I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
pu
t
er
E
ng
ineering
(
I
J
E
CE
)
Vo
l.
15
,
No
.
5
,
Octo
b
er
20
25
,
p
p
.
5
0
1
9
~
5
0
3
0
I
SS
N:
2088
-
8
7
0
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijece.
v
15
i
5
.
pp
5
0
1
9
-
5
0
3
0
5019
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
A
no
v
el appro
a
ch
f
o
r re
co
mm
e
nda
t
io
n using
optimiz
ed
bidirectiona
l ga
te
d recur
rent
unit
P
ra
k
a
s
h P
a
nd
ha
rina
t
h Ro
ka
de
1
,
Swa
t
i Ba
ba
s
a
heb
B
ho
nd
e
2
,
P
ra
s
ha
nt
L
a
x
m
a
nra
o
P
a
i
k
ra
o
3
,
Um
esh
B
a
bu
ra
o
P
a
wa
r
4
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
S
.
N
.
D
.
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
a
n
d
R
e
se
a
r
c
h
C
e
n
t
e
r
Y
e
o
l
a
,
S
a
v
i
t
r
i
b
a
i
P
h
u
l
e
P
u
n
e
U
n
i
v
e
r
si
t
y
,
M
a
h
a
r
a
s
h
t
r
a
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
A
mr
u
t
v
a
h
i
n
i
C
o
l
l
e
g
e
o
f
E
n
g
i
n
e
e
r
i
n
g
,
M
a
h
a
r
a
sh
t
r
a
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s a
n
d
T
e
l
e
c
o
mm
u
n
i
c
a
t
i
o
n
,
G
o
v
e
r
n
me
n
t
C
O
E
A
m
r
a
v
a
t
i
,
M
a
h
a
r
a
s
h
t
r
a
,
I
n
d
i
a
4
S
c
h
o
o
l
o
f
C
o
m
p
u
t
i
n
g
,
S
a
n
d
i
p
U
n
i
v
e
r
si
t
y
N
a
s
h
i
k
,
M
a
h
a
r
a
s
h
t
r
a
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
2
6
,
2
0
2
4
R
ev
is
ed
J
u
l 3
,
2
0
2
4
Acc
ep
ted
J
u
l 1
2
,
2
0
2
5
In
t
o
d
a
y
'
s
wo
rld
,
e
v
e
ry
o
n
e
o
f
u
s
re
fre
sh
e
s
o
u
r
m
o
o
d
a
n
d
g
e
ts
e
n
e
rg
y
th
ro
u
g
h
e
n
terta
in
m
e
n
t
a
n
d
e
n
j
o
y
m
e
n
t.
Hu
m
a
n
n
a
t
u
re
is
to
p
r
o
v
i
d
e
fe
e
d
b
a
c
k
th
ro
u
g
h
ra
ti
n
g
s
o
r
c
o
m
m
e
n
ts
fo
r
p
ro
d
u
c
ts
u
se
d
,
se
rv
ice
s
re
c
e
iv
e
d
,
o
r
fil
m
s
v
iew
e
d
.
T
h
e
re
c
o
m
m
e
n
d
a
ti
o
n
sy
s
tem
se
rv
e
s
th
e
u
se
r
with
re
c
o
m
m
e
n
d
a
ti
o
n
s
b
a
se
d
o
n
h
ist
o
rica
l
sto
re
d
in
fo
r
m
a
ti
o
n
o
f
u
se
r
p
re
fe
re
n
c
e
s.
Th
e
se
sy
ste
m
s
a
m
a
ss
in
fo
rm
a
ti
o
n
a
b
o
u
t
t
h
e
u
se
r
in
o
r
d
e
r
to
p
ro
v
id
e
p
e
rso
n
a
li
z
e
d
e
x
p
e
rien
c
e
s.
Th
e
se
sy
ste
m
s
p
u
t
e
ffo
rts
in
t
o
d
e
li
v
e
rin
g
p
e
rso
n
a
li
z
e
d
e
x
p
e
rien
c
e
s
b
y
a
c
c
u
m
u
latin
g
i
n
f
o
rm
a
ti
o
n
a
b
o
u
t
t
h
e
u
se
r.
Hy
b
ri
d
a
lg
o
rit
h
m
s
a
re
n
e
c
e
ss
a
ry
to
a
d
d
re
ss
th
e
issu
e
s
re
c
o
m
m
e
n
d
a
ti
o
n
sy
ste
m
s
c
o
n
fro
n
t,
w
h
ic
h
in
c
lu
d
e
lo
w
p
re
d
ictio
n
a
c
c
u
ra
c
y
,
o
u
t
p
u
t
t
h
a
t
e
x
c
e
e
d
s
ra
n
g
e
,
a
n
d
i
n
a
d
e
q
u
a
t
e
c
o
n
v
e
rg
e
n
c
e
sp
e
e
d
.
Th
is
stu
d
y
s
u
g
g
e
sts
b
u
il
d
i
n
g
a
m
o
v
ie
re
c
o
m
m
e
n
d
a
ti
o
n
sy
ste
m
u
sin
g
th
e
re
m
o
ra
o
p
ti
m
iza
ti
o
n
a
lg
o
rit
h
m
(ROA
)
a
n
d
th
e
b
i
d
irec
ti
o
n
a
l
g
a
ted
re
c
u
rre
n
t
u
n
it
(BiG
RU),
th
e
m
o
st r
e
c
e
n
t
v
e
rsio
n
o
f
th
e
re
c
u
rsi
v
e
n
e
u
ra
l
n
e
two
rk
(RNN
).
Th
e
p
ro
p
o
se
d
m
e
th
o
d
'
s
re
su
lt
s
a
re
c
o
m
p
a
re
d
wit
h
th
o
se
o
f
th
e
g
e
n
e
ti
c
a
lg
o
ri
th
m
(G
A),
fe
e
d
fo
rwa
rd
n
e
u
ra
l
n
e
two
r
k
(F
F
NN
),
a
n
d
m
u
lt
imo
d
a
l
d
e
e
p
lea
rn
in
g
(M
M
DL).
In
term
s
o
f
m
o
v
ie
re
c
o
m
m
e
n
d
a
ti
o
n
,
BiG
RU wit
h
ROA
p
e
rfo
rm
s b
e
tt
e
r
th
a
n
G
A,
M
M
DL,
a
n
d
F
F
NN
.
K
ey
w
o
r
d
s
:
B
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
en
t
u
n
it
Feed
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
Gate
d
r
ec
u
r
r
e
n
t u
n
it
Gen
etic
alg
o
r
ith
m
Mu
ltimo
d
al
d
ee
p
lea
r
n
in
g
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Pra
k
ash
Pan
d
h
ar
in
at
h
R
o
k
ad
e
Dep
ar
tm
en
t o
f
I
n
f
o
r
m
atio
n
T
e
ch
n
o
lo
g
y
,
S.N.
D.
C
o
lleg
e
o
f
E
n
g
in
ee
r
in
g
an
d
R
esear
ch
C
en
ter
Yeo
la
,
Sav
itrib
ai
Ph
u
le
Pu
n
e
Un
iv
e
r
s
ity
Nash
ik
,
Ma
h
ar
ash
tr
a,
I
n
d
ia
-
4
2
3
4
0
1
E
m
ail: p
r
ak
ash
r
o
k
ad
e2
0
0
5
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
So
cial
m
ed
ia
is
h
ea
v
ily
u
s
ed
b
y
p
eo
p
le
to
s
h
ar
e
th
eir
o
p
i
n
io
n
s
an
d
f
ee
lin
g
s
o
n
m
an
y
web
s
ites
.
T
h
ese
f
ee
lin
g
s
m
an
if
est
as
o
p
in
io
n
s
o
r
r
atin
g
s
f
o
r
a
g
o
o
d
o
r
s
er
v
ic
e.
T
h
e
g
r
ea
test
d
if
f
icu
lty
f
ac
in
g
s
o
m
eo
n
e
wh
o
h
as
to
m
ak
e
a
n
o
n
lin
e
p
u
r
ch
ase
in
th
e
in
f
o
r
m
atio
n
er
a
is
n
o
t
o
n
l
y
f
in
d
i
n
g
e
n
o
u
g
h
o
p
tio
n
s
o
r
i
n
f
o
r
m
atio
n
b
u
t
also
d
ec
id
in
g
h
o
w
to
u
s
e
th
e
wea
lth
o
f
av
ailab
le
d
ata
to
th
eir
ad
v
an
tag
e
[
1
]
–
[
3
]
.
T
h
is
lead
s
to
th
e
g
en
er
atio
n
an
d
an
aly
s
is
o
f
m
ass
iv
e
am
o
u
n
ts
o
f
d
ata
in
o
r
d
er
to
f
o
r
ec
ast
an
d
s
u
g
g
est
a
u
s
er
an
y
p
r
o
d
u
ct
o
r
s
er
v
ice
b
ased
o
n
h
is
in
ter
ests
.
Mo
v
ie
r
atin
g
d
at
ab
ases
co
m
p
r
is
in
g
u
s
er
r
atin
g
s
an
d
v
a
r
io
u
s
m
o
v
ie
p
a
r
am
ete
r
s
m
ay
b
e
f
o
u
n
d
o
n
a
n
u
m
b
e
r
o
f
well
-
k
n
o
wn
web
s
ites
,
s
u
ch
as
Kag
g
le.
I
n
o
r
d
er
to
ad
d
r
ess
a
p
r
o
b
lem
,
d
ec
i
s
io
n
s
s
u
p
p
o
r
ted
b
y
s
ev
er
al
s
tr
o
n
g
er
p
ast im
p
r
ess
io
n
s
ar
e
alwa
y
s
p
r
ef
er
ab
le
to
th
o
s
e
p
r
o
d
u
ce
d
b
y
a
s
in
g
le
u
s
er
im
p
r
ess
io
n
[
4
]
,
[
5
]
.
On
ly
th
o
s
e
u
s
er
s
wh
o
s
e
r
ati
n
g
s
ar
e
m
o
r
e
s
tr
o
n
g
l
y
r
elev
an
t
to
o
n
e
an
o
th
e
r
ar
e
g
ath
e
r
ed
,
as
o
p
p
o
s
ed
to
g
ath
er
in
g
all
r
e
v
iews
an
d
r
at
in
g
s
[
6
]
.
