TELK
O
M
N
I
K
A
Te
l
e
c
ommu
n
i
c
at
i
on
,
C
omp
u
ti
n
g
,
El
e
c
t
r
on
i
c
s
an
d
C
on
tr
o
l
V
ol
.
18
,
N
o
.
2
,
A
pri
l
2020
,
p
p.
752
~
760
IS
S
N
:
1693
-
6930
,
a
c
c
r
e
di
t
e
d
F
i
rs
t
G
r
a
de
by
K
e
m
e
n
ri
s
t
e
kdi
k
t
i
,
D
e
c
r
e
e
N
o:
21
/
E
/
K
P
T
/
2
018
D
O
I:
10
.
129
28/
T
E
L
K
O
M
N
IK
A
.
v1
8i
2
.
1391
2
752
Jou
r
n
al
h
o
m
e
pa
ge
:
ht
t
p:
/
/
j
our
nal
.
uad
.
a
c
.
i
d/
i
nd
e
x
.
php
/
T
E
L
K
O
MNIKA
S
e
n
t
i
me
n
t
a
n
a
l
y
si
s by
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
S
r
e
e
vi
d
ya
P
.
1
,
O
.
V
.
R
a
m
an
a
M
u
r
t
h
y
2
,
S
.
V
e
n
i
3
1
,
3
D
e
p
a
r
t
m
e
n
t
of
E
l
e
c
t
r
oni
c
s
a
n
d
C
om
m
u
ni
c
a
t
i
o
n
E
ngi
ne
e
r
i
ng
,
A
m
r
i
t
a
S
c
hoo
l
of
E
n
gi
n
e
e
r
i
ng
,
A
m
ri
t
a
V
i
s
hw
a
V
i
d
ya
p
e
e
t
ha
m
,
I
nd
i
a
2
D
e
pa
r
t
m
e
n
t
o
f
E
l
e
c
t
r
i
c
a
l
a
nd
E
l
e
c
t
r
on
i
c
s
E
n
gi
n
e
e
r
i
ng,
A
m
r
i
t
a
S
c
h
ool
of
E
n
gi
ne
e
r
i
ng,
A
m
ri
t
a
V
i
s
hw
a
V
i
d
ya
p
e
e
t
ha
m
,
I
nd
i
a
A
r
ti
c
l
e
I
n
fo
A
BS
TR
A
C
T
Ar
t
i
c
l
e
h
i
s
t
or
y
:
Re
c
e
i
v
e
d
A
ug
15
,
201
9
Re
vi
s
e
d
J
a
n
9
,
2020
A
c
c
e
pt
e
d
F
e
b
21
,
202
0
W
e
pr
o
pos
e
a
m
ode
l
f
o
r
c
a
r
r
y
i
ng
ou
t
d
e
e
p
l
e
a
r
ni
ng
b
a
s
e
d
m
ul
t
i
m
od
a
l
s
e
n
t
i
m
e
nt
a
n
a
l
ys
i
s
.
T
he
M
O
U
D
da
t
a
s
e
t
i
s
t
a
ke
n
f
or
e
xpe
r
i
m
e
n
t
a
t
i
on
pur
p
os
e
s
.
W
e
d
e
ve
l
op
e
d
t
w
o
pa
r
a
l
l
e
l
t
e
x
t
ba
s
e
d
a
nd
a
u
di
o
ba
s
e
d
m
ode
l
s
a
n
d
f
ur
t
he
r
,
f
us
e
d
t
he
s
e
he
t
e
r
oge
n
e
ou
s
f
e
a
t
u
r
e
m
a
p
s
t
a
ke
n
f
r
o
m
i
nt
e
r
m
e
d
i
a
t
e
l
a
y
e
r
s
t
o
c
o
m
p
l
e
t
e
t
he
a
r
c
hi
t
e
c
t
u
r
e
.
P
e
r
f
or
m
a
nc
e
m
e
a
s
ur
e
s
–
A
c
c
ur
a
c
y
,
pr
e
c
i
s
i
o
n,
r
e
c
a
l
l
a
nd
F
1
-
s
c
o
r
e
–
a
r
e
ob
s
e
r
ve
d
t
o
out
pe
r
f
o
r
m
t
he
e
xi
s
t
i
ng
m
od
e
l
s
.
Ke
y
w
or
d
s
:
B
i
m
o
da
l
CN
N
l
a
ye
rs
MOUD
M
ul
t
i
m
od
a
l
W
ord
e
m
b
e
ddi
n
gs
T
hi
s
i
s
an
op
e
n
ac
c
e
s
s
ar
t
i
c
l
e
u
nde
r
t
he
C
C
B
Y
-
SA
l
i
c
e
ns
e
.
Cor
r
e
s
pon
di
n
g
Au
t
h
or
:
S
re
e
v
i
dya
P
.
,
D
e
pa
r
t
m
e
nt
of
E
l
e
c
t
r
oni
c
s
a
nd
Com
m
un
i
c
a
t
i
on
E
ng
i
n
e
e
r
i
ng
,
A
m
ri
t
a
S
c
hoo
l
of
E
ng
i
ne
e
ri
ng,
A
m
ri
t
a
V
i
s
hw
a
V
i
d
ya
p
e
e
t
ha
m
,
C
oi
m
ba
t
ore
,
64111
2
,
Ind
i
a
.
E
m
a
i
l
:
s
re
e
vi
dy
a
p
m
e
n
on@
m
a
i
l
.
c
om
1.
I
N
TR
O
D
U
C
TI
O
N
G
re
a
t
e
f
fort
s
a
r
e
n
e
e
d
e
d
t
o
de
v
e
l
op
m
a
c
hi
n
e
s
t
h
a
t
c
a
n
m
i
m
i
c
t
h
e
na
t
ura
l
a
bi
l
i
t
y
of
hu
m
a
n
be
i
ngs
t
o
unde
rs
t
a
nd
e
m
ot
i
ons
,
a
na
l
yz
e
s
i
t
ua
t
i
o
ns
a
nd
u
nde
rs
t
a
nd
t
he
s
e
nt
i
m
e
nt
s
a
s
s
oc
i
a
t
e
d
w
i
t
h
t
h
e
c
ont
e
xt
.
T
he
s
e
nt
i
m
e
nt
a
na
l
ys
i
s
i
s
a
n
e
ff
e
c
t
i
v
e
m
e
c
h
a
ni
s
m
t
o
e
xp
l
o
re
t
he
s
oc
i
o
-
e
c
o
nom
i
c
or
d
e
m
ogr
a
phi
c
i
nf
l
u
e
nc
e
i
n
hum
a
n
r
e
c
i
pro
c
a
t
i
on
.
W
i
t
h
t
he
a
v
a
i
l
a
b
i
l
i
t
y
of
a
pl
e
t
hora
of
opi
ni
ona
t
e
d
v
i
de
os
i
n
s
oc
i
a
l
m
e
di
a
,
m
ul
t
i
m
oda
l
a
ppro
a
c
h
e
s
i
n
t
h
e
s
e
nt
i
m
e
nt
a
n
a
l
ys
i
s
i
s
g
a
i
n
i
ng
a
t
t
e
n
t
i
on
.
O
pi
ni
o
na
t
e
d
vi
d
e
os
a
re
h
i
gh
l
y
uns
t
ruc
t
ur
e
d;
he
n
c
e
ve
rba
l
a
nd
non
-
v
e
rb
a
l
c
u
e
s
a
re
c
o
m
pl
e
m
e
nt
a
ry
i
n
t
h
e
s
e
nt
i
m
e
n
t
a
na
l
ys
i
s
a
t
t
h
i
s
j
u
nc
t
ure
.
T
h
a
t
m
e
a
ns
a
na
l
yz
i
ng
t
h
e
c
om
m
un
i
c
a
t
i
on
i
n
a
ud
i
o
,
v
i
s
ua
l
a
l
ong
w
i
t
h
t
e
x
t
m
oda
l
i
t
i
e
s
h
a
s
t
o
b
e
i
nc
or
pora
t
e
d
for
a
c
h
i
e
v
i
ng
e
ffe
c
t
i
ve
s
ol
u
t
i
ons
.
M
os
t
of
t
he
e
x
i
s
t
i
ng
fr
a
m
e
w
orks
for
c
l
a
s
s
i
fyi
ng
t
h
e
s
e
n
t
i
m
e
nt
s
a
re
ba
s
e
d
on
t
r
a
ns
c
r
i
pt
i
ons
ba
s
e
d
a
n
a
l
ys
i
s
[1]
a
nd
t
he
us
e
of
l
e
x
i
c
ons
,
bu
t
no
t
m
uc
h
of
t
h
e
l
i
t
e
r
a
t
ur
e
m
i
n
e
s
t
hrough
t
h
e
vo
c
a
l
a
nd
vi
s
ua
l
c
ue
s
e
m
b
e
dde
d
i
n
t
h
e
vi
d
e
os
.
T
he
vo
i
c
e
d
c
o
m
m
u
ni
c
a
t
i
on
c
a
n
gi
v
e
m
or
e
i
n
for
m
a
t
i
on
re
g
a
rd
i
ng
t
he
hum
a
n
e
m
pa
t
he
t
i
c
c
ond
i
t
i
ons
[2]
.
T
hi
s
w
ork
a
i
m
s
i
n
f
us
i
ng
i
nfor
m
a
t
i
on
fro
m
d
i
ff
e
re
n
t
m
od
a
l
i
t
i
e
s
for
t
he
s
e
n
t
i
m
e
n
t
a
na
l
ys
i
s
.
T
he
pri
m
a
r
y
be
ne
f
i
t
o
f
a
n
a
l
yz
i
ng
v
i
de
os
a
l
ong
w
i
t
h
t
e
xt
s
i
s
t
ha
t
t
h
e
r
i
c
h
s
e
t
of
b
e
ha
v
i
or
a
l
c
ue
s
p
re
s
e
nt
i
n
a
udi
o
a
nd
v
i
de
o
re
c
ord
i
ngs
c
a
n
yi
e
l
d
e
n
ha
nc
e
d
m
od
e
l
s
.
T
he
voc
a
l
m
od
ul
a
t
i
ons
,
fa
c
i
a
l
e
xpre
s
s
i
ons
a
nd
ge
s
t
ure
s
i
n
t
h
e
v
i
s
ua
l
d
a
t
a
,
a
l
ong
w
i
t
h
t
e
x
t
ua
l
da
t
a
,
h
e
l
p
t
o
a
na
l
yz
e
t
h
e
a
ffe
c
t
i
ve
do
m
a
i
n
of
t
he
op
i
ni
on
ho
l
de
r
i
n
a
be
t
t
e
r
w
a
y
.
T
hus
,
a
c
o
m
bi
ne
d
t
e
x
t
,
v
oc
a
l
a
nd
vi
s
u
a
l
d
a
t
a
he
l
p
t
o
c
r
e
a
t
e
a
m
or
e
rob
us
t
a
nd
e
m
o
t
i
on
s
p
e
c
i
fi
c
s
e
n
t
i
m
e
n
t
a
na
l
ys
i
s
m
ode
l
[3]
.
T
h
e
r
e
i
s
a
n
a
rr
a
y
of
t
e
c
hn
i
qu
e
s
a
v
a
i
l
a
b
l
e
fo
r
c
a
rryi
ng
ou
t
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
Se
nt
i
m
e
n
t
a
nal
y
s
i
s
by
de
e
p
l
e
ar
ni
ng
appr
o
ac
h
e
s
(Sr
e
e
v
i
d
y
a
P
)
753
t
he
s
e
nt
i
m
e
n
t
a
na
l
ys
i
s
,
t
hrough
i
n
c
orp
ora
t
i
ng
m
a
c
hi
n
e
l
e
a
rni
ng
a
nd
d
e
e
p
l
e
a
rn
i
ng
pa
ra
d
i
gm
s
.
T
he
re
a
r
e
m
ul
t
i
-
f
a
c
e
t
e
d
c
ha
l
l
e
nge
s
a
s
s
o
c
i
a
t
e
d
w
i
t
h
e
x
t
ra
c
t
i
ng
i
n
for
m
a
t
i
on
fr
om
d
i
ffe
re
n
t
m
od
a
l
i
t
i
e
s
a
nd
t
o
f
us
e
t
h
e
m
t
oge
t
he
r
fo
r
t
he
a
na
l
ys
i
s
.
W
e
prop
os
e
a
bi
m
od
a
l
a
p
proa
c
h
f
or
pr
e
di
c
t
i
ng
t
h
e
s
e
nt
i
m
e
nt
s
us
i
n
g
d
e
e
p
l
e
a
rni
n
g
ba
s
e
d
t
e
c
h
ni
qu
e
s
.
T
he
prop
os
e
d
de
e
p
s
e
n
t
i
m
e
n
t
a
na
l
ys
i
s
fra
m
e
w
ork
i
n
c
l
ud
e
s
:
a.
A
Co
nvol
u
t
i
ona
l
N
e
ur
a
l
N
e
t
w
or
k
(CN
N
)
b
a
s
e
d
m
od
e
l
w
i
t
h
m
a
x
-
poo
l
i
ng
,
a
nd
d
e
ns
e
l
a
y
e
rs
t
o
proc
e
s
s
fe
a
t
ure
s
e
x
t
ra
c
t
e
d
fro
m
s
e
nt
e
nc
e
l
e
ve
l
u
t
t
e
ra
n
c
e
s
.
b.
