I
nte
rna
t
io
na
l J
o
urna
l o
f
Adv
a
nces in Applie
d Science
s
(
I
J
AAS)
Vo
l.
6
,
No
.
4
,
Dec
em
b
er
2
0
1
7
,
p
p
.
2
8
3
~2
9
2
I
SS
N:
2252
-
8814
283
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
jo
u
r
n
a
l.c
o
m/o
n
lin
e/in
d
ex
.
p
h
p
/I
J
AAS
Co
m
pa
ra
tive
Stu
dy
of Vario
us Ne
ura
l Net
w
o
rk
Ar
c
hitectu
res
for MPE
G
-
4
Vid
e
o
Tra
ff
ic
Predic
ti
o
n
J.
P.
K
ha
r
a
t
*
DK
T
E
Ic
h
a
lk
a
r
a
n
ji
,
In
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Sep
1
5
,
2
0
1
7
R
ev
i
s
ed
No
v
1
0
,
2
0
1
7
A
cc
ep
ted
No
v
2
0
,
2
0
1
7
Ne
tw
o
rk
tra
ff
ic
a
s
it
is
V
BR
i
n
n
a
tu
re
e
x
h
i
b
it
s
str
o
n
g
c
o
rre
lati
o
n
s
w
h
ich
m
a
k
e
it
su
it
a
b
le
f
o
r
p
re
d
ictio
n
.
R
e
a
l
-
ti
m
e
f
o
re
c
a
stin
g
o
f
n
e
t
w
o
rk
t
ra
ff
ic
lo
a
d
a
c
c
u
ra
tel
y
a
n
d
in
a
c
o
m
p
u
tatio
n
a
ll
y
e
ff
icie
n
t
m
a
n
n
e
r
is
th
e
k
e
y
e
le
m
e
n
t
o
f
p
ro
a
c
ti
v
e
n
e
tw
o
rk
m
a
n
a
g
e
m
e
n
t
a
n
d
c
o
n
g
e
stio
n
c
o
n
tr
o
l.
T
h
is
p
a
p
e
r
c
o
m
m
e
n
ts
o
n
t
h
e
M
P
EG
-
4
v
id
e
o
traf
f
ic
p
re
d
ictio
n
s
e
v
a
lu
a
ted
b
y
d
if
fe
re
n
t
ty
p
e
s
o
f
n
e
u
ra
l
n
e
tw
o
rk
a
rc
h
it
e
c
tu
re
s
a
n
d
c
o
m
p
a
re
s
th
e
p
e
r
f
o
rm
a
n
c
e
o
f
th
e
sa
m
e
in
term
s
o
f
m
e
a
n
sq
u
a
re
e
rro
r
f
o
r
th
e
sa
m
e
v
id
e
o
f
ra
m
e
s.
F
o
r
th
a
t
th
r
e
e
t
y
p
e
s
o
f
n
e
u
ra
l
a
rc
h
it
e
c
tu
re
s
a
re
u
se
d
n
a
m
e
l
y
F
e
e
d
f
o
r
w
a
rd
,
Ca
sc
a
d
e
d
F
e
e
d
f
o
rw
a
rd
a
n
d
T
im
e
D
e
la
y
N
e
u
ra
l
Ne
tw
o
rk
.
T
h
e
re
su
lt
s
sh
o
w
th
a
t
c
a
sc
a
d
e
fe
e
d
f
o
rwa
rd
n
e
tw
o
rk
p
ro
d
u
c
e
s
m
in
im
u
m
e
rro
r
a
s
c
o
m
p
a
re
d
to
o
th
e
r
n
e
tw
o
rk
s.
T
h
is
p
a
p
e
r
a
lso
c
o
m
p
a
re
s
th
e
re
su
lt
s
o
f
trad
it
io
n
a
l
p
re
d
icti
o
n
m
e
th
o
d
o
f
a
v
e
ra
g
in
g
o
f
f
ra
m
e
s
f
o
r
f
u
tu
re
f
ra
m
e
p
re
d
i
c
ti
o
n
w
it
h
n
e
u
ra
l
b
a
se
d
m
e
th
o
d
s.
T
h
e
e
x
p
e
ri
m
e
n
tal
re
su
lt
s
sh
o
w
th
a
t
n
o
n
li
n
e
a
r
p
re
d
ictio
n
b
a
se
d
o
n
NN
s
is
b
e
tt
e
r
su
it
e
d
f
o
r
traf
f
ic p
re
d
ictio
n
p
u
r
p
o
se
s th
a
n
li
n
e
a
r
f
o
re
c
a
stin
g
m
o
d
e
ls.
K
ey
w
o
r
d
:
C
ascad
ed
f
ee
d
f
o
r
w
ar
d
n
eu
r
al
Feed
f
o
r
w
ar
d
n
e
u
r
al
n
et
w
o
r
k
L
i
n
ea
r
p
r
ed
ictio
n
MP
E
G
-
4
v
id
eo
tr
af
f
ic
n
et
w
o
r
k
No
n
li
n
ea
r
p
r
ed
ictio
n
T
im
e
d
ela
y
n
eu
r
al
n
et
w
o
r
k
Co
p
y
rig
h
t
©
201
7
In
s
t
it
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
J.
P.
Kh
ar
at,
DKT
E
I
ch
alk
ar
an
j
i
,
I
n
d
ia.
1.
I
NT
RO
D
UCT
I
O
N
Ma
n
y
m
u
lti
m
ed
ia
ap
p
licatio
n
s
,
s
u
c
h
as
v
id
eo
-
co
n
f
er
en
c
in
g
an
d
v
id
eo
-
b
ased
en
ter
tain
m
e
n
t
s
er
v
ices,
r
el
y
o
n
th
e
e
f
f
icien
t
tr
an
s
m
i
s
s
i
o
n
o
f
liv
e
o
r
s
to
r
ed
v
id
eo
.
T
h
ese
v
id
eo
s
ar
e
co
m
p
r
es
s
ed
b
ef
o
r
e
tr
an
s
m
i
s
s
io
n
i
n
o
r
d
er
t
o
r
ed
u
ce
th
e
s
ize
o
f
v
id
eo
s
.
Fo
r
th
is
p
u
r
p
o
s
e
MP
E
G
-
4
(
Mo
v
in
g
P
ictu
r
e
E
x
p
er
t
Gr
o
u
p
)
v
id
eo
co
m
p
r
es
s
io
n
s
tan
d
ar
d
is
u
s
ed
.
T
h
is
ME
G
-
4
s
tan
d
ar
d
g
en
er
at
es
v
ar
iab
le
b
it
r
ate
tr
af
f
ic
w
h
i
ch
is
v
er
y
b
u
r
s
t
y
i
n
n
atu
r
e,
d
u
e
to
th
e
f
r
a
m
e
s
tr
u
c
tu
r
e
o
f
th
e
en
co
d
in
g
s
c
h
e
m
e
an
d
n
atu
r
al
v
ar
iat
io
n
s
w
it
h
i
n
an
d
b
et
w
ee
n
s
ce
n
e
s
[1
-
7
]
.
I
f
o
n
e
h
as
to
tr
an
s
m
it
th
is
tr
a
f
f
ic
o
v
er
n
et
w
o
r
k
,
th
i
s
tr
af
f
ic
w
ill
b
e
tr
an
s
m
it
ted
b
y
p
ea
k
r
ate.
B
u
t
as
MP
E
G
-
4
tr
af
f
ic
i
s
VB
R
i
n
n
a
t
u
r
e,
I
-
f
r
a
m
e
is
lar
g
e
i
n
s
ize
as
co
m
p
ar
ed
to
P
an
d
B
f
r
am
e
s
i
ze
s
.
Hen
ce
e
f
f
ic
ien
t
u
tili
za
t
io
n
o
f
b
an
d
w
id
th
w
i
ll
n
o
t
tak
e
p
lace
.
A
l
s
o
Var
iab
le
-
b
it
-
r
ate
(
VB
R
)
tr
af
f
ic
co
m
p
li
ca
tes
th
e
d
esi
g
n
o
f
ef
f
icien
t
r
ea
l
-
ti
m
e
s
to
r
ag
e,
r
etr
iev
al,
tr
an
s
p
o
r
t,
an
d
p
r
o
v
is
io
n
i
n
g
m
ec
h
a
n
i
s
m
s
ca
p
ab
le
o
f
ac
h
iev
i
n
g
h
ig
h
r
eso
u
r
ce
u
tili
za
tio
n
T
o
o
v
er
co
m
e
t
h
i
s
p
r
o
b
lem
,
g
en
er
all
y
VB
R
tr
af
f
ic
is
s
m
o
o
th
ed
.
Fo
r
s
m
o
o
t
h
in
g
o
f
tr
af
f
i
c,
v
ar
io
u
s
ap
p
r
o
ac
h
es h
av
e
b
ee
n
s
u
g
g
e
s
t
ed
.
Ma
in
l
y
t
h
er
e
ar
e
t
w
o
t
y
p
e
s
: L
i
n
ea
r
Me
th
o
d
an
d
No
n
-
li
n
e
ar
m
e
th
o
d
[
8
]
.
I
n
lin
ea
r
m
e
th
o
d
,
v
ar
io
u
s
ap
p
r
o
ac
h
es
h
a
v
e
b
ee
n
s
u
g
g
es
ted
.
Fe
w
o
f
th
e
m
ar
e
:
T
h
e
f
ir
s
t
ap
p
r
o
ac
h
is
to
co
n
v
er
t
VB
R
to
C
B
R
.
I
n
th
is
ap
p
r
o
ac
h
,
f
ir
s
t
f
e
w
f
r
a
m
es a
r
e
tak
e
n
f
r
o
m
s
o
u
r
ce
i
n
b
u
f
f
er
.
T
h
en
th
e
av
er
ag
e
o
f
t
h
ese
f
r
am
es
is
ta
k
e
n
.
Fin
al
l
y
f
r
a
m
e
s
w
il
l
b
e
tr
an
s
m
itted
at
th
is
a
v
er
ag
e
r
ate.
B
u
t
th
i
s
s
c
h
e
m
e
i
n
tr
o
d
u
ce
s
f
i
n
ite
d
ela
y
i
n
p
la
y
b
ac
k
at
r
ec
eiv
er
.
A
ls
o
t
h
er
e
ar
e
ch
a
n
ce
s
th
at
f
e
w
f
r
a
m
e
s
m
i
g
h
t
g
et
lo
s
t d
u
e
to
b
u
f
f
er
f
u
ll
n
ess
,
an
d
m
a
y
a
f
f
ec
t t
h
e
q
u
alit
y
o
f
p
ictu
r
e
an
d
h
e
n
ce
QO
C
.
T
h
e
s
ec
o
n
d
ap
p
r
o
ac
h
is
as
f
r
a
m
es
co
m
e
f
r
o
m
t
h
e
s
o
u
r
ce
,
th
e
r
ate
is
u
p
d
ated
p
er
f
r
am
e
.
B
u
t
th
i
s
ap
p
r
o
ac
h
also
in
tr
o
d
u
ce
s
p
la
y
b
ac
k
d
ela
y
a
t
r
ec
ei
v
er
.