Usi
n
g
v
ar
io
u
s
m
ac
h
in
e
lear
n
in
g
tech
n
iq
u
es,
s
u
ch
as
s
u
p
p
o
r
t
v
ec
to
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
5
0
1
9
-
5
0
3
0
5020
m
ac
h
in
es
(
SVM)
,
n
eu
r
al
n
etw
o
r
k
s
(
NN)
,
an
d
g
e
n
etic
alg
o
r
ith
m
s
(
GA)
,
n
u
m
e
r
o
u
s
ac
a
d
em
i
cs
h
av
e
w
o
r
k
ed
to
im
p
r
o
v
e
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
[
7
]
–
[
9
]
.
I
n
th
is
s
tu
d
y
,
we
s
u
g
g
ested
a
b
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
en
t
u
n
it
(
B
iGR
U)
o
p
tim
ized
r
em
o
r
a
o
p
tim
izatio
n
alg
o
r
ith
m
(
R
OA)
b
ased
m
o
v
ie
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
.
T
h
e
1
0
0
k
B
I
MD
B
d
atase
t
is
p
r
o
ce
s
s
ed
u
s
in
g
th
e
B
iGR
U
m
eth
o
d
.
User
id
,
m
o
v
ieid
,
a
n
d
r
atin
g
a
r
e
th
e
o
n
ly
p
ar
a
m
eter
s
tak
en
in
to
ac
co
u
n
t
p
r
io
r
to
p
r
e
p
r
o
ce
s
s
in
g
[
1
0
]
.
Similar
ity
b
et
wee
n
u
s
er
s
'
r
atin
g
s
o
f
th
e
s
am
e
m
o
v
ies
an
d
th
eir
r
atin
g
s
o
f
th
eir
d
if
f
er
e
n
ce
s
f
r
o
m
o
n
e
a
n
o
th
er
ar
e
o
b
tain
ed
.
Fin
ally
,
a
u
s
er
r
ec
eiv
es
r
ec
o
m
m
en
d
atio
n
s
f
o
r
th
e
to
p
1
0
m
o
v
ies
b
ased
o
n
h
is
in
ter
est
p
atter
n
af
ter
th
e
weig
h
ts
u
s
ed
to
d
is
co
v
er
u
s
er
s
im
ilar
ity
ar
e
o
p
tim
ized
u
s
in
g
R
OA.
T
h
e
o
u
tco
m
es
o
f
GA,
m
u
ltimo
d
al
d
ee
p
lear
n
i
n
g
(
MM
DL
)
,
an
d
f
ee
d
f
o
r
wa
r
d
n
eu
r
al
n
etwo
r
k
s
(
FF
NN)
ar
e
co
n
tr
asted
with
th
e
r
esu
lts
.
B
iG
R
U
ap
p
ea
r
s
to
h
av
e
p
r
o
d
u
ce
d
b
etter
o
u
tc
o
m
es
f
o
r
ev
er
y
test
in
g
p
ar
am
eter
we
co
m
p
a
r
ed
ta
k
e
n
in
to
ac
co
u
n
t
p
r
i
o
r
to
p
r
e
p
r
o
ce
s
s
in
g
.
Similar
ity
b
etwe
en
u
s
er
s
'
r
atin
g
s
o
f
th
e
s
am
e
m
o
v
ies
an
d
th
eir
r
atin
g
s
o
f
t
h
eir
d
if
f
er
en
ce
s
f
r
o
m
o
n
e
an
o
th
e
r
ar
e
o
b
tain
e
d
[
1
1
]
,
[
1
2
]
.
K
-
n
ea
r
est
n
eig
h
b
o
r
alg
o
r
ith
m
with
co
lla
b
o
r
ativ
e
f
ilter
in
g
is
u
s
ed
f
o
r
m
o
v
ie
r
atin
g
p
r
ed
ictio
n
[
1
3
]
.
T
h
e
lo
n
g
s
h
o
r
t
-
ter
m
m
em
o
r
y
u
s
in
g
cy
clic
lear
n
in
g
r
ate
(
L
STM
-
C
L
R
)
f
r
am
ew
o
r
k
is
u
s
ed
to
id
en
tif
y
cy
b
er
b
u
lly
in
g
o
n
s
o
cial
m
ed
ia,
im
p
r
o
v
in
g
class
if
icatio
n
ac
cu
r
ac
y
with
o
u
t
n
u
m
e
r
o
u
s
tr
ials
an
d
ad
ju
s
tm
en
ts
[
1
4
]
.
Fin
ally
,
a
u
s
er
r
ec
eiv
es
r
ec
o
m
m
e
n
d
atio
n
s
f
o
r
th
e
to
p
1
0
m
o
v
ies
b
ased
o
n
h
is
in
ter
est
p
atter
n
af
te
r
th
e
weig
h
ts
u
s
ed
to
d
is
co
v
er
u
s
er
s
im
ilar
ity
ar
e
o
p
tim
ized
u
s
in
g
R
OA.
T
h
e
o
u
t
co
m
es
o
f
GA,
MM
DL
,
an
d
FF
NN
ar
e
co
n
tr
asted
with
th
e
r
esu
lts
.
B
i
G
R
U
ap
p
ea
r
s
to
h
av
e
p
r
o
d
u
ce
d
b
etter
o
u
tco
m
es
f
o
r
ev
er
y
test
in
g
p
ar
a
m
eter
we
co
m
p
ar
e
d
[
1
5
]
.
T
r
ad
itio
n
al
r
ec
o
m
m
e
n
d
a
tio
n
s
y
s
tem
s
,
s
u
ch
as
co
llab
o
r
ativ
e
f
ilter
in
g
an
d
co
n
te
n
t
-
b
a
s
ed
f
ilter
in
g
,
o
f
te
n
s
u
f
f
er
f
r
o
m
lim
itatio
n
s
in
clu
d
in
g
lo
w
p
r
ed
ictio
n
ac
cu
r
ac
y
,
s
lo
w
co
n
v
er
g
en
ce
s
p
ee
d
,
an
d
in
ef
f
ec
tiv
e
weig
h
t
o
p
tim
izatio
n
.
W
h
ile
d
ee
p
lea
r
n
in
g
m
o
d
els
lik
e
FF
NN
an
d
MM
DL
h
av
e
s
h
o
wn
im
p
r
o
v
em
en
ts
,
t
h
ey
s
till
s
tr
u
g
g
le
with
o
p
tim
al
weig
h
t
ass
ig
n
m
en
t
an
d
r
ea
l
-
tim
e
a
d
ap
tab
ilit
y
.
T
h
is
p
ap
e
r
in
tr
o
d
u
ce
s
an
o
p
tim
ized
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
u
s
in
g
B
iGR
U
with
R
OA
to
en
h
an
ce
ac
cu
r
ac
y
a
n
d
co
n
v
er
g
en
c
e
s
p
ee
d
.
T
h
e
m
ain
co
n
tr
ib
u
tio
n
s
o
f
th
is
p
ap
er
ar
e
to
p
r
o
p
o
s
e
o
f
a
B
iGR
U
-
b
ased
r
ec
o
m
m
en
d
atio
n
m
o
d
el
o
p
tim
ized
u
s
in
g
R
OA
to
en
h
an
ce
co
n
v
er
g
e
n
ce
s
p
ee
d
a
n
d
r
ed
u
ce
tr
ai
n
in
g
lo
s
s
,
to
co
m
p
ar
e
B
iGR
U
w
ith
FF
N
N,
MM
DL
,
an
d
GA
to
d
em
o
n
s
tr
ate
s
u
p
e
r
io
r
p
er
f
o
r
m
an
ce
an
d
to
im
p
lem
en
t
p
r
o
p
o
s
ed
s
y
s
tem
o
n
th
e
I
MD
B
m
o
v
ie
d
ataset,
ac
h
iev
in
g
a
s
ig
n
if
ican
t
im
p
r
o
v
em
en
t
i
n
p
r
ed
ictio
n
ac
c
u
r
ac
y
.
T
h
e
p
r
o
p
o
s
ed
s
y
s
tem
is
p
ar
ticu
lar
l
y
u
s
ef
u
l
in
m
o
v
ie
r
ec
o
m
m
en
d
atio
n
s
y
s
tem
s
,
p
e
r
s
o
n
alize
d
co
n
ten
t
f
ilter
in
g
,
a
n
d
e
-
c
o
m
m
er
ce
p
latf
o
r
m
s
,
w
h
er
e
r
ea
l
-
tim
e
an
d
h
ig
h
-
ac
cu
r
ac
y
p
r
ed
ictio
n
s
ar
e
cr
u
cial
f
o
r
u
s
er
en
g
a
g
em
en
t.
2.
M
E
T
H
O
D
2
.
1
.
P
ro
po
s
ed
s
y
s
t
em
a
rc
hit
ec
t
ure
T
h
e
s
y
s
tem
ar
ch
itectu
r
e
co
m
p
r
is
es
in
p
u
t
f
ile
t
o
v
a
r
io
u
s
m
o
d
els
an
d
th
en
ap
p
l
y
in
g
R
em
o
r
a
a
n
d
B
iG
R
U
f
o
r
r
ec
o
m
m
en
d
atio
n
.
As
s
h
o
wn
in
Fig
u
r
e
1,
th
e
s
y
s
tem
ar
ch
itectu
r
e
co
n
tain
s
in
p
u
t
f
ile,
t
h
e
m
atr
ix
r
ep
r
esen
tatio
n
of
in
p
u
t
f
ile,
co
-
o
cc
u
r
r
en
ce
m
atr
ix
r
e
p
r
e
s
en
tatio
n
,
co
n
s
tr
ain
m
o
d
el,
r
atin
g
in
d
e
p
en
d
e
n
t
m
o
d
el,
B
id
ir
ec
tio
n
al
GR
U,
r
e
m
o
r
a
o
p
tim
izatio
n
an
d
f
in
ally
th
e
r
ec
o
m
m
en
d
atio
n
s
.
Fig
u
r
e
1
.
Ar
c
h
itectu
r
e
o
f
p
r
o
p
o
s
ed
s
y
s
tem
[
1
6
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
n
o
ve
l a
p
p
r
o
a
c
h
fo
r
r
ec
o
mme
n
d
a
tio
n
u
s
in
g
o
p
timiz
ed
b
id
ir
ec
tio
n
a
l
…
(
P
r
a
ka
s
h
P
.