A
m
ode
l
for
pro
c
e
s
s
i
n
g
t
r
a
ns
c
r
i
pt
i
ons
w
hi
c
h
i
s
t
r
a
i
n
e
d
w
i
t
h
CN
N
l
a
ye
rs
.
T
h
e
s
e
n
t
e
n
c
e
l
e
v
e
l
t
e
x
t
i
s
m
a
pp
e
d
i
n
t
o
a
ve
c
t
or
s
p
a
c
e
us
i
ng
a
w
ord
re
pr
e
s
e
n
t
a
t
i
o
n
l
e
a
rn
e
d
b
y
w
or
d
e
m
b
e
dd
i
ng
.
c.
A
fus
i
o
n
m
od
e
l
c
on
t
a
i
ni
ng
t
he
f
e
a
t
ur
e
s
e
xt
ra
c
t
e
d
fro
m
s
p
e
c
i
fi
c
l
a
ye
rs
o
f
bot
h
a
ud
i
o
a
nd
t
ra
ns
c
ri
pt
i
on
m
ode
l
s
.
Conve
n
t
i
o
na
l
l
y
,
t
h
e
prob
l
e
m
of
s
e
n
t
i
m
e
n
t
a
n
a
l
ys
i
s
i
s
ba
s
e
d
on
t
e
x
t
ua
l
i
nf
orm
a
t
i
on
.
T
h
e
a
n
a
l
ys
i
s
i
s
c
a
rr
i
e
d
out
a
t
w
ord
l
e
ve
l
,
s
e
nt
e
nc
e
l
e
v
e
l
or
do
c
u
m
e
n
t
l
e
ve
l
.
P
re
-
proc
e
s
s
i
ng
s
t
e
ps
i
nc
l
ud
e
c
l
e
a
n
i
ng
of
t
e
xt
s
,
re
m
o
va
l
o
f
w
hi
t
e
s
pa
c
e
s
,
e
xp
a
ndi
ng
t
h
e
a
bbr
e
v
i
a
t
i
ons
,
s
t
e
m
m
i
ng
,
re
m
ov
a
l
of
s
t
op
w
or
ds
,
n
e
ga
t
i
o
n
ha
nd
l
i
n
g
fol
l
ow
e
d
b
y
fe
a
t
ure
s
e
l
e
c
t
i
on
a
n
d
fi
na
l
l
y
c
l
a
s
s
i
fi
c
a
t
i
o
n
t
e
c
h
ni
que
s
[4]
.
T
he
c
l
a
s
s
i
fi
c
a
t
i
o
n
t
e
c
h
ni
qu
e
s
c
a
n
be
di
vi
d
e
d
i
n
t
o
m
a
c
hi
ne
l
e
a
rn
i
ng
(M
L
)
ba
s
e
d
a
ppro
a
c
he
s
a
nd
l
e
xi
c
on
ba
s
e
d
a
ppro
a
c
he
s
.
T
he
M
L
ba
s
e
d
s
upe
rvi
s
e
d
l
e
a
rn
i
ng
a
ppro
a
c
he
s
i
n
c
l
ud
e
prob
a
bi
l
i
s
t
i
c
m
o
de
l
s
s
uc
h
a
s
N
a
i
v
e
Ba
y
e
s
c
l
a
s
s
i
fi
e
rs
[5
]
or
B
a
ye
s
i
a
n
c
l
a
s
s
i
fi
e
rs
[6]
.
Be
c
a
us
e
of
t
h
e
s
pa
rs
e
n
a
t
ur
e
of
t
h
e
t
e
xt
d
a
t
a
,
t
he
S
u
ppor
t
V
e
c
t
or
M
a
c
hi
n
e
s
(S
V
M
s
)
a
r
e
e
ffe
c
t
i
ve
l
y
us
e
d
for
c
l
a
s
s
i
fy
i
ng
t
ra
ns
c
ri
pt
i
on
s
e
n
t
i
m
e
nt
s
,
bo
t
h
for
m
u
l
t
i
-
c
l
a
s
s
a
n
d
bi
na
ry
c
l
a
s
s
pro
bl
e
m
s
.
L
i
a
nd
L
i
[7]
us
e
d
S
V
M
for
c
l
a
s
s
i
fy
i
ng
s
e
nt
i
m
e
n
t
s
i
n
m
i
c
ro
bl
ogs
.
N
e
ur
a
l
ne
t
w
ork
a
nd
S
V
M
w
e
re
a
ppl
i
e
d
for
s
e
nt
i
m
e
nt
a
n
a
l
ys
i
s
a
nd
c
o
m
pa
r
e
d
by
M
ora
e
s
e
t
a
l
.
[8]
.
T
he
a
ut
o
m
a
t
e
d
l
e
x
i
c
on
ba
s
e
d
a
pp
roa
c
he
s
a
re
s
pl
i
t
i
nt
o
d
i
c
t
i
ona
r
y
ba
s
e
d
a
ppro
a
c
he
s
a
n
d
c
orp
us
ba
s
e
d
a
ppro
a
c
he
s
[9]
.
T
he
di
c
t
i
ona
r
y
ba
s
e
d
a
ppro
a
c
h
e
s
fo
c
us
on
f
i
ndi
ng
t
h
e
op
i
ni
on
s
e
e
d
w
ord
,
w
h
e
re
a
s
c
orpus
b
a
s
e
d
a
p
proa
c
h
be
g
i
ns
w
i
t
h
a
s
e
e
d
l
i
s
t
of
opi
ni
on
w
o
rds
.
T
he
c
or
pus
b
a
s
e
d
a
ppr
oa
c
h
i
s
l
i
m
i
t
e
d
du
e
t
o
t
he
di
ff
i
c
ul
t
y
i
n
pr
e
p
a
ri
ng
hug
e
c
orp
us
a
n
d
no
rm
a
l
l
y
e
m
p
l
oys
e
i
t
he
r
s
t
a
t
i
s
t
i
c
a
l
ba
s
e
d
t
e
c
hn
i
que
s
[10]
or
s
e
m
a
nt
i
c
b
a
s
e
d
t
e
c
h
ni
qu
e
s
[11]
.
W
i
t
h
t
he
i
nc
re
a
s
e
d
pr
e
s
e
n
c
e
of
m
ul
t
i
m
e
d
i
a
t
ool
s
,
e
s
p
e
c
i
a
l
l
y
on
s
o
c
i
a
l
m
e
d
i
a
pl
a
t
for
m
s
,
s
e
n
t
i
m
e
nt
a
na
l
ys
i
s
c
o
ul
d
n
ot
b
e
re
s
t
ri
c
t
e
d
t
o
t
ra
n
s
c
ri
pt
i
on
b
a
s
e
d
a
n
a
l
ys
i
s
.
T
hi
s
ha
s
p
a
v
e
d
w
a
ys
t
o
m
ul
t
i
m
od
a
l
a
ppro
a
c
h
e
s
i
n
s
e
nt
i
m
e
n
t
a
n
a
l
ys
i
s
.
W
hi
l
e
t
he
u
ni
m
o
da
l
t
e
xt
b
a
s
e
d
a
n
a
l
ys
i
s
w
a
s
foc
us
e
d
a
t
t
e
x
t
pre
-
pro
c
e
s
s
i
ng
a
nd
s
e
l
e
c
t
i
ng
s
u
i
t
a
bl
e
m
e
t
hods
for
a
n
a
l
ys
i
s
,
t
he
r
e
w
e
re
gr
e
a
t
e
r
c
ha
l
l
e
nge
s
i
n
m
ul
t
i
m
oda
l
a
ppro
a
c
h
e
s
.
In
c
onv
e
n
t
i
on
a
l
a
n
a
l
ys
i
s
,
rul
e
ba
s
e
d
m
e
t
hods
us
i
ng
l
e
xi
c
ons
a
nd
da
t
a
dr
i
ve
n
m
e
t
hods
us
i
ng
l
a
rge
,
a
nnot
a
t
e
d
d
a
t
a
ba
s
e
s
[12
,
13]
a
re
popul
a
r
.
Bu
t
i
n
m
ul
t
i
m
od
a
l
a
na
l
ys
i
s
,
t
h
e
he
t
e
ro
ge
n
e
ous
d
i
m
e
ns
i
ons
f
rom
i
m
a
ge
,
t
e
xt
a
n
d
a
udi
o
s
i
gna
l
s
a
re
t
o
b
e
c
o
m
b
i
ne
d
t
oge
t
he
r.
T
he
r
e
a
r
e
t
hr
e
e
s
t
r
a
t
e
gi
e
s
po
pul
a
r
for
m
u
l
t
i
m
o
da
l
fus
i
on,
vi
z
,
e
a
r
l
y
fus
i
on
l
a
t
e
fus
i
o
na
nd
i
nt
e
r
m
i
t
t
e
nt
fus
i
on
.
T
he
w
ork
i
n
[14]
a
pp
l
y
e
a
rl
y
fus
i
on
of
l
ow
l
e
v
e
l
a
nd
m
i
d
l
e
v
e
l
f
e
a
t
ur
e
s
e
x
t
ra
c
t
e
d
fro
m
hu
m
a
n
fa
c
e
s
t
o
ha
ve
group
l
e
v
e
l
e
m
ot
i
on
d
e
t
e
c
t
i
on
.
A
m
a
j
or
s
hort
c
o
m
i
ng
of
e
a
r
l
y
fus
i
o
n
t
e
c
hni
que
i
s
t
h
e
a
bs
e
nc
e
of
d
e
t
a
i
l
e
d
m
ode
l
i
ng
for
v
i
e
w
-
s
p
e
c
i
fi
c
dyna
m
i
c
s
,
w
hi
c
h
w
i
l
l
a
ff
e
c
t
t
h
e
m
od
e
l
i
ng
of
i
n
t
e
r
-
vi
e
w
dyn
a
m
i
c
s
w
h
i
c
h
c
a
us
e
s
ove
rfi
t
t
i
ng
of
i
npu
t
d
a
t
a
a
nd
m
od
e
l
s
b
a
s
e
d
on
l
a
t
e
f
us
i
on
a
r
e
no
r
m
a
l
l
y
go
od
i
n
m
od
e
l
i
ng
v
i
e
w
-
s
p
e
c
i
fi
c
dyna
m
i
c
s
.
L
a
t
e
fus
i
ons
h
a
v
e
s
hor
t
c
o
m
i
ngs
i
n
m
ode
l
i
n
g
t
he
c
ros
s
-
vi
e
w
dyn
a
m
i
c
s
s
i
nc
e
t
h
e
s
e
c
ros
s
-
m
oda
l
i
t
y
dy
na
m
i
c
s
a
re
c
ons
i
d
e
re
d
t
o
be
m
ore
di
ffi
c
ul
t
[15]
.
T
h
e
t
r
a
di
t
i
on
a
l
h
a
nd
c
ra
f
t
e
d
f
e
a
t
ure
e
xt
r
a
c
t
i
on
m
e
t
ho
ds
p
a
ve
d
w
a
ys
t
o
d
e
e
p
l
e
a
rni
n
g
t
e
c
hni
q
ue
s
,
a
d
di
t
i
on
a
l
l
y,
t
h
e
R
e
c
u
rre
n
t
N
e
ura
l
N
e
t
w
orks
(
R
N
N
)
a
nd
L
ong
S
hort
t
i
m
e
M
e
m
or
y
(
L
S
T
M
)
c
ou
l
d
t
a
k
e
u
p
t
he
s
pa
t
i
a
l
a
nd
t
e
m
por
a
l
i
nfo
rm
a
t
i
on
di
r
e
c
t
l
y
fro
m
t
h
e
r
a
w
d
a
t
a
[16]
.
2.
R
ES
EA
R
C
H
M
ET
H
O
D
A
bi
m
oda
l
a
pproa
c
h
w
i
t
h
ut
t
e
r
a
nc
e
s
t
a
k
e
n
i
n
a
u
di
o
a
nd
t
e
x
t
for
m
a
t
s
i
s
pr
opos
e
d
he
r
e
fo
rs
e
nt
i
m
e
n
t
a
na
l
ys
i
s
.
T
he
M
O
U
D
d
a
t
a
s
e
t
c
o
nt
a
i
n
i
ng
opi
n
i
on
a
t
e
d
u
t
t
e
r
a
nc
e
s
i
n
s
e
nt
e
nc
e
l
e
v
e
l
[13
]
i
s
t
a
ke
n
fo
r
e
xpe
r
i
m
e
n
t
s
.
T
h
e
a
r
c
h
i
t
e
c
t
ure
d
e
ve
l
op
e
d
i
s
s
how
n
i
n
F
i
g
ure
1
.
U
t
t
e
r
a
nc
e
s
a
ud
i
o
a
n
dt
e
xt
a
r
e
t
h
e
i
np
ut
s
of
t
he
fr
a
m
e
w
ork
a
nd
t
he
ou
t
pu
t
i
s
bi
n
a
ry
c
l
a
s
s
i
fi
c
a
t
i
on
-
pos
i
t
i
ve
or
n
e
ga
t
i
v
e
po
l
a
r
i
t
y
.