T
h
e
m
o
s
t
co
m
m
o
n
n
o
n
l
in
ea
r
f
o
r
ec
asti
n
g
m
et
h
o
d
s
in
v
o
l
v
e
n
e
u
r
al
n
e
t
w
o
r
k
s
(
NN)
[
9
-
1
1
]
.
A
lth
o
u
g
h
s
o
m
e
ar
ticle
s
s
tate
t
h
at
li
n
ea
r
p
r
ed
ictio
n
m
o
d
els
ar
e
u
n
ab
le
to
d
escr
ib
e
th
e
ch
ar
ac
ter
is
tics
o
f
n
et
w
o
r
k
tr
a
f
f
ic
[
1
1
]
,
o
th
er
s
tu
d
ie
s
co
n
f
ir
m
th
e
p
r
ac
tica
l
u
s
ab
ilit
y
o
f
lin
ea
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8814
IJ
AA
S
Vo
l.
6
,
No
.
4
,
Dec
em
b
er
201
7
:
2
8
3
–
2
9
2
284
p
r
ed
icto
r
s
f
o
r
r
ea
l
-
ti
m
e
tr
af
f
i
c
p
r
ed
ictio
n
[
1
2
]
.
T
h
u
s
,
it
r
em
ai
n
s
u
n
clea
r
w
h
ic
h
p
r
ed
icto
r
s
p
r
o
v
id
e
th
e
b
est
p
er
f
o
r
m
a
n
ce
,
b
ein
g
i
n
th
e
s
a
m
e
ti
m
e
s
i
m
p
le,
ad
ap
tab
le
an
d
ac
cu
r
ate.
T
h
e
r
est
o
f
p
ap
er
is
o
r
g
a
n
iz
ed
as
b
elo
w
:
Sectio
n
I
d
escr
i
b
es
th
e
th
r
ee
d
i
f
f
er
en
t
t
y
p
e
s
o
f
n
eu
r
al
ar
ch
itect
u
r
es
u
s
ed
f
o
r
p
r
ed
icti
o
n
.
Sectio
n
I
I
f
o
cu
s
e
s
o
n
th
e
t
r
ain
in
g
an
d
T
esti
n
g
o
f
Neu
r
al
n
et
w
o
r
k
s
.
Sectio
n
I
I
I
d
ea
ls
w
it
h
s
i
m
u
lat
io
n
r
e
s
u
l
ts
o
f
t
h
r
ee
ar
ch
itect
u
r
es.
Sect
i
o
n
I
V
p
r
ese
n
ts
t
h
e
co
m
p
ar
ati
v
e
an
al
y
s
is
o
f
al
l
t
h
e
th
r
ee
ar
ch
i
tectu
r
e
s
.
Sectio
n
V
co
m
p
ar
es
t
h
e
Neu
r
al
Net
w
o
r
k
r
esu
lts
w
it
h
t
h
e
co
n
v
e
n
tio
n
al
m
eth
o
d
s
an
d
f
i
n
all
y
,
s
ec
tio
n
V
m
a
k
es t
h
e
f
i
n
al
co
n
cl
u
s
io
n
.
2.
NE
URA
L
N
E
T
WO
RK
A
NN
h
as
m
u
lt
ip
le
in
ter
co
n
n
ec
ted
p
r
o
ce
s
s
in
g
ele
m
en
t
s
g
r
o
u
p
ed
in
to
la
y
er
s
.
E
ac
h
la
y
er
h
as
s
ev
er
al
n
o
d
es.
T
h
e
in
p
u
ts
to
o
n
e
n
o
d
e
in
a
la
y
er
ar
e
t
h
e
o
u
tp
u
t
s
o
f
all
o
th
er
n
o
d
es
in
th
e
p
r
ev
io
u
s
la
y
er
.
T
h
e
n
o
d
es
alg
eb
r
aica
ll
y
s
u
m
th
e
s
e
w
ei
g
h
ted
s
ig
n
als
an
d
p
as
s
th
e
m
th
r
o
u
g
h
a
n
o
n
l
in
ea
r
s
q
u
as
h
i
n
g
f
u
n
ctio
n
to
p
r
o
d
u
ce
a
n
et
o
u
tp
u
t.
T
h
e
f
u
n
cti
o
n
is
u
s
u
all
y
a
s
i
g
m
o
id
f
u
n
ctio
n
o
r
a
h
y
p
er
b
o
lic
ta
n
g
e
n
t.
B
ased
o
n
th
e
s
tr
u
ct
u
r
e
o
f
t
h
e
n
et
w
o
r
k
o
r
th
e
w
a
y
th
e
n
o
d
es
ar
e
in
ter
co
n
n
ec
ted
,
t
h
er
e
ar
e
t
w
o
b
r
o
ad
ca
teg
o
r
ies
o
f
N
Ns;
t
h
e
f
ee
d
f
o
r
w
ar
d
m
u
lti
-
la
y
er
p
er
ce
p
tr
o
n
(
FML
P
)
an
d
th
e
r
ec
u
r
r
e
n
t
m
u
lti
-
la
y
er
p
er
ce
p
tr
o
n
(
R
M
L
P
)
.
FML
P
is
d
i
f
f
er
en
t
f
r
o
m
R
M
L
P
in
t
h
e
s
e
n
s
e
t
h
at
th
er
e
is
n
o
cr
o
s
s
tal
k
b
et
w
ee
n
t
h
e
n
o
d
es
o
f
a
g
i
v
en
la
y
er
.
E
ac
h
la
y
er
i
n
a
m
u
ltil
a
y
er
n
eu
r
al
n
et
w
o
r
k
h
as it
s
o
w
n
s
p
ec
if
ic
f
u
n
ctio
n
.
T
h
e
in
p
u
t la
y
e
r
ac
ce
p
ts
in
p
u
t si
g
n
als
f
r
o
m
t
h
e
o
u
ts
id
e
w
o
r
ld
a
n
d
r
ed
is
tr
ib
u
tes
t
h
ese
s
i
g
n
als
to
all
n
e
u
r
o
n
s
i
n
t
h
e
h
id
d
en
l
a
y
er
.
T
h
e
in
p
u
t
la
y
er
r
ar
el
y
in
cl
u
d
es
co
m
p
u
ti
n
g
n
eu
r
o
n
s
,
an
d
th
u
s
d
o
es
n
o
t
p
r
o
ce
s
s
in
p
u
t
p
atter
n
s
.
T
h
e
o
u
tp
u
t
la
y
er
ac
ce
p
ts
o
u
tp
u
t
s
ig
n
als
,
a
s
ti
m
u
l
u
s
p
atter
n
,
f
r
o
m
t
h
e
h
id
d
en
la
y
er
a
n
d
est
ab
lis
h
es
t
h
e
o
u
tp
u
t
p
atter
n
o
f
th
e
e
n
tire
n
et
w
o
r
k
.
An
y
co
n
t
i
n
u
o
u
s
f
u
n
c
tio
n
ca
n
b
e
ex
p
r
ess
ed
w
it
h
o
n
e
h
id
d
en
la
y
er
.
T
w
o
h
id
d
en
la
y
er
s
ca
n
p
r
ed
ict
d
is
co
n
tin
u
o
u
s
f
u
n
ctio
n
s
to
o
[
1
3
]
.
2
.
1
.
F
ee
dfo
rwa
rd
Neura
l N
et
wo
rk
T
h
e
n
et
w
o
r
k
is
co
m
p
o
s
ed
o
f
an
i
n
p
u
t
la
y
er
,
a
s
er
ies
o
f
h
id
d
en
la
y
er
s
a
n
d
a
n
o
u
tp
u
t
la
y
e
r
.
I
n
t
h
is
n
et
w
o
r
k
,
th
e
s
i
g
n
als
f
r
o
m
ea
ch
n
o
d
e
ar
e
tr
an
s
m
itted
to
all
th
e
n
o
d
es
in
th
e
n
ex
t
la
y
er
,
an
d
o
n
ly
t
h
e
h
id
d
en
la
y
er
s
h
av
e
a
s
ig
m
o
id
-
t
y
p
e
d
i
s
cr
i
m
i
n
ato
r
y
f
u
n
ct
io
n
.
I
n
t
h
is
w
o
r
k
,
a
h
y
p
er
b
o
lic
tan
g
en
t
h
a
s
b
ee
n
u
s
ed
as
t
h
e
d
is
cr
i
m
i
n
ato
r
y
f
u
n
ctio
n
.
T
h
e
in
p
u
t
a
n
d
th
e
o
u
tp
u
t
la
y
er
s
h
av
e
li
n
ea
r
d
is
cr
i
m
i
n
ato
r
y
f
u
n
ctio
n
s
a
n
d
th
e
i
n
p
u
t
la
y
er
h
as
n
o
b
iases
.
FM
L
P
s
w
it
h
ap
p
r
o
p
r
iate
s
ig
n
al
s
in
t
h
e
in
p
u
t
la
y
er
ar
e
g
o
o
d
at
ap
p
r
o
x
im
a
tin
g
s
ta
tic
n
o
n
li
n
ea
r
itie
s
,
i.e
.
m
e
m
o
r
y
-
le
s
s
n
o
n
li
n
e
ar
f
u
n
ctio
n
s
.
E
ac
h
o
f
th
e
p
r
o
ce
s
s
i
n
g
ele
m
en
ts
o
f
an
FM
L
P
n
et
w
o
r
k
is
g
o
v
er
n
ed
b
y
t
h
e
f
o
llo
w
i
n
g
eq
u
atio
n
.
=
[
,
]
(
∑
[
−
1
,
]
[
,
]
|
−
1
|
=
1
[
−
1
,
]
+
[
,
]
)
(
1
)
W
h
er
e
x
[
l,
i]
is
th
e
i
th
n
o
d
e
o
u
tp
u
t
o
f
t
h
e
l
st
la
y
er
f
o
r
s
a
m
p
le
t,
w
[
l−1
,
j
]
[
l,
i]
is
th
e
w
ei
g
h
t,
th
e
ad
j
u
s
tab
le
p
ar
a
m
eter
,
co
n
n
ec
ti
n
g
t
h
e
j
th
n
o
d
e
o
f
th
e
(
l
−
1
)
th
la
y
er
to
t
h
e
i
th
n
o
d
e
o
f
th
e
l
th
l
a
y
er
,
b
[
l,
i]
i
s
t
h
e
b
ias,
also
an
ad
j
u
s
tab
le
p
ar
am
eter
,
o
f
th
e
i
th
n
o
d
e
in
t
h
e
l
th
la
y
er
.
W
i
11
W
h
11
X1
Y1
X2
Y2
X
n
W
i
nm
W
i
m
p
Y
n
Fig
u
r
e
1.
Feed
-
f
o
r
w
ar
d
Neu
r
al
Net
w
o
r
k
2
.
2
.
Ca
s
ca
de
-
F
o
rwa
rd
Neura
l N
et
w
o
rk
(
CF
)
T
h
is
n
et
w
o
r
k
is
a
Feed
-
Fo
r
w
ar
d
n
et
w
o
r
k
w
i
th
m
o
r
e
t
h
a
n
o
n
e
h
id
d
en
la
y
er
.