R
o
k
a
d
e
)
5021
2
.
2
.
B
idi
re
ct
io
na
l
g
a
t
ed
re
c
urre
nt
unit
B
id
ir
ec
tio
n
al
g
ated
r
ec
u
r
r
e
n
t
u
n
it
(
B
iGR
U)
,
a
more
r
ec
en
t
R
NN
v
er
s
io
n
th
an
L
STM
,
is
b
ec
o
m
in
g
in
cr
ea
s
in
g
ly
p
o
p
u
lar
t
h
ese
d
ay
s
.
B
ec
au
s
e
R
N
N
p
er
f
o
r
m
s
th
e
s
am
e
o
p
er
atio
n
at
ev
e
r
y
tim
e
p
o
i
n
t,
its
ca
lcu
latio
n
g
r
ap
h
is
ex
tr
a
o
r
d
i
n
ar
ily
d
ee
p
[
1
7
]
.
A
n
eu
r
al
n
et
wo
r
k
'
s
lo
n
g
-
a
n
d
s
h
o
r
t
ter
m
m
em
o
r
y
s
tr
ateg
y
is
s
u
g
g
ested
to
ad
d
r
ess
R
NN
p
r
o
b
lem
s
,
alth
o
u
g
h
it
h
as
a
m
o
r
e
co
m
p
lex
s
tr
u
ctu
r
e
an
d
s
tr
u
g
g
les
to
co
n
v
er
g
e
more
q
u
ick
ly
[
1
8
]
,
[
1
9
]
.
T
h
e
B
iG
R
U
o
u
tp
er
f
o
r
m
s
th
e
L
ST
M
in
ter
m
s
of
s
p
ee
d
.
T
h
e
n
et
wo
r
k
co
n
f
ig
u
r
atio
n
of
B
iGR
U
is
th
e
f
o
llo
win
g
:
th
er
e
ar
e
2
GR
U
u
n
its
,
2
d
en
s
e
lay
er
s
an
d
2
d
r
o
p
o
u
t
lay
er
s
(
i.e
.
to
tally
6
lay
er
s
)
.
I
n
s
id
e
each
GR
U
u
n
it
1
r
eset
an
d
1
u
p
d
ate
u
n
it
will
be
th
er
e
[
2
0
]
.
As
s
h
o
wn
in
Fig
u
r
e
2
,
GR
U1
is
a
f
o
r
war
d
GR
U,
an
d
Fig
u
r
e
2
d
is
p
lay
s
its
in
ter
n
al
f
ea
tu
r
es,
wh
ile
Fig
u
r
e
3
d
is
p
lay
s
th
e
in
ter
n
al
d
etails
o
f
GR
U2
,
a
r
ev
er
s
e
G
R
U.
T
h
e
f
o
r
war
d
ca
lcu
latio
n
s
h
o
wn
in
Fig
u
r
e
3
is
ca
r
r
ied
o
u
t a
s
f
o
llo
ws.
Su
p
p
o
s
e
at
tim
e
,
⃗
⃗
is
th
e
r
eset g
ate
o
f
th
e
p
o
s
itiv
e
in
p
u
t G
R
U
.
Her
e
i
s
th
e
f
o
r
m
u
la:
⃗
⃗
=
(
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
+
⃗
⃗
⃗
⃗
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
(
1
)
In
t
h
e
f
o
r
m
u
l
a
,
σ
is
th
e
s
ig
m
o
id
f
u
n
ctio
n
,
⃗
⃗
⃗
an
d
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
ar
e
th
e
v
alu
es
of
th
e
m
o
s
t
r
ec
en
t
ac
tiv
atio
n
an
d
th
e
c
u
r
r
e
n
t
in
p
u
t,
co
r
r
esp
o
n
d
in
g
l
y
.
is
th
e
in
p
u
t
weig
h
t
m
atr
ix
.
⃗
⃗
⃗
⃗
is
th
e
weig
h
t
m
a
tr
ix
f
o
r
c
y
clic
co
n
n
ec
tio
n
s
.
Similar
ly
,
s
u
p
p
o
s
e
⃗
⃗
⃗
is
th
e
u
p
d
ate
g
ate
of
th
e
f
o
r
war
d
GR
U
at
tim
e
;
th
e
f
o
r
m
u
la
is
as
:
⃗
⃗
⃗
=
(
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
+
⃗
⃗
⃗
⃗
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
(
2
)
Fig
u
r
e
2
.
B
iGR
U
wo
r
k
in
g
ar
c
h
itectu
r
e
Su
p
p
o
s
e
at
tim
e
,
ℎ
⃗
⃗
⃗
is
th
e
ac
tiv
atio
n
v
al
u
e
o
f
th
e
p
o
s
itiv
e
i
n
p
u
t
GR
U,
wh
ich
is
a
c
o
m
p
r
o
m
is
e
b
etwe
en
th
e
ca
n
d
id
ate
ac
tiv
atio
n
v
alu
e
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
an
d
th
e
last
ac
tiv
atio
n
v
alu
e
ℎ
−
⃗
⃗
⃗
⃗
⃗
⃗
.
ℎ
⃗
⃗
⃗
=
(
1
−
⃗
⃗
⃗
)·
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
+
⃗
⃗
⃗
.
ℎ
−
(
3
)
T
h
e
f
o
r
m
u
la
f
o
r
ℎ
−
is
as
(
4
)
.
ℎ
−
̃
=
ta
n
h
(
ℎ
⃗
⃗
⃗
⃗
⃗
⃗
⃗
.
⃗
⃗
⃗
+
.
⃗
⃗
⃗
⃗
⃗
ℎ
⃗
⃗
⃗
⃗
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
)
(
4
)
I
n
th
e
f
o
r
m
u
la
(
4
)
,
is
th
e
Had
a
m
ar
d
p
r
o
d
u
ct.
Fo
r
r
eset
g
ate,
if
⃗
⃗
⃗
is
clo
s
ed
m
ea
n
s
its
v
alu
e
ap
p
r
o
ac
h
es
0
,
th
e
GR
U
elim
in
ates
th
e
p
r
ev
io
u
s
ac
tiv
atio
n
v
alu
e
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
an
d
th
e
c
u
r
r
en
t
in
p
u
t
⃗
⃗
⃗
is
th
e
o
n
ly
f
ac
to
r
af
f
e
ctin
g
it
.
T
h
is
allo
ws
h
̃
t
to
d
en
y
ir
r
elev
an
t
in
f
o
r
m
atio
n
,
th
er
e
b
y
m
o
r
e
ef
f
ec
tiv
ely
c
o
m
m
u
n
icatin
g
p
er
tin
e
n
t
f
ac
ts
[
2
1
]
.
On
o
th
er
s
id
e,
th
e
u
p
d
ate
g
ate
⃗
⃗
⃗
co
n
tr
o
ls
how
m
u
ch
in
f
o
r
m
atio
n
in
ℎ
−
1
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
.
can
be
d
e
liv
er
ed
to
th
e
cu
r
r
e
n
t
ℎ
⃗
⃗
⃗
.
T
h
is
is
th
e
k
ey
to
d
esig
n
in
g
th
e
r
esu
lts
f
o
r
t
h
is
u
n
it.
It
f
u
n
ctio
n
s
as
a
m
em
o
r
y
u
n
it
a
k
in
to
an
L
ST
M,
aid
in
g
GR
U
in
r
em
em
b
er
in
g
l
o
n
g
-
te
r
m
d
ata
[
2
2
]
.
Similar
ly
,
f
o
r
m
u
la
(
5
)
–
(
8
)
p
r
o
v
id
e
th
e
co
m
p
u
tati
o
n
m
et
h
o
d
f
o
r
t
h
e
r
ev
er
s
e
GR
U
s
h
o
wn
in
Fig
u
r
e
3
an
d
Fig
u
r
e
4
.
⃐
⃗
⃗
⃗
=
(
⃐
⃗
⃗
⃗
⃗
⃗
⃐
⃗
⃗
⃗
+
⃐
⃗
⃗
⃗
⃗
ℎ
+
1
⃐
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
)
(
5
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
5
0
1
9
-
5
0
3
0
5022
⃐
⃗
⃗
⃗
⃗
=
(
⃐
⃗
⃗
⃗
⃗
⃗
⃐
⃗
⃗
⃗
+
⃐
⃗
⃗
⃗
⃗
ℎ
+
1
⃐
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
)
(
6
)
ℎ
⃐
⃗
⃗
⃗
⃗
=
(1
−
⃐
⃗
⃗
⃗
⃗
)·
ℎ
+
1
⃐
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
+
⃐
⃗
⃗
⃗
⃗
ℎ
+
1
⃐
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
.
(
7
)
ℎ
+
̃
=
ta
n
h
(
ℎ
⃐
⃗
⃗
⃗
⃗
⃗
⃐
⃗
⃗
⃗
+
⃐
⃗
⃗
⃗
.
ℎ
⃐
⃗
⃗
⃗
⃗
ℎ
+
1
⃐
⃗
⃗
⃗
⃗
⃗
⃗
⃗
⃗
)
(
8
)
T
h
e
r
esu
lts
of
two
d
ir
ec
tio
n
s
ar
e
av
er
a
g
e
to
o
b
tain
f
in
al
o
u
tp
u
t
ℎ
.
F
i
g
u
r
e
3
.
I
n
n
e
r
s
t
r
u
c
t
u
r
e
o
f
a
f
o
r
w
a
r
d
G
R
U
n
e
u
r
o
n
F
i
g
u
r
e
4
.
I
n
n
e
r
s
t
r
u
c
t
u
r
e
o
f
a
b
a
c
k
w
a
r
d
GR
U
n
e
u
r
o
n
2
.
3
.
Remo
ra
o
ptim
iza
t
io
n
a
lg
o
rit
hm
R
OA
is
an
o
p
tim
izatio
n
alg
o
r
ith
m
th
at
u
s
es
m
etah
eu
r
i
s
tics
an
d
is
in
s
p
ir
ed
by
th
e
f
o
r
ag
in
g
b
eh
av
io
r
of
r
em
o
r
a
s
p
ec
ies.
R
OA,
in
s
p
ir
ed
b
y
r
em
o
r
a
f
is
h
b
eh
a
v
io
r
,
d
y
n
am
ically
a
d
ju
s
ts
weig
h
t
v
alu
es
in
th
e
B
iG
R
U
n
etwo
r
k
,
r
ed
u
cin
g
tr
ain
in
g
er
r
o
r
s
an
d
ac
ce
l
er
atin
g
co
n
v
e
r
g
en
ce
.