T
h
e
a
rc
h
i
t
e
c
t
ura
l
pi
p
e
l
i
ne
i
nc
l
ude
s
t
w
o
pa
r
a
l
l
e
l
i
n
de
p
e
nd
e
nt
de
e
p
l
e
a
rni
ng
fr
a
m
e
w
ork
s
ha
vi
ng
un
i
m
od
a
l
pro
c
e
s
s
i
n
g
of
a
ud
i
o
a
nd
t
e
xt
ut
t
e
ra
n
c
e
s
.
T
he
d
e
e
p
ne
ura
l
f
e
a
t
ure
s
e
xt
r
a
c
t
e
dfro
m
t
h
e
s
e
i
n
di
vi
d
ua
l
m
o
da
l
i
t
i
e
s
a
re
fus
e
d
t
og
e
t
h
e
r
a
nd
g
i
ve
n
a
s
i
n
put
t
o
t
he
fi
n
a
l
CN
N
l
a
ye
rs
t
o
a
pp
l
y
t
he
b
i
m
o
da
l
fus
i
on.
2.
1
.
U
n
i
mod
al
ap
p
r
oa
c
h
e
s
T
he
propos
e
d
s
ys
t
e
m
i
nt
e
nds
t
o
de
v
e
l
op
i
n
di
v
i
du
a
l
m
od
e
l
s
for
t
r
a
ns
c
r
i
pt
i
ons
a
n
d
a
u
di
o
s
i
gna
l
s
a
t
t
he
fi
rs
t
s
t
a
g
e
.
L
a
t
e
r,
a
bi
m
od
a
l
a
r
c
hi
t
e
c
t
ur
e
i
s
d
e
ve
l
ope
d
by
i
n
t
e
gr
a
t
i
ng
t
h
e
i
nde
pe
nd
e
nt
m
ode
l
s
.
E
a
c
h
s
t
a
g
e
i
s
d
e
s
c
r
i
be
d
a
s
fo
l
l
ow
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
1
693
-
69
30
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
,
V
ol
.
18
,
N
o.
2
,
A
pri
l
2
020:
752
-
7
60
754
2.
2
.
A
u
d
i
o
fe
atu
r
e
s
A
na
l
y
z
i
ng
t
h
e
s
p
e
e
c
h
a
s
s
ound
w
i
l
l
h
e
l
p
t
h
e
s
ys
t
e
m
t
o
foc
us
on
c
l
a
s
s
i
fyi
ng
t
he
pol
a
ri
t
yof
t
he
s
e
nt
e
nc
e
e
i
t
he
r
a
s
pos
i
t
i
v
e
or
ne
g
a
t
i
ve
by
e
l
i
m
i
na
t
i
ng
t
he
l
a
ngu
a
ge
ba
r
ri
e
r.
A
s
f
or
t
h
e
a
udi
o
ut
t
e
r
a
nc
e
s
a
re
c
on
c
e
rne
d
,
t
he
a
u
di
o
fe
a
t
ur
e
s
a
re
e
x
t
ra
c
t
e
d
from
t
h
e
i
np
ut
a
udi
o
s
i
gn
a
l
byt
he
a
pp
l
i
c
a
t
i
on
of
a
t
hi
r
d
p
a
rt
y
a
c
ous
t
i
c
f
e
a
t
ur
e
e
x
t
ra
c
t
i
on
t
oo
l
c
a
l
l
e
d
O
p
e
nE
a
r
[17
,
18]
.
T
he
f
e
a
t
ure
s
e
xt
r
a
c
t
e
d
a
re
us
i
ng
S
M
IL
E
fe
a
t
ur
e
e
xt
r
a
c
t
or
a
nd
L
ow
L
e
ve
l
D
e
s
c
ri
p
t
ors
(
L
L
D
s
)
i
nc
l
udi
n
g
13
M
e
l
-
/
B
a
rk
-
F
re
que
n
c
y
-
C
e
ps
t
ra
l
Coe
ffi
c
i
e
nt
s
(M
F
CC)
w
h
i
c
h
t
yp
i
c
a
l
l
y
ra
ng
e
s
be
t
w
e
e
n
300H
z
t
o5K
H
z
,
p
ros
ody,
e
ne
r
gy,
vo
i
c
e
prob
a
bi
l
i
t
i
e
s
a
nd
s
pe
c
t
ra
l
c
oe
ff
i
c
i
e
n
t
s
re
s
u
l
t
i
ng
i
n
a
fe
a
t
ur
e
v
e
c
t
ors
e
t
27
f
or
e
a
c
h
ut
t
e
ra
nc
e
s
.
T
h
e
fe
a
t
u
re
s
a
re
e
xt
r
a
c
t
e
d
w
i
t
h
a
fra
m
e
s
a
m
pl
e
ra
t
e
of
25
m
s
a
ndz
-
s
t
a
nd
a
rd
i
z
a
t
i
on
i
s
p
e
rfo
rm
e
d
for
s
p
e
a
k
e
r
norm
a
l
i
z
a
t
i
on
.
T
hi
s
f
e
a
t
ur
e
s
e
t
i
s
a
pp
l
i
e
d
t
o
a
d
e
e
p
l
e
a
rn
i
ng
fra
m
e
w
o
rk
s
t
a
r
t
i
n
g
w
i
t
h
a
c
onvo
l
ut
i
on
l
a
y
e
r
i
nc
l
udi
n
g256
f
i
l
t
e
rs
of
s
i
z
e
t
hre
e
.
T
he
c
onvo
l
ut
i
on
l
a
y
e
rs
a
r
e
i
n
t
e
rl
e
a
v
e
d
w
i
t
h
a
m
a
x
-
p
ool
i
ng
l
a
ye
r
s
o
t
ha
t
t
h
e
fi
l
t
e
r
out
put
s
i
z
e
i
s
r
e
du
c
e
d
by
fa
c
t
o
r
of
t
w
o.
T
h
e
ne
t
w
or
k
go
e
s
d
e
e
p
e
r
i
n
t
h
i
s
f
a
s
hi
on
by
i
m
p
l
e
m
e
nt
i
ng
c
onvo
l
ut
i
ona
l
l
a
y
e
rs
of
s
i
z
e
3
w
i
t
h
t
he
num
b
e
r
of
f
i
l
t
e
rs
a
s
128
a
n
d
64
re
s
p
e
c
t
i
v
e
l
y
.
A
ft
e
r
t
he
c
o
nvol
ut
i
ona
l
ope
ra
t
i
ons
,
t
h
re
e
c
ons
e
c
u
t
i
v
e
de
ns
e
l
a
y
e
rs
a
r
e
a
dd
e
d
t
o
f
l
a
t
t
e
n
t
h
e
n
e
t
w
o
rka
n
d
gra
d
ua
l
l
y
re
duc
e
t
h
e
out
pu
t
.
T
he
m
a
x
-
poo
l
i
n
g
l
a
ye
r
w
i
l
l
r
e
du
c
e
t
h
e
di
m
e
ns
i
ons
o
f
t
h
e
s
e
t
of
f
e
a
t
ur
e
by
a
f
a
c
t
or
o
f
2
.
N
e
x
t
,
a
dropou
t
i
s
a
ppl
i
e
d
a
s
a
re
gul
a
ri
z
a
t
i
on
t
e
c
hni
que
i
n
ord
e
rt
o
r
e
du
c
e
t
he
num
b
e
r
of
ne
t
w
ork
c
onn
e
c
t
i
ons
w
hi
c
h
h
e
l
ps
t
o
a
voi
d
t
h
e
o
ve
rf
i
t
t
i
ng
of
t
he
ne
t
w
ork
l
a
y
e
rs
j
us
t
b
e
for
e
f
l
a
t
t
e
ni
ng
t
he
l
a
ye
rs
.
T
he
non
-
l
i
n
e
a
r
R
e
c
t
i
f
i
e
d
L
i
ne
a
r
U
ni
t
(R
e
L
U
)
i
s
a
pp
l
i
e
d
a
s
t
he
a
c
t
i
v
a
t
i
on
fu
nc
t
i
on
f
or
t
h
e
h
i
dde
n
l
a
ye
rs
a
nd
f
i
na
l
de
c
i
s
i
on
on
t
ype
of
t
he
s
e
n
t
i
m
e
n
t
i
s
b
a
s
e
do
n
t
he
ou
t
put
of
t
h
e
s
oft
m
a
x
fun
c
t
i
on
.
T
he
t
r
a
ns
c
r
i
be
d
ut
t
e
r
a
n
c
e
s
i
n
t
he
M
O
U
D
da
t
a
s
e
t
,
w
hi
c
h
i
s
a
nnot
a
t
e
d
i
s
c
o
m
bi
n
e
d
i
nt
oa
s
i
ngl
e
CS
V
fi
l
e
.
In
i
t
i
a
l
l
y
t
he
da
t
a
b
a
s
e
un
de
rg
oe
s
c
e
rt
a
i
n
pre
-
proc
e
s
s
i
ng
s
t
e
ps
s
o
a
s
t
o
a
v
oi
d
t
he
out
l
i
e
rs
.
S
ubs
e
que
n
t
l
y,
t
he
d
a
t
a
i
s
gi
ve
n
t
o
a
t
oke
ni
z
e
r
t
o
c
r
e
a
t
e
t
he
v
oc
a
bul
a
ry
.
T
he
w
or
de
m
be
ddi
ngs
a
re
us
e
d
t
o
ge
t
t
h
e
w
ord
ve
c
t
ors
.
It
i
s
l
i
ke
t
h
e
c
on
c
a
t
e
na
t
i
on
of
w
ords
.
T
hi
s
fe
a
t
ur
e
s
e
t
s
a
re
t
ra
i
ne
d
i
n
a
d
e
e
p
n
e
ur
a
l
fra
m
e
w
ork
II
t
o
c
a
rry
ou
t
t
he
ou
t
pu
t
t
he
s
e
n
t
i
m
e
n
t
s
c
l
a
s
s
i
fi
c
a
t
i
on
.
F
i
gure
1
.
T
he
pro
pos
e
d
b
i
m
od
a
l
a
rc
hi
t
e
c
t
ur
e
f
or
s
e
nt
i
m
e
nt
a
n
a
l
ys
i
s
2.
3
.
Te
xtu
al
f
e
atu
r
e
s
P
ri
m
a
ri
l
y,
t
e
xt
da
t
a
m
us
t
b
e
e
nc
ode
d
a
s
ve
c
t
ors
be
fore
a
ppl
yi
ng
i
t
t
o
t
h
e
d
e
e
p
l
e
a
r
ni
ng
m
o
de
l
.
F
or
h
a
t
(
i
)
s
e
n
t
e
n
c
e
s
a
re
pre
-
p
roc
e
s
s
e
d
a
nd
t
o
ke
n
i
z
e
d
t
o
g
e
t
t
he
i
nt
e
ge
r
r
e
pr
e
s
e
n
t
a
t
i
on
.
T
he
s
t
a
rt
a
nd
s
t
op
w
ords
a
s
w
e
l
l
a
s
t
h
e
w
i
l
d
c
h
a
ra
c
t
e
rs
a
r
e
re
m
ov
e
d
dur
i
ng
pr
e
-
pro
c
e
s
s
i
ng
.
A
t
t
he
s
a
m
e
t
i
m
e
a
l
l
t
h
e
w
ords
a
re
onve
r
t
e
d
t
o
l
ow
e
rc
a
s
e
l
e
t
t
e
rs
.
K
e
r
a
s
T
o
ke
n
i
z
a
t
i
on
A
P
I
i
s
us
e
dfor
t
oke
n
i
z
i
ng
t
he
s
e
nt
e
nc
e
s
.
(
i
i
)
F
i
n
a
l
l
y
,
T
he
w
ord
e
m
b
e
ddi
n
gs
a
re
a
p
pl
i
e
d
t
o
c
onve
rt
t
h
e
p
os
i
t
i
ve
i
n
t
e
g
e
rs
t
o
d
e
ns
e
v
e
c
t
ors
o
f
f
i
xe
d
s
i
z
e
.
T
h
e
de
ns
e
ve
c
t
ors
re
pre
s
e
nt
t
he
proj
e
c
t
i
on
of
t
he
w
ords
i
nt
o
a
c
ont
i
nu
ous
v
e
c
t
or
s
p
a
c
e
w
he
r
e
by
e
a
c
h
w
or
d
w
i
l
l
h
a
ve
a
u
ni
qu
e
ve
c
t
o
r
re
pr
e
s
e
n
t
a
t
i
on
.
A
s
a
re
s
u
l
t
,
t
h
e
w
ords
w
i
l
l
be
i
n
a
c
oord
i
na
t
e
s
ys
t
e
m
,
w
he
r
e
,
re
l
a
t
e
d
w
ords
ba
s
e
d
on
t
he
c
orpus
re
l
a
t
i
o
ns
hi
ps
w
i
l
l
b
e
p
l
a
c
e
d
c
l
os
e
d
t
o
e
a
c
h
o
t
he
r.
T
h
e
v
e
c
t
or
va
l
ue
s
a
r
e
l
e
a
rne
d
i
n
a
w
a
y
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
Se
nt
i
m
e
n
t
a
nal
y
s
i
s
by
de
e
p
l
e
ar
ni
ng
appr
o
ac
h
e
s
(Sr
e
e
v
i
d
y
a
P
)
755
t
ha
t
re
s
e
m
b
l
e
s
t
o
t
he
m
e
t
hod
of
l
e
a
rni
ng
i
n
a
t
yp
i
c
a
l
ne
ura
l
n
e
t
w
ork
[1
9].