M
u
lt
ip
le
la
y
er
s
o
f
n
eu
r
o
n
s
w
i
th
n
o
n
li
n
ea
r
tr
an
s
f
er
f
u
n
ctio
n
s
a
llo
w
t
h
e
n
et
w
o
r
k
to
lear
n
m
o
r
e
co
m
p
lex
n
o
n
lin
ea
r
r
elatio
n
s
h
ip
s
b
et
w
ee
n
in
p
u
t
a
n
d
o
u
tp
u
t
v
e
cto
r
s
.
T
h
is
n
et
w
o
r
k
ca
n
b
e
u
s
ed
as
a
g
en
er
al
f
u
n
ctio
n
ap
p
r
o
x
i
m
ato
r
.
I
t
ca
n
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
AA
S
I
SS
N:
2252
-
8814
C
o
mp
a
r
a
tive
S
tu
d
y
o
f V
a
r
io
u
s
N
eu
r
a
l Netw
o
r
k
A
r
ch
itectu
r
e
s
...
(
J.
P
K
h
a
r
a
t
)
285
ap
p
r
o
x
im
a
te
an
y
f
u
n
ctio
n
w
it
h
a
f
in
i
te
n
u
m
b
er
o
f
d
is
co
n
ti
n
u
itie
s
,
ar
b
itra
r
ily
w
ell,
g
i
v
e
n
s
u
f
f
icie
n
t
n
e
u
r
o
n
s
in
th
e
h
id
d
en
la
y
er
.
Fi
g
u
r
e
2
s
h
o
w
s
th
e
ar
c
h
itect
u
r
e
o
f
a
ca
s
ca
d
e
-
f
o
r
w
ar
d
Net
w
o
r
k
w
it
h
t
w
o
h
id
d
en
-
la
y
er
s
.
Fig
u
r
e
2.
C
ascad
ed
Feed
-
Fo
r
w
ar
d
Neu
r
al
Net
w
o
r
k
2
.
3
.
F
ee
d
-
F
o
rwa
rd
w
it
h T
a
pp
ed
T
i
m
e
Dela
y
s
(
F
F
T
D)
A
tap
p
ed
d
ela
y
li
n
e
ca
n
b
e
u
s
ed
w
i
th
li
n
ea
r
n
e
u
r
o
n
s
to
m
o
d
if
y
o
n
l
y
t
h
e
i
n
p
u
t
la
y
er
(
to
allo
w
f
o
r
d
elay
ed
in
p
u
ts
)
.
T
h
e
tap
p
ed
d
ela
y
li
n
e
s
en
d
s
th
e
cu
r
r
e
n
t
s
ig
n
al,
i
n
ad
d
itio
n
to
a
n
u
m
b
er
o
f
d
ela
y
ed
v
er
s
io
n
s
,
to
th
e
w
ei
g
h
t
m
atr
ix
.
T
h
e
r
est
o
f
t
h
e
n
et
w
o
r
k
,
b
e
y
o
n
d
th
e
in
p
u
t
la
y
er
,
i
s
t
h
e
s
a
m
e
a
s
th
e
f
ee
d
-
f
o
r
w
ar
d
n
et
w
o
r
k
.
Fi
g
u
r
e
3
s
h
o
w
s
a
tap
p
ed
d
elay
li
n
e
ap
p
lied
to
a
s
i
n
g
le
i
n
p
u
t.
T
h
e
s
a
m
e
co
n
ce
p
t
c
an
b
e
ap
p
lied
to
a
ll
n
et
w
o
r
k
i
n
p
u
t
s
.
0
X
(
n
-
1)
Y
X
(
n
-
2)
X
(
n
-
p)
1
Fig
u
r
e
3.
Feed
-
Fo
r
w
ar
d
w
it
h
T
ap
p
ed
T
im
e
Dela
y
s
(
FF
T
D)
3.
P
RE
DIC
T
I
O
N
O
F
M
P
E
G
-
CO
DE
D
VI
DE
O
T
RAC
E
S
3
.
1
Appro
a
ch
Neu
r
al
n
et
w
o
r
k
s
ar
e
g
en
er
all
y
co
n
s
id
er
ed
to
b
e
o
n
e
o
f
th
e
m
o
s
t
ef
f
ec
tiv
e
to
o
ls
f
o
r
p
r
ed
icti
o
n
.
Du
e
to
th
eir
an
alo
g
y
w
i
th
b
io
lo
g
ical
n
eu
r
al
n
et
w
o
r
k
s
(
h
u
m
a
n
b
r
ain
)
,
th
e
y
s
ee
m
s
to
b
e
s
u
itab
le
to
s
o
lv
e
p
r
ed
ictio
n
r
elate
d
task
s
.
Ou
r
p
r
ed
ictio
n
s
w
er
e
b
ased
o
n
tak
in
g
N
p
r
ev
io
u
s
p
atter
n
s
to
p
r
ed
ict
o
n
e
f
o
llo
w
i
n
g
p
atter
n
.
Z
-
1
Z
-
1
Y
3
Y
2
Y
1
X
2
X
3
X
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8814
IJ
AA
S
Vo
l.
6
,
No
.
4
,
Dec
em
b
er
201
7
:
2
8
3
–
2
9
2
286
Sin
ce
w
e
u
s
ed
a
s
u
p
er
v
is
ed
l
ea
r
n
in
g
p
ar
ad
ig
m
,
o
u
r
n
et
w
o
r
k
s
w
o
r
k
ed
w
it
h
d
esire
d
v
al
u
e
s
o
f
tar
g
et
p
atter
n
s
d
u
r
in
g
th
e
tr
ai
n
i
n
g
p
r
o
ce
s
s
.
3
.
2
.
M
et
ho
do
lo
g
y
Used
Fo
r
n
eu
r
al
n
et
w
o
r
k
s
,
it
is
n
o
t
a
ch
allen
g
e
to
p
r
ed
ict
p
atter
n
s
ex
is
ti
n
g
o
n
a
s
eq
u
en
ce
w
it
h
w
h
ic
h
th
e
y
w
er
e
tr
ai
n
ed
.
T
h
e
r
ea
l
ch
a
llen
g
e
i
s
to
p
r
ed
ict
s
eq
u
e
n
ce
s
o
r
m
o
v
ie
s
t
h
at
t
h
e
n
et
w
o
r
k
d
id
n
o
t
u
s
e
f
o
r
tr
ai
n
in
g
.
Ho
w
e
v
er
,
th
e
p
ar
t
o
f
th
e
s
eq
u
en
ce
to
b
e
u
s
ed
f
o
r
tr
ain
i
n
g
s
h
o
u
ld
b
e
“
r
ich
en
o
u
g
h
”
to
eq
u
i
p
th
e
n
et
w
o
r
k
w
it
h
en
o
u
g
h
p
o
w
er
to
r
ec
o
n
s
tr
u
ct
o
r
ex
tr
ap
o
late
p
atter
n
s
t
h
at
m
a
y
e
x
i
s
t
i
n
o
t
h
er
s
eq
u
en
c
es
o
r
m
o
v
ies.
T
h
is
r
eq
u
ir
es
a
p
r
o
p
e
r
s
elec
tio
n
o
f
th
e
m
o
v
ie
to
b
e
u
s
ed
f
o
r
tr
a
in
i
n
g
i
n
ad
d
itio
n
to
th
e
p
r
o
p
er
s
elec
tio
n
o
f
th
e
n
u
m
b
er
o
f
tr
ai
n
i
n
g
p
o
in
ts
.
T
h
en
,
th
e
tr
ain
ed
n
et
w
o
r
k
s
ar
e
te
s
ted
w
it
h
a
d
if
f
er
en
t
p
o
r
tio
n
o
f
t
h
e
tr
ain
in
g
m
o
v
ie
in
ad
d
itio
n
to
a
n
u
m
b
er
o
f
s
e
g
m
e
n
t
s
f
r
o
m
th
e
r
e
m
ai
n
i
n
g
m
o
v
ies
(
n
o
t
u
s
ed
d
u
r
i
n
g
tr
ai
n
i
n
g
)
.
T
h
e
o
th
er
i
s
s
u
e
th
at
n
ee
d
s
to
b
e
ad
d
r
ess
ed
is
h
o
w
lo
n
g
t
h
e
tr
ain
i
n
g
s
eq
u
e
n
ce
s
h
o
u
l
d
b
e
in
o
r
d
er
t
o
ca
p
tu
r
e
“
j
u
s
t
e
n
o
u
g
h
”
u
s
e
f
u
l
p
atter
n
s
f
r
o
m
t
h
e
tr
ain
in
g
m
o
v
ie.
T
h
is
is
s
u
e
is
i
m
p
o
r
tan
t
b
ec
au
s
e
a
lo
n
g
er
-
t
h
a
n
-
n
ec
es
s
ar
y
tr
ain
in
g
s
eq
u
en
ce
w
i
ll g
iv
e
g
o
o
d
r
esu
lt
s
f
o
r
th
e
m
o
v
ie
u
s
ed
d
u
r
i
n
g
tr
a
in
i
n
g
a
n
d
p
o
o
r
r
esu
lts
f
o
r
th
e
r
e
m
ai
n
in
g
m
o
v
ie
s
.
Fo
r
all
p
r
ed
icto
r
s
d
esig
n
ed
i
n
th
is
s
t
u
d
y
,
o
n
l
y
a
s
e
g
m
e
n
t
o
f
t
h
e
J
u
r
as
s
ic
P
ar
k
v
id
eo
s
eq
u
en
ce
is
u
s
e
d
f
o
r
tr
ain
i
n
g
an
d
cr
o
s
s
-
v
alid
ati
o
n
,
to
d
eter
m
in
e
th
e
s
to
p
p
in
g
p
o
in
t
f
o
r
th
e
tr
ain
i
n
g
p
r
o
ce
s
s
.
Af
ter
f
i
x
in
g
all
o
f
th
e
n
et
w
o
r
k
p
ar
a
m
eter
s
,
t
h
e
d
ev
elo
p
ed
p
r
e
d
icto
r
s
ar
e
test
ed
o
n
th
e
r
e
m
ai
n
i
n
g
s
eg
m
e
n
t
o
f
th
e
J
u
r
ass
ic
P
ar
k
.
A
ll
ti
m
e
-
s
er
ie
s
ar
e
s
ca
led
b
y
a
s
in
g
le
s
ca
li
n
g
f
ac
to
r
s
o
th
at
th
e
y
lie
m
o
s
tl
y
in
t
h
e
r
an
g
e
f
r
o
m
1
to
-
1
,
m
ak
i
n
g
th
e
m
s
u
itab
le
f
o
r
p
r
o
ce
s
s
in
g
b
y
t
h
e
n
e
u
r
al
n
et
w
o
r
k
.
So
m
e
o
f
th
e
d
etails
o
f
t
h
e
v
id
eo
d
ata
tr
ac
es
u
s
ed
in
t
h
e
cu
r
r
en
t r
esear
ch
,
alo
n
g
w
i
th
t
h
e
s
eg
m
en
ts
u
s
ed
in
tr
ai
n
i
n
g
a
n
d
cr
o
s
s
-
v
al
id
atio
n
,
ar
e
as f
o
llo
w
s
:
T
r
ain
in
g
Seq
u
e
n
ce
co
n
s
is
t
o
f
f
ir
s
t
3
0
0
0
f
r
a
m
es
f
r
o
m
m
o
v
i
e
J
u
r
ass
ic
p
ar
k
en
co
d
ed
in
h
i
g
h
q
u
alit
y
m
o
d
e.
