icall
y
i
n
Fi
g
u
r
e
4
.
Fig
u
r
e
3
.
P
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
Fig
u
r
e
4
.
Mic
r
o
-
av
er
ag
e
p
er
f
o
r
m
an
ce
o
f
p
r
o
p
o
s
ed
m
o
d
el
v
s
Oth
er
DNE
R
m
o
d
els
T
ab
le
2
s
h
o
w
s
t
h
e
r
esu
lts
o
f
t
h
e
p
er
f
o
r
m
a
n
ce
m
etr
ics
o
b
tai
n
ed
f
o
r
ea
ch
cla
s
s
lab
el
f
o
r
th
e
p
r
o
p
o
s
ed
m
o
d
el
as
w
el
l
as
o
th
er
DNE
R
s
y
s
te
m
s
.
I
n
L
ST
M
-
C
R
F,
t
h
e
f
ea
tu
r
es
ar
e
b
ased
o
n
b
o
th
w
o
r
d
an
d
ch
ar
ac
ter
lev
el
e
m
b
ed
d
in
g
.
T
h
e
p
er
ce
n
tag
e
o
f
i
m
p
r
o
v
e
m
e
n
t
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el
o
v
er
L
ST
M
-
C
R
F
w
it
h
r
esp
ec
t
to
m
icr
o
-
a
v
er
ag
e
p
r
ec
is
io
n
,
r
ec
all
an
d
F1
-
s
co
r
e
ar
e
9
.
2
2
,
1
1
.
1
9
,
an
d
1
1
.
1
7
r
esp
ec
tiv
el
y
.
L
i
u
et
a
l
.
[
2
6
]
,
h
av
e
ex
p
er
i
m
e
n
ted
a
C
R
F
b
ased
m
o
d
el
(
L
I
U)
w
it
h
s
e
m
a
n
tic
f
ea
t
u
r
es
b
as
ed
o
n
w
o
r
d
e
m
b
ed
d
in
g
.
O
n
co
m
p
ar
i
s
o
n
w
it
h
t
h
i
s
s
y
s
te
m
,
o
u
r
p
r
o
p
o
s
e
d
m
o
d
el
th
at
u
s
e
s
s
e
n
te
n
ce
e
m
b
ed
d
i
n
g
in
s
tac
k
ed
Bi
-
L
ST
M
an
d
r
esid
u
al
L
ST
M
h
as
i
m
p
r
o
v
ed
th
e
m
icr
o
-
a
v
er
ag
e
p
r
ec
is
io
n
,
r
ec
all
an
d
f
1
-
s
co
r
e
b
y
4
.
1
8
%,
1
4
.
6
3
%
,
an
d
8
.
8
0
%
r
esp
ec
tiv
el
y
.
R
o
ck
tä
s
ch
e
l
et
a
l.
[
2
7
]
,
s
tu
d
ied
a
m
o
d
el
th
a
t
r
an
k
ed
f
i
r
s
t
in
th
e
DDI
2
0
1
3
ch
alle
n
g
e
s
t
u
d
ies
t
h
e
i
m
p
ac
t
o
f
d
o
m
ai
n
s
p
ec
i
f
ic
f
ea
t
u
r
es
u
s
in
g
li
n
e
ar
ch
ai
n
C
R
F
(
W
B
I
)
f
o
r
id
en
tify
i
n
g
d
r
u
g
n
a
m
e
s
.
W
h
ile
co
m
p
ar
i
n
g
w
i
th
t
h
i
s
m
o
d
el,
a
s
i
g
n
i
f
ican
t
i
m
p
r
o
v
e
m
e
n
t
o
f
1
8
.
8
%,
1
8
.
1
2
%
,
an
d
1
7
.
6
4
%
r
esp
ec
tiv
el
y
f
o
r
m
icr
o
-
av
er
a
g
e
p
r
ec
is
io
n
,
r
ec
all
an
d
F1
-
s
co
r
e
is
s
h
o
w
n
b
y
t
h
e
p
r
o
p
o
s
ed
m
o
d
el.
B
ased
o
n
t
h
e
ab
o
v
e
r
es
u
lt
s
,
i
t
h
a
s
b
ee
n
o
b
s
er
v
ed
t
h
at
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
p
er
f
o
r
m
s
b
et
ter
as
t
h
e
s
en
te
n
ce
e
m
b
ed
d
in
g
in
cl
u
d
ed
in
t
h
e
ar
ch
itect
u
r
e
co
n
s
id
er
s
th
e
co
m
p
le
te
s
e
n
te
n
ce
f
o
r
s
y
n
ta
x
a
n
d
s
e
m
a
n
ti
c
f
ea
t
u
r
es
u
n
l
ik
e
w
o
r
d
e
m
b
ed
d
i
n
g
w
h
ich
m
a
y
i
g
n
o
r
e
s
o
m
e
o
f
th
e
c
h
ar
ac
ter
f
ea
t
u
r
es.