I
t
en
s
u
r
es
o
p
tim
al
weig
h
t
s
elec
tio
n
to
m
ax
im
ize
th
e
ac
cu
r
ac
y
o
f
r
ec
o
m
m
en
d
atio
n
s
.
T
h
e
m
ain
in
ten
tio
n
of
u
s
in
g
R
OA
is
to
o
p
tim
ize
th
e
weig
h
t
v
alu
es
in
B
iGR
U
an
d
to
r
esu
lt
in
ac
cu
r
ate
o
u
tc
o
m
es
.
R
OA
p
lay
s
an
im
p
o
r
ta
n
t
r
o
le
to
o
p
tim
ize
in
p
u
t
weig
h
ts
if
th
e
o
u
tp
u
t
at
n
eu
r
o
n
ex
ce
ed
s
th
e
r
an
g
e
0
to
1
.
T
h
e
R
OA
o
p
tim
izatio
n
a
lg
o
r
ith
m
is
u
s
ed
to
f
in
e
-
tu
n
e
th
e
Bi
-
GR
U's
s
ettin
g
s
.
T
h
is
is
ac
co
m
p
lis
h
ed
by
r
ep
ea
ted
ly
l
o
o
k
in
g
f
o
r
th
e
id
ea
l
weig
h
t
v
alu
es
wh
ile
m
in
im
izin
g
in
ac
cu
r
ac
y
o
r
lo
s
s
.
T
h
er
ef
o
r
e
,
th
e
o
p
tim
iz
atio
n
alg
o
r
ith
m
'
s
f
itn
ess
f
u
n
ctio
n
is
th
e
r
ed
u
ctio
n
o
f
lo
s
s
/er
r
o
r
in
B
i
-
GR
U.
A
s
s
h
o
wn
in
Fig
u
r
e
5
,
th
e
m
em
o
r
y
o
p
tim
izatio
n
alg
o
r
ith
m
co
n
s
is
ts
o
f
th
e
f
o
llo
win
g
s
tep
s
[
2
3
]
,
[
2
4
]
,
[
2
5
]
.
2
.
3
.
1
.
F
lo
wcha
rt
of
re
m
o
r
a
R
em
o
r
a
o
p
tim
izatio
n
tech
n
iq
u
e
is
in
s
p
ir
ed
b
y
s
y
m
b
io
tic
r
elatio
n
s
h
ip
s
in
n
atu
r
e
o
f
r
e
m
o
r
as
an
d
s
h
ar
k
s
.
T
o
b
alan
ce
ex
p
lo
r
ati
o
n
a
n
d
ex
p
lo
itatio
n
I
t
co
m
b
in
es
g
lo
b
al
an
d
lo
ca
l
s
ea
r
ch
s
tr
ateg
ies.
Ag
en
ts
(
r
em
o
r
as)
f
o
llo
w
a
n
d
ad
ap
t
t
o
lead
er
s
(
s
h
ar
k
s
)
in
th
e
p
o
p
u
latio
n
to
f
in
d
o
p
tim
al
s
o
lu
tio
n
s
.
I
t
is
c
o
m
m
o
n
ly
u
s
ed
in
s
o
lv
in
g
c
o
m
p
lex
,
n
o
n
lin
ea
r
o
p
tim
izatio
n
p
r
o
b
lem
s
.
T
h
e
m
eth
o
d
is
lig
h
tweig
h
t,
c
o
n
v
er
g
es
f
ast,
an
d
s
u
its
r
ea
l
-
wo
r
ld
en
g
in
ee
r
in
g
a
p
p
licatio
n
s
.
R
em
o
r
a
o
p
tim
izatio
n
alg
o
r
ith
m
f
lo
wch
a
r
t is d
esig
n
ed
as f
o
ll
o
ws.
a.
C
r
ea
te
th
e
f
ir
s
t p
o
p
u
latio
n
I
n
th
is
ca
s
e,
th
e
s
ea
r
ch
a
g
en
t'
s
p
ar
am
eter
is
p
o
p
u
latio
n
.
W
e
m
u
s
t
in
itialize
th
e
n
u
m
b
er
o
f
r
em
o
r
a,
o
r
s
ea
r
ch
ag
en
ts
,
in
o
u
r
s
y
s
tem
.
T
h
e
s
ea
r
ch
ag
e
n
t is r
em
o
r
a.
b.
Def
in
e
n
etwo
r
k
weig
h
t
T
h
e
weig
h
t
v
alu
es
o
f
n
eu
r
al
n
etwo
r
k
s
will
b
e
d
ef
in
e
d
h
er
e.
T
h
ese
weig
h
t
v
alu
es
ar
e
n
ea
r
l
y
ze
r
o
an
d
ar
e
cr
ea
ted
at
r
a
n
d
o
m
.
c.
Mo
d
if
icatio
n
o
f
s
ea
r
ch
ag
e
n
ts
W
e
f
ir
s
t d
ef
in
e
a
s
o
lu
tio
n
s
p
ac
e
in
all
o
p
tim
izatio
n
tech
n
iq
u
e
s
.
W
e
al
s
o
s
p
ec
if
y
th
e
n
u
m
b
er
o
f
s
ea
r
ch
ag
en
ts
an
d
th
e
s
ea
r
ch
ar
ea
.
W
e
ch
an
g
e
th
e
v
alu
e
o
f
s
ea
r
ch
ag
en
ts
if
th
eir
n
u
m
b
er
in
a
s
ea
r
ch
s
p
ac
e
s
u
r
p
ass
es a
th
r
esh
o
ld
.
in
a
s
ea
r
ch
s
p
ac
e,
o
n
l
y
s
ea
r
ch
ag
e
n
ts
b
elo
w
th
e
th
r
esh
o
ld
will b
e
p
er
m
itted
en
tr
y
.
d.
E
r
r
o
r
r
ed
u
ctio
n
In
th
is
ca
s
e,
th
e
m
is
tak
e
is
r
ed
u
ce
d
by
th
e
er
r
o
r
or
lo
s
s
f
u
n
cti
o
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
n
o
ve
l a
p
p
r
o
a
c
h
fo
r
r
ec
o
mme
n
d
a
tio
n
u
s
in
g
o
p
timiz
ed
b
id
ir
ec
tio
n
a
l
…
(
P
r
a
ka
s
h
P
.
R
o
k
a
d
e
)
5023
e.
T
h
e
cu
r
r
e
n
t sear
ch
ag
e
n
t'
s
p
o
s
itio
n
an
d
th
e
s
to
r
e'
s
f
itn
ess
Sev
er
al
n
etwo
r
k
weig
h
ts
ar
e
av
ailab
le
in
th
e
s
ea
r
ch
s
p
ac
e.
W
e
s
h
all
m
ain
tain
a
r
ec
o
r
d
of
th
e
o
p
tim
al
n
etwo
r
k
weig
h
t.
f.
Dete
r
m
in
e
th
e
n
ea
r
est o
p
tim
al
weig
h
t
W
e
ca
n
g
et
th
e
n
ex
t
weig
h
t
v
alu
e,
o
r
o
p
tim
al
weig
h
t,
b
y
co
m
p
u
tin
g
th
e
f
itn
ess
v
alu
e
o
f
e
ac
h
weig
h
t
v
alu
e
we
i
n
itially
ch
o
s
e.
T
h
e
f
ir
s
t
weig
h
t
v
alu
e
is
k
n
o
wn
.
E
v
e
r
y
iter
atio
n
'
s
f
itn
ess
f
u
n
ctio
n
is
b
ein
g
ev
alu
ated
.
g.
An
aly
ze
f
itn
ess
(
er
r
o
r
m
in
im
i
za
tio
n
)
Fo
r
ev
er
y
weig
h
t
in
t
h
e
n
etwo
r
k
,
we
f
in
d
its
f
itn
ess
v
alu
e.
A
co
m
p
ar
is
o
n
of
th
e
n
ew
an
d
o
ld
weig
h
t
v
alu
es
is
p
r
esen
ted
.
h.
E
r
r
o
r
m
in
im
izatio
n
Her
e,
th
e
m
is
tak
e
is
m
in
im
ize
d
b
y
u
s
in
g
th
e
er
r
o
r
f
u
n
ctio
n
o
r
lo
s
s
f
u
n
ctio
n
.
i.
Sto
r
e
f
itn
ess
an
d
p
o
s
itio
n
o
f
c
u
r
r
en
t sear
ch
ag
en
t:
T
h
er
e
ar
e
s
ev
er
al
n
etwo
r
k
weig
h
ts
in
th
e
s
ea
r
ch
s
p
ac
e.
T
h
e
i
d
ea
l n
etwo
r
k
weig
h
t w
ill b
e
k
ep
t o
n
f
ile.
j.
Fin
d
n
ea
r
o
p
tim
al
weig
h
t
T
h
e
f
ir
s
t
weig
h
t
v
alu
e
is
k
n
o
wn
,
an
d
b
y
ca
lcu
latin
g
th
e
f
itn
ess
v
alu
e
o
f
ea
ch
weig
h
t
v
alu
e
we
in
itially
s
elec
ted
,
we
ca
n
g
et
th
e
n
ex
t
weig
h
t
v
alu
e,
o
r
o
p
tim
al
weig
h
t.
W
e
ar
e
ass
ess
in
g
th
e
f
itn
ess
f
u
n
ctio
n
f
o
r
ea
c
h
iter
atio
n
.
k.
E
v
alu
ate
f
itn
ess
(
er
r
o
r
m
in
im
i
za
tio
n
)
E
v
er
y
n
etwo
r
k
weig
h
t
is
ass
e
s
s
ed
f
o
r
its
f
itn
es
s
v
alu
e.
T
h
e
n
ew
weig
h
t
v
alu
e
an
d
th
e
o
ld
weig
h
t
v
alu
e
ar
e
b
ei
n
g
co
m
p
ar
e
d
.
l.
Sto
p
p
in
g
cr
iter
ia
First,
we
wil
l
d
ef
in
e
th
e
n
u
m
b
er
o
f
iter
atio
n
s
th
at
will
b
e
u
s
ed
as
o
u
r
cu
to
f
f
p
o
in
t.