T
esti
n
g
Seq
u
e
n
ce
co
n
s
is
t
s
o
f
First
1
0
0
0
,
Mid
d
le
1
0
0
0
an
d
L
ast
4
0
0
f
r
a
m
es
f
r
o
m
t
h
e
s
a
m
e
m
o
v
ie
s
eq
u
en
ce
.
I
n
o
r
d
er
to
ac
h
iev
e
g
o
o
d
r
esu
lts
,
p
r
o
b
ab
l
y
o
n
e
o
f
t
h
e
m
o
s
t
i
m
p
o
r
tan
t
p
r
o
b
le
m
s
is
to
c
h
o
o
s
e
th
e
ap
p
r
o
p
r
iate
co
n
f
i
g
u
r
atio
n
o
f
n
eu
r
al
n
e
t
w
o
r
k
.
D
u
r
in
g
tr
ai
n
i
n
g
o
f
M
L
P
s
,
w
e
tr
ied
m
an
y
t
y
p
e
s
o
f
co
n
f
i
g
u
r
atio
n
s
f
o
r
o
u
r
p
r
ed
ictio
n
s
.
T
h
er
e
w
e
r
e
n
o
tab
le
d
if
f
er
e
n
ce
s
o
f
p
r
ed
ictio
n
er
r
o
r
s
a
m
o
n
g
t
h
e
m
.
W
e
ch
o
s
e
th
e
R
M
SE
(
r
o
o
t m
ea
n
s
q
u
ar
e
er
r
o
r
)
as a
n
o
b
j
ec
tiv
e
cr
iter
io
n
to
co
m
p
ar
e
th
e
m
.
W
e
m
ad
e
e
x
p
er
i
m
e
n
t
s
w
ith
th
e
n
u
m
b
er
o
f
in
p
u
t
n
e
u
r
o
n
s
ch
a
n
g
in
g
f
r
o
m
1
to
2
0
an
d
also
ex
p
er
i
m
en
ts
w
it
h
v
ar
io
u
s
n
u
m
b
er
o
f
h
i
d
d
en
n
e
u
r
o
n
s
a
n
d
n
u
m
b
er
o
f
h
id
d
en
la
y
er
s
.
W
e
ac
h
ie
v
ed
t
h
e
b
est
r
es
u
lt
s
o
f
tr
ai
n
in
g
th
e
M
L
P
n
et
w
o
r
k
u
s
i
n
g
n
et
w
o
r
k
co
n
f
i
g
u
r
atio
n
2
0
-
15
-
10
-
1
(
wh
ich
m
ea
n
s
:
2
0
i
n
p
u
t
n
eu
r
o
n
s
,
1
0
an
d
1
5
n
eu
r
o
n
s
in
h
id
d
en
la
y
er
,
1
o
u
tp
u
t
n
eu
r
o
n
)
,
L
e
v
e
n
b
er
g
-
Ma
r
q
u
ar
d
t
tr
ain
in
g
alg
o
r
ith
m
an
d
lear
n
in
g
-
r
ate
p
ar
am
e
ter
0
.
0
0
1
.
W
e
also
m
ad
e
ex
p
er
i
m
e
n
ts
w
it
h
t
h
e
lear
n
i
n
g
r
ate
an
d
n
u
m
b
er
o
f
iter
atio
n
s
w
h
ile
tr
ai
n
in
g
t
h
e
n
et
w
o
r
k
.
W
e
g
o
t
th
e
b
est
r
esu
lt
f
o
r
L
r
(
l
ea
r
n
in
g
r
ate)
=0
.
0
1
an
d
E
p
o
c
h
s
=1
5
0
0
.
A
ls
o
w
e
tr
ied
t
h
e
p
r
ed
ictio
n
f
o
r
v
ar
io
u
s
w
i
n
d
o
w
s
izes.
T
h
e
w
i
n
d
o
w
s
ize
v
ar
iatio
n
,
w
e
p
r
ef
er
r
ed
is
5
to
1
2
.
T
h
e
r
esu
lts
o
f
th
e
p
r
ed
ictio
n
f
o
r
th
e
tr
ain
in
g
a
n
d
test
s
et
ar
e
s
h
o
w
n
in
n
ex
t sectio
n
.
3
.
3
.
P
er
f
o
r
m
a
nce
M
et
rics
Her
e
w
e
d
ef
i
n
e
th
e
p
er
f
o
r
m
a
n
ce
m
etr
ic
s
u
s
ed
to
co
m
p
ar
e
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
d
if
f
er
en
t
m
o
d
els
d
ev
elo
p
ed
in
t
h
i
s
r
esear
c
h
.
T
h
r
ee
t
y
p
es
o
f
er
r
o
r
s
ca
n
b
e
u
s
ed
as
p
er
f
o
r
m
an
ce
m
etr
ic
f
o
r
t
h
e
p
r
ed
ictio
n
s
ch
e
m
es d
ev
e
lo
p
ed
in
th
i
s
w
o
r
k
.
T
h
e
th
r
ee
p
er
f
o
r
m
an
ce
s
m
et
r
ic
ar
e
d
ef
in
ed
as:
a.
Me
an
Sq
u
ar
e
E
r
r
o
r
(
MSE
)
:
MSE
is
th
e
r
atio
b
et
w
ee
n
t
h
e
s
u
m
o
f
th
e
s
q
u
ar
e
o
f
t
h
e
p
r
ed
ictio
n
er
r
o
r
an
d
th
e
s
u
m
o
f
t
h
e
s
q
u
ar
e
o
f
t
h
e
i
n
p
u
t
d
ata.
I
t is r
ep
r
esen
ted
b
y
t
h
e
f
o
llo
w
i
n
g
E
q
u
a
tio
n
:
MSE
=
∑
(
(
)
−
(
)
̂
)
2
=
1
∑
=
1
(
)
2
×
100
(
2
)
W
h
er
e
N
i
s
t
h
e
len
g
t
h
o
f
t
h
e
m
o
v
i
n
g
a
v
er
ag
e
ti
m
e
-
s
er
ies,
X
MA
i
s
t
h
e
ac
t
u
al
s
ize
o
f
th
e
j
-
t
h
ele
m
e
n
t
o
f
th
e
m
o
v
in
g
av
er
ag
e
ti
m
e
-
s
er
ies
an
d
X
MA
is
t
h
e
p
r
ed
ictio
n
o
f
th
e
j
-
th
ele
m
en
t.
MSE
is
an
in
d
icato
r
o
f
th
e
o
v
er
a
ll q
u
alit
y
o
f
t
h
e
p
r
ed
ictio
n
.
b
.
Ma
x
i
m
u
m
A
b
s
o
lu
te
E
r
r
o
r
(
MA
E
)
:
M
A
E
i
s
t
h
e
m
a
x
i
m
u
m
er
r
o
r
b
et
w
ee
n
th
e
ac
t
u
al
m
o
v
in
g
a
v
er
ag
e
o
f
t
h
e
VOP
s
izes a
n
d
th
e
p
r
ed
icted
m
o
v
i
n
g
av
er
a
g
e
o
f
th
e
V
OP
s
izes.
I
t is g
i
v
en
b
y
th
e
f
o
llo
w
in
g
E
q
u
atio
n
:
=
ma
x
1
≤
≤
|
(
)
−
̂
(
)
|
(
3
)
MA
E
is
th
e
m
ax
i
m
u
m
p
r
ed
ict
io
n
er
r
o
r
an
d
p
r
o
v
id
es
th
e
in
f
o
r
m
at
io
n
ab
o
u
t
t
h
e
w
o
r
s
t
ca
s
e
o
f
f
ail
u
r
e
o
f
th
e
p
r
ed
ictio
n
m
o
d
el.
c.
Ma
x
i
m
u
m
R
elati
v
e
E
r
r
o
r
(
MRE)
:
MRE
i
s
t
h
e
m
ax
i
m
u
m
o
f
th
e
r
atio
b
et
w
ee
n
t
h
e
p
r
ed
ictio
n
e
r
r
o
r
an
d
th
e
ac
tu
al
i
n
p
u
t d
ata
an
d
is
g
i
v
en
b
y
t
h
e
eq
u
atio
n
:
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
AA
S
I
SS
N:
2252
-
8814
C
o
mp
a
r
a
tive
S
tu
d
y
o
f V
a
r
io
u
s
N
eu
r
a
l Netw
o
r
k
A
r
ch
itectu
r
e
s
...
(
J.
P
K
h
a
r
a
t
)
287
MRE=
ma
x
1
≤
≤
|
(
)
−
̂
(
)
|
|
(
)
|
(
4
)
MRE
is
a
m
ea
s
u
r
e
o
f
t
h
e
r
el
ativ
e
co
m
p
ar
i
s
o
n
b
et
w
ee
n
t
h
e
p
r
ed
ictio
n
er
r
o
r
an
d
th
e
co
r
r
esp
o
n
d
in
g
ac
tu
al
m
o
v
i
n
g
a
v
er
ag
e
v
al
u
e
o
f
th
e
V
OP
s
ize.
3
.
4
.
Sca
lin
g
o
f
t
he
Da
t
a
Fo
r
th
e
p
u
r
p
o
s
e
o
f
n
o
n
-
li
n
ea
r
m
o
d
elin
g
u
s
i
n
g
NNs,
t
h
e
i
n
p
u
t
d
ata
m
u
s
t
b
e
s
ca
led
to
lie
w
it
h
i
n
ce
r
tain
b
o
u
n
d
s
.
T
h
e
s
ca
lin
g
o
f
in
p
u
t
d
ata
is
a
v
er
y
i
m
p
o
r
tan
t
asp
ec
t
o
f
tr
ain
in
g
th
e
n
et
w
o
r
k
.
I
n
o
r
d
er
t
o
p
r
ev
en
t t
h
e
s
at
u
r
atio
n
o
f
n
o
d
es in
NN
s
,
th
e
i
n
p
u
t d
ata
is
f
o
r
ce
d
to
lie
b
etw
ee
n
−1
an
d
1
.
4.
SI
M
UL
AT
I
O
N
R
E
S
UL
T
S
4
.
1
.
F
ee
d
-
f
o
rwa
rd
Neura
l N
et
wo
rk
a.
P
r
ed
ictio
n
R
esu
lt
s
f
o
r
Var
io
u
s
W
in
d
o
w
Size
s
I
n
t
h
is
s
ec
tio
n
w
e
w
ill
d
i
s
cu
s
s
th
e
s
i
m
u
latio
n
r
es
u
lts
.
I
n
t
h
i
s
s
tu
d
y
3
n
eu
r
al
n
et
w
o
r
k
s
w
er
e
test
ed
f
o
r
th
e
s
a
m
e
i
n
p
u
t
p
atter
n
s
.