E
v
e
n
wh
en
c
h
ar
ac
ter
le
v
el
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
1
9
,
No
.
6
,
Decem
b
er
2021
:
18
84
-
1
891
1890
e
m
b
ed
d
in
g
is
co
m
b
in
ed
w
i
th
w
o
r
d
e
m
b
ed
d
in
g
as
in
[
9
]
,
th
e
s
en
te
n
ce
le
v
el
e
m
b
ed
d
in
g
p
er
f
o
r
m
s
b
etter
in
th
e
p
r
o
p
o
s
ed
m
o
d
el
i
n
ter
m
s
o
f
all
t
h
e
m
icr
o
-
a
v
er
ag
ed
p
er
f
o
r
m
an
ce
m
etr
ics.
I
n
ad
d
itio
n
,
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
s
h
o
w
s
t
h
e
p
o
w
er
o
f
u
s
i
n
g
B
i
-
L
ST
M
la
y
er
s
th
a
t
r
ea
d
s
th
e
co
n
tex
t
b
ac
k
an
d
f
o
r
th
i
n
th
e
s
e
n
ten
ce
a
n
d
c
ap
tu
r
es
th
e
co
n
tex
t
w
el
l
r
at
h
er
t
h
an
u
s
i
n
g
a
s
i
n
g
le
L
ST
M
la
y
er
.
A
l
s
o
,
t
h
e
a
u
to
m
at
ic
e
x
tr
ac
tio
n
o
f
f
ea
t
u
r
es
u
s
in
g
s
tack
ed
B
i
-
L
ST
M
la
y
er
s
i
n
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
is
u
s
ed
to
r
ec
o
g
n
ize
e
n
tit
ies
b
etter
t
h
a
n
u
s
in
g
ML
al
g
o
r
it
h
m
s
lik
e
C
R
F a
s
in
[
2
6
]
,
[
2
7
]
.
I
n
ad
d
itio
n
to
th
e
ab
o
v
e
r
e
s
u
l
ts
,
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
is
ev
al
u
ated
t
o
ad
d
r
ess
th
e
m
aj
o
r
co
n
ce
r
n
in
th
e
DN
E
R
f
ield
in
id
en
ti
f
y
i
n
g
t
h
e
n
-
g
r
a
m
d
r
u
g
en
titi
e
s
w
h
er
e
n
>1
an
d
th
e
r
esu
lts
ar
e
s
h
o
w
n
i
n
T
ab
le
3
.
T
h
e
r
es
u
lt
s
ar
e
p
r
o
m
i
s
in
g
in
id
en
t
if
y
i
n
g
t
h
e
2
-
g
r
a
m
a
n
d
3
-
g
r
a
m
d
r
u
g
e
n
ti
ties
.
Ho
w
e
v
er
,
th
e
r
esu
lt
s
ca
n
b
e
f
u
r
th
er
i
m
p
r
o
v
e
d
.
T
ab
le
2
.
C
o
m
p
ar
is
o
n
o
f
p
er
f
o
r
m
an
ce
m
etr
ic
s
-
p
r
o
p
o
s
ed
m
o
d
el
vs
o
th
er
DNE
R
s
y
s
te
m
s
C
l
a
ss L
a
b
e
l
P
r
o
p
o
se
d
M
o
d
e
l
L
S
T
M
-
CRF
LI
U
W
B
I
Pr
Re
Fs
Pr
Re
Fs
Pr
Re
Fs
Pr
Re
Fs
D
r
u
g
9
1
.
2
3
8
8
.
4
2
9
0
.
3
7
8
5
.
7
8
8
0
.
8
6
8
2
.
5
9
9
2
.
3
4
8
5
.
6
7
8
9
.
5
4
7
4
.
3
8
5
.
5
9
7
9
.
3
2
B
r
a
n
d
8
9
.
3
2
9
0
.
1
0
8
9
.
3
4
8
8
.
2
2
7
7
.
8
3
8
2
.
1
4
1
0
0
9
5
.
3
2
9
7
.
2
1
8
1
.
2
7
8
6
.
7
1
8
4
.
7
7
G
r
o
u
p
8
9
.