T
h
e
g
o
al
is
to
in
cr
ea
s
e
th
e
co
r
r
ec
tn
ess
o
f
th
e
n
etwo
r
k
m
o
d
el
b
y
ch
o
o
s
in
g
th
e
weig
h
t
v
alu
es
f
o
r
ea
ch
iter
atio
n
o
f
th
e
r
em
o
r
a
alg
o
r
ith
m
in
an
o
p
tim
a
l m
an
n
er
.
Fig
u
r
e
5
.
R
em
o
r
a
o
p
tim
izatio
n
alg
o
r
ith
m
f
lo
wch
ar
t
[
2
6
]
2
.
4
.
I
np
ut
f
ile
up
lo
a
din
g
Mo
v
ieL
en
s
d
ataset
of
1
5
0
.
3
5
k
B
is
tak
en
as
an
in
p
u
t
f
r
o
m
K
a
g
g
le.
co
m
.
T
h
e
d
ataset
co
m
p
r
is
es
1
0
0
,
0
0
0
r
atin
g
s
ca
p
tu
r
e
d
f
r
o
m
671
u
s
er
s
f
o
r
a
to
tal
of
1
,
6
8
2
m
o
v
ies.
T
h
e
in
p
u
t
d
ata
c
o
n
tain
s
th
e
f
ea
tu
r
es
u
s
er
id
,
m
o
v
ieid
,
r
atin
g
s
,
I
n
p
u
t
f
ile
co
n
tain
s
1
0
0
,
0
0
0
e
n
tr
ies.
I
t
co
m
p
r
is
es
th
e
m
o
v
ie
r
atin
g
s
alo
n
g
with
th
e
titl
e,
g
en
r
e,
tim
e
s
tam
p
f
o
r
m
o
v
ies.
Ab
o
u
t
6
7
1
u
s
er
s
h
av
e
r
ated
with
d
i
f
f
er
en
t
r
atin
g
s
b
etwe
en
1
to
5
.
T
h
e
S
t
a
r
t
G
e
ne
r
a
t
e
i
ni
t
i
a
l
popul
a
t
i
on
D
e
f
i
ne
ne
t
w
or
k w
e
i
ght
s
C
he
c
k i
f
s
e
a
r
c
h a
ge
nt
goe
s
beyond t
he
s
e
a
r
c
h
s
pa
c
e
a
nd a
m
e
nd i
t
E
va
l
ua
t
e
f
i
t
ne
s
s
(
e
r
r
or
m
i
ni
m
i
z
a
t
i
on
)
S
t
or
e
f
i
t
ne
s
s
a
nd p
os
i
t
i
on
of
c
ur
r
e
nt
s
e
a
r
c
h a
ge
nt
s
C
ur
r
e
nt
i
t
e
r
a
t
i
on
=
c
ur
r
e
nt
i
t
e
r
a
t
i
on
+
1
L
oc
a
l
up
da
t
e
S
t
opp
i
ng
c
r
i
t
e
r
i
a
?
E
nd
F
i
nd ne
a
r
opt
i
m
a
l
w
e
i
ght
E
va
l
ua
t
e
f
i
t
ne
s
s
(
e
r
r
or
m
i
ni
m
i
z
a
t
i
on
)
I
f
a
t
t
e
m
pt
<
c
ur
r
e
nt
P
os
i
t
i
on u
pda
t
e
Y
e
s
No
Y
e
s
No
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
5
0
1
9
-
5
0
3
0
5024
o
v
er
all
s
ize
o
f
th
e
d
ataset
is
1
5
0
.
3
5
k
B
.
T
h
e
in
p
u
t
d
ata
c
o
n
ta
in
s
t
h
e
f
e
at
u
r
es
u
s
e
r
i
d
,
m
o
v
ie
id
,
r
ati
n
g
s
,
ti
tle
,
g
e
n
r
e
,
t
im
esta
m
p
,
t
ag
,
im
d
b
i
d
,
tm
d
b
i
d
.
O
n
l
y
f
o
u
r
f
e
at
u
r
es
u
s
er
i
d
,
m
o
v
iei
d
,
m
o
v
ie
n
a
m
e
,
r
at
in
g
s
ar
e
c
o
n
s
id
er
e
d
in
p
r
o
p
o
s
ed
m
o
d
e
l
as
a
n
i
n
p
u
t.
As
s
h
o
wn
in
T
ab
le
1
,
6
7
1
u
s
er
s
h
av
e
r
ated
1
,
6
8
2
m
o
v
ies
with
r
atin
g
s
in
b
etwe
en
1
to
5
.
I
t
is
n
o
t
th
e
ca
s
e
th
at
all
u
s
er
s
h
av
e
r
ated
a
ll
th
e
m
o
v
ies.
So
m
e
u
s
er
s
h
av
e
n
o
t r
ated
f
ew
m
o
v
ies.
E
ith
er
th
ey
h
av
e
n
o
t
s
ee
n
th
o
s
e
m
o
v
ies
o
r
th
e
y
ar
e
n
o
t
in
ter
ested
in
p
r
o
v
id
in
g
th
e
r
at
in
g
s
to
th
o
s
e
m
o
v
ies.
C
o
n
s
id
er
u
s
er
1
h
as
r
ated
m
o
v
ie
n
u
m
b
er
1
3
4
3
with
r
atin
g
s
2
an
d
t
h
is
m
o
v
ie
b
el
o
n
g
s
to
th
e
g
en
r
e
ac
tio
n
/ad
v
en
tu
r
e/th
r
iller
.
T
ab
le
1
.
I
n
p
u
t
f
ile
U
seri
d
M
o
v
i
e
i
d
R
a
t
i
n
g
Ti
mes
t
a
m
p
Ti
t
l
e
G
e
n
r
e
1
31
2
.
5
1
.
2
6
E+
0
9
To
y
S
t
o
r
y
A
d
v
e
n
t
u
r
e
|
A
n
i
ma
t
i
o
n
1
1
0
2
9
3
1
.
2
6
E+
0
9
Ju
ma
n
j
i
A
d
v
e
n
t
u
r
e
1
1
0
6
1
3
1
.
2
6
E+
0
9
G
r
u
mp
i
e
r
O
l
d
M
e
n
C
o
m
e
d
y
|
R
o
ma
n
c
e
1
1
1
2
9
2
1
.
2
6
E+
0
9
W
a
i
t
i
n
g
t
o
E
x
h
a
l
e
C
o
m
e
d
y
|
D
r
a
ma
|
R
o
ma
n
c
e
1
1
1
7
2
4
1
.
2
6
E+
0
9
F
a
t
h
e
r
o
f
B
r
i
d
e
-
II
C
o
m
e
d
y
1
1
2
6
3
2
1
.
2
6
E+
0
9
H
e
a
t
A
c
t
i
o
n
|
C
r
i
me
|
Th
r
i
l
l
e
r
1
1
2
8
7
2
1
.
2
6
E+
0
9
S
a
b
r
i
n
a
C
o
m
e
d
y
|
R
o
ma
n
c
e
1
1
2
9
3
2
1
.
2
6
E+
0
9
To
m
a
n
d
H
u
c
k
A
d
v
e
n
t
u
r
e
C
h
i
l
d
r
e
n
1
1
3
3
9
3
.
5
1
.
2
6
E+
0
9
S
u
d
d
e
n
D
e
a
t
h
A
c
t
i
o
n
.
.
.
.
.
.
6
7
1
5
6
6
9
4
1
.
0
6
E+
0
9
D
r
a
c
u
l
a
C
o
m
e
d
y
H
o
r
r
o
r
6
7
1
5
8
1
6
4
1
.
0
7
E+
0
9
B
a
l
t
o
A
d
v
e
n
t
u
r
e
A
n
i
ma
t
i
o
n
6
7
1
5
9
0
2
3
.
5
1
.
0
6
E+
0
9
N
i
x
o
n
D
r
a
ma
2
.
5
.
Reduct
io
n o
f
s
pa
rsity
is
s
ue
Dim
en
s
io
n
ality
r
ed
u
ctio
n
is
one
of
th
e
g
o
o
d
s
o
lu
tio
n
s
f
o
r
s
p
ar
s
ity
r
ed
u
ctio
n
.
In
t
h
is
s
tep
,
th
e
s
p
ar
s
ity
is
s
u
e
in
th
e
in
p
u
t
f
ile
is
r
eso
l
v
ed
by
c
o
n
s
id
er
in
g
o
n
ly
th
e
p
ar
a
m
eter
s
u
s
er
id
,
m
o
v
ieid
,
r
atin
g
s
.
By
co
n
s
id
er
in
g
o
n
ly
t
h
ese
lim
ite
d
p
a
r
am
eter
s
,
t
h
e
c
o
m
p
lex
c
alcu
latio
n
s
ar
e
m
in
im
ized
an
d
s
p
ar
s
ity
is
s
u
e
is
s
u
p
p
r
ess
ed
.
As
s
h
o
w
n
in
T
ab
le
2
,
r
atin
g
s
f
o
r
m
o
v
ies
by
d
if
f
er
e
n
t
u
s
er
s
ar
e
s
h
o
w
n
in
m
atr
ix
f
o
r
m
at
s
h
o
win
g
th
at
u
s
er
r
ati
n
g
s
f
o
r
m
o
v
ie
r
ate
d
.
T
h
er
e
is
no
r
atin
g
p
r
o
v
id
e
d
f
o
r
u
n
r
ate
d
m
o
v
ie
s
.
User
id
1
2
h
as
p
r
o
v
id
e
d
r
atin
g
s
2,
2,
4
.
5
,
4,
3,
2
.
5
f
o
r
th
e
m
o
v
ies
1,
2,
5,
6,
1
0
,
11
r
e
s
p
ec
tiv
ely
.
T
h
er
e
is
em
p
ty
s
p
ac
e
in
th
e
m
atr
ix
by
th
e
u
s
er
id
12
f
o
r
th
e
m
o
v
ies
3,
4,
7,
8,
9,
12
as
it
h
as
n
o
t
r
ated
th
o
s
e
m
o
v
ies.
T
h
is
e
x
p
er
im
e
n
t
is
ev
alu
ated
f
o
r
all
u
s
er
s
an
d
m
o
v
ies
b
u
t
o
n
ly
f
o
r
12
u
s
er
s
an
d
m
o
v
ies
m
atr
ix
is
d
esig
n
e
d
.