W
e
h
av
e
tak
e
n
t
h
e
p
r
ev
io
u
s
f
r
a
m
e
s
izes
to
p
r
ed
ict
f
u
tu
r
e
s
in
g
le
f
r
a
m
e
s
ize.
T
h
is
ca
n
b
e
r
ef
er
r
ed
as
s
in
g
le
s
tep
ah
ea
d
p
r
ed
ictio
n
.
T
h
e
n
u
m
b
er
o
f
p
r
ev
io
u
s
d
ata
p
o
in
t
s
u
s
ed
to
p
r
ed
ict
th
e
n
e
x
t
d
ata
p
o
in
t
is
ter
m
e
d
a
s
W
i
n
d
o
w
Si
ze
.
Fo
r
o
u
r
ex
p
er
i
m
e
n
tatio
n
w
e
h
a
v
e
v
ar
ied
t
h
e
w
i
n
d
o
w
s
i
ze
f
r
o
m
5
f
r
a
m
e
s
to
1
2
f
r
a
m
es
a
n
d
co
m
p
ar
ed
th
e
r
esu
lt
s
f
o
r
v
ar
io
u
s
ar
ch
itect
u
r
es.
I
n
t
h
is
w
o
r
k
,
f
iv
e
e
m
p
ir
ic
al
MP
E
G
-
4
v
id
eo
tr
ac
es g
en
er
ated
b
y
F
itzek
et
a
l
.
an
d
av
ailab
le
o
n
t
h
e
p
u
b
lic
d
o
m
ai
n
[
1
4
]
ar
e
u
s
ed
.
Fig
u
r
e
4
.
r
ep
r
esen
ts
t
h
e
tr
ai
n
i
n
g
s
et
f
o
r
1
2
w
i
n
d
o
w
s
ize.
H
er
e
1
2
w
in
d
o
w
s
izes
r
ep
r
esen
ts
t
h
at
f
ir
s
t
1
2
f
r
a
m
es
in
th
e
s
eq
u
e
n
ce
h
a
v
e
b
ee
n
tak
e
n
to
p
r
ed
ict
t
h
e
1
3
th
f
r
a
m
e
in
t
h
e
s
eq
u
en
ce
.
Su
c
h
f
ir
s
t
3
0
0
0
f
r
a
m
es
f
r
o
m
th
e
m
o
v
ie
J
u
r
ass
ic
P
ar
k
h
av
e
ta
k
en
as
a
tr
ain
i
n
g
s
et
to
tr
ain
th
e
n
e
u
r
al
n
et
w
o
r
k
.
Fro
m
Fig
u
r
e
4
it
is
clea
r
th
at
t
h
e
p
r
ed
icted
f
r
a
m
e
s
an
d
ac
tu
al
f
r
a
m
e
s
ar
e
v
er
y
m
u
c
h
clo
s
er
.
I
n
th
i
s
ca
s
e
th
e
p
r
ed
ictio
n
er
r
o
r
is
v
er
y
s
m
al
l.
Fig
u
r
e
4
.
T
r
ain
in
g
Seq
u
e
n
ce
f
o
r
1
2
W
in
d
o
w
Size
Fig
u
r
e
5
.
T
est Seq
u
en
ce
f
o
r
Fi
r
s
t 1
0
0
0
Fra
m
es
Fig
u
r
e
6.
T
est
Seq
u
en
ce
f
o
r
Mid
d
le
1
0
0
0
Fra
m
es
Fig
u
r
e
7.
T
est
Seq
u
en
ce
f
o
r
L
ast 4
0
0
Fra
m
e
s
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8814
IJ
AA
S
Vo
l.
6
,
No
.
4
,
Dec
em
b
er
201
7
:
2
8
3
–
2
9
2
288
Fig
u
r
e
5
to
7
r
ep
r
esen
t
s
t
h
e
t
est
r
es
u
lts
.
T
h
e
d
if
f
er
en
t
f
r
a
m
e
s
eq
u
e
n
ce
s
f
r
o
m
t
h
e
s
a
m
e
m
o
v
ie
h
a
v
e
b
ee
n
tak
e
n
t
o
te
s
t
t
h
e
tr
ai
n
ed
n
et
w
o
r
k
.
F
ig
u
r
e
5
r
ep
r
esen
t
s
t
h
e
tes
t
r
es
u
lt
f
o
r
f
ir
s
t
1
0
0
0
f
r
a
m
es
in
th
e
m
o
v
ie
“
J
u
r
as
s
ic
P
ar
k
”.
T
h
e
R
ed
in
d
icate
s
p
r
ed
icted
f
r
a
m
es
w
h
ile
b
l
u
e
in
d
icate
s
o
r
ig
in
a
l
f
r
a
m
e
s
eq
u
en
ce
.
F
ig
u
r
e
6
r
ep
r
esen
ts
t
h
e
test
r
e
s
u
l
t
f
o
r
t
h
e
m
id
d
le
1
0
0
0
f
r
a
m
es
i
n
t
h
e
m
o
v
ie
“
J
u
r
a
s
s
ic
P
ar
k
”.
Fi
g
u
r
e
7
r
ep
r
esen
ts
th
e
test
r
esu
lt
s
f
o
r
th
e
la
s
t 4
0
0
f
r
a
m
es
in
th
e
m
o
v
ie
“
J
u
r
ass
ic
P
ar
k
”.
I
n
th
e
s
a
m
e
w
a
y
t
h
e
n
et
w
o
r
k
is
e
v
a
lu
a
ted
f
o
r
th
e
s
a
m
e
test
s
eq
u
e
n
ce
s
s
i
m
p
l
y
b
y
c
h
an
g
i
n
g
th
e
w
i
n
d
o
w
s
izes.
Fo
r
ea
ch
w
in
d
o
w
s
ize
th
e
r
o
o
t
m
ea
n
s
q
u
ar
e
er
r
o
r
is
ca
lcu
lated
.
A
s
t
h
e
w
i
n
d
o
w
s
ize
c
h
an
g
es
,
th
e
p
r
ed
ictio
n
er
r
o
r
also
ch
an
g
es.
Fig
u
r
e
8.
B
ar
C
h
ar
t Rep
r
esen
t
atio
n
o
f
E
r
r
o
r
s
o
f
Var
io
u
s
W
i
n
d
o
w
S
izes
Fig
u
r
e
8
r
ep
r
esen
ts
th
e
b
ar
ch
ar
r
ep
r
ese
n
tatio
n
o
f
er
r
o
r
s
o
f
v
ar
io
u
s
w
i
n
d
o
w
s
iz
es.
W
h
il
e
tr
an
s
m
is
s
io
n
o
f
th
e
m
o
v
ie,
i
n
o
r
d
er
to
r
e
d
u
ce
th
e
d
ela
y
ti
m
e,
it
is
r
eq
u
ir
ed
to
tak
e
th
e
m
in
i
m
u
m
n
u
m
b
er
o
f
f
r
a
m
e
s
to
p
r
ed
ict
th
e
f
u
t
u
r
e
f
r
a
m
es.
Mi
n
i
m
u
m
t
h
e
w
i
n
d
o
w
s
ize,
less
er
w
il
l
b
e
th
e
d
ela
y
.
Hen
ce
w
e
ev
a
lu
ated
th
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
f
ee
d
f
o
r
w
ar
d
n
e
u
r
al
n
e
t
w
o
r
k
f
o
r
v
ar
i
o
u
s
f
r
a
m
e
s
izes.
T
h
e
w
i
n
d
o
w
s
ize,
w
e
co
n
s
id
er
ed
is
f
r
o
m
5
to
1
2
.
Fro
m
ab
o
v
e
f
ig
.
it
ca
n
b
e
co
n
cl
u
d
ed
th
at,
n
et
w
o
r
k
g
i
v
es
m
i
n
i
m
u
m
er
r
o
r
f
o
r
1
2
w
i
n
d
o
w
s
ize.
Gen
er
all
y
in
MP
E
G
-
4
s
ta
n
d
ar
d
o
n
e
GOP
co
n
s
is
t
o
f
1
2
f
r
a
m
es.
I
f
f
u
r
t
h
er
w
i
n
d
o
w
s
ize
is
i
n
cr
ea
s
ed
,
th
e
p
r
ed
ictio
n
er
r
o
r
w
ill
b
e
d
ec
r
ea
s
ed
f
u
r
t
h
er
.
B
u
t
it
w
i
ll
in
c
r
ea
s
e
th
e
d
ela
y
ti
m
e.
B
y
co
n
s
id
er
in
g
b
o
th
th
e
s
e
p
ar
am
eter
s
,
w
e
ca
n
s
a
y
th
a
t n
e
t
w
o
r
k
g
i
v
es t
h
e
b
est p
r
ed
ictio
n
r
esu
lts
f
o
r
1
2
w
i
n
d
o
w
s
ize.
4
.
2
.
Ca
s
ca
ded F
ee
d
-
f
o
rw
a
rd
Neu
ra
l N
et
wo
rk
4
.
2
.
1
.
P
re
dict
io
n Re
s
ults f
o
r
Va
rio
us
Wind
o
w
Sizes
Fig
u
r
e
9
.
r
ep
r
esen
ts
th
e
tr
ai
n
in
g
s
et
f
o
r
1
2
w
i
n
d
o
w
s
ize.
Su
c
h
f
ir
s
t
3
0
0
0
f
r
am
e
s
f
r
o
m
th
e
m
o
v
ie
J
u
r
ass
ic
P
ar
k
h
a
v
e
tak
e
n
as a
t
r
ain
in
g
s
et
to
tr
ai
n
th
e
n
e
u
r
al
n
et
w
o
r
k
.
Fig
u
r
e
9
.
T
r
ain
in
g
Seq
u
e
n
ce
f
o
r
1
2
W
in
d
o
w
Size
Fig
u
r
e
1
0
.
T
est Seq
u
en
ce
f
o
r
First 1
0
0
0
Fra
m
e
s
5
6
7
8
9
10
11
12
0
0
.
0
0
2
0
.
0
0
4
0
.
0
0
6
0
.
0
0
8
0
.
0
1
0
.
0
1
2
0
.
0
1
4
0
.
0
1
6
0
.
0
1
8
0
.
0
2
E
R
R
O
R
C
O
M
P
A
R
I
S
O
N
f
o
r
V
A
R
I
O
U
S
W
I
N
D
O
W
S
I
Z
E
S
W
I
N
D
O
W
S
I
Z
E
R
S
M
E
E
R
R
O
R
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
AA
S
I
SS
N:
2252
-
8814
C
o
mp
a
r
a
tive
S
tu
d
y
o
f V
a
r
io
u
s
N
eu
r
a
l Netw
o
r
k
A
r
ch
itectu
r
e
s
...
(
J.
P
K
h
a
r
a
t
)
289
Fig
u
r
e
1
1
.
T
est Seq
u
en
ce
f
o
r
Mid
d
le
1
0
0
0
Fra
m
es
Fig
u
r
e
1
2
.
T
est
Seq
u
en
ce
f
o
r
L
ast 4
0
0
Fra
m
e
s
Fig
u
r
e
10
to
12
r
ep
r
esen
ts
t
h
e
test
r
es
u
lt
s
.