3
4
7
9
.
3
1
8
4
.
5
6
8
6
.
4
3
8
9
.
2
9
8
7
.
9
8
9
.
4
2
8
2
.
4
9
8
6
.
1
7
9
.
4
7
6
.
2
2
7
8
.
6
7
D
r
u
g
_
n
1
0
0
4
0
.
4
5
5
7
.
6
7
7
8
.
2
1
5
7
.
6
4
6
3
.
4
8
8
9
.
3
9
1
4
.
5
6
2
4
.
7
5
3
1
.
0
2
9
0
.
4
1
1
4
.
2
M
i
c
r
o
-
A
v
e
r
a
g
e
9
1
.
1
2
8
6
.
0
0
8
8
.
0
0
8
3
.
6
2
7
8
.
0
0
7
9
.
2
6
8
7
.
4
6
7
5
.
2
2
8
0
.
8
8
7
6
.
7
7
3
.
0
0
7
4
.
8
T
ab
le
3
.
P
er
ce
n
tag
e
o
f
n
-
g
r
a
m
en
titi
e
s
r
ec
o
g
n
ized
u
s
i
n
g
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
Ty
p
e
o
f
n
-
g
r
a
m e
n
t
i
t
y
%
R
e
c
o
g
n
i
z
e
d
2
-
g
r
a
m
8
3
.
8
9
%
3
-
g
r
a
m
7
6
.
6
7
%
4
-
g
r
a
m
4
0
%
4.
CO
NCLU
SI
O
N
I
n
t
h
is
p
ap
er
,
w
e
h
av
e
p
r
o
p
o
s
ed
a
n
o
v
e
l
DNE
R
ar
ch
itec
tu
r
e
u
s
i
n
g
t
h
e
late
s
t
a
n
d
ad
v
an
ce
d
D
L
m
o
d
el
s
.
I
t
in
cl
u
d
es
s
tack
ed
B
i
-
L
ST
M
an
d
a
r
esid
u
al
L
STM
la
y
er
s
.
T
h
e
ar
ch
itect
u
r
e
ta
k
es
t
h
e
i
n
p
u
t
i
n
t
h
e
f
o
r
m
o
f
v
ec
to
r
f
r
o
m
s
e
n
te
n
ce
lev
el
e
m
b
ed
d
in
g
m
o
d
el
a
n
d
o
u
tp
u
t
s
th
e
d
esire
d
d
r
u
g
lab
el
s
eq
u
en
ce
w
it
h
B
I
O
tag
g
in
g
s
ch
e
m
e.
W
e
co
n
d
u
ct
ed
ex
p
er
i
m
e
n
ts
u
s
in
g
D
DI
2
0
1
3
d
r
u
g
b
an
k
d
ataset.
O
u
r
p
r
o
p
o
s
ed
m
o
d
el
h
as
ac
h
iev
ed
h
i
g
h
er
p
er
f
o
r
m
a
n
ce
th
an
t
h
e
r
es
u
lts
o
b
tain
ed
u
s
i
n
g
t
h
e
s
a
m
e
d
ata
s
et
w
it
h
p
r
ev
i
o
u
s
s
tate
-
of
-
th
e
-
ar
t
m
o
d
el
s
.
B
esid
es,
th
e
p
r
o
p
o
s
ed
m
o
d
el
h
as
s
h
o
w
n
g
o
o
d
r
es
u
lts
in
r
ec
o
g
n
izi
n
g
2
-
a
n
d
3
-
g
r
a
m
e
n
tit
ies.
T
h
e
f
u
tu
r
e
r
esear
c
h
m
a
y
b
e
o
r
ien
ted
to
w
ar
d
s
f
u
r
t
h
er
i
m
p
r
o
v
i
n
g
t
h
e
p
er
f
o
r
m
a
n
ce
u
s
i
n
g
o
th
e
r
latest
e
m
b
ed
d
i
n
g
tech
n
iq
u
es a
n
d
co
n
te
x
t a
w
ar
e
DL
ar
ch
i
tectu
r
e
s
.
RE
F
E
R
E
NC
E
S
[1
]
L
.
He
,
Z.
Ya
n
g
,
H.
L
in
,
a
n
d
Y.
L
i
,
“
Dru
g
n
a
m
e
re
c
o
g
n
it
io
n
in
b
i
o
m
e
d
ica
l
tex
ts:
a
m
a
c
h
in
e
-
lea
rn
in
g
-
b
a
se
d
m
e
th
o
d
,
”
Dr
u
g
Disc
o
v
.