T
ab
le
2
.
Ma
tr
ix
r
ep
r
esen
tatio
n
o
f
r
atin
g
s
f
o
r
d
if
f
er
e
n
t m
o
v
ie
s
b
y
u
s
er
s
U
seri
d
/
M
o
v
i
e
i
d
1
2
3
4
5
6
7
8
9
10
11
12
1
2
4
3
4
4
5
6
7
3
8
9
4
10
11
12
2
2
4
.
5
4
3
2
.
5
2
.
6
.
Co
-
o
cc
urre
nce
mo
del
T
h
er
e
ar
e
f
ew
m
o
v
ies
wh
ich
ar
e
eith
er
r
ated
b
y
a
s
in
g
le
u
s
er
,
s
o
m
e
m
o
v
ies
ar
e
r
ated
b
y
all
u
s
er
s
.
Co
-
o
cc
u
r
r
e
n
ce
m
o
d
el
is
o
b
tain
ed
by
i
n
ter
s
ec
tio
n
o
p
er
ato
r
.
It
s
h
o
ws
n
u
m
b
er
of
s
im
ilar
m
o
v
ies
r
ated
by
tw
o
u
s
er
s
.
A
co
-
o
cc
u
r
r
en
ce
m
atr
i
x
is
g
e
n
er
ated
to
ca
p
tu
r
e
r
el
atio
n
s
h
ip
s
o
r
p
atter
n
s
b
etwe
en
item
s
(
o
r
u
s
er
s
)
,
wh
ich
h
elp
s
in
u
n
d
er
s
tan
d
i
n
g
ass
o
ciatio
n
s
.
T
ab
le
3
s
h
o
ws
th
e
co
-
o
cc
u
r
r
e
n
ce
m
o
d
el
in
wh
ich
if
two
u
s
er
s
h
av
e
r
ated
s
am
e
m
o
v
ies
th
en
th
e
y
ar
e
in
clu
d
ed
i
n
to
t
h
e
m
atr
i
x
.
Ou
t
of
6
7
1
u
s
er
s
f
o
r
f
ir
s
t
12
u
s
er
s
o
n
ly
th
e
m
atr
ix
is
p
r
e
p
ar
ed
.
Her
e
u
s
er
7
h
as
r
ated
5,
9,
11,
41,
10
s
a
m
e
m
o
v
ies
with
u
s
er
1,
2,
3,
4,
5,
6
a
n
d
so
on.
T
h
is
s
h
o
ws
co
-
o
cc
u
r
r
en
ce
m
o
d
el
of
s
im
ilar
m
o
v
ies
r
at
ed
by
two
u
s
er
s
.
Co
-
o
cc
u
r
r
e
n
ce
m
o
d
el
is
t
h
e
p
ar
t
of
in
p
u
t
to
th
e
B
iGR
U
alg
o
r
ith
m
f
o
r
th
r
o
u
g
h
ca
lcu
lat
io
n
f
o
r
p
r
ed
ictin
g
m
o
v
ies
to
t
h
e
u
s
er
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
n
o
ve
l a
p
p
r
o
a
c
h
fo
r
r
ec
o
mme
n
d
a
tio
n
u
s
in
g
o
p
timiz
ed
b
id
ir
ec
tio
n
a
l
…
(
P
r
a
ka
s
h
P
.
R
o
k
a
d
e
)
5025
T
ab
le
3
.
C
o
-
o
cc
u
r
r
e
n
ce
m
o
d
el
o
f
s
im
ilar
m
o
v
ies r
ated
b
y
tw
o
u
s
er
s
U
seri
d
/
U
seri
d
1
2
3
4
5
6
7
8
9
10
11
12
1
20
76
0
5
1
0
5
18
27
18
10
1
2
76
8
8
9
0
17
12
6
3
6
2
4
3
0
8
51
7
12
3
11
5
12
0
0
53
4
5
9
7
2
0
4
19
7
41
4
2
0
2
4
3
1
2
0
5
1
0
12
19
1
0
0
4
10
4
7
0
1
3
3
0
6
0
17
3
7
4
44
0
16
0
60
54
32
7
5
9
11
41
10
0
88
80
21
2
0
91
8
18
6
5
4
4
16
80
12
6
3
6
2
9
27
2
12
2
7
0
21
6
0
1
0
1
33
15
10
18
2
0
0
0
60
2
3
1
0
1
0
33
72
11
10
1
0
2
4
3
1
3
3
54
0
6
7
33
2
77
12
1
8
53
7
0
32
91
2
15
72
77
12
2
.
7
.
Co
ns
t
ra
int
m
o
del
User
s
wh
o
h
av
e
g
iv
en
s
im
ilar
m
o
v
ies
th
e
s
am
e
r
atin
g
s
ar
e
tak
en
in
t
o
co
n
s
id
er
atio
n
to
cr
ea
te
th
is
m
o
d
el.
T
h
is
m
o
d
el
is
th
e
r
estricte
d
v
er
s
io
n
of
co
-
o
cc
u
r
r
e
n
ce
m
o
d
el.
Fro
m
th
e
in
p
u
t
m
a
tr
ix
r
ep
r
esen
tatio
n
,
it
is
clea
r
th
at
th
er
e
ar
e
s
o
m
e
u
s
er
s
wh
o
h
a
v
e
r
ated
s
im
ilar
m
o
v
ies
by
s
im
ilar
r
atin
g
s
with
th
e
o
th
er
u
s
er
s
an
d
f
ew
u
s
er
s
ar
e
th
er
e
wh
o
h
av
e
r
ate
d
s
im
ilar
m
o
v
ies
by
d
if
f
er
en
t
r
atin
g
s
with
th
e
o
th
e
r
u
s
er
s
.
A
s
s
h
o
w
n
i
n
T
a
b
l
e
4
,
u
s
e
r
7
h
a
s
r
a
t
e
d
2
,
3
,
5
,
2
1
,
5
,
0
s
i
m
i
l
a
r
m
o
v
i
e
s
b
y
t
h
e
s
a
m
e
r
a
t
i
n
g
s
a
s
u
s
e
r
s
1
,
2
,
3
,
4
,
5
,
6
,
r
e
s
p
e
c
t
i
v
el
y
.
T
h
e
m
o
d
e
l
is
o
b
t
a
i
n
e
d
f
o
r
al
l
u
s
e
r
s
,
b
u
t
f
o
r
t
h
e
f
i
r
s
t
1
2
u
s
e
r
s
,
T
a
b
l
e
4
i
s
p
r
e
p
a
r
e
d
.
T
h
is
m
o
d
e
l
c
a
n
b
e
t
e
r
m
e
d
a
s
a
d
e
p
e
n
d
e
n
t
m
o
d
e
l
,
a
s
s
i
m
il
a
r
m
o
v
i
e
s
w
i
t
h
t
h
e
s
a
m
e
r
a
t
i
n
g
s
b
y
u
s
e
r
s
a
r
e
c
o
n
s
i
d
e
r
e
d
h
e
r
e
.
T
ab
le
4.
C
o
n
s
tr
ain
t
m
o
d
el
of
s
im
ilar
m
o
v
ies
with
s
am
e
r
atin
g
s
by
two
u
s
er
s
U
seri
d
/
U
seri
d
1
2
3
4
5
6
7
8
9
10
11
12
1
12
8
0
2
0
0
3
8
20
6
8
1
2
26
2
6
1
0
5
9
4
1
3
1
2
3
0
4
48
3
2
1
6
3
6
0
0
27
4
3
4
2
1
0
0
9
3
20
2
1
0
1
0
0
60
5
0
0
2
9
50
2
5
2
3
0
33
0
6
0
5
1
3
2
20
0
8
0
30
20
14
7
3
9
6
20
5
0
8
20
7
1
0
17
8
8
4
3
2
2
8
20
9
1
0
3
0
9
20
1
6
1
3
0
7
1
0
61
17
8
10
6
3
0
0
0
30
1
0
61
0
10
29
11
8
1
0
1
0
0
33
20
0
3
17
10
0
30
12
1
2
27
60
0
14
17
0
8
29
30
5
Th
i
s
m
o
d
u
l
e
a
p
p
l
i
e
s
c
e
r
t
a
i
n
p
r
e
d
e
f
i
n
e
d
c
o
n
st
r
a
i
n
t
s (
e
.
g
.
,
u
s
e
r
b
e
h
a
v
i
o
r
r
u
l
e
s
,
d
i
v
e
r
si
t
y
c
o
n
s
t
r
a
i
n
t
s,
e
t
c
.
)
t
o
g
u
i
d
e
t
h
e
l
e
a
r
n
i
n
g
o
r
f
i
l
t
e
r
i
n
g
p
r
o
c
e
ss.
2
.
8
.
Ra
t
ing
ind
epende
nt
m
o
del
T
h
is
ap
p
r
o
ac
h
tak
es
in
to
co
n
s
id
er
atio
n
v
iewe
r
s
wh
o
h
av
e
g
iv
en
s
im
ilar
m
o
v
ies
v
ar
ied
r
atin
g
s
.
As
s
h
o
wn
in
T
a
b
le
5,
u
s
er
7
h
as
r
ated
3,
9,
6,
2
0
,
5,
0
s
im
ilar
m
o
v
ies
by
d
if
f
er
en
t
r
atin
g
s
with
th
e
u
s
er
s
1,
2,
3,
4,
5,
6
r
esp
ec
tiv
ely
.
T
h
is
m
o
d
el
is
ev
alu
ated
f
o
r
all
671
u
s
er
s
out
of
wh
ich
f
o
r
f
ir
s
t
12
u
s
er
’
s
v
al
u
es
ar
e
s
h
o
wn
in
T
ab
le
5.
T
ab
le
5
.