T
he
f
e
a
t
ur
e
ve
c
t
ors
ob
t
a
i
ne
da
r
e
pa
dd
e
d
t
o
a
w
i
ndow
of
s
t
a
nd
a
rd
l
e
ngt
h
of
60
.
T
hi
s
s
t
a
nda
r
di
z
e
d
v
e
c
t
ors
a
r
e
g
i
ve
n
a
s
i
n
put
oft
he
de
e
p
l
e
a
rn
i
ng
m
ode
l
.
T
h
e
f
i
rs
t
i
npu
t
w
a
s
g
i
v
e
n
t
o
t
h
e
c
onvo
l
ut
i
on
a
l
l
a
y
e
r
of
s
i
z
e
3
,
c
ons
i
s
t
i
ngof
128
fi
l
t
e
rs
,
fo
l
l
ow
e
d
b
y
a
gl
o
ba
l
m
a
x
po
ol
i
ng
l
a
ye
r.
T
he
c
on
vol
u
t
i
o
na
l
l
a
ye
r
s
ys
t
e
m
a
t
i
c
a
l
l
y
a
ppl
y
l
e
a
r
ne
d
fi
l
t
e
rs
t
o
t
he
i
npu
t
d
a
t
a
s
o
a
s
t
o
c
r
e
a
t
e
fe
a
t
ur
e
m
a
ps
t
h
a
t
s
u
m
m
a
ri
z
e
t
h
e
pr
e
s
e
n
c
e
of
t
he
s
t
rong
f
e
a
t
ure
s
e
t
i
n
t
h
e
i
n
put
da
t
a
.
T
h
e
g
l
ob
a
l
max
-
pool
i
ng
l
a
ye
r
w
i
l
l
dow
n
s
a
m
pl
e
e
a
c
h
f
e
a
t
ur
e
m
a
p
i
nt
o
a
s
i
ngl
e
v
a
l
u
e
w
h
i
c
h
i
s
t
he
m
a
xi
m
u
m
v
a
l
u
e
of
t
he
p
a
t
c
h
e
s
o
f
t
h
e
fe
a
t
ur
e
s
e
t
[20]
.
In
t
h
i
s
w
a
y
t
he
pr
obl
e
m
s
d
ue
t
o
ove
rfi
t
t
i
ng
of
t
he
fu
l
l
y
c
onn
e
c
t
e
d
l
a
y
e
rs
c
a
n
b
e
m
i
ni
m
i
z
e
d
.
S
ubs
e
que
n
t
l
y
,
t
he
r
e
a
r
e
t
w
o
d
e
ns
e
l
a
y
e
r
s
.
A
l
l
l
a
y
e
rs
e
x
c
e
p
t
t
h
e
fi
n
a
l
d
e
ns
e
l
a
y
e
r
i
s
w
i
t
h
Re
L
U
a
c
t
i
va
t
i
o
n
fu
nc
t
i
on
,
w
he
r
e
a
s
t
h
e
f
i
n
a
l
de
c
i
s
i
on
m
a
ki
ng
l
a
y
e
r
h
a
s
s
of
t
m
a
x
a
s
t
h
e
a
c
t
i
v
a
t
i
on
fun
c
t
i
on.
T
he
m
ode
l
ha
s
96
,
796
pa
ra
m
e
t
e
rs
t
o
b
e
l
e
a
rn
e
d
duri
n
g
t
h
e
t
ra
i
ni
n
g.
T
ypi
c
a
l
l
y
,
i
f
t
h
e
re
a
re
n
w
ords
i
n
a
s
e
nt
e
nc
e
[2
1],
i
t
c
a
n
b
e
t
ok
e
ni
z
e
d
a
s
a
n
i
nt
e
ge
r
v
e
c
t
or
T
,
w
h
e
re
(
1
)
60
},
,..
{
2
,
1
=
=
n
T
T
T
T
n
(1)
T
1xd
di
m
e
ns
i
on
,
d
de
no
t
e
s
t
h
e
w
ord
l
e
ngt
h.
By
a
ppl
y
i
ng
t
he
w
o
rd
e
m
be
d
di
ngs
e
a
c
h
t
ok
e
ns
w
i
l
l
b
e
ve
c
t
ori
z
e
d
c
ons
i
s
t
i
ng
of
t
h
e
f
e
a
t
ur
e
r
e
pr
e
s
e
n
t
a
t
i
ons
of
t
h
e
r
e
q
ui
re
d
t
ra
ns
c
ri
p
t
i
ons
.
It
i
s
gi
ve
n
a
s
e
m
be
ddi
n
gs
,
E
T
W
e
,
=
(2)
w
he
re
W
e
i
s
t
he
p
a
ra
m
e
t
e
rs
t
o
b
e
t
un
e
d
a
nd
T
dX
e
W
.
T
he
hi
d
de
n
l
a
ye
r
out
put
i
s
re
pr
e
s
e
n
t
e
d
a
s
)
,
(
t
i
E
f
h
=
(3)
w
he
re
t
i
s
t
he
w
e
i
ght
s
a
nd
bi
a
s
p
a
ra
m
e
t
e
rs
a
nd
t
h
e
fi
n
a
l
a
c
t
i
v
a
t
i
on
l
a
ye
r
i
s
t
h
e
s
oft
m
a
x
l
a
y
e
r
[
22]
.
F
or
a
gi
ve
n
c
l
a
s
s
hi
t
h
e
s
of
t
m
a
x
fun
c
t
i
on
i
s
r
e
pr
e
s
e
n
t
e
d
a
s
:
=
=
c
j
h
h
i
j
i
e
e
h
1
(4)
w
he
re
hj
a
re
t
h
e
va
l
ue
s
i
nf
e
rr
e
d
b
y
t
he
n
e
t
for
e
a
c
h
c
l
a
s
s
i
n
C
.
2.
4
.
Bi
mod
a
l
F
r
a
me
w
o
r
k
In
t
h
e
p
ropos
e
d
m
ode
l
,
i
n
di
v
i
dua
l
,
p
a
ra
l
l
e
l
ne
t
w
o
rks
w
e
r
e
t
ra
i
ne
d
i
n
i
t
i
a
l
l
y
.
L
a
t
e
r
,
t
he
i
nt
e
r
m
i
t
t
e
nt
l
a
y
e
rs
of
bot
h
t
h
e
s
e
ne
t
w
orks
a
re
e
x
t
ra
c
t
e
d
a
s
fe
a
t
ur
e
i
nput
for
t
h
e
bi
m
od
a
l
fra
m
e
w
ork.
In
t
hi
s
w
a
y,
t
he
c
o
m
p
l
e
m
e
n
t
a
r
y
i
nfor
m
a
t
i
o
n
from
bo
t
h
t
h
e
m
o
da
l
i
t
i
e
s
c
oul
d
b
e
t
a
ke
n
e
ffe
c
t
i
ve
l
y.
T
he
3
rd
l
a
ye
r
o
f
t
he
t
e
x
t
ua
l
m
ode
l
a
n
d
6t
h
l
a
ye
r
o
f
t
h
e
a
u
di
o
m
od
e
l
a
r
e
opt
i
m
a
l
l
y
s
e
l
e
c
t
e
d
a
nd
e
x
t
ra
c
t
e
d
a
s
fe
a
t
ur
e
s
for
t
he
fi
n
a
l
fus
i
on
m
od
e
l
.
T
h
e
g
l
ob
a
l
m
a
xpoo
l
i
n
g
l
a
y
e
r
i
n
t
e
xt
m
oda
l
i
t
y
s
i
gni
f
i
c
a
nt
l
yr
e
duc
e
d
t
h
e
s
i
z
e
of
t
h
e
f
e
a
t
ur
e
m
a
p
a
nd
t
h
e
s
a
m
e
w
a
s
don
e
i
n
a
ud
i
o
m
o
da
l
i
t
y
t
ho
rough
dow
ns
a
m
p
l
i
ng
t
h
e
d
e
ns
e
l
a
ye
r.
F
e
a
t
u
re
s
f
rom
t
h
e
s
e
t
w
o
l
a
y
e
rs
a
re
c
o
nc
a
t
e
na
t
e
d
a
n
d
i
t
i
s
a
pp
l
i
e
d
a
s
i
n
put
t
o
t
he
t
h
i
rd
c
o
m
bi
n
e
d
m
od
e
l
.
T
he
fe
a
t
u
re
s
e
t
s
a
r
e
a
ppl
i
e
d
d
i
re
c
t
l
y
w
i
t
hou
t
a
ny
pre
-
pro
c
e
s
s
i
n
g.
T
hi
s
m
od
e
l
i
s
a
l
s
o
a
de
e
p
ne
ur
a
l
n
e
t
w
ork
c
ons
i
s
t
i
ng
of
c
o
nvol
u
t
i
o
na
l
l
a
y
e
rs
a
nd
m
a
x
-
p
ool
i
ng
l
a
y
e
rs
.
T
he
ou
t
put
fr
om
t
h
e
m
od
e
l
w
i
l
l
c
l
a
s
s
i
f
y
t
he
ut
t
e
ra
n
c
e
s
a
s
pos
i
t
i
v
e
or
ne
ga
t
i
v
e
pol
a
ri
t
y.
T
he
de
c
i
s
i
on
ve
c
t
or
form
e
d
by
c
om
b
i
ni
ng
t
h
e
t
e
x
t
a
nd
a
ud
i
o
m
od
a
l
i
t
i
e
s
a
re
i
m
prov
i
ng
t
h
e
p
e
rfor
m
a
nc
e
of
s
e
nt
i
m
e
nt
a
l
a
n
a
l
ys
i
s
c
ons
i
d
e
ra
b
l
y
c
o
m
pa
re
d
t
o
i
ndi
v
i
du
a
l
m
od
a
l
i
t
i
e
s
a
l
on
e
.
T
h
e
fi
na
l
de
c
i
s
i
onon
s
e
nt
i
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
i
s
t
a
ke
n
ba
s
e
d
on
t
h
e
s
oft
m
a
x
a
c
t
i
v
a
t
i
on
f
unc
t
i
on
.
T
he
e
xp
e
ri
m
e
nt
s
a
re
c
o
nduc
t
e
d
on
M
O
U
D
da
t
a
s
e
t
b
o
t
h
o
n
i
nd
i
vi
d
ua
l
a
nd
c
om
b
i
n
e
d
m
oda
l
i
t
i
e
s
.
D
uri
ng
t
h
e
t
r
a
i
n
i
ng
ph
a
s
e
of
t
h
e
prop
os
e
d
m
ode
l
,
t
he
w
e
i
ght
s
a
re
a
dj
us
t
e
d
t
o
m
i
ni
m
i
z
e
t
he
l
os
s
fun
c
t
i
on
.
T
he
hyp
e
r
-
p
a
r
a
m
e
t
e
rs
of
t
he
pro
pos
e
d
n
e
ura
l
ne
t
w
ork
m
o
d
e
l
a
r
e
t
un
e
d
fut
z
i
n
g
w
i
t
h
t
he
w
e
i
ght
s
t
o
fu
rt
h
e
r
opt
i
m
i
z
e
t
he
re
s
u
l
t
s
.
T
h
e
r
ol
e
o
pt
i
m
i
z
e
r
i
n
de
e
p
ne
u
ra
l
n
e
t
w
ork
m
od
e
l
s
a
r
e
t
o
m
i
ni
m
i
z
e
t
he
c
os
t
func
t
i
on
J
(
ϴ
)
,
w
he
re
t
he
p
a
ra
m
e
t
e
r
,
w
i
t
h
r
e
s
pe
c
t
t
o
t
h
e
pe
rf
orm
a
nc
e
m
e
a
s
ure
P
,
und
e
r
c
o
ns
i
d
e
ra
t
i
on
b
y
a
pp
l
yi
ng
gra
di
e
n
t
d
e
s
c
e
nt
a
l
gor
i
t
h
m
s
.
T
h
e
c
os
t
fun
c
t
i
on
c
a
n
be
re
pre
s
e
nt
e
d
a
s
:
(
)
;
,
l
o
g
ˆ
~
m
o
d
,
y
x
P
P
J
el
d
a
t
a
y
x
=
(5)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
1
693
-
69
30
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
,
V
ol
.
18
,
N
o.
2
,
A
pri
l
2
020:
752
-
7
60
756
A
fa
s
t
e
r
c
on
ve
rg
e
n
c
e
of
t
h
e
e
m
ode
l
i
s
a
c
h
i
e
v
e
d
by
s
e
l
e
c
t
i
ng
a
pr
ope
r
l
e
a
rni
ng
r
a
t
e
a
s
i
n
:
(
)
J
−
=
0
(6)
w
he
re
η
0
re
pr
e
s
e
n
t
s
t
he
l
e
a
rni
ng
ra
t
e
of
t
he
gr
a
di
e
nt
d
e
s
c
e
nt
a
l
gor
i
t
h
m
.
T
he
o
pt
i
m
i
z
e
r
a
l
g
ori
t
hm
w
e
us
e
d
for
c
om
p
a
r
i
s
on
purpos
e
a
r
e
s
t
oc
ha
s
t
i
c
gr
a
d
i
e
n
t
de
s
c
e
n
t
a
l
gori
t
hm
(S
G
D
),
R
oot
M
e
a
n
S
qu
a
re
d
prop(RM
S
P
ro
p)
a
nd
A
D
A
pt
i
v
e
M
om
e
n
t
e
s
t
i
m
a
t
i
on(A
D
A
M
).