T
o
d
a
y
,
v
o
l.
1
9
,
n
o
.
5
,
p
p
.
6
1
0
-
6
1
7
,
M
a
y
2
0
1
4
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
d
ru
d
is.
2
0
1
3
.
1
0
.
0
0
6
.
[2
]
M
.
Ch
o
,
J.
Ha
,
C.
P
a
rk
,
a
n
d
S
.
P
a
rk
,
“
Co
m
b
in
a
to
rial
f
e
a
tu
re
e
m
b
e
d
d
i
n
g
b
a
se
d
o
n
CNN
a
n
d
L
S
T
M
f
o
r
b
io
m
e
d
ica
l
n
a
m
e
d
e
n
ti
ty
r
e
c
o
g
n
it
io
n
,
”
J
.
Bi
o
me
d
.
In
f
o
rm
,
v
o
l.
1
0
3
,
M
a
r.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
jb
i
.
2
0
2
0
.
1
0
3
3
8
1
.
[3
]
M
.
G
rid
a
c
h
,
“
Ch
a
ra
c
ter
-
lev
e
l
n
e
u
ra
l
n
e
tw
o
rk
f
o
r
b
io
m
e
d
ica
l
n
a
m
e
d
e
n
ti
ty
re
c
o
g
n
it
io
n
,
”
J
.
Bi
o
me
d
.
In
fo
rm
,
v
o
l.
7
0
,
p
p
.
8
5
-
9
1
,
Ju
n
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
j
b
i.
2
0
1
7
.
0
5
.
0
0
2
.
[4
]
B.
Bh
a
su
ra
n
,
G
.
M
u
ru
g
e
sa
n
,
S
.
A
b
d
u
lk
a
d
h
a
r
,
a
n
d
J.
Na
tara
jan
,
“
S
tac
k
e
d
e
n
se
m
b
le
c
o
m
b
in
e
d
w
it
h
f
u
z
z
y
m
a
tch
in
g
f
o
r
b
io
m
e
d
ica
l
n
a
m
e
d
e
n
ti
ty
r
e
c
o
g
n
it
io
n
o
f
d
ise
a
se
s,”
J
.
Bi
o
me
d
.
In
f
o
rm
,
v
o
l.
6
4
,
p
p
.
1
-
9
,
De
c
.
2
0
1
6
,
d
o
i:
1
0
.
1
0
1
6
/j
.
jb
i.
2
0
1
6
.
0
9
.
0
0
9
.
[5
]
X
.
L
i,
H.
Zh
a
n
g
,
a
n
d
X
.
H.
Zh
o
u
,
“
Ch
in
e
se
c
li
n
ica
l
n
a
m
e
d
e
n
ti
t
y
re
c
o
g
n
it
io
n
w
it
h
v
a
rian
t
n
e
u
ra
l
stru
c
tu
re
s
b
a
se
d
o
n
BERT
m
e
th
o
d
s,”
J
.
Bi
o
me
d
.
I
n
fo
rm
,
v
o
l
.
1
0
7
,
Ju
l
.
2
0
2
0
,
d
o
i:
1
0
.
1
0
1
6
/
j.
jb
i.
2
0
2
0
.
1
0
3
4
2
2
.
[6
]
A
.
Ba
sh
a
r,
“
S
u
rv
e
y
o
n
Ev
o
lv
in
g
De
e
p
L
e
a
rn
in
g
Ne
u
ra
l
Ne
t
w
o
rk
A
rc
h
it
e
c
tu
re
s,”
J
.
Arti
f.
In
tell.
Ca
p
su
l.
Ne
tw
o
rk
s
,
v
o
l.
1
,
n
o
.
2
,
p
p
.
7
3
-
8
2
,
De
c
.
2
0
1
9
,
d
o
i:
1
0
.
3
6
5
4
8
/
jaic
n
.
2
0
1
9
.
2
.
0
0
3
.
[7
]
J.
L
i,
A
.
S
u
n
,
J.
Ha
n
,
a
n
d
C.
L
i,
“
A
S
u
rv
e
y
o
n
De
e
p
Lea
rn
in
g
f
o
r
Na
m
e
d
En
ti
t
y
Re
c
o
g
n
it
io
n
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
K
n
o
wle
d
g
e
a
n
d
Da
t
a
E
n
g
i
n
e
e
rin
g
,
v
o
l.
8
,
M
a
r.
2
0
2
0
,
d
o
i:
1
0
.
1
1
0
9
/T
KD
E.
2
0
2
0
.
2
9
8
1
3
1
4
.