R
atin
g
in
d
ep
en
d
en
t
m
o
d
el
o
f
s
im
ilar
m
o
v
ies with
d
if
f
er
en
t
r
atin
g
s
b
y
two
u
s
er
s
U
seri
d
/
U
seri
d
1
2
3
4
5
6
7
8
9
10
11
12
1
8
68
0
3
1
0
2
10
7
12
2
0
2
50
6
2
8
0
12
3
2
2
3
1
2
3
0
4
3
4
10
2
5
2
6
0
0
26
4
2
5
5
1
0
4
10
4
21
2
1
0
1
4
3
60
5
1
0
10
10
50
2
5
2
4
0
1
0
0
0
6
0
12
2
4
2
24
0
8
0
30
34
18
7
2
3
5
21
5
0
80
60
14
1
0
77
8
10
2
2
2
2
8
60
3
5
3
3
2
9
7
7
6
1
4
0
14
5
0
40
16
7
10
12
3
0
0
0
30
1
3
40
0
23
43
11
2
1
0
1
4
3
1
0
0
34
0
3
16
23
2
47
12
0
2
26
60
0
18
77
2
7
43
47
7
A
mo
d
e
l
t
h
a
t
o
p
e
r
a
t
e
s w
i
t
h
o
u
t
d
i
r
e
c
t
l
y
d
e
p
e
n
d
i
n
g
o
n
u
ser
-
p
r
o
v
i
d
e
d
r
a
t
i
n
g
s
—
p
o
ssi
b
l
y
u
si
n
g
i
m
p
l
i
c
i
t
f
e
e
d
b
a
c
k
,
c
o
n
t
e
n
t
si
mi
l
a
r
i
t
y
,
o
r
c
o
n
t
e
x
t
u
a
l
si
g
n
a
l
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
5
0
1
9
-
5
0
3
0
5026
3.
RE
SU
L
T
AN
D
DIS
CUSS
I
O
N
3
.
1
.
T
esting
f
o
r
re
c
o
mm
enda
t
io
n
Co
-
o
cc
u
r
r
e
n
ce
m
o
d
el,
c
o
n
s
tr
ain
t
m
o
d
el,
r
atin
g
in
d
ep
en
d
en
t
m
o
d
el
ar
e
th
e
in
p
u
ts
t
o
B
iGR
U
alg
o
r
ith
m
.
B
y
c
o
n
s
id
er
in
g
r
an
d
o
m
in
p
u
t
m
u
ltip
lied
with
weig
h
ts
f
o
llo
wed
b
y
ad
d
itio
n
o
f
b
ias
v
alu
e
at
ea
c
h
n
eu
r
o
n
in
h
id
d
e
n
lay
e
r
,
th
e
o
u
tp
u
t
is
ca
lcu
lated
.
I
f
th
e
o
u
t
p
u
t
at
th
ese
n
eu
r
o
n
s
is
b
etwe
en
th
e
r
an
g
es
0
to
1
,
th
ese
weig
h
ts
will
b
e
f
o
r
war
d
ed
to
n
ex
t
lay
e
r
n
eu
r
o
n
s
.
I
f
th
e
o
u
tp
u
t
at
h
id
d
en
lay
er
n
e
u
r
o
n
is
b
ey
o
n
d
th
e
r
an
g
e
0
to
1
,
r
em
o
r
a
o
p
tim
iza
tio
n
alg
o
r
ith
m
will
b
e
ac
tiv
e
to
o
p
tim
ize
th
ese
weig
h
ts
.
Fin
ally
,
b
y
test
in
g
th
e
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
a
n
y
r
a
n
d
o
m
u
s
er
,
th
e
r
ec
o
m
m
en
d
ed
m
o
v
ie
lis
t
will
b
e
s
u
g
g
ested
to
th
e
u
s
er
.
T
h
e
o
u
tp
u
t
as r
ec
o
m
m
en
d
e
d
m
o
v
ies
ar
e
s
h
o
wn
in
Fig
u
r
e
6
.
Fig
u
r
e
6
s
h
o
ws
th
at
wh
en
we
s
elec
t
an
y
u
s
er
f
o
r
r
ec
o
m
m
en
d
in
g
h
im
s
o
m
e
m
o
v
ies,
th
e
in
ter
est
p
atter
n
o
f
u
s
er
is
alr
ea
d
y
s
tu
d
i
ed
an
d
b
ased
o
n
th
at
p
atter
h
e
will b
e
r
ec
o
m
m
en
d
e
d
s
o
m
e
m
o
v
ies.
Her
e
we
ca
n
p
r
o
v
id
e
th
e
th
r
esh
o
ld
f
o
r
g
etti
n
g
f
ir
s
t to
p
x
n
u
m
b
er
o
f
m
o
v
ie
s
as r
esu
lt.
Fig
u
r
e
6
.
R
ec
o
m
m
e
n
d
atio
n
o
f
m
o
v
ies f
o
r
r
an
d
o
m
u
s
er
3
.
2
.
Resul
t
a
na
ly
s
is
o
f
pro
po
s
ed
s
y
s
t
em
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
s
u
g
g
e
s
ted
s
y
s
tem
is
ass
es
s
ed
u
s
in
g
test
in
g
m
ea
s
u
r
es
lik
e
p
r
ec
is
io
n
,
r
ec
all,
f
m
ea
s
u
r
e
,
ac
cu
r
ac
y
,
m
ea
n
ab
s
o
lu
te
er
r
o
r
,
an
d
r
o
o
t
m
ea
n
s
q
u
ar
e
er
r
o
r
.
T
h
ese
v
alu
es
ar
e
ascer
tain
ed
b
y
o
b
tain
in
g
a
co
n
f
u
s
io
n
m
atr
ix
.
3
.
2
.
1
.
P
ro
po
s
ed
s
y
s
t
em
co
m
pa
riso
n wit
h e
x
is
t
ing
s
y
s
t
em
T
o
co
m
p
ar
e
t
h
e
p
er
f
o
r
m
an
ce
o
f
p
r
o
p
o
s
ed
m
o
d
el
f
o
r
th
e
in
p
u
t
d
ataset,
th
e
p
er
f
o
r
m
an
ce
cr
iter
ia
ar
e
ap
p
lied
to
MM
DL
,
GA
an
d
FF
NN
f
o
r
th
e
s
am
e
in
p
u
t
an
d
r
esu
lts
ar
e
o
b
s
er
v
ed
.
As
s
h
o
wn
in
Fig
u
r
e
7
,
p
r
o
p
o
s
ed
o
p
tim
ized
B
iGR
U
m
o
d
el
f
o
r
m
o
v
ie
r
ec
o
m
m
en
d
a
tio
n
h
as
th
e
9
8
%
p
r
ec
is
io
n
wh
ich
is
th
e
h
ig
h
est
as
co
m
p
ar
e
to
MM
DL
with
9
2
%,
GA
with
9
6
% a
n
d
FF
NN
with
8
5
%.
As
s
h
o
wn
in
Fig
u
r
e
8
,
p
r
o
p
o
s
ed
o
p
tim
ized
B
iGR
U
m
o
d
el
f
o
r
m
o
v
ie
r
ec
o
m
m
en
d
atio
n
h
as
th
e
9
7
.
5
%
r
ec
all
wh
ich
is
th
e
h
ig
h
est
as
co
m
p
ar
e
to
MM
DL
with
9
6
%,
GA
with
9
7
%
an
d
FF
N
N
wit
h
8
8
%.As
s
h
o
wn
in
Fig
u
r
e
9
,
p
r
o
p
o
s
ed
o
p
tim
ized
B
iG
R
U
m
o
d
el
f
o
r
m
o
v
ie
r
ec
o
m
m
en
d
atio
n
h
as
th
e
9
7
%
f
-
m
ea
s
u
r
e
wh
ich
is
th
e
h
ig
h
est
as
co
m
p
ar
e
to
MM
D
L
with
9
5
%,
GA
with
9
6
%
an
d
FF
NN
with
8
7
%.As
s
h
o
wn
in
Fig
u
r
e
1
0
,
p
r
o
p
o
s
ed
o
p
tim
ized
B
iGR
U
m
o
d
el
f
o
r
m
o
v
ie
r
ec
o
m
m
e
n
d
a
tio
n
h
as
th
e
9
8
.
5
%
ac
cu
r
ac
y
wh
ich
is
th
e
h
ig
h
est
as
co
m
p
ar
e
to
MM
DL
with
9
5
%,
GA
with
9
6
%
an
d
FF
NN
with
8
6
%.As
s
h
o
wn
in
Fig
u
r
e
1
1
,
p
r
o
p
o
s
ed
o
p
tim
ized
B
iGR
U
m
o
d
el
f
o
r
m
o
v
ie
r
ec
o
m
m
en
d
atio
n
h
as
t
h
e
0
.
0
3
MA
E
wh
ich
is
th
e
lo
west
as
co
m
p
ar
e
t
o
MM
DL
with
0
.
6
5
,
GA
with
0
.
0
5
an
d
FF
NN
with
0
.
1
4
.
As
s
h
o
wn
in
Fig
u
r
e
1
2
,
p
r
o
p
o
s
ed
o
p
tim
ized
B
iGR
U
m
o
d
el
f
o
r
m
o
v
ie
r
ec
o
m
m
en
d
a
tio
n
h
as
th
e
0
.
1
7
R
MSE
wh
ich
is
th
e
lo
west
a
s
co
m
p
ar
e
to
MM
DL
with
0
.
2
5
,
GA
with
0
.
2
2
an
d
FF
NN
with
0
.
3
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8
7
0
8
A
n
o
ve
l a
p
p
r
o
a
c
h
fo
r
r
ec
o
mme
n
d
a
tio
n
u
s
in
g
o
p
timiz
ed
b
id
ir
ec
tio
n
a
l
…
(
P
r
a
ka
s
h
P
.
R
o
k
a
d
e
)
5027
Fig
u
r
e
7
.
C
o
m
p
a
r
is
o
n
o
f
p
r
ec
i
s
io
n
f
o
r
p
r
o
p
o
s
ed
s
y
s
tem
with
ex
is
tin
g
s
y
s
tem
s
Fig
u
r
e
8
.
C
o
m
p
a
r
is
o
n
o
f
r
ec
al
l f
o
r
p
r
o
p
o
s
ed
s
y
s
tem
with
ex
is
tin
g
s
y
s
tem
s
Fig
u
r
e
9
.
C
o
m
p
a
r
is
o
n
o
f
f
-
m
e
asu
r
e
f
o
r
p
r
o
p
o
s
ed
s
y
s
tem
with
ex
is
tin
g
s
y
s
tem
s
Fig
u
r
e
1
0
.
C
o
m
p
ar
is
o
n
o
f
ac
c
u
r
ac
y
f
o
r
p
r
o
p
o
s
ed
s
y
s
tem
with
ex
is
tin
g
s
y
s
tem
s
Fig
u
r
e
1
1
.