T
he
S
G
D
d
oe
s
t
he
p
a
ra
m
e
t
e
r
u
pda
t
e
s
for
a
l
l
t
h
e
t
r
a
i
n
i
ng
e
xa
m
pl
e
s
i
n
t
h
e
t
ra
i
ni
ng
s
e
t
w
i
t
h
a
pr
e
f
i
xe
d
l
e
a
rn
i
ng
ra
t
e
[23]
.
In
t
h
e
RM
S
P
rop
a
l
gori
t
h
m
pro
pos
e
d
by
G
e
offre
y
H
i
nt
on,
i
ns
t
e
a
d
of
l
e
t
t
i
n
g
a
l
l
of
t
h
e
gr
a
di
e
nt
s
t
o
a
c
c
um
u
l
a
t
e
t
he
v
a
l
u
e
of
i
t
s
m
o
m
e
n
t
um
,
RM
S
P
rop
a
l
gor
i
t
h
m
o
nl
y
a
c
c
u
m
u
l
a
t
e
s
gr
a
d
i
e
n
t
s
i
n
a
fi
x
e
d
w
i
ndow
.
A
da
m
opt
i
m
i
z
e
r
c
o
m
put
e
s
a
da
p
t
i
v
e
l
e
a
rni
ng
r
a
t
e
s
for
e
a
c
h
pa
r
a
m
e
t
e
r
c
ons
i
de
re
d
i
n
t
he
a
l
gor
i
t
h
m
a
nd
i
t
s
t
or
e
s
t
he
e
xpo
ne
n
t
i
a
l
l
y
d
e
c
a
y
i
ng
a
v
e
ra
ge
of
t
h
e
s
qua
r
e
of
t
h
e
gra
di
e
nt
s
of
t
he
pr
e
vi
ous
v
a
l
u
e
s
[24]
.
2.
5
.
Th
e
M
O
U
D
D
atas
e
t
T
he
M
u
l
t
i
m
o
da
l
U
t
t
e
r
a
nc
e
O
pi
n
i
on
D
a
t
a
ba
s
e
(M
O
U
D
)
i
nt
rodu
c
e
d
by
P
e
r
e
z
e
t
.
a
l
.
[13]
i
s
a
n
opi
ni
ona
t
e
d
da
t
a
s
e
t
i
n
S
p
a
ni
s
h
l
a
ngu
a
ge
.
It
c
ons
i
s
t
s
of
produ
c
t
r
e
v
i
e
w
a
nd
re
c
om
m
e
nda
t
i
o
ns
i
n
ut
t
e
r
a
nc
e
l
e
v
e
l
from
80
s
p
e
a
k
e
rs
c
o
l
l
e
c
t
e
d
t
hrou
gh
Y
ouT
ube
v
i
d
e
os
.
F
r
om
t
h
e
a
v
a
i
l
a
b
l
e
4
98vi
de
os
w
e
s
e
l
e
c
t
e
d
43
8
re
c
or
di
ngs
for
ou
r
w
o
rk,
w
h
i
c
h
s
how
e
d
c
ons
i
s
t
e
nc
y
a
m
on
g
s
pe
e
c
h
a
nd
t
e
x
t
m
od
a
l
i
t
i
e
s
a
n
d
on
a
n
a
ve
r
a
ge
,
e
a
c
h
one
of
t
he
v
i
de
o
h
a
s
6
ut
t
e
r
a
nc
e
s
of
5
s
e
c
o
nds
dur
a
t
i
onw
i
t
h
a
s
t
a
nda
r
d
de
v
i
a
t
i
on
of
1
.
2
s
e
c
onds
.
Th
e
c
o
nt
e
nt
s
of
e
a
c
h
one
of
t
h
e
v
i
de
o
c
l
i
ps
w
e
r
e
t
r
a
ns
c
ri
be
d
t
h
rough
m
a
nu
a
l
l
y
p
roc
e
s
s
i
ng
t
h
e
v
e
rb
a
l
s
t
a
t
e
m
e
nt
s
for
i
t
s
c
onno
t
a
t
i
ons
.
A
nno
t
a
t
i
ons
of
t
he
d
a
t
a
s
e
t
w
a
s
don
e
us
i
n
g
E
l
a
n
t
oo
l
fo
r
s
e
n
t
i
m
e
n
t
a
na
l
ys
i
s
.
Bot
h
a
ud
i
oa
nd
v
i
de
o
m
od
e
s
a
r
e
a
nno
t
a
t
e
d
us
i
n
g
t
he
t
ool
.
T
w
o
a
nnot
a
t
o
rs
i
n
de
p
e
nd
e
nt
l
y
a
nno
t
a
t
e
d
t
he
po
l
a
ri
t
y
of
t
he
ut
t
e
r
a
n
c
e
s
a
s
p
os
i
t
i
ve
,
ne
g
a
t
i
ve
or
n
e
ut
ra
l
.
I
n
ou
r
c
l
a
s
s
i
fi
c
a
t
i
on
pro
bl
e
m
,
p
os
i
t
i
ve
a
nd
ne
ga
t
i
v
e
s
e
nt
i
m
e
nt
s
w
e
r
e
onl
y
c
ons
i
d
e
re
d.
3.
R
ES
U
LTS
A
N
D
A
N
A
LY
S
I
S
T
he
obj
e
c
t
i
v
e
i
s
t
o
c
l
a
s
s
i
f
y
t
he
s
e
nt
i
m
e
n
t
s
i
n
t
h
e
v
i
de
os
b
a
s
e
d
on
t
he
po
l
a
r
i
t
y
a
s
p
os
i
t
i
ve
or
ne
g
a
t
i
ve
t
hrough
a
na
l
y
z
i
ng
t
he
M
O
U
D
da
t
a
s
e
t
.
A
c
o
m
bi
n
e
d
a
u
di
o
a
nd
t
e
x
t
m
o
de
l
w
a
s
de
v
e
l
o
pe
d
by
i
m
p
l
e
m
e
nt
i
ng
de
e
p
ne
ur
a
l
ne
t
w
orks
.
T
h
e
da
t
a
s
e
t
w
a
s
opt
i
m
a
l
l
y
di
vi
d
e
d
i
nt
o
a
t
r
a
i
n
-
t
e
s
t
r
a
t
i
o
of
80
:
20
f
or
de
ve
l
opi
n
g
t
he
m
o
de
l
a
nd
t
e
s
t
i
ng
t
h
e
da
t
a
.
T
h
e
c
a
t
e
g
ori
c
a
l
c
ros
s
e
nt
ro
p
y,
w
h
i
c
h
i
s
a
c
o
m
b
i
na
t
i
on
of
s
oft
m
a
x
a
nd
c
ros
s
e
nt
ro
py
fun
c
t
i
on
w
a
s
t
a
ke
n
a
s
t
h
e
l
os
s
f
unc
t
i
on
for
t
ra
i
ni
ng
t
he
m
od
e
l
.
T
he
uni
m
od
a
l
fe
a
t
ur
e
s
a
r
e
a
pp
l
i
e
d
t
o
t
he
t
w
o
pa
ra
l
l
e
l
s
u
bne
t
s
a
nd
t
h
e
o
ut
pu
t
s
of
fro
m
i
n
t
e
r
m
e
d
i
a
t
e
hi
dde
n
l
a
y
e
rs
a
re
op
t
i
m
a
l
l
y
s
e
l
e
c
t
e
d.
T
he
s
e
s
e
l
e
c
t
e
d
v
a
l
u
e
s
a
r
e
fus
e
d
t
o
g
e
t
a
nd
t
he
s
a
m
e
w
i
l
l
b
e
a
c
t
i
ng
a
s
t
he
i
npu
t
t
o
t
h
e
f
i
na
l
s
ubn
e
t
.
S
e
ve
r
a
l
e
xp
e
ri
m
e
nt
s
w
e
re
c
ond
uc
t
e
d
be
fo
re
f
i
xi
ng
t
he
prop
os
e
d
a
rc
h
i
t
e
c
t
ure
.
W
e
c
o
m
p
i
l
e
d
t
h
e
m
od
e
l
w
i
t
h
di
ffe
r
e
n
t
hyp
e
r
-
pa
ra
m
e
t
e
rs
a
l
s
o.
T
h
e
re
w
a
s
s
o
m
e
s
i
gni
f
i
c
a
nt
c
ha
n
ge
s
ba
s
e
d
on
t
he
o
pt
i
m
i
z
e
r
s
e
l
e
c
t
i
on
.
T
he
m
i
ni
b
a
t
c
h
e
s
c
a
n
o
ffe
r
t
h
e
e
ffe
c
t
of
re
gu
l
a
r
i
z
a
t
i
on
.
T
h
e
m
i
ni
ba
t
c
h
s
e
l
e
c
t
e
d
w
a
s
32
fo
r
t
he
propos
e
d
m
ode
l
.
T
h
e
re
w
e
r
e
s
i
gn
i
fi
c
a
n
t
c
ha
ng
e
s
i
n
t
h
e
pe
r
form
a
n
c
e
of
t
h
e
m
o
de
l
ba
s
e
d
on
t
he
opt
i
m
i
z
e
r
s
e
l
e
c
t
i
on.
T
he
out
p
ut
of
t
h
e
pro
pos
e
d
s
ys
t
e
m
i
s
one
hot
e
n
c
ode
d.
T
he
re
s
u
l
t
s
of
t
h
e
e
xpe
r
i
m
e
n
t
s
a
r
e
t
a
bu
l
a
t
e
d
i
n
T
a
bl
e
1
.
T
he
p
e
rfor
m
a
n
c
e
of
t
he
pr
opos
e
d
m
od
e
l
w
a
s
e
v
a
l
ua
t
e
d
us
i
ng
di
ff
e
re
nt
p
e
rfor
m
a
n
c
e
m
a
t
ri
c
e
s
vi
z
,
a
c
c
ura
c
y
,
pr
e
c
i
s
i
on
,
re
c
a
l
l
a
n
d
F
1
-
s
c
ore
.
T
a
b
l
e
1
.
T
he
p
e
rfor
m
a
n
c
e
c
om
p
i
l
a
t
i
on
of
t
e
xt
,
a
udi
o
a
n
d
t
e
xt
+
a
ud
i
o
m
od
a
l
i
t
i
e
s
M
o
d
e
o
p
t
i
m
i
z
e
r
a
c
c
u
ra
c
y
p
re
c
i
s
i
o
n
re
c
a
l
l
F
-
1
s
c
o
re
a
u
d
i
o
A
d
a
m
0
.
7
6
0
.
8
0
.
7
1
0
.
7
5
S
G
D
0
.
7
2
0
.
7
1
0
.
7
5
0
.
7
Rm
s
p
ro
p
0
.
7
0
.
6
8
0
.
7
5
0
.
7
1
t
e
x
t
A
d
a
m
0
.
7
1
0
.
7
1
0
.
7
1
0
.
7
1
S
G
D
0
.
7
3
0
.
7
4
0
.
6
0
.
6
7
Rm
s
p
ro
p
0
.
7
2
0
.
6
8
0
.
6
8
0
.
6
8
a
u
d
i
o
+
t
e
x
t
A
d
a
m
0
.
8
4
0
.
8
6
0
.
8
2
0
.
8
4
S
G
D
0
.
7
5
0
.
8
6
0
.
6
0
.
7
1
Rm
s
p
ro
p
0
.
7
2
0
.
7
5
0
.
6
8
0
.
7
2
T
he
gr
a
ph
i
c
a
l
r
e
pre
s
e
nt
a
t
i
on
of
a
c
c
ura
c
y
on
t
r
a
i
n
i
ng
e
a
c
h
e
po
c
h
i
s
s
how
n
i
n
F
i
g
ur
e
s
2
-
10.
T
he
e
ff
e
c
t
s
of
di
ff
e
r
e
nt
opt
i
m
i
z
e
rs
a
re
hi
gh
l
i
g
ht
e
d
h
e
re
.
T
he
A
D
A
M
opt
i
m
i
z
e
r
i
s
s
how
i
ng
t
h
e
op
t
i
m
a
l
r
e
s
ul
t
s
.
T
he
S
G
D
i
s
gi
vi
ng
f
l
uc
t
ua
t
i
ons
dur
i
ng
c
on
ve
rg
e
n
c
e
.
T
he
pa
r
a
m
e
t
e
rs
s
e
l
e
c
t
e
d
for
S
G
D
a
re
a
s
l
e
a
rn
i
ng
ra
t
e
=
0
.
001
a
nd
m
om
e
nt
u
m
=
0
.
9
.
I
n
t
he
c
a
s
e
of
RM
S
P
rop
a
l
gori
t
h
m
,
t
h
e
v
a
l
ue
s
a
r
e
a
l
s
o
t
h
e
s
a
m
e
.
F
or
A
da
m
opt
i
m
i
z
e
r
,
i
n
a
d
di
t
i
on
t
o
t
he
a
bo
ve
va
l
u
e
s
,
t
h
e
d
e
c
a
y
ra
t
e
s
a
r
e
a
l
s
o
f
i
xe
d
a
s
0
.