T
h
e
f
r
a
m
e
s
eq
u
en
ce
s
f
r
o
m
t
h
e
s
a
m
e
m
o
v
ie
h
a
v
e
b
ee
n
tak
en
t
o
test
th
e
tr
ain
ed
n
et
w
o
r
k
.
Fig
u
r
e
10
r
ep
r
esen
ts
t
h
e
tes
t
r
es
u
lt
f
o
r
f
ir
s
t
1
0
0
0
f
r
a
m
es
i
n
t
h
e
m
o
v
ie
“
J
u
r
as
s
ic
P
ar
k
”.
T
h
e
R
ed
in
d
icate
s
p
r
ed
icted
f
r
a
m
es
w
h
ile
b
lu
e
in
d
icate
s
o
r
ig
i
n
al
f
r
a
m
e
s
eq
u
en
ce
.
F
ig
u
r
e
11
r
ep
r
esen
ts
th
e
te
s
t
r
esu
l
t
f
o
r
th
e
m
id
d
le
1
0
0
0
f
r
am
e
s
in
t
h
e
m
o
v
ie
“
J
u
r
ass
ic
P
ar
k
”.
F
ig
u
r
e
12
r
ep
r
esen
ts
th
e
test
r
esu
lts
f
o
r
th
e
last
4
0
0
f
r
a
m
es i
n
t
h
e
m
o
v
ie
“
J
u
r
ass
ic
P
a
r
k
”.
Fig
u
r
e
13.
B
ar
C
h
ar
t Rep
r
esen
tatio
n
o
f
E
r
r
o
r
s
o
f
Var
io
u
s
W
i
n
d
o
w
S
izes
Fig
u
r
e
13.
s
h
o
w
s
th
e
er
r
o
r
c
o
m
p
ar
is
o
n
f
o
r
v
a
r
io
u
s
w
i
n
d
o
w
s
ize
s
.
Fro
m
ab
o
v
e
F
ig
u
r
e
it
ca
n
b
e
co
n
clu
d
ed
th
at,
n
et
w
o
r
k
g
iv
e
s
m
i
n
i
m
u
m
er
r
o
r
f
o
r
1
2
w
i
n
d
o
w
s
ize.
4
.
3
.
T
i
m
e
Dela
y
Neura
l N
et
w
o
rk
4
.3
.
1.
P
re
dict
io
n Re
s
ults f
o
r
Va
rio
us
Wind
o
w
Sizes
Fig
u
r
e
1
4
r
ep
r
esen
t
s
t
h
e
tr
ain
in
g
s
e
t
f
o
r
1
2
w
i
n
d
o
w
s
ize.
Su
c
h
f
ir
s
t
3
0
0
0
f
r
a
m
e
s
f
r
o
m
th
e
m
o
v
ie
J
u
r
ass
ic
P
ar
k
h
a
v
e
tak
e
n
as a
t
r
ain
in
g
s
et
to
tr
ai
n
th
e
n
e
u
r
al
n
et
w
o
r
k
.
Fig
u
r
e
1
4
.
T
r
ain
in
g
Seq
u
e
n
ce
f
o
r
1
2
W
in
d
o
w
S
ize
Fig
u
r
e
1
5
.
T
est Seq
u
en
ce
f
o
r
First 1
0
0
0
Fra
m
e
s
5
6
7
8
9
10
11
12
0
0
.
0
0
2
0
.
0
0
4
0
.
0
0
6
0
.
0
0
8
0
.
0
1
0
.
0
1
2
0
.
0
1
4
0
.
0
1
6
E
R
R
O
R
C
O
M
P
A
R
I
S
O
N
f
o
r
V
A
R
I
O
U
S
W
I
N
D
O
W
S
I
Z
E
S
W
I
N
D
O
W
S
I
Z
E
R
S
M
E
E
R
R
O
R
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8814
IJ
AA
S
Vo
l.
6
,
No
.
4
,
Dec
em
b
er
201
7
:
2
8
3
–
2
9
2
290
Fig
u
r
e
1
6
.
T
est Seq
u
en
ce
f
o
r
Mid
d
le
1
0
0
0
Fra
m
es
Fig
u
r
e
1
7
.
T
est Seq
u
en
ce
f
o
r
L
ast 4
0
0
Fra
m
e
s
Fig
u
r
e
15
to
17
r
ep
r
esen
ts
t
h
e
test
r
es
u
lt
s
.
T
h
e
f
r
a
m
e
s
eq
u
en
ce
s
f
r
o
m
t
h
e
s
a
m
e
m
o
v
ie
h
a
v
e
b
ee
n
tak
en
to
test
th
e
tr
ain
ed
n
et
w
o
r
k
.
Fig
u
r
e
15
r
ep
r
esen
ts
t
h
e
tes
t
r
es
u
lt
f
o
r
f
ir
s
t
1
0
0
0
f
r
a
m
es
i
n
t
h
e
m
o
v
ie
“
J
u
r
as
s
ic
P
ar
k
”.
T
h
e
R
ed
in
d
icate
s
p
r
ed
icted
f
r
a
m
es
w
h
ile
b
lu
e
in
d
icate
s
o
r
ig
i
n
al
f
r
a
m
e
s
eq
u
en
ce
.
F
ig
u
r
e
16
r
ep
r
esen
ts
th
e
te
s
t
r
esu
l
t
f
o
r
th
e
m
id
d
le
1
0
0
0
f
r
am
e
s
in
t
h
e
m
o
v
ie
“
J
u
r
ass
ic
P
ar
k
”.
F
ig
u
r
e
17
r
ep
r
esen
ts
th
e
test
r
esu
lts
f
o
r
th
e
last
4
0
0
f
r
a
m
es i
n
t
h
e
m
o
v
ie
“
J
u
r
ass
ic
P
a
r
k
”.
Fig
u
r
e
1
8
s
h
o
w
s
t
h
e
er
r
o
r
co
m
p
ar
i
s
o
n
.
Fro
m
ab
o
v
e
f
i
g
u
r
e
it
ca
n
b
e
co
n
c
lu
d
ed
t
h
at,
n
e
t
w
o
r
k
g
i
v
es
m
i
n
i
m
u
m
er
r
o
r
f
o
r
1
2
w
i
n
d
o
w
s
ize.
Fig
u
r
e
18.
B
ar
C
h
ar
t
R
ep
r
esen
tatio
n
o
f
E
r
r
o
r
s
o
f
Var
io
u
s
W
i
n
d
o
w
S
izes
5.
E
VA
L
UA
T
I
O
N
O
F
T
H
E
P
E
RF
O
RM
ANCES
O
F
3
NE
URAL AR
CH
I
T
E
C
T
UR
E
S
Fig
u
r
e
19
r
ep
r
esen
t
s
t
h
e
er
r
o
r
w
is
e
co
m
p
ar
is
o
n
o
f
th
r
ee
n
e
u
r
al
ar
ch
itectu
r
e
s
.
Fo
r
t
h
e
s
ai
d
p
r
o
b
lem
th
r
ee
n
e
u
r
al
n
et
w
o
r
k
ar
ch
ite
ctu
r
es
ar
e
co
n
s
id
er
ed
n
a
m
el
y
f
ee
d
-
f
o
r
w
ar
d
,
ca
s
ca
d
ed
f
ee
d
-
f
o
r
w
ar
d
an
d
t
i
m
e
d
elay
n
e
u
r
al
n
et
w
o
r
k
.
All
th
e
th
r
ee
ar
ch
itect
u
r
es
ar
e
tr
ain
e
d
an
d
test
ed
f
o
r
th
e
s
a
m
e
f
r
am
e
s
eq
u
e
n
ce
s
.
T
h
e
tr
ain
i
n
g
a
n
d
te
s
ti
n
g
d
ata
f
o
r
al
l
th
e
ar
c
h
itect
u
r
es
ar
e
s
a
m
e.
A
l
s
o
t
h
e
la
y
er
s
a
n
d
n
o
o
f
n
e
u
r
o
n
s
i
n
ea
ch
la
y
er
is
also
k
ep
t
co
n
s
tan
t.
T
h
e
la
y
er
w
i
s
e
an
d
n
eu
r
o
n
w
i
s
e
all
ar
ch
itect
u
r
es
ar
e
h
a
v
i
n
g
s
a
m
e
co
n
f
i
g
u
r
atio
n
.
T
h
e
tr
ain
i
n
g
p
ar
a
m
e
ter
s
,
s
u
ch
a
s
n
o
o
f
ep
o
ch
s
,
lear
n
i
n
g
r
ate
a
n
d
p
er
f
o
r
m
a
n
ce
g
o
al
ar
e
also
s
a
m
e
f
o
r
all
th
e
t
h
r
ee
ar
ch
itect
u
r
es.
A
l
l
th
e
ar
ch
itec
tu
r
es
ar
e
ev
alu
a
ted
in
ter
m
s
o
f
r
o
o
t
m
ea
n
s
q
u
ar
e
er
r
o
r
.
T
h
e
er
r
o
r
is
ca
lcu
lated
f
o
r
all
t
h
e
w
i
n
d
o
w
s
izes.
Fro
m
t
h
e
ab
o
v
e
f
i
g
u
r
e
,
it
i
s
clea
r
th
at,
f
o
r
ca
s
ca
d
ed
f
ee
d
-
f
o
r
w
ar
d
n
eu
r
al
n
et
w
o
r
k
t
h
e
er
r
o
r
p
er
w
in
d
o
w
s
ize
i
s
v
er
y
s
m
all
a
s
co
m
p
ar
ed
to
r
e
m
ai
n
i
n
g
ar
ch
i
tectu
r
e
s
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
f
ee
d
-
f
o
r
w
ar
d
n
et
w
o
r
k
is
s
u
p
er
io
r
to
th
at
o
f
ti
m
e
d
ela
y
n
eu
r
al
n
et
wo
r
k
.
Hen
ce
b
y
r
ef
er
r
i
n
g
th
e
F
ig
u
r
e
4
,
w
e
ca
n
co
n
cl
u
d
e
th
at,
th
e
o
u
t
o
f
th
e
t
h
r
ee
ar
ch
ite
ctu
r
es,
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
ca
s
ca
d
ed
f
ee
d
-
f
o
r
w
ar
d
n
et
w
o
r
k
is
h
ig
h
e
s
t
a
s
co
m
p
ar
ed
to
r
em
ai
n
i
n
g
t
w
o
ar
ch
itect
u
r
es.
A
ls
o
w
e
ca
n
s
a
y
th
at
t
h
e
n
eu
r
al
n
et
w
o
r
k
s
ca
n
b
e
u
s
ed
as th
e
e
f
f
ec
tiv
e
to
o
l f
o
r
v
id
eo
tr
af
f
ic
p
r
ed
ictio
n
.
5
6
7
8
9
10
11
12
0
0
.
0
0
2
0
.
0
0
4
0
.
0
0
6
0
.
0
0
8
0
.
0
1
0
.
0
1
2
0
.
0
1
4
0
.
