I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
201
7
,
p
p
.
3
5
7
0
~
3
5
7
7
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v7
i
6
.
pp
3
5
7
0
-
3577
3570
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
JE
C
E
Co
m
pa
ra
tive
Stu
dy
of Neural
Net
w
o
rk
s
A
lg
o
rith
ms
f
o
r Cloud
Co
m
pu
ting C
PU
Scheduling
G
ibet
T
a
ni H
icha
m
,
E
l A
m
r
a
ni Cha
k
er
,
E
la
a
cha
k
L
o
t
f
i
L
a
b
o
ra
to
r
y
o
f
In
f
o
rm
a
ti
c
s S
y
ste
m
s an
d
T
e
lec
o
m
m
u
n
ica
ti
o
n
s (L
IS
T)
,
A
b
d
e
lm
a
l
e
k
Essa
a
d
i
Un
iv
e
rsity
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
y
24
,
2
0
1
7
R
ev
i
s
ed
J
u
n
2
2
,
2
0
1
7
A
cc
ep
ted
J
u
l 1
0
,
2
0
1
7
Clo
u
d
Co
m
p
u
ti
n
g
is
th
e
m
o
st
p
o
w
e
r
f
u
l
c
o
m
p
u
ti
n
g
m
o
d
e
l
o
f
o
u
r
ti
m
e
.
W
h
il
e
th
e
m
a
jo
r
IT
p
ro
v
id
e
rs
a
n
d
c
o
n
su
m
e
rs
a
re
c
o
m
p
e
ti
n
g
to
e
x
p
lo
it
t
h
e
b
e
n
e
f
it
s
o
f
th
is
c
o
m
p
u
ti
n
g
m
o
d
e
l
in
o
r
d
e
r
to
th
riv
e
t
h
e
ir
p
ro
f
it
s,
m
o
st
o
f
th
e
c
l
o
u
d
c
o
m
p
u
ti
n
g
p
latf
o
rm
s
a
re
stil
l
b
u
i
l
t
o
n
o
p
e
ra
ti
n
g
sy
ste
m
s
th
a
t
u
se
s
b
a
sic
CP
U
(Co
re
P
r
o
c
e
ss
in
g
Un
it
)
sc
h
e
d
u
li
n
g
a
lg
o
rit
h
m
s
th
a
t
lac
k
s
th
e
i
n
telli
g
e
n
c
e
n
e
e
d
e
d
f
o
r
su
c
h
i
n
n
o
v
a
ti
v
e
c
o
m
p
u
ti
n
g
m
o
d
e
l.
Co
rre
sp
d
o
n
d
in
g
ly
,
th
is
p
a
p
e
r
p
re
se
n
ts
th
e
b
e
n
e
f
it
s
o
f
a
p
p
ly
in
g
A
rti
f
icia
l
N
e
u
ra
l
Ne
t
w
o
rk
s
a
l
g
o
r
it
h
m
s
in
re
g
a
rd
s
to
e
n
h
a
n
c
in
g
CP
U
sc
h
e
d
u
li
n
g
f
o
r
Clo
u
d
Co
m
p
u
ti
n
g
m
o
d
e
l.
F
u
rth
e
rm
o
re
,
a
se
t
o
f
c
h
a
ra
c
terist
ics
a
n
d
th
e
o
re
ti
c
a
l
m
e
tri
c
s
a
r
e
p
ro
p
o
se
d
f
o
r
th
e
sa
k
e
o
f
c
o
m
p
a
rin
g
th
e
d
iff
e
r
e
n
t
A
rti
f
icia
l
N
e
u
ra
l
Ne
t
w
o
rk
s
a
lg
o
rit
h
m
s
a
n
d
f
in
d
i
n
g
th
e
m
o
st
a
c
c
u
ra
te
a
lg
o
rit
h
m
f
o
r
Clo
u
d
Co
m
p
u
t
in
g
C
P
U
S
c
h
e
d
u
l
in
g
.
K
ey
w
o
r
d
:
C
lo
u
d
c
o
m
p
u
tin
g
C
P
U
s
ch
ed
u
lin
g
Neu
r
al
n
et
w
o
r
k
s
Co
p
y
rig
h
t
©
2
0
1
7
I
n
stit
u
te o
f
Ad
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
:
Gib
et
T
an
i H
ich
a
m
,
L
ab
o
r
ato
r
y
o
f
I
n
f
o
r
m
atics
S
y
s
te
m
s
a
n
d
T
elec
o
m
m
u
n
icat
io
n
s
(
L
I
ST
)
,
A
b
d
el
m
a
lek
E
s
s
aa
d
i U
n
iv
er
s
it
y
,
R
o
u
te
Z
iate
n
,
B
.