999
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
Se
nt
i
m
e
n
t
a
nal
y
s
i
s
by
de
e
p
l
e
ar
ni
ng
appr
o
ac
h
e
s
(Sr
e
e
v
i
d
y
a
P
)
757
F
i
gure
2
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
t
e
x
t
d
a
t
a
w
i
t
hA
D
A
M
opt
i
m
i
z
e
r
F
i
gure
3
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
a
udi
o
d
a
t
a
w
i
t
hA
D
A
M
opt
i
m
i
z
e
r
F
i
gure
4
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
t
e
x
t
+
a
ud
i
o
d
a
t
a
w
i
t
hA
D
A
M
opt
i
m
i
z
e
r
F
i
gure
5
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
t
e
x
t
d
a
t
a
w
i
t
h
SGD
opt
i
m
i
z
e
r
F
i
gure
6
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
a
udi
o
d
a
t
a
w
i
t
h
SGD
opt
i
m
i
z
e
r
F
i
gure
7
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
t
e
x
t
+
a
ud
i
o
d
a
t
a
w
i
t
h
SGD
opt
i
m
i
z
e
r
F
i
gure
8
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
t
e
x
t
d
a
t
a
w
i
t
h
RM
S
P
rop
opt
i
m
i
z
e
r
F
i
gure
9
.
T
ra
i
n
a
nd
t
e
s
t
a
c
c
ura
c
y
i
n
a
udi
o
d
a
t
a
w
i
t
h
RM
S
P
rop
op
t
i
m
i
z
e
r
F
i
gure
10
.
T
r
a
i
n
a
n
d
t
e
s
t
a
c
c
ur
a
c
y
i
n
t
e
x
t
+
a
u
di
o
d
a
t
a
w
i
t
h
RM
S
P
rop
opt
i
m
i
z
e
r
F
urt
he
r
,
w
e
c
om
p
a
re
d
t
he
p
e
rfor
m
a
nc
e
of
our
a
l
gor
i
t
h
m
w
i
t
h
s
om
e
of
t
h
e
e
xi
s
t
i
ng
a
l
go
ri
t
hm
s
a
nd
t
he
s
u
pe
r
i
ori
t
y
o
f
t
h
e
c
om
bi
n
e
d
a
udi
o
a
nd
t
e
xt
prop
o
s
e
d
a
r
c
hi
t
e
c
t
ure
i
s
qu
i
e
t
e
vi
d
e
nt
from
T
a
bl
e
2
.
T
he
prop
os
e
d
m
ode
l
w
a
s
c
om
pa
r
e
d
w
i
t
h
fo
ur
o
f
t
he
e
xi
s
t
i
ng
s
t
a
t
e
of
t
h
e
a
rt
m
e
t
hods
.
P
or
i
a
e
t
.
a
l
.
,
[25]
propos
e
d
a
s
pe
a
ke
r
e
x
c
l
us
i
ve
t
e
c
h
ni
qu
e
f
or
a
na
l
yz
i
ng
t
h
e
s
e
n
t
i
m
e
n
t
s
e
m
b
e
dde
d
i
n
t
h
e
u
t
t
e
ra
n
c
e
s
.
W
a
n
g
e
t
.
a
l
.
,
[26]
propos
e
d
t
o
m
i
t
i
g
a
t
e
t
h
e
prob
l
e
m
of
ge
ne
r
a
l
i
z
a
bi
l
t
y
t
o
a
l
a
rge
r
m
a
r
gi
n
.
P
ori
a
e
t
.
a
l
.
,
[2
7]
propos
e
d
t
he
Co
nvol
ut
i
ona
l
Re
c
urr
e
nt
M
ul
t
i
K
e
r
na
l
L
e
a
rn
i
ng
(
CR
M
K
L
)
m
od
e
l
us
i
ng
CN
N
n
e
t
w
o
rks
e
x
c
l
us
i
ve
l
y
for
t
r
a
i
n
i
ng
t
h
e
m
od
e
l
,
a
nd
t
he
c
o
m
b
i
ne
d
ne
t
w
ork
t
a
ke
s
t
he
b
e
s
t
f
e
a
t
ur
e
s
on
l
y
b
y
P
ri
nc
i
pa
l
Co
m
pon
e
n
t
A
nl
ys
i
s
(P
CA
)
a
nd
t
h
e
y
us
e
d
S
V
M
for
t
h
e
de
c
i
s
i
on
m
a
ki
ng
.
T
he
re
s
ul
t
s
of
t
h
e
e
x
pe
r
i
m
e
nt
s
w
e
re
t
a
bul
a
t
e
d
i
n
T
a
b
l
e
2.
It
s
how
s
t
he
t
e
s
t
re
s
u
l
t
s
w
i
t
h
a
nd
w
i
t
h
out
fe
a
t
ure
s
e
l
e
c
t
i
on
.
O
ur
re
s
ul
t
s
a
n
d
t
h
e
r
e
s
ul
t
s
o
bt
a
i
n
e
d
by
P
o
ri
a
a
nd
h
i
s
t
e
a
m
a
r
e
c
om
p
a
t
i
bl
e
.
Ca
m
br
i
a
e
t
.
a
l
.
,
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
1
693
-
69
30
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
,
V
ol
.
18
,
N
o.
2
,
A
pri
l
2
020:
752
-
7
60
758
[28]
pr
e
s
e
n
t
e
d
a
d
e
e
p
l
e
a
rn
i
ng
a
r
c
hi
t
e
c
t
ure
foc
us
i
ng
on
s
p
e
a
ke
r
i
nd
e
pe
n
de
n
t
s
ys
t
e
m
s
.
O
u
r
m
e
t
hod
pe
r
f
orm
e
d
m
uc
h
be
t
t
e
r
t
ha
n
t
hi
s
pr
opos
e
d
w
ork
.
T
s
a
i
,
e
t
.
a
l
.
,
[29]
prop
os
e
d
a
m
ul
t
i
m
oda
l
f
a
c
t
ori
z
e
d
m
oda
l
(M
F
M
)
w
i
t
h
m
ul
t
i
m
oda
l
di
s
c
ri
m
i
n
a
t
i
ve
a
n
d
m
od
a
l
i
t
ys
p
e
c
i
fi
c
g
e
ne
ra
t
i
v
e
f
a
c
t
ors
.
T
a
b
l
e
2
.
P
rop
os
e
d
a
r
c
hi
t
e
c
t
u
re
s
ve
rs
us
s
t
a
t
e
of
t
h
e
a
rt
m
e
t
ho
ds
M
o
d
e
P
o
ri
a
2
0
1
8
W
a
n
g
2
0
1
7
P
o
ri
a
2
0
1
6
Ca
m
b
ri
a
2
0
1
7
O
u
r
M
e
t
h
o
d
T
e
x
t
4
8
.
4
5
2
.
2
7
4
.
5
5
3
.
7
7
1
.
1
A
u
d
i
o
5
3
.
7
5
4
.
4
7
9
.
8
5
3
.
7
76
T
e
x
t
+
A
u
d
i
o
5
7
.
1
5
7
.
4
8
3
.
8
5
7
.
1
84
S
e
nt
i
m
e
nt
p
re
d
i
c
t
i
on
re
s
u
l
t
s
on
M
O
U
D
a
re
de
p
i
c
t
e
d
i
n
T
a
bl
e
2
.
T
he
be
s
t
re
s
u
l
t
s
a
re
hi
gh
l
i
g
ht
e
d
i
nbol
d
a
nd
S
O
T
A
s
how
s
t
h
e
c
ha
ng
e
s
i
n
pe
rf
orm
a
nc
e
o
ve
r
pr
e
vi
o
us
s
t
a
t
e
o
f
t
h
e
a
r
t
(S
O
T
A
)
re
s
ul
t
s
.
T
he
i
m
prov
e
m
e
n
t
s
a
re
hi
ghl
i
ght
e
d
i
n
bo
l
d
i
n
T
a
bl
e
3
.
T
a
b
l
e
3
.
P
e
rfor
m
a
n
c
e
c
om
p
a
ri
s
on
w
i
t
h
s
t
a
t
e
of
t
h
e
a
rt
r
e
s
ul
t
s
[29]
M
e
t
h
o
d
A
c
c
u
ra
c
y
F1
-
s
c
o
re
M
a
j
o
ri
t
y
6
0
.
4
4
5
.
5
RF
6
4
.
2
6
3
.
3
S
V
M
-
MD
5
9
.
4
4
5
.
5
T
H
M
M
6
1
.
3
57
EF
-
H
C
RF
5
4
.
7
5
4
.
7
EF
-
L
D
H
CRF
5
2
.
8
4
9
.
3
MV
-
H
CR
F
6
0
.
4
4
5
.
5
MV
-
L
D
H
CRF
5
3
.
8
4
6
.
9
CM
V
-
H
C
RF
6
0
.
4
4
5
.
5
CM
V
-
L
D
H
CR
F
5
3
.
8
4
7
.
8
DF
67
6
7
.
1
EF
-
L
S
T
M
67
6
4
.
3
EF
-
S
L
S
T
M
5
6
.
6
5
1
.
4
EF
-
B
L
S
T
M
5
8
.
5
5
8
.
9
EF
-
S
BL
S
T
M
6
3
.
2
6
3
.
3
MV
-
L
S
T
M
5
7
.
6
4
8
.
2
BC
-
L
S
T
M
7
2
.
6
7
2
.
9
T
F
N
6
3
.
2
6
1
.
7
M
A
RN
8
1
.
1
8
1
.
2
MFN
8
1
.
1
8
0
.
4
M
F
M
8
2
.
1
8
1
.
7
P
ro
p
o
s
e
d
M
e
t
h
o
d
8
4
.
1
8
4
.
1
S
O
T
A
2
2
.
4
4.
C
O
N
C
LU
S
I
O
N
A
n
a
n
a
l
ys
i
s
of
t
he
e
x
i
s
t
i
n
g
m
e
t
hodol
ogi
e
s
for
s
e
nt
i
m
e
nt
a
na
l
ys
i
s
a
nd
t
he
c
o
m
pa
ri
s
on
w
i
t
h
t
he
p
ropos
e
d
bi
m
od
a
l
s
e
nt
i
m
e
nt
a
n
a
l
ys
i
s
s
ys
t
e
m
i
s
c
a
rr
i
e
d
out
h
e
re
.
T
he
propos
e
d
fr
a
m
e
w
ork
e
s
t
a
bl
i
s
he
s
t
he
s
upe
r
i
or
i
t
y
of
bi
m
od
a
l
a
p
proa
c
h
e
s
ov
e
r
uni
m
od
a
l
a
ppr
oa
c
h
e
s
.
W
e
a
r
e
i
n
c
orpor
a
t
i
ng
t
h
e
t
he
pow
e
rfu
l
CN
N
ba
s
e
d
d
e
e
p
l
e
a
rn
i
ng
t
e
c
hn
i
qu
e
s
for
t
h
e
t
e
s
t
c
a
s
e
.
T
he
i
n
t
e
r
m
e
di
a
t
e
l
e
ve
l
f
e
a
t
ur
e
fus
i
on
m
e
t
hod
i
s
a
da
p
t
e
d
h
e
re
.
T
h
e
s
e
qu
e
nt
i
a
l
a
nd
c
orre
l
a
t
e
d
i
nfor
m
a
t
i
o
n
i
s
c
o
l
l
e
c
t
e
d
t
hroug
h
w
or
d
e
m
be
dd
i
ngs
i
n
t
he
t
e
x
t
ua
l
d
a
t
a
a
nd
a
ud
i
o
f
e
a
t
ur
e
e
xt
r
a
c
t
i
ons
.
F
or
fur
t
he
r
a
na
l
ys
i
s
,
s
o
a
s
t
o
i
nc
re
a
s
e
t
he
a
c
c
ura
c
y
o
f
t
he
pe
rfor
m
a
nc
e
of
t
he
m
ode
l
n
on
ve
rb
a
l
c
om
m
un
i
c
a
t
i
ons
l
i
k
e
j
e
s
t
e
rs
a
n
d
i
m
a
ge
s
c
a
n
be
i
nc
or
pora
t
e
d
.
T
h
e
m
u
l
t
i
m
oda
l
a
ppr
oa
c
h
c
a
n
i
nt
e
gr
a
t
e
a
l
l
t
he
i
nf
orm
a
t
i
on
re
l
a
t
e
d
t
o
t
he
c
o
m
m
uni
c
a
t
i
o
n,
w
hi
c
h
i
n
t
ur
n
c
a
n
m
a
k
e
t
he
hum
a
n
c
om
put
e
r
i
nt
e
r
a
c
t
i
ons
m
o
re
re
a
l
i
s
t
i
c
a
nd
m
e
a
ni
ng
ful
.
R
EF
ER
EN
C
ES
[
1]
W
.
M
e
dha
t
,
e
t
a
l
.
,
“
S
e
n
t
i
m
e
nt
a
na
l
y
s
i
s
a
l
g
or
i
t
h
m
s
a
nd
a
pp
l
i
c
a
t
i
o
ns
:
A
s
ur
v
e
y
,”
A
i
n
Sham
s
E
n
gi
n
e
e
r
i
n
g
J
ou
r
na
l
,
vol
.
5
,
no
.
4
,
p
p.
10
93
-
1113
,
D
e
c
.
20
14
.
[
2]
M
.
W
.