[8
]
Y.
L
e
Cu
n
,
Y.
Be
n
g
io
,
a
n
d
G
.
Hin
to
n
,
“
De
e
p
lea
rn
in
g
,
”
Na
t
u
r
e
,
v
o
l.
5
2
1
,
n
o
.
7
5
5
3
,
p
p
.
4
3
6
-
4
4
4
,
M
a
y
.
2
0
1
5
,
d
o
i:
1
0
.
1
0
3
8
/n
a
t
u
re
1
4
5
3
9
.
[9
]
D.
Zen
g
,
C.
S
u
n
,
L
.
L
in
,
a
n
d
B.
L
iu
,
“
L
S
T
M
-
CRF
f
o
r
Dru
g
-
Na
m
e
d
En
t
it
y
Re
c
o
g
n
it
i
o
n
,
”
En
tr
o
p
y
,
v
o
l.
1
9
,
n
o
.
6
,
p
p
.
2
8
3
,
J
u
n
.
2
0
1
7
,
d
o
i
:
1
0
.
3
3
9
0
/
e
1
9
0
6
0
2
8
3
.
[1
0
]
A
.
G
.
Ag
irre,
M
.
M
a
rim
o
n
,
A
.
In
tx
a
u
rr
o
n
d
o
,
O.
Ra
b
a
l,
M
.
V
il
leg
a
s
,
a
n
d
M
.
Kra
ll
i
n
g
e
r,
“
P
h
a
rm
a
Co
NER:
P
h
a
rm
a
c
o
lo
g
ica
l
S
u
b
sta
n
c
e
s,
Co
m
p
o
u
n
d
s a
n
d
p
r
o
tein
s Na
m
e
d
En
ti
ty
R
e
c
o
g
n
it
io
n
trac
k
,
”
in
Pro
c
e
e
d
in
g
s o
f
T
h
e
5
t
h
W
o
rk
sh
o
p
o
n
Bi
o
NL
P
Op
e
n
S
h
a
r
e
d
T
a
sk
s,
No
v
.
2
0
1
9
,
p
p
.
1
–
1
0
,
d
o
i:
1
0
.
1
8
6
5
3
/v
1
/
d
1
9
-
5
7
0
1
.
[1
1
]
M
.
E.
P
e
ters
e
t
a
l.
,
“
De
e
p
c
o
n
tex
tu
a
li
z
e
d
w
o
rd
re
p
re
se
n
tatio
n
s,”
a
r
Xi
v
Pre
p
r.
a
rXiv
1
8
0
2
.
0
5
3
6
5
,
F
e
b
.
2
0
1
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
A
n
o
ve
l d
ee
p
lea
r
n
in
g
a
r
ch
itectu
r
e
fo
r
d
r
u
g
n
a
med
en
tity reco
g
n
itio
n
(
T.
Ma
th
u
)
1891
[1
2
]
Z.
Cu
i
,
R.
Ke
,
Z
.
P
u
,
a
n
d
Y.
W
a
n
g
,
“
De
e
p
b
id
irec
ti
o
n
a
l
a
n
d
u
n
i
d
irec
ti
o
n
a
l
L
S
T
M
re
c
u
rre
n
t
n
e
u
r
a
l
n
e
tw
o
rk
f
o
r
n
e
tw
o
rk
-
w
id
e
traff
i
c
sp
e
e
d
p
re
d
ic
ti
o
n
,
”
a
rXiv
p
re
p
rin
t
a
rX
iv:1
8
0
1
.
0
2
1
4
3
,
Ja
n
.
2
0
1
8
.
[1
3
]
J.
Ki
m
,
M
.
El
-
Kh
a
m
y
,
a
n
d
J.
Lee
,
“
Re
sid
u
a
l
L
S
T
M
:
De
si
g
n
o
f
a
d
e
e
p
re
c
u
rre
n
t
a
rc
h
it
e
c
tu
re
f
o
r
d
istan
t
sp
e
e
c
h
re
c
o
g
n
it
io
n
,
”
a
rXiv P
re
p
r.
a
rXiv
1
7
0
1
.
0
3
3
6
0
,
Ja
n
.
2
0
1
7
.
[1
4
]
Y.
W
a
n
g
,
X
.
Zh
a
n
g
,
M
.
L
u
,
H.
W
a
n
g
,
a
n
d
Y.
Ch
o
e
,
“
A
tt
e
n
ti
o
n
a
u
g
m
e
n
tatio
n
w
it
h
m
u
lt
i
-
re
sid
u
a
l
in
b
id
irec
ti
o
n
a
l
L
S
T
M
,
”
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l.