0
1
6
E
R
R
O
R
C
O
M
P
A
R
I
S
O
N
f
o
r
V
A
R
I
O
U
S
W
I
N
D
O
W
S
I
Z
E
S
W
I
N
D
O
W
S
I
Z
E
R
S
M
E
E
R
R
O
R
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
AA
S
I
SS
N:
2252
-
8814
C
o
mp
a
r
a
tive
S
tu
d
y
o
f V
a
r
io
u
s
N
eu
r
a
l Netw
o
r
k
A
r
ch
itectu
r
e
s
...
(
J.
P
K
h
a
r
a
t
)
291
Fig
u
r
e
19
.
P
er
f
o
r
m
an
ce
E
v
alu
atio
n
o
f
A
r
ch
i
tectu
r
e
s
I
n
T
er
m
s
o
f
E
r
r
o
r
6.
CO
M
P
ARIS
O
N
O
F
E
URAL
AP
P
RAO
CH
WI
T
H
CO
N
VE
N
T
I
O
NA
L
AP
P
RAO
CH
Gen
er
all
y
t
h
e
co
n
v
e
n
tio
n
al
m
et
h
o
d
o
f
p
r
ed
ictio
n
is
Av
e
r
ag
in
g
m
e
th
o
d
.
T
h
is
is
ca
lle
d
as
lin
ea
r
m
et
h
o
d
o
f
p
r
ed
ictio
n
.
I
n
th
is
m
et
h
o
d
th
e
av
er
ag
e
o
f
p
r
ev
io
u
s
f
r
a
m
e
s
ize
s
is
ta
k
en
to
p
r
ed
ict
th
e
n
e
x
t
f
r
a
m
e
s
ize.
I
n
o
u
r
p
r
o
b
l
e
m
o
f
p
r
ed
ictio
n
,
w
e
h
ad
ta
k
e
n
t
h
is
l
in
ea
r
a
p
p
r
o
ac
h
f
o
r
co
m
p
ar
i
n
g
r
es
u
lt
s
an
d
f
o
r
ev
a
lu
at
in
g
th
e
p
er
f
o
r
m
a
n
ce
o
f
n
e
u
r
al
n
e
t
w
o
r
k
.
F
ir
s
t
f
o
r
av
er
ag
i
n
g
m
eth
o
d
,
w
e
co
n
s
id
er
ed
1
2
w
i
n
d
o
w
s
izes.
B
y
t
h
i
s
ap
p
r
o
ac
h
,
w
e
p
r
ed
icted
th
e
f
r
a
m
es.
Fig
u
r
e
20
in
d
icate
s
th
e
co
m
p
ar
is
o
n
b
et
w
ee
n
NN
ap
p
r
o
ac
h
an
d
Av
er
ag
i
n
g
ap
p
r
o
ac
h
.
B
lu
e
in
d
icate
s
th
e
er
r
o
r
b
etw
ee
n
ac
tu
al
f
r
am
es
a
n
d
p
r
ed
icted
f
r
am
es
b
y
NN
ap
p
r
o
ac
h
.
W
h
ile
R
ed
in
d
icate
s
th
e
er
r
o
r
b
et
w
ee
n
t
h
e
ac
t
u
al
f
r
a
m
e
s
an
d
p
r
ed
icted
f
r
am
e
s
b
y
a
v
er
ag
i
n
g
m
et
h
o
d
.
Fro
m
t
h
e
ab
o
v
e
f
i
g
u
r
e
,
it
is
clea
r
t
h
a
t
th
e
les
s
er
r
o
r
is
g
iv
e
n
b
y
n
eu
r
a
l n
et
w
o
r
k
as c
o
m
p
ar
ed
to
Av
e
r
ag
in
g
m
eth
o
d
er
r
o
r
o
u
tp
u
t.
Fig
u
r
e
20
.
E
r
r
o
r
C
o
m
p
ar
is
o
n
f
o
r
NN
A
p
p
r
o
ac
h
an
d
T
r
a
d
itio
n
al
A
p
p
r
o
ac
h
7.
CO
NCLU
SI
O
NS
T
r
af
f
ic
p
r
ed
ictio
n
m
o
d
els
u
s
in
g
n
e
u
r
al
n
et
w
o
r
k
s
ar
e
u
s
ef
u
l
t
o
o
ls
f
o
r
tr
af
f
ic
m
a
n
ag
e
m
e
n
t
al
g
o
r
ith
m
s
,
f
lo
w
co
n
tr
o
l
m
ec
h
a
n
is
m
s
,
co
n
g
e
s
tio
n
co
n
tr
o
l
s
ch
e
m
e
s
,
d
y
n
a
m
ic
b
an
d
w
id
th
al
lo
ca
tio
n
,
an
d
QOS
co
n
tr
o
l
o
f
liv
e
VB
R
v
id
eo
s
an
d
o
th
er
r
ea
l
-
ti
m
e
v
id
eo
ap
p
licatio
n
s
w
h
er
e
th
e
v
id
eo
s
tr
ea
m
is
n
o
t
k
n
o
w
n
in
ad
v
a
n
ce
.
W
h
ile
s
ev
e
r
al
p
ap
er
s
h
av
e
d
ea
lt
w
it
h
tr
af
f
ic
p
r
ed
ictio
n
m
o
d
els
th
at
ca
n
ca
p
tu
r
e
s
o
m
e
s
tati
s
t
ical
ch
ar
ac
ter
is
tic
s
o
f
th
e
tr
af
f
ic
an
d
th
e
i
n
h
er
en
t
n
o
n
-
s
tatio
n
ar
ities
a
n
d
n
o
n
li
n
ea
r
ities
as
s
o
ciate
d
w
it
h
MP
E
G
v
id
eo
tr
af
f
ic,
a
d
etailed
s
tu
d
y
o
f
d
i
f
f
er
en
t
n
eu
r
al
n
et
w
o
r
k
s
t
ec
h
n
iq
u
es
h
as
n
o
t b
ee
n
d
o
n
e.
I
n
th
is
w
o
r
k
,
w
e
h
av
e
e
x
a
m
in
ed
th
r
ee
d
if
f
er
en
t
n
e
u
r
al
n
et
wo
r
k
tech
n
iq
u
es
(
C
F,
FF
,
an
d
F
FT
D)
an
d
ev
alu
a
ted
th
e
m
in
ter
m
s
o
f
th
e
ir
p
er
f
o
r
m
a
n
ce
in
p
r
ed
ictin
g
MP
E
G
-
4
v
id
eo
tr
af
f
ic.
T
h
is
w
o
r
k
ca
n
b
e
co
n
clu
d
ed
with
t
h
e
f
o
llo
w
i
n
g
s
ta
te
m
e
n
ts
.
1.
W
e
h
av
e
tr
ied
m
a
n
y
co
n
f
i
g
u
r
a
tio
n
s
an
d
t
y
p
es
o
f
n
e
u
r
al
n
et
wo
r
k
s
f
o
r
v
id
eo
s
tr
ea
m
d
ata
p
r
e
d
ictio
n
.
Firs
t,
w
e
tr
ied
to
f
i
n
d
s
u
itab
le
n
et
w
o
r
k
co
n
f
i
g
u
r
atio
n
s
.
T
h
is
p
r
o
ce
s
s
led
u
s
to
th
e
n
e
t
w
o
r
k
ar
ch
itect
u
r
e
2
0
-
10
-
6
-
1.
0
500
1000
1500
2000
2500
3000
-1
0
1
2
x
1
0
4
0
500
1000
1500
2000
2500
3000
-1
-
0
.
5
0
0
.
5
1
x
1
0
4
E
R
R
O
R
B
E
T
W
E
E
N
N
N
O
U
T
P
U
T
A
N
D
A
C
T
U
A
L
F
R
A
M
E
S
E
R
R
O
R
B
E
T
W
E
E
N
A
V
G
O
U
T
P
U
T
A
N
D
A
C
T
U
A
L
F
R
A
M
E
S
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8814
IJ
AA
S
Vo
l.
6
,
No
.
4
,
Dec
em
b
er
201
7
:
2
8
3
–
2
9
2
292
2.
Fo
r
co
m
p
ar
is
o
n
p
u
r
p
o
s
es,
w
e
tr
ied
to
p
r
ed
ict
th
e
d
ata
u
s
i
n
g
v
ar
io
u
s
i
n
p
u
t
p
atter
n
s
.
W
e
ch
o
s
e
5
to
1
2
in
p
u
t
p
atter
n
s
.
Fro
m
f
i
g
s
.
6
.
5
,
7
.
5
,
8
.
5
,
th
e
b
est
s
u
itab
le
p
att
er
n
w
as
1
2
in
p
u
t
p
atter
n
s
.
I
n
o
r
d
er
to
m
ak
e
th
e
p
r
ed
ictio
n
m
o
r
e
ef
f
ec
ti
v
e,
it
is
p
o
s
s
ib
le
to
tak
e
also
th
e
ch
ar
ac
ter
o
f
th
e
ti
m
e
s
er
ie
s
in
to
ac
co
u
n
t.
Fo
r
o
u
r
d
ata,
ap
p
r
o
x
i
m
atel
y
ea
c
h
1
2th
p
atter
n
f
o
r
m
s
a
p
ea
k
(
in
o
th
er
w
o
r
d
s
,
t
h
e
d
is
ta
n
ce
o
f
t
h
e
co
n
s
ec
u
ti
v
e
p
ea
k
s
is
m
o
s
t
l
y
1
2
p
atter
n
s
)
.
T
h
is
is
w
h
y
f
o
r
1
2
in
p
u
t
p
atte
r
n
;
w
e
g
e
t
th
e
m
in
i
m
u
m
p
r
ed
ict
io
n
er
r
o
r
as
co
m
p
ar
ed
to
o
th
er
in
p
u
t p
atter
n
s
.
3.
W
e
h
av
e
e
v
alu
a
ted
th
r
ee
ar
c
h
itect
u
r
es.
T
h
e
b
est
p
er
f
o
r
m
a
n
ce
w
e
g
et
f
o
r
ca
s
ca
d
ed
f
ee
d
-
f
o
r
w
ar
d
an
d
f
ee
d
-
f
o
r
w
ar
d
n
eu
r
al
n
et
w
o
r
k
i
n
ter
m
s
o
f
er
r
o
r
m
ea
s
u
r
e.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
ti
m
e
d
ela
y
n
eu
r
al
n
et
w
o
r
k
is
f
air
l
y
lo
w
as
co
m
p
ar
ed
to
p
r
ev
io
u
s
t
w
o
ar
c
h
itect
u
r
es.
B
u
t
if
w
e
co
m
p
ar
e
in
ter
m
s
o
f
s
i
m
u
latio
n
t
i
m
e,
ti
m
e
d
ela
y
n
e
u
r
al
n
et
w
o
r
k
i
s
s
i
m
u
lated
in
le
s
s
ti
m
e
as c
o
m
p
ar
ed
to
p
r
ev
io
u
s
t
w
o
ar
ch
itect
u
r
es.
4.
B
y
co
n
s
id
er
i
n
g
all
ab
o
v
e
p
o
in
ts
w
e
ca
n
co
n
clu
d
e
th
at,
ac
c
u
r
ate
tr
af
f
ic
p
r
ed
ictio
n
u
s
in
g
n
e
u
r
al
n
et
w
o
r
k
s
is
in
d
ee
d
p
o
s
s
ib
le.