K
r
a
u
s
,
“
V
o
i
c
e
-
o
nl
y
c
o
m
m
u
ni
c
a
t
i
o
n
e
nha
n
c
e
s
e
m
p
a
t
hi
c
a
c
c
ur
a
c
y
,”
A
m
e
r
i
c
a
n
P
s
y
c
hol
ogi
s
t
,
vo
l
.
72
,
n
o.
7
,
pp.
64
4
-
6
54
,
2
017
.
[
3]
S
.
P
o
r
i
a
,
e
t
a
l
.
,
“
A
r
e
v
i
e
w
o
f
a
f
f
e
c
t
i
ve
c
o
m
pu
t
i
ng:
F
r
o
m
un
i
m
od
a
l
a
n
a
l
ys
i
s
t
o
m
ul
t
i
m
oda
l
f
u
s
i
on
,”
I
n
f
o
r
m
at
i
o
n
F
us
i
on
,
v
ol
.
37
,
p
p.
98
-
12
5
,
S
e
p
.
2017
.
[
4]
E
.
H
a
dd
i
,
e
t
a
l
.
,
“
T
h
e
r
o
l
e
o
f
t
e
x
t
pr
e
-
p
r
o
c
e
s
s
i
ng
i
n
s
e
n
t
i
m
e
n
t
a
na
l
y
s
i
s
,”
P
r
oc
e
d
i
a
C
om
put
e
r
S
c
i
e
nc
e
,
v
ol
.
17
,
pp.
26
-
32
,
201
3.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
Se
nt
i
m
e
n
t
a
nal
y
s
i
s
by
de
e
p
l
e
ar
ni
ng
appr
o
ac
h
e
s
(Sr
e
e
v
i
d
y
a
P
)
759
[
5]
V
.
N
a
r
a
ya
na
n
,
e
t
a
l
.
,
“
F
a
s
t
a
nd
a
c
c
ur
a
t
e
s
e
n
t
i
m
e
n
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
u
s
i
ng
a
n
e
nh
a
nc
e
d
N
a
ï
ve
B
a
ye
s
m
ode
l
,”
i
n
I
nt
e
r
n
at
i
on
al
C
on
f
e
r
e
n
c
e
on
I
n
t
e
l
l
i
g
e
n
t
D
a
t
a
E
ng
i
n
e
e
r
i
n
g
a
nd
A
ut
o
m
at
e
d
L
e
ar
n
i
n
g
,
S
p
r
i
nge
r
,
pp
.
194
-
20
1
,
20
13
.
[
6]
J
.
O
r
t
i
g
os
a
-
H
e
r
n
´
a
nd
e
z
,
e
t
a
l
.
,
“
A
pp
r
oa
c
h
i
ng
s
e
nt
i
m
e
n
t
a
na
l
ys
i
s
b
y
us
i
ng
s
e
m
i
-
s
up
e
r
v
i
s
e
d
l
e
a
r
ni
n
g
o
f
m
u
l
t
i
-
di
m
e
ns
i
ona
l
c
l
a
s
s
i
f
i
e
r
s
,”
N
e
ur
oc
om
p
ut
i
ng
,
vo
l
.
92
,
pp
.
98
-
11
5
,
S
e
p
20
12
.
[
7]
Y
.
M
.
L
i
a
n
d
T
.
Y
.
Li
,
“
D
e
r
i
v
i
ng
m
a
r
k
e
t
i
nt
e
l
l
i
ge
n
c
e
f
r
om
m
i
c
r
o
b
l
ogs
,”
D
e
c
i
s
i
o
n
S
uppo
r
t
Sy
s
t
e
m
s
,
v
ol
.
55
,
no
.
1
,
pp.
20
6
-
2
17
,
A
p
r
.
2013
.
[
8]
R
.
M
o
r
a
e
s
,
e
t
a
l
.
,
“
D
oc
um
e
n
t
-
l
e
v
e
l
s
e
n
t
i
m
e
nt
c
l
a
s
s
i
f
i
c
a
t
i
on
:
A
n
e
m
p
i
r
i
c
a
l
c
om
pa
r
i
s
on
be
t
w
e
e
n
S
V
M
a
nd
A
N
N
,”
E
x
pe
r
t
Sy
s
t
e
m
s
w
i
t
h
A
pp
l
i
c
a
t
i
o
ns
:
A
n
I
n
t
e
r
nat
i
o
nal
J
our
nal
,
vo
l
.
40
,
no.
2
,
pp
.
621
-
633
,
F
e
b
.
20
13
.
[
9]
M
.
T
a
b
oa
d
a
,
e
t
a
l
.
,
“
L
e
x
i
c
o
n
-
ba
s
e
d
m
e
t
ho
ds
f
o
r
s
e
n
t
i
m
e
nt
a
n
a
l
y
s
i
s
,”
C
om
pu
t
a
t
i
ona
l
l
i
n
gu
i
s
t
i
c
s
,
vo
l
.
37
,
no.
2
,
pp.
26
7
-
3
07
,
J
un
.
2
011
.
[
10]
N
.
Hu
,
e
t
a
l
.
,
“
M
a
n
i
pu
l
a
t
i
on
o
f
on
l
i
ne
r
e
v
i
e
w
s
:
A
n
a
n
a
l
y
s
i
s
o
f
r
a
t
i
n
gs
,
r
e
a
da
bi
l
i
t
y
,
a
nd
s
e
n
t
i
m
e
nt
s
,”
D
e
c
i
s
i
on
s
uppo
r
t
s
y
s
t
e
m
s
,
vo
l
.
52
,
no
.
3
,
pp
.
6
74
-
684
,
F
e
b
.
20
12
.
[
11]
S
.
M
.
K
i
m
a
nd
E
.
H
ov
y
,
“
D
e
t
e
r
m
i
n
i
ng
t
h
e
s
e
nt
i
m
e
n
t
o
f
o
pi
n
i
o
ns
,”
i
n
P
r
oc
e
e
d
i
ngs
of
t
h
e
20
t
h
i
n
t
e
r
na
t
i
ona
l
c
onf
e
r
e
nc
e
o
n
C
om
pu
t
at
i
on
al
L
i
ngu
i
s
t
i
c
s
,
p
p
.
1367
-
13
73
,
A
u
g
.
2
00
4
.
[
12]
L
.
P
.
M
or
e
nc
y
,
e
t
a
l
.
,
“
T
ow
a
r
ds
m
u
l
t
i
m
o
da
l
s
e
nt
i
m
e
n
t
a
na
l
ys
i
s
:
H
a
r
ve
s
t
i
n
g
op
i
n
i
on
s
f
r
o
m
t
he
w
e
b
,”
i
n
P
r
o
c
e
e
di
n
gs
of
t
h
e
13t
h
i
n
t
e
r
na
t
i
o
nal
c
on
f
e
r
e
nc
e
o
n
m
ul
t
i
m
o
da
l
i
n
t
e
r
f
a
c
e
s
,
p
p.
1
69
-
1
76
,
N
ov
.
201
1
.
[
13]
V
.
P
´
e
r
e
z
-
R
o
s
a
s
,
e
t
a
l
.
,
“
U
t
t
e
r
a
nc
e
-
l
e
ve
l
m
ul
t
i
m
oda
l
s
e
nt
i
m
e
n
t
a
na
l
ys
i
s
,”
i
n
P
r
o
c
e
e
di
n
gs
o
f
t
h
e
51s
t
A
nn
ual
M
e
e
t
i
ng
o
f
t
he
A
s
s
o
c
i
a
t
i
on
f
or
C
om
pu
t
a
t
i
ona
l
L
i
ng
ui
s
t
i
c
s
,
v
ol
.
1
,
p
p
.
973
-
982
,
A
ug
.
201
3
.
[
14]
B
.
B
a
l
a
j
i
a
nd
V
.
R
.
M
.
O
r
uga
n
t
i
,
“
M
u
l
t
i
-
l
e
v
e
l
f
e
a
t
ur
e
f
us
i
o
n
f
or
g
r
o
up
-
l
e
v
e
l
e
m
o
t
i
on
r
e
c
ogn
i
t
i
on
,”
i
n
P
r
oc
e
e
d
i
ng
s
of
t
he
19
t
h
A
C
M
I
n
t
e
r
nat
i
o
nal
C
o
nf
e
r
e
n
c
e
on
M
u
l
t
i
m
oda
l
I
n
t
e
r
ac
t
i
on
,
p
p
.
583
-
586
,
N
ov
.
201
7
.
[
15]
A
.
Z
a
de
h
,
e
t
a
l
.
,
“
M
u
l
t
i
-
a
t
t
e
nt
i
on
r
e
c
u
r
r
e
nt
ne
t
w
or
k
f
or
hu
m
a
n
c
o
m
m
un
i
c
a
t
i
on
c
o
m
p
r
e
h
e
ns
i
on
,”
i
n
T
h
i
r
t
y
-
Se
c
ond
A
A
A
I
C
on
f
e
r
e
nc
e
o
n
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
nc
e
,
pp
.
564
2
-
5
649
,
A
pr
.
20
18.
[
16]
S
.
P
or
i
a
,
e
t
a
l
.
,
“
C
on
t
e
xt
-
de
pe
nd
e
n
t
s
e
n
t
i
m
e
nt
a
na
l
ys
i
s
i
n
u
s
e
r
-
g
e
ne
r
a
t
e
d
v
i
d
e
os
,”
i
n
P
r
oc
e
e
d
i
ng
s
of
t
he
55
t
h
A
nnua
l
M
e
e
t
i
ng
o
f
t
he
A
s
s
o
c
i
a
t
i
on
f
or
C
om
pu
t
a
t
i
ona
l
L
i
ng
ui
s
t
i
c
s
,
vo
l
.
1
,
p
p
.
873
-
883
,
J
u
l
.
2
017
.
[
17]
F
.
E
y
be
n
,
e
t
a
l
.
,
“
O
p
e
nE
A
R
i
nt
r
odu
c
i
ng
t
he
M
u
ni
c
h
o
pe
n
-
s
o
u
r
c
e
e
m
ot
i
o
n
a
n
d
a
f
f
e
c
t
r
e
c
o
gni
t
i
on
t
oo
l
k
i
t
,”
i
n
2
009
3r
d
i
nt
e
r
n
at
i
on
al
c
on
f
e
r
e
n
c
e
on
af
f
e
c
t
i
v
e
c
om
put
i
ng
an
d
i
n
t
e
l
l
i
g
e
nt
i
nt
e
r
a
c
t
i
on
and
w
or
k
s
h
ops
,
vol
.
1,
p
p.
1
-
6
,
2
009
.
[
18]
S
.
G
.
A
j
a
y
,
e
t
a
l
.
,
“
E
xpl
or
i
ng
t
he
s
i
gni
f
i
c
a
n
c
e
o
f
l
ow
f
r
e
qu
e
nc
y
r
e
g
i
ons
i
n
e
l
e
c
t
r
og
l
o
t
t
ogr
a
ph
i
c
s
i
g
na
l
s
f
or
e
m
ot
i
on
r
e
c
o
gn
i
t
i
on
,”
i
n
I
nt
e
r
n
at
i
on
al
S
y
m
po
s
i
um
on
S
i
gna
l
P
r
oc
e
s
s
i
ng
and
I
n
t
e
l
l
i
g
e
n
t
R
e
c
ogn
i
t
i
on
Sy
s
t
e
m
s
,
Spr
i
ng
e
r
,
p
p.
31
9
-
3
27
,
2
017
.
[
19]
Y
.
B
e
ngi
o
,
e
t
a
l
.
,
“
A
n
e
u
r
a
l
p
r
ob
a
bi
l
i
s
t
i
c
l
a
ng
ua
g
e
m
od
e
l
,”
J
our
na
l
of
m
ac
h
i
n
e
l
e
a
r
ni
ng
r
e
s
e
ar
c
h
,
v
ol
.
3
,
pp.
11
37
-
1155
,
F
e
b
20
03
.
[
20]
M
.
O
qua
b
,
e
t
a
l
.
,
“
I
s
ob
j
e
c
t
l
oc
a
l
i
z
a
t
i
on
f
o
r
f
r
e
e
?
-
w
e
a
k
l
y
-
s
u
pe
r
vi
s
e
d
l
e
a
r
n
i
ng
w
i
t
h
c
onvo
l
u
t
i
ona
l
n
e
ur
a
l
ne
t
w
or
k
s
,”
i
n
P
r
o
c
e
e
di
n
gs
o
f
t
he
I
E
E
E
C
onf
e
r
e
nc
e
o
n
C
om
pu
t
e
r
V
i
s
i
on
an
d
P
a
t
t
e
r
n
R
e
c
ogn
i
t
i
o
n
,p
p.
6
85
-
694
,
2015
.
[
21]
C
.
d
os
S
a
nt
o
s
a
nd
M
.
G
a
t
t
i
,
“
D
e
e
p
c
on
vol
ut
i
on
a
l
n
e
ur
a
l
n
e
t
w
or
ks
f
or
s
e
n
t
i
m
e
nt
a
na
l
ys
i
s
o
f
s
ho
r
t
t
e
x
t
s
,
”
i
n
P
r
oc
e
e
d
i
ng
s
o
f
C
O
L
I
N
G
201
4,
t
h
e
25
t
h
I
n
t
e
r
na
t
i
o
na
l
C
on
f
e
r
e
n
c
e
o
n
C
om
pu
t
a
t
i
o
nal
L
i
ngu
i
s
t
i
c
s
:
T
e
c
hn
i
c
a
l
P
ape
r
s
,
p
p.