3
8
5
,
p
p
.
3
4
0
-
3
4
7
,
A
p
r.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
1
6
/j
.
n
e
u
c
o
m
.
2
0
1
9
.
1
0
.
0
6
8
.
[1
5
]
Ş
.
Öz
tü
rk
a
n
d
U.
Öz
k
a
y
a
,
“
Re
si
d
u
a
l
L
S
T
M
la
y
e
re
d
CNN
f
o
r
c
las
sif
i
c
a
ti
o
n
o
f
g
a
stro
in
tes
ti
n
a
l
tr
a
c
t
d
ise
a
se
s,”
J
.
Bi
o
me
d
.
I
n
f
o
rm
.
,
v
o
l.
1
1
3
,
p
p
.
1
0
3
6
3
8
,
Ja
n
.
2
0
2
1
,
d
o
i:
1
0
.
1
0
1
6
/
j.
j
b
i.
2
0
2
0
.
1
0
3
6
3
8
.
[1
6
]
J.
P
e
n
n
in
g
t
o
n
,
R.
S
o
c
h
e
r
,
a
n
d
C.
D.
M
a
n
n
i
n
g
,
“
G
lo
V
e
:
G
lo
b
a
l
v
e
c
to
rs
f
o
r
w
o
rd
re
p
re
se
n
tatio
n
,
”
in
Pro
c
e
e
d
in
g
s
o
f
th
e
2
0
1
4
c
o
n
fer
e
n
c
e
o
n
e
mp
iric
a
l
me
th
o
d
s
i
n
n
a
t
u
ra
l
l
a
n
g
u
a
g
e
p
ro
c
e
ss
in
g
(
EM
NL
P)
,
p
p
.
1
5
3
2
–
1
5
4
3
,
Oc
t.
2
0
1
4
,
d
o
i:
1
0
.
3
1
1
5
/v
1
/
d
1
4
-
1
1
6
2
.
[1
7
]
Z.
L
iu
e
t
a
l.
,
“
En
ti
ty
re
c
o
g
n
it
io
n
f
ro
m
c
li
n
ica
l
tex
ts
v
ia
re
c
u
rre
n
t
n
e
u
ra
l
n
e
tw
o
rk
,
”
BM
C
M
e
d
.
In
fo
r
m.
De
c
is.
M
a
k
.
,
v
o
l.
1
7
,
n
o
.
2
,
p
p
.
5
3
-
6
1
,
Ju
l.
2
0
1
7
,
d
o
i:
1
0
.
1
1
8
6
/S
1
2
9
11
-
0
1
7
-
0
4
6
8
-
7.
[1
8
]
C.
W
a
n
g
,
H.
Ya
n
g
,
a
n
d
C.
M
e
in
e
l,
“
I
m
a
g
e
Ca
p
ti
o
n
in
g
w
it
h
De
e
p
Bid
irec
ti
o
n
a
l
L
S
T
M
s
a
n
d
M
u
lt
i
-
T
a
s
k
L
e
a
rn
in
g
,
”
ACM
T
ra
n
s.
M
u
l
ti
me
d
.
C
o
mp
u
t.
Co
mm
u
n
.
Ap
p
l.
,
v
o
l.
1
4
,
n
o
.
2
s,
p
p
.
1
-
2
0
,
A
p
r.
2
0
1
8
,
d
o
i
:
1
0
.
1
1
4
5
/
3
1
1
5
4
3
2
.
[1
9
]
A
.
S
n
ieg
u
la,
A
.
P
.
M
a
ra
rid
a
,
a
n
d
L
.
Ch
o
m
a
t
e
k
,
“
S
tu
d
y
o
f
n
a
m
e
d
e
n
ti
ty
re
c
o
g
n
it
io
n
m
e
th
o
d
s in
b
i
o
m
e
d
ica
l
f
ield
,
”
in
Pro
c
e
d
ia
C
o
mp
u
ter
S
c
ien
c
e
,
v
o
l.
1
6
0
,
p
p
.
2
6
0
-
2
6
5
,
Ja
n
.
2
0
1
9
,
d
o
i:
1
0
.
1
0
1
6
/
j.
p
r
o
c
s.2
0
1
9
.
0
9
.
4
6
6
.
[2
0
]
M
.
H
.
Zaz
o
,
I.
S
.
Be
d
m
a
r,
P
.
M
a
rti
n
e
z
,
a
n
d
T
.