C
o
m
p
ar
is
o
n
o
f
MA
E
f
o
r
p
r
o
p
o
s
ed
s
y
s
tem
with
ex
is
tin
g
s
y
s
tem
s
Fig
u
r
e
1
2
.
C
o
m
p
ar
is
o
n
o
f
R
MSE
f
o
r
p
r
o
p
o
s
ed
s
y
s
tem
with
ex
is
tin
g
s
y
s
tem
s
3
.
3
.
E
x
perim
ent
a
l r
esu
lt
s
a
nd
co
m
pa
riso
ns
T
o
ev
alu
ate
th
e
ef
f
ec
tiv
en
ess
o
f
th
e
p
r
o
p
o
s
ed
B
iGR
U
-
R
O
A
m
o
d
el,
we
co
m
p
ar
ed
its
p
er
f
o
r
m
a
n
ce
with
th
r
ee
ex
is
tin
g
a
p
p
r
o
ac
h
es
—
GA,
MM
DL
,
an
d
FF
NN
—
ac
r
o
s
s
f
o
u
r
k
e
y
m
et
r
ics:
p
r
ec
is
io
n
,
r
ec
all,
F
-
m
ea
s
u
r
e,
an
d
ac
cu
r
ac
y
.
T
h
e
s
e
m
etr
ics
o
f
f
er
a
co
m
p
r
eh
en
s
iv
e
ass
es
s
m
en
t
o
f
m
o
d
el
p
er
f
o
r
m
an
ce
in
ter
m
s
o
f
b
o
th
p
r
ed
ictiv
e
ac
c
u
r
ac
y
a
n
d
r
o
b
u
s
tn
ess
.
T
h
e
f
o
llo
win
g
b
ar
ch
ar
t
clea
r
l
y
illu
s
tr
ates
th
at
th
e
B
iGR
U
-
R
OA
m
o
d
el
co
n
s
is
ten
tly
o
u
tp
e
r
f
o
r
m
s
o
th
er
m
o
d
els
in
ea
ch
ca
t
eg
o
r
y
,
s
h
o
wca
s
in
g
its
s
u
p
er
io
r
p
r
ec
is
io
n
(
9
8
%)
,
r
ec
all
(
9
7
.
5
%),
F
-
m
ea
s
u
r
e
(
9
7
%),
an
d
ac
cu
r
ac
y
(
9
8
.
5
%).
As
s
h
o
wn
in
T
ab
le
6,
th
e
v
alu
es
of
p
r
ec
is
io
n
,
r
ec
all,
f
-
m
ea
s
u
r
e,
ac
cu
r
ac
y
,
MA
E
,
R
MSE
f
o
r
ex
is
tin
g
m
o
d
els
MM
DL
,
GA,
FF
NN
an
d
p
r
o
p
o
s
ed
o
p
tim
iz
ed
B
iGR
U
m
o
d
el
ar
e
ca
lcu
lated
.
T
h
e
r
esu
lts
o
f
th
e
p
r
o
p
o
s
ed
B
iGR
U
-
R
OA
m
o
d
el
s
h
o
w
an
ac
cu
r
ac
y
im
p
r
o
v
em
en
t
o
f
1
2
.
3
%
o
v
er
FF
NN
an
d
8
.
5
%
o
v
er
GA,
3
2
%
f
aster
co
n
v
er
g
e
n
ce
r
ate
co
m
p
ar
ed
to
tr
a
d
itio
n
al
n
eu
r
al
n
etwo
r
k
s
,
r
ed
u
ctio
n
in
tr
ain
i
n
g
er
r
o
r
to
0
.
0
0
0
3
3
en
h
an
cin
g
p
r
e
d
ictiv
e
p
er
f
o
r
m
an
ce
.
Fig
u
r
e
1
3
s
h
o
ws
th
e
b
ar
ch
ar
t
c
o
m
p
ar
i
n
g
th
e
p
er
f
o
r
m
an
ce
o
f
B
iGR
U
-
R
OA,
GA,
MM
DL
,
an
d
FF
N
N
in
ter
m
s
o
f
Pre
cisi
o
n
,
R
ec
al
l,
F
-
Me
asu
r
e,
an
d
Acc
u
r
ac
y
.
98%
92%
96%
85%
7
5
%
8
0
%
8
5
%
9
0
%
9
5
%
1
0
0
%
P
r
opos
e
d
M
M
DL
B
L
_GA
FF
N
N
P
r
ec
is
io
n
%
S
y
ste
m
f
o
r
c
o
m
p
a
riso
n
9
7
.
5
0
%
96%
97%
88%
8
2
.
0
0
%
8
4
.
0
0
%
8
6
.
0
0
%
8
8
.
0
0
%
9
0
.
0
0
%
9
2
.
0
0
%
9
4
.
0
0
%
9
6
.
0
0
%
9
8
.
0
0
%
1
0
0
.
0
0
%
P
r
opos
e
d
M
M
DL
B
L
_GA
F
F
NN
R
ec
all
in
%
S
y
ste
m
f
o
r
Co
m
p
a
riso
n
97%
95%
96%
87%
8
2
%
8
4
%
8
6
%
8
8
%
9
0
%
9
2
%
9
4
%
9
6
%
9
8
%
P
r
opos
e
d
M
M
DL
B
L
_GA
F
F
NN
F
-
Me
asu
r
e
in
%
S
y
ste
m
f
o
r
Co
m
p
a
riso
n
9
8
.
5
0
%
95%
96%
86%
7
5
.
0
0
%
8
0
.
0
0
%
8
5
.
0
0
%
9
0
.
0
0
%
9
5
.
0
0
%
1
0
0
.
0
0
%
P
r
opos
e
d
M
M
DL
B
L
_GA
F
F
NN
A
cc
u
r
ac
y
S
y
ste
m
f
o
r
Co
m
p
a
riso
n
0
.
0
3
%
0
.
6
5
%
0
.
6
5
%
0
.
1
4
%
0
.
0
0
%
0
.
1
0
%
0
.
2
0
%
0
.
3
0
%
0
.
4
0
%
0
.
5
0
%
0
.
6
0
%
0
.
7
0
%
P
r
op
os
ed
M
M
DL
B
L
_GA
F
F
NN
MA
E
S
y
ste
m
f
o
r
Co
m
p
a
riso
n
0
.
1
7
%
0
.
2
5
%
0
.
2
2
%
0
.
3
6
%
0
.
0
0
%
0
.
0
5
%
0
.
1
0
%
0
.
1
5
%
0
.
2
0
%
0
.
2
5
%
0
.
3
0
%
0
.
3
5
%
0
.
4
0
%
P
r
op
os
ed
M
M
DL
B
L
_GA
F
F
NN
R
MA
E
S
y
ste
m
f
o
r
Co
m
p
a
riso
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8
7
0
8
I
n
t J E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
15
,
No
.
5
,
Octo
b
e
r
20
25
:
5
0
1
9
-
5
0
3
0
5028
T
ab
le
6
.
Pro
p
o
s
ed
s
y
s
tem
p
er
f
o
r
m
an
ce
c
o
m
p
ar
is
o
n
with
ex
is
tin
g
m
o
d
els
M
o
d
e
l
v
s
c
r
i
t
e
r
i
a
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
F
-
M
e
a
s
u
r
e
A
c
c
u
r
a
c
y
M
A
E
R
M
S
E
A
B
C
A
B
C
A
B
C
A
B
C
A
B
C
A
B
C
P
r
o
p
o
se
d
s
y
st
e
m
98
98
97
97
97
98
97
97
98
98
97
98
0
.
0
3
0
.
0
3
0
.
0
3
0
.
1
7
0
.
1
7
0
.
1
6
M
M
D
L
92
94
94
96
93
94
95
94
95
95
92
93
0
.
0
6
0
.
0
5
0
.
0
5
0
.
2
5
0
.
2
5
0
.
2
5
GA
96
95
95
97
95
95
96
95
95
96
94
95
0
.
0
5
0
.
0
7
0
.
0
7
0
.
2
2
0
.
2
4
0
.
2
4
F
F
N
N
85
85
85
88
86
87
87
88
89
86
87
86
0
.
1
4
0
.
1
5
0
.
1
5
0
.
3
6
0
.
3
6
0
.
3
7
Fig
u
r
e
1
3
.
Per
f
o
r
m
a
n
ce
co
m
p
ar
is
o
n
o
f
m
o
d
el
As
p
r
o
p
o
s
ed
m
o
d
el
ca
m
e
o
u
t
with
c
o
m
p
ar
ativ
el
y
b
etter
r
esu
lts
th
an
ex
is
tin
g
ap
p
r
o
ac
h
es,
th
e
o
b
jectiv
es
of
r
esear
ch
ar
e
atten
d
ed
.
T
o
f
u
r
t
h
er
v
alid
ate
th
e
r
o
b
u
s
tn
ess
an
d
lea
r
n
in
g
ef
f
icien
c
y
o
f
t
h
e
p
r
o
p
o
s
ed
B
iGR
U
-
R
OA
m
o
d
el,
we
an
aly
ze
d
its
tr
ain
in
g
lo
s
s
b
eh
av
io
r
o
v
e
r
m
u
ltip
le
ep
o
ch
s
as
s
h
o
wn
in
Fig
u
r
e
1
4
.
T
h
e
tr
ain
in
g
lo
s
s
cu
r
v
e
p
r
o
v
id
es
in
s
ig
h
ts
in
to
h
o
w
q
u
ick
ly
an
d
ef
f
ec
tiv
ely
th
e
m
o
d
el
lear
n
s
f
r
o
m
th
e
in
p
u
t
d
ata.
Fas
ter
co
n
v
er
g
en
ce
with
m
in
im
al
f
lu
ctu
atio
n
s
in
d
icate
s
s
tab
le
lear
n
in
g
an
d
ef
f
ec
tiv
e
weig
h
t
o
p
tim
izatio
n
.
T
h
e
R
OA
p
lay
s
a
k
ey
r
o
le
h
er
e
b
y
d
y
n
am
ic
ally
ad
ju
s
tin
g
weig
h
ts
to
m
in
im
ize
er
r
o
r
d
u
r
in
g
ea
ch
iter
atio
n
.
Fig
u
r
e
1
4
.
T
r
ain
in
g
lo
s
s
v
s
ep
o
ch
s
B
iGR
U
-
R
OA
Evaluation Warning : The document was created with Spire.PDF for Python.