69
-
78
,
A
ug
.
201
4
.
[
22]
I
.
G
ood
f
e
l
l
ow
,
e
t
a
l
.
,
“
D
e
e
p
l
e
a
r
n
i
ng
,”
M
I
T
p
r
e
s
s
,
2
016
.
[
23]
L
.
B
o
t
t
o
u,
e
t
a
l
.
,
“
O
pt
i
m
i
z
a
t
i
on
m
e
t
hod
s
f
or
l
a
r
ge
-
s
c
a
l
e
m
a
c
h
i
ne
l
e
a
r
n
i
ng
,”
Si
am
R
e
v
i
e
w
,
vo
l
.
60
,
no
.
2
,
pp.
22
3
-
3
11
,
2
018
.
[
24]
D
.
P
.
K
i
n
gm
a
a
n
d
J
.
Ba
,
“
A
d
a
m
:
A
m
e
t
h
od
f
o
r
s
t
oc
h
a
s
t
i
c
o
pt
i
m
i
z
a
t
i
o
n
,”
a
r
X
i
v
.
o
r
g
,
a
r
X
i
v
:
141
2698
0
,
20
14
.
[
25]
S
.
P
o
r
i
a
,
e
t
a
l
.
,
“
M
u
l
t
i
m
oda
l
S
e
n
t
i
m
e
nt
A
na
l
ys
i
s
:
A
d
dr
e
s
s
i
ng
K
e
y
I
s
s
u
e
s
a
nd
S
e
t
t
i
ng
U
p
t
he
B
a
s
e
l
i
ne
s
,”
I
E
E
E
I
nt
e
l
l
i
ge
n
t
Sy
s
t
e
m
s
,
v
ol
.
33
,
n
o.
6
,
pp
.
17
-
25
,
20
18
.
[
26]
H
.
W
a
ng
,
e
t
a
l
.
,
“
S
e
l
e
c
t
-
a
d
di
t
i
ve
l
e
a
r
n
i
n
g:
I
m
p
r
ov
i
ng
ge
n
e
r
a
l
i
z
a
t
i
on
i
n
m
u
l
t
i
m
od
a
l
s
e
nt
i
m
e
n
t
a
na
l
y
s
i
s
,
”
i
n
2
017
I
E
E
E
I
n
t
e
r
na
t
i
o
na
l
C
o
nf
e
r
e
nc
e
on
M
u
l
t
i
m
e
d
i
a
an
d
E
x
po
(
I
C
M
E
)
,
p
p.
949
-
95
4
,
20
17
.
[
27]
S
.
P
o
r
i
a
,
e
t
a
l
.
,
“
C
o
nvo
l
u
t
i
o
na
l
M
K
L
b
a
s
e
d
m
ul
t
i
m
oda
l
e
m
o
t
i
o
n
r
e
c
ogn
i
t
i
on
a
nd
s
e
n
t
i
m
e
nt
a
na
l
ys
i
s
,
”
i
n
20
16
I
E
E
E
16t
h
i
nt
e
r
n
at
i
on
al
c
o
nf
e
r
e
nc
e
on
d
at
a
m
i
ni
ng
(
I
C
D
M
)
,
p
p.
43
9
-
4
48
,
201
6
.
[
28]
E
.
C
a
m
b
r
i
a
,
e
t
a
l
.
,
“
B
e
nc
hm
a
r
ki
n
g
m
ul
t
i
m
o
da
l
s
e
nt
i
m
e
n
t
a
n
a
l
y
s
i
s
,”
i
n
I
nt
e
r
n
at
i
on
al
C
o
nf
e
r
e
n
c
e
o
n
C
om
pu
t
a
t
i
o
na
l
L
i
ngu
i
s
t
i
c
s
a
nd
I
n
t
e
l
l
i
ge
nt
T
e
x
t
P
r
oc
e
s
s
i
ng
,
Sp
r
i
n
ge
r
,
p
p
.
166
-
179
,
2017
.
[
29]
Y
.
H
.
H
.
T
s
a
i
,
e
t
a
l
.
,
“
L
e
a
r
ni
n
g
f
a
c
t
o
r
i
z
e
d
m
u
l
t
i
m
od
a
l
r
e
pr
e
s
e
nt
a
t
i
o
ns
,”
a
r
X
i
v
.
or
g,
a
r
X
i
v
:
1
8060
6176
,
201
8
.
BI
O
G
R
A
P
H
I
ES
O
F
A
U
T
H
O
R
S
S
r
e
e
v
i
d
y
a
P
.
i
s
c
u
r
r
e
nt
l
y
a
R
e
s
e
a
r
c
h
S
c
h
ol
a
r
i
n
t
h
e
de
pa
r
t
m
e
nt
of
E
l
e
c
t
r
o
ni
c
s
a
nd
I
ns
t
r
um
e
n
t
a
t
i
o
n
a
t
A
m
r
i
t
a
V
i
s
w
a
V
i
dy
a
pe
e
t
h
a
m
,
C
oi
m
ba
t
or
e
.
S
he
w
a
s
gr
a
du
a
t
e
d
f
r
o
m
C
a
l
i
c
ut
U
ni
ve
r
s
i
t
y
a
nd
d
i
d
he
r
M
a
s
t
e
r
s
a
t
P
S
G
c
ol
l
e
ge
of
T
e
c
h
no
l
ogy
,
C
oi
m
ba
t
o
r
e
.
S
h
e
h
a
s
got
m
or
e
t
ha
n
10
p
ubl
i
c
a
t
i
on
s
i
n
he
r
c
r
e
di
t
.
H
e
r
m
a
j
o
r
a
r
e
a
s
o
f
i
nt
e
r
e
s
t
a
r
e
a
r
t
i
f
i
c
i
a
l
i
nt
e
l
l
i
ge
n
c
e
,
a
f
f
e
c
t
i
ve
c
om
pu
t
i
n
g,
m
u
l
t
i
m
oda
l
a
i
t
y
a
nd
i
m
a
ge
p
r
oc
e
s
s
i
n
g.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
1
693
-
69
30
T
E
L
K
O
M
N
IK
A
T
e
l
e
c
o
m
m
u
n
Co
m
pu
t
E
l
Co
nt
ro
l
,
V
ol
.
18
,
N
o.
2
,
A
pri
l
2
020:
752
-
7
60
760
D
r
.
O
.
V
.
R
a
m
an
a
M
u
r
t
h
y
c
u
r
r
e
n
t
l
y
s
e
r
v
e
s
a
s
A
s
s
i
s
t
a
n
t
P
r
o
f
e
s
s
or
(
S
G
)
a
t
t
h
e
D
e
pa
r
t
m
e
n
t
of
E
l
e
c
t
r
i
c
a
l
a
nd
E
l
e
c
t
r
o
ni
c
s
E
n
gi
ne
e
r
i
ng
,
S
c
h
oo
l
of
E
ng
i
ne
e
r
i
ng
,
A
m
r
i
t
a
V
i
s
hw
a
V
i
d
ya
p
e
e
t
ha
m
,
C
oi
m
ba
t
o
r
e
C
a
m
pus
.
I
nd
i
a
.
H
e
h
a
s
m
o
r
e
t
h
a
n
9
ye
a
r
s
of
r
e
s
e
a
r
c
h
e
xpe
r
i
e
nc
e
i
n
r
e
pu
t
e
d
U
ni
ve
r
s
i
t
i
e
s
i
n
c
l
u
di
n
g
N
a
t
i
on
a
l
U
ni
ve
r
s
i
t
y
o
f
S
i
ng
a
po
r
e
,
N
a
nya
ng
T
e
c
hn
ol
o
gi
c
a
l
U
n
i
ve
r
s
i
t
y,
S
i
ng
a
po
r
e
a
n
d
U
ni
v
e
r
s
i
t
y
o
f
C
a
nb
e
r
r
a
,
A
u
s
t
r
a
l
i
a
.
H
e
ha
s
m
o
r
e
t
ha
n
20
pu
bl
i
c
a
t
i
on
s
i
n
m
a
j
or
i
nt
e
r
na
t
i
o
na
l
j
o
ur
n
a
l
s
a
nd
c
on
f
e
r
e
n
c
e
s
t
o
h
i
s
c
r
e
di
t
.
H
e
ha
s
p
a
r
t
i
c
i
p
a
t
e
d
i
n
a
nu
m
b
e
r
o
f
w
or
ks
hop
s
o
n
B
i
g
da
t
a
,
de
e
p
l
e
a
r
n
i
n
g,
s
t
a
t
i
s
t
i
c
s
.
D
r
.
S
.
V
e
n
i
j
oi
n
e
d
A
m
r
i
t
a
V
i
s
hw
a
V
i
dy
a
pe
e
t
ha
m
a
s
a
f
a
c
u
l
t
y
i
n
t
h
e
y
e
a
r
20
01
u
nde
r
t
he
D
e
pa
r
t
m
e
nt
o
f
E
l
e
c
t
r
oni
c
s
a
nd
C
o
m
m
un
i
c
a
t
i
on
E
ng
i
ne
e
r
i
ng
.
S
he
i
s
c
u
r
r
e
n
t
l
y
w
o
r
k
i
ng
a
s
A
s
s
oc
i
a
t
e
P
r
o
f
e
s
s
o
r
.
S
he
r
e
c
e
i
ve
d
he
r
A
M
I
E
de
g
r
e
e
f
r
o
m
I
ns
t
i
t
ut
i
on
o
f
E
n
gi
ne
e
r
s
,
C
a
l
c
ut
t
a
,
i
n
t
h
e
ye
a
r
199
4
a
nd
M
.
E
.
de
g
r
e
e
(
A
pp
l
i
e
d
E
l
e
c
t
r
o
ni
c
s
)
f
r
o
m
B
ha
r
a
t
hi
a
r
U
n
i
ve
r
s
i
t
y,
C
oi
m
ba
t
o
r
e
,
i
n
t
h
e
y
e
a
r
1998
.
S
he
o
bt
a
i
n
e
d
h
e
r
P
h.
D
.
d
e
gr
e
e
i
n
t
he
a
r
e
a
of
I
m
a
g
e
P
r
oc
e
s
s
i
ng
f
r
o
m
A
m
r
i
t
a
V
i
s
hw
a
V
i
dy
a
pe
e
t
h
a
m
i
n
J
a
nu
a
r
y
201
2.
H
e
r
a
r
e
a
s
of
i
nt
e
r
e
s
t
i
n
c
l
ude
S
i
g
na
l
a
nd
I
m
a
ge
P
r
oc
e
s
s
i
ng
,
H
a
r
dw
a
r
e
I
m
p
l
e
m
e
n
t
a
t
i
on
o
f
S
i
gn
a
l
a
nd
I
m
a
g
e
P
r
oc
e
s
s
i
n
g
A
l
gor
i
t
h
m
s
.
S
he
h
a
s
pub
l
i
s
he
d
n
e
a
r
l
y
50
pa
pe
r
s
i
n
i
n
t
e
r
na
t
i
ona
l
j
ou
r
n
a
l
s
a
nd
c
onf
e
r
e
nc
e
s
.
S
he
ha
s
r
e
c
e
i
ve
d
be
s
t
p
a
pe
r
a
w
a
r
d
f
ou
r
t
i
m
e
s
f
o
r
t
h
e
p
a
pe
r
s
s
he
ha
d
p
r
e
s
e
n
t
e
d
i
n
t
he
c
on
f
e
r
e
n
c
e
s
.
.
S
he
i
s
a
n
A
s
s
oc
i
a
t
e
M
e
m
be
r
i
n
t
he
I
ns
t
i
t
u
t
i
o
n
of
E
ng
i
ne
e
r
s
,
M
e
m
b
e
r
i
n
I
S
T
E
,
M
e
m
be
r
i
n
I
E
T
E
,
B
O
S
m
e
m
b
e
r
,
D
oc
t
o
r
a
l
C
om
m
i
t
t
e
e
M
e
m
be
r
(
A
m
r
i
t
a
V
i
s
hw
a
V
i
dy
a
pe
e
t
h
a
m
,
A
nn
a
U
ni
ve
r
s
i
t
y
,
K
a
r
un
ya
U
n
i
v
e
r
s
i
t
y
a
nd
K
e
r
a
l
a
A
g
r
i
c
ul
t
u
r
a
l
U
n
i
ve
r
s
i
t
y)
,
A
c
a
de
m
i
c
C
o
m
m
u
ni
t
y
M
e
m
b
e
r
i
n
I
nt
e
r
n
a
t
i
o
na
l
C
ong
r
e
s
s
f
or
g
l
o
ba
l
S
c
i
e
n
c
e
a
nd
T
e
c
hno
l
og
y
(
I
C
G
S
T
)
.
S
he
h
a
s
s
e
r
ve
d
a
s
a
r
e
v
i
e
w
e
r
a
nd
s
e
s
s
i
on
c
ha
i
r
i
n
t
he
I
nt
e
r
na
t
i
o
na
l
c
on
f
e
r
e
n
c
e
s
,
r
e
v
i
e
w
e
r
i
n
r
e
pu
t
e
d
j
ou
r
na
l
s
l
i
ke
I
nde
r
s
c
i
e
nc
e
a
nd
I
E
T
I
m
a
ge
P
r
oc
e
s
s
i
n
g.
Evaluation Warning : The document was created with Spire.PDF for Python.