T
h
is
i
s
es
p
ec
iall
y
tr
u
e
f
o
r
MP
E
G
-
4
v
id
eo
tr
af
f
ic
w
h
ic
h
i
s
m
o
r
e
d
if
f
i
cu
lt
to
p
r
ed
ict
th
an
o
t
h
er
MP
E
G
v
id
eo
s
tan
d
ar
d
s
b
ec
au
s
e
it
is
b
u
r
s
tier
o
v
e
r
a
w
id
e
r
an
g
e
o
f
ti
m
e
s
ca
les
an
d
h
as
h
ig
h
er
d
eg
r
ee
o
f
s
elf
-
s
i
m
ilar
itie
s
.
8
.
T
RACK
S F
O
R
F
E
AA
T
U
RE
WO
RK
So
m
e
r
ec
o
m
m
en
d
atio
n
s
f
o
r
f
u
tu
r
e
w
o
r
k
ar
e:
1.
Use
o
f
m
o
r
e
t
h
an
o
n
e
m
o
d
el
f
o
r
m
u
lti
-
s
tep
-
ah
ea
d
p
r
ed
ictio
n
o
f
t
h
e
s
o
u
r
ce
v
id
eo
tr
af
f
ic.
T
h
is
r
eq
u
ir
e
s
th
e
d
esi
g
n
o
f
a
s
c
h
e
m
e
w
h
ich
s
w
itc
h
es
b
et
w
ee
n
t
h
e
p
r
ed
icti
o
n
s
m
o
d
el
s
d
ep
en
d
in
g
o
n
t
h
e
b
it
r
ate
o
f
t
h
e
v
id
eo
tr
af
f
ic.
2.
Desig
n
o
f
n
o
n
-
li
n
ea
r
p
r
ed
ictio
n
m
o
d
els
w
h
ic
h
ca
n
b
e
ad
a
p
ted
o
n
lin
e.
T
ill
n
o
w
r
e
s
ea
r
ch
er
s
h
av
e
u
s
ed
lin
ea
r
,
n
o
n
-
li
n
ea
r
an
d
ad
ap
ti
v
e
lin
ea
r
m
o
d
els
f
o
r
t
h
e
p
r
ed
ictio
n
o
f
MP
E
G
-
co
d
ed
v
id
eo
s
o
u
r
ce
tr
af
f
ic.
T
h
e
d
o
m
ain
n
o
n
-
li
n
ea
r
m
o
d
el
in
g
tec
h
n
iq
u
e
s
w
h
ic
h
ca
n
b
e
ad
ap
ted
o
n
lin
e
f
o
r
th
e
p
r
ed
ictio
n
o
f
MP
E
G
-
co
d
ed
v
id
eo
s
o
u
r
ce
tr
af
f
ic
h
a
s
n
o
t b
ee
n
ex
p
lo
r
ed
.
3.
Desig
n
o
f
a
co
n
tr
o
l
s
ch
e
m
e
f
o
r
ef
f
icie
n
t
d
eli
v
er
y
o
f
m
u
lti
m
e
d
ia
tr
af
f
ic
u
s
i
n
g
t
h
e
o
u
tp
u
t
o
f
th
e
e
m
p
ir
ical
m
o
d
el
s
d
escr
ib
ed
in
th
i
s
r
esea
r
ch
w
o
r
k
.
RE
F
E
R
ANC
E
S
[1
]
E.
P
.
Ra
th
g
e
b
,
“
P
o
li
c
in
g
o
f
Re
a
li
stic
Vb
r
V
id
e
o
T
ra
ff
ic
in
a
n
A
t
m
Ne
tw
o
rk
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
n
Dig
it
a
l
a
n
d
An
a
l
o
g
Co
mm
u
n
ica
ti
o
n
S
y
ste
ms
,
V
o
l
.
6
,
P
p
.
2
1
3
–
2
2
6
,
Oc
to
b
e
r
–
De
c
e
m
b
e
r
1
9
9
3
.
[2
]
W
.
E.
L
e
lan
d
,
M
.
S
.
T
a
q
q
u
,
W
.
W
il
li
n
g
e
r,
A
n
d
D.
V
.
W
il
so
n
,
“
On
th
e
S
e
lf
-
S
i
m
il
a
r
Na
tu
re
o
f
Et
h
e
rn
e
t
T
ra
f
f
i
c
(Ex
ten
d
e
d
V
e
rsio
n
),
”
Ie
e
e
/A
c
m T
ra
n
s.
Ne
two
rk
i
n
g
,
Vo
l.
2
,
P
p
.
1
–
1
5
,
F
e
b
ru
a
ry
1
9
9
4
.
[3
]
A
.
R.
Re
ib
m
a
n
A
n
d
A
.
W
.
Be
rg
e
r,
“
T
ra
ff
ic
De
s
c
rip
to
rs
F
o
r
V
b
r
V
i
d
e
o
T
e
lec
o
n
f
e
re
n
c
in
g
Ov
e
r
A
t
m
Ne
t
w
o
rk
s,”
Ie
e
e
/A
c
m T
ra
n
s.
Ne
two
rk
in
g
,
V
o
l
.
3
,
P
p
.
3
2
9
–
3
3
9
,
Ju
n
e
1
9
9
5
.
[4
]
M
.
G
ro
ss
g
lau
se
r,
S
.
Ke
sh
a
v
,
A
n
d
D.
T
se
,
“
R
c
b
r:
A
S
i
m
p
le
A
n
d
Ef
f
icie
n
t
S
e
rv
ice
F
o
r
M
u
lt
ip
le T
im
e
-
S
c
a
le T
ra
f
f
i
c
,
”
Ie
e
e
/A
c
m T
ra
n
s.
Ne
two
rk
in
g
,
De
c
e
m
b
e
r
1
9
9
7
.
[5
]
M
.
Kru
n
z
A
n
d
S
.
K.
T
rip
a
t
h
i,
“
O
n
T
h
e
C
h
a
r
a
c
ter
isti
c
s
Of
Vb
r
M
p
e
g
S
tre
a
ms
,
”
In
Pro
c
.
Acm
S
ig
me
trics
,
P
p
.
1
9
2
–
2
0
2
,
Ju
n
e
1
9
9
7
.
[6
]
M
.
G
a
rr
e
tt
A
n
d
W
.
W
il
li
n
g
e
r,
“
An
a
lys
is,
M
o
d
e
li
n
g
a
n
d
Ge
n
e
ra
t
io
n
o
f
S
e
lf
-
S
imil
a
r
Vb
r
Vi
d
e
o
T
r
a
ff
ic
,
”
In
Pro
c
.
Acm
S
ig
c
o
mm
,
S
e
p
tem
b
e
r
1
9
9
4
.
[7
]
M
.
Kru
n
z
A
n
d
S
.
K
.
T
rip
a
th
i,
“
On
T
h
e
Ch
a
r
a
c
ter
isti
c
s
o
f
V
b
r
M
p
e
g
S
tre
a
ms
,
”
In
Pro
c
.
Acm
S
i
g
me
trics
,
P
p
.
1
9
2
–
2
0
2
,
Ju
n
e
1
9
9
7
.
[8
]
A
d
e
l
A
b
d
e
n
n
o
u
r
“
Ev
a
lu
a
ti
o
n
o
f
Ne
u
ra
l
Ne
t
w
o
rk
Arc
h
it
e
c
tu
re
s
F
o
r
M
p
e
g
-
4
V
i
d
e
o
T
ra
ff
ic
P
re
d
ictio
n
”
,
Ie
e
e
T
ra
n
sa
c
ti
o
n
o
n
Bro
a
d
c
a
stin
g
,
Vo
l.
5
2
,
No
2
,
J
u
n
e
2
0
0
6
.
[9
]
P
.
Co
rtez
,
M
.
Rio
,
M
.
Ro
c
h
a
,
P
.
S
o
u
sa
,
In
tern
e
t
T
ra
ff
ic
F
o
re
c
a
st
in
g
Us
in
g
Ne
u
ra
l
Ne
t
w
o
rk
s,
In
tern
a
ti
o
n
a
l
Jo
in
t
Co
n
f
e
re
n
c
e
o
n
Ne
u
ra
l
Ne
tw
o
rk
s,
P
p
.
2
6
3
5
–
2
6
4
2
.
V
a
n
c
o
u
v
e
r,
Ca
n
a
d
a
,
2
0
0
6
.
[1
0
]
V
.
B.
Dh
a
rm
a
d
h
ik
a
ri,
J.
D.
G
a
v
a
d
e
,
A
n
Nn
A
p
p
ro
a
c
h
F
o
r
M
p
e
g
V
id
e
o
T
ra
ff
ic
P
re
d
icti
o
n
,
2
n
d
In
ter
n
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
S
o
f
twa
re
Tec
h
n
o
l
o
g
y
a
n
d
En
g
i
n
e
e
rin
g
,
P
p
.
V
1
-
57
–
V1
-
6
1
.
S
a
n
Ju
a
n
,
Us
a
,
2
0
1
0
.
[1
1
]
H.
F
e
n
g
,
Y.
S
h
u
,
S
tu
d
y
o
n
N
e
tw
o
rk
T
r
a
ff
ic
P
re
d
ictio
n
T
e
c
h
n
i
q
u
e
s,
In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
W
irele
ss
Co
m
m
u
n
ica
ti
o
n
s,
Ne
tw
o
rk
in
g
a
n
d
M
o
b
il
e
C
o
m
p
u
ti
n
g
,
P
p
.
1
0
4
1
–
1
0
4
4
.
W
u
h
a
n
,
Ch
in
a
,
2
0
0
5
.
[1
2
]
L
.
Ca
i,
J.
W
a
n
g
,
C.
W
a
n
g
,
L
.
Ha
n
,
A
No
v
e
l
F
o
rw
a
rd
in
g
A
l
g
o
rit
h
m
Ov
e
r
M
u
lt
ip
a
th
Ne
tw
o
r
k
,
In
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
Co
m
p
u
ter De
sig
n
a
n
d
A
p
p
li
c
a
ti
o
n
s,
P
p
.
V5
-
3
5
3
–
V5
-
3
5
7
.
Qi
n
h
u
a
n
g
d
a
o
,
C
h
i
n
a
,
2
0
1
0
.
[1
3
]
S
.
Ha
y
k
in
,
Ne
u
ra
l
Ne
tw
o
rk
s
-
A
Co
m
p
re
h
e
n
siv
e
F
o
u
n
d
a
ti
o
n
.
U
p
p
e
r
S
a
d
d
le Ri
v
e
r,
Ne
w
Je
rse
y
:
P
re
n
ti
c
e
Ha
ll
,
1
9
9
4
.
[1
4
]
M
p
e
g
-
4
A
n
d
H.
2
6
3
V
id
e
o
T
ra
c
e
s
F
o
r
Ne
tw
o
rk
P
e
rf
o
rm
a
n
c
e
Ev
a
lu
a
ti
o
n
,
Http
:/
/
Www
-
T
k
n
.
Ee.T
u
Be
rli
n
.
De
/Res
e
a
rc
h
/T
ra
c
e
/
T
ra
c
e
.
Htm
1
.
Evaluation Warning : The document was created with Spire.PDF for Python.