De
c
lerc
k
,
“
T
h
e
DD
I
c
o
rp
u
s:
A
n
a
n
n
o
tate
d
c
o
rp
u
s
w
it
h
p
h
a
rm
a
c
o
lo
g
ica
l
su
b
sta
n
c
e
s
a
n
d
d
ru
g
–
d
ru
g
in
tera
c
ti
o
n
s
,
”
J
.
Bi
o
me
d
.
In
fo
rm
.
,
v
o
l.
4
6
,
n
o
.
5
,
p
p
.
9
1
4
–
9
2
0
,
Oc
t.
2
0
1
3
,
d
o
i:
1
0
.
1
0
1
6
/
j.
j
b
i.
2
0
1
3
.
0
7
.
0
1
1
.
[2
1
]
A
.
Ku
tu
z
o
v
a
n
d
E.
Ku
z
m
e
n
k
o
,
“
T
o
le
m
m
a
ti
z
e
o
r
n
o
t
to
le
m
m
a
ti
z
e
:
h
o
w
w
o
rd
n
o
rm
a
li
sa
ti
o
n
a
ffe
c
ts
E
L
M
o
p
e
rf
o
r
m
a
n
c
e
in
w
o
rd
se
n
se
d
isa
m
b
ig
u
a
ti
o
n
,
”
in
Pro
c
e
e
d
in
g
s
o
f
t
h
e
Fi
rs
t
NL
PL
W
o
rk
sh
o
p
o
n
De
e
p
L
e
a
rn
in
g
fo
r
Na
tu
ra
l
L
a
n
g
u
a
g
e
Pr
o
c
e
ss
in
g
,
S
e
p
.
2
0
1
9
,
p
p
.
2
2
-
28
.
[2
2
]
A
.
G
ra
v
e
s,
N.
Ja
it
l
y
,
a
n
d
A
.
R.
M
o
h
a
m
e
d
,
“
H
y
b
rid
sp
e
e
c
h
re
c
o
g
n
it
io
n
w
it
h
De
e
p
Bid
irec
ti
o
n
a
l
LS
T
M
,
”
in
2
0
1
3
IEE
E
W
o
rk
sh
o
p
o
n
Au
t
o
ma
ti
c
S
p
e
e
c
h
Rec
o
g
n
it
i
o
n
a
n
d
Un
d
e
rs
ta
n
d
in
g
,
A
S
RU
2
0
1
3
-
Pro
c
e
e
d
i
n
g
s
,
De
c
.
2
0
1
3
,
p
p
.
2
7
3
–
2
7
8
,
d
o
i:
1
0
.
1
1
0
9
/A
S
RU.2
0
1
3
.
6
7
0
7
7
4
2
.
[2
3
]
A
.
G
ra
v
e
s
a
n
d
J
.
S
c
h
m
i
d
h
u
b
e
r
,
“
F
r
a
m
e
w
is
e
P
h
o
n
e
m
e
C
l
a
s
s
if
ic
a
t
io
n
w
i
t
h
B
i
d
i
r
e
c
t
i
o
n
a
l
L
S
T
M
a
n
d
O
t
h
e
r
N
e
u
r
a
l
N
e
tw
o
rk
A
r
c
h
i
te
c
t
u
r
e
s
,
”
Ne
u
r
a
l
n
e
t
w
o
r
k
s
,
v
o
l
.
1
8
,
n
o
.
5
-
6
,
p
p
.
6
0
2
-
6
1
0
,
J
u
l
.
2
0
0
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
n
e
t
.
2
0
0
5
.
0
6
.
0
4
2
.
[2
4
]
T
.
L
i
u
,
S
.
Y
u
,
B
.
Xu
,
a
n
d
H
.
Yi
n
,
“
R
e
c
u
r
r
e
n
t
n
e
tw
o
rk
s
w
i
t
h
a
t
te
n
t
i
o
n
a
n
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
tw
o
rk
s
f
o
r
se
n
t
e
n
c
e
r
e
p
r
e
se
n
t
a
t
i
o
n
a
n
d
c
l
a
s
s
if
ic
a
t
i
o
n
,
”
A
p
p
l
.
I
n
t
e
l
l
.
,
v
o
l
.
4
8
,
n
o
.
1
0
,
p
p
.
3
7
9
7
–
3
8
0
6
,
O
c
t
.
2
0
1
8
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
0
4
8
9
-
0
1
8
-
1176
-
4.
[2
5
]
V
.
V
a
n
A
sc
h
,
“
M
a
c
ro
-
a
n
d
m
icro
-
a
v
e
ra
g
e
d
e
v
a
lu
a
ti
o
n
m
e
a
su
re
s,”
Belg
iu
m: CL
iP
S
,
p
p
.
1
–
2
7
,
S
e
p
.
2
0
1
3
.
[2
6
]
S
.
L
iu
,
B.
T
a
n
g
,
Q.
Ch
e
n
,
X
.
Wan
g
,
Y.
Yu
,
a
n
d
Y.
Wan
g
,
“
E
ff
e
c
ts
o
f
S
e
m
a
n
ti
c
F
e
a
tu
re
s
o
n
M
a
c
h
in
e
L
e
a
rn
in
g
-
Ba
se
d
Dru
g
Na
m
e
Re
c
o
g
n
it
io
n
S
y
ste
m
s:
W
o
rd
Em
b
e
d
d
in
g
s
v
s.
M
a
n
u
a
ll
y
Co
n
stru
c
ted
Dic
ti
o
n
a
ries
,
”
In
fo
rm
a
ti
o
n
,
v
o
l.
6
,
n
o
.
4
,
p
p
.
8
4
8
–
8
6
5
,
De
c
.
2
0
1
5
,
d
o
i:
1
0
.
3
3
9
0
/i
n
f
o
6
0
4
0
8
4
8
.
[2
7
]
T
.
Ro
c
k
tä
sc
h
e
l,
T
.
Hu
b
e
r,
M
.
W
e
id
li
c
h
,
a
n
d
U.
L
e
s
e
r,
“
W
BI
-
NE
R:
T
h
e
i
m
p
a
c
t
o
f
d
o
m
a
in
-
sp
e
c
if
i
c
f
e
a
tu
re
s
o
n
th
e
p
e
rf
o
r
m
a
n
c
e
o
f
id
e
n
ti
fy
in
g
a
n
d
c
las
si
fy
in
g
m
e
n
ti
o
n
s
o
f
d
ru
g
s,”
in
Pro
c
e
e
d
in
g
s
o
f
th
e
S
e
v
e
n
t
h
In
ter
n
a
t
io
n
a
l
W
o
rk
sh
o
p
o
n
S
e
ma
n
ti
c
Eva
l
u
a
t
io
n
(
S
e
mEv
a
l
2
0
1
3
)
,
v
o
l
.
2
,
Ju
n
.
2
0
1
3
,
p
p
.
3
5
6
-
3
6
3
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
T.
M
a
th
u
is an
A
s
sista
n
t
P
r
o
f
e
ss
o
r
in
th
e
De
p
a
rtm
e
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
a
t
Ka
ru
n
y
a
In
stit
u
te
o
f
Tec
h
n
o
lo
g
y
a
n
d
S
c
ien
c
e
s,
Co
im
b
a
to
re
,
In
d
i
a
.
S
h
e
is
a
lso
p
u
rsu
in
g
h
e
r
P
h
.
D
d
e
g
re
e
a
t
th
e
sa
m
e
u
n
iv
e
rs
it
y
.
He
r
re
se
a
rc
h
in
tere
sts
in
c
l
u
d
e
d
a
ta
m
in
in
g
,
te
x
t
m
in
in
g
,
n
a
tu
ra
l
lan
g
u
a
g
e
p
r
o
c
e
ss
in
g
,
m
a
c
h
in
e
lea
rn
i
n
g
a
n
d
d
e
e
p
lea
rn
i
n
g
.
K
u
m
u
d
h
a
R
a
i
m
o
n
d
is a P
ro
f
e
ss
o
r
in
th
e
De
p
a
rtm
e
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
E
n
g
in
e
e
rin
g
a
t
Ka
ru
n
y
a
In
stit
u
te
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
S
c
ien
c
e
s,
Co
im
b
a
to
re
,
In
d
ia.
He
r
a
re
a
s
o
f
e
x
p
e
rti
se
in
c
lu
d
e
m
a
c
h
in
e
lea
rn
in
g
,
in
telli
g
e
n
t
s
y
ste
m
s,
b
io
m
e
tri
c
s,
b
io
i
n
f
o
rm
a
ti
c
s,
b
io
m
e
d
ica
l
a
p
p
li
c
a
ti
o
n
s,
sa
telli
te
im
a
g
e
p
ro
c
e
ss
in
g
,
w
a
ter
m
a
rk
in
g
,
c
o
m
p
re
ss
io
n
a
n
d
w
irele
ss
se
n
so
r
n
e
tw
o
rk
s.
Evaluation Warning : The document was created with Spire.PDF for Python.