P
.
4
1
6
T
an
g
ier
,
Mo
r
o
cc
o
E
m
ail:
g
ib
et.
tan
i.
h
ic
h
a
m
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
C
P
U
(
C
o
r
e
P
r
o
ce
s
s
in
g
U
n
it)
s
ch
ed
u
li
n
g
is
t
h
e
p
r
o
ce
s
s
o
f
as
s
ig
n
in
g
co
m
p
u
te
r
e
s
o
u
r
ce
s
to
a
p
ar
ticu
lar
j
o
b
o
r
task
s
u
b
m
itted
f
o
r
ex
ec
u
tio
n
o
n
a
s
p
ec
if
ic
s
y
s
te
m
.
T
h
e
allo
ca
tio
n
o
f
co
m
p
u
te
r
eso
u
r
ce
s
f
o
llo
w
s
a
p
r
ed
ef
in
ed
s
ch
ed
u
lin
g
alg
o
r
it
h
m
o
n
t
h
e
s
y
s
te
m
(
g
e
n
er
all
y
d
ec
lar
ed
o
n
th
e
o
p
e
r
atin
g
s
y
s
te
m
k
er
n
el)
.
Up
to
th
is
m
o
m
en
t,
t
h
e
m
aj
o
r
ity
o
f
c
o
m
p
u
ter
s
y
s
te
m
s
ar
e
u
s
i
n
g
s
i
m
p
le
s
c
h
ed
u
li
n
g
alg
o
r
it
h
m
s
t
h
at
w
er
e
d
ef
i
n
ed
o
n
th
e
p
ast
2
0
y
ea
r
s
(
s
u
ch
as
First
C
o
m
e
Fir
s
t
Ser
v
ed
,
R
o
u
n
d
R
o
b
in
o
r
P
r
io
r
ity
s
ch
ed
u
li
n
g
)
an
d
th
a
t
s
t
ill
g
i
v
e
r
e
m
ar
k
ab
le
r
es
u
lt
s
f
o
r
d
ail
y
u
s
e
b
ases
w
it
h
s
o
m
e
m
i
n
o
r
m
o
d
i
f
icat
io
n
s
.
Ho
w
ev
er
,
w
i
t
h
t
h
e
e
x
tr
ao
r
d
in
ar
y
ad
v
an
ce
o
f
co
m
p
u
ter
e
n
g
i
n
ee
r
in
g
,
th
e
m
aj
o
r
s
h
i
f
t
o
f
t
h
e
wo
r
ld
in
to
t
h
e
i
n
ter
n
e
t
a
n
d
w
it
h
t
h
e
b
ir
t
h
o
f
C
lo
u
d
C
o
m
p
u
tin
g
,
t
h
ese
b
asic CP
U
s
ch
ed
u
li
n
g
alg
o
r
it
h
m
s
ar
e
s
tar
ti
n
g
to
b
ec
o
m
e
d
ep
r
ec
ated
.
T
h
e
m
aj
o
r
p
r
o
b
lem
w
it
h
th
e
ex
is
t
in
g
C
P
U
s
ch
ed
u
li
n
g
al
g
o
r
ith
m
s
is
t
h
e
lo
w
p
er
f
o
r
m
a
n
ce
r
elate
d
to
th
e
ti
m
e
-
co
n
s
u
m
in
g
j
o
b
s
th
at
co
m
e
s
w
it
h
th
e
C
lo
u
d
m
o
d
el
o
f
co
m
p
u
ti
n
g
(
Of
f
er
in
g
I
T
r
es
o
u
r
ce
s
as
s
er
v
ice
s
:
I
n
f
r
as
tr
u
ct
u
r
es,
p
lat
f
o
r
m
s
an
d
A
p
p
licatio
n
s
)
,
th
er
e
f
o
r
e
th
e
y
p
r
o
d
u
ce
a
p
o
o
r
r
esp
o
n
s
e
ti
m
e
th
at
i
s
n
o
t
s
u
itab
le
f
o
r
lar
g
e
-
s
ca
le
en
v
ir
o
n
m
e
n
ts
.
On
th
e
s
a
m
e
co
n
tex
t,
m
an
y
i
n
v
esti
g
ato
r
s
ar
e
p
r
o
m
o
ti
n
g
A
r
ti
f
icial
Ne
u
r
al
Net
w
o
r
k
s
(
A
NN)
as
a
s
o
lu
tio
n
to
o
p
tim
ize
th
e
e
x
is
tin
g
al
g
o
r
ith
m
s
,
t
h
u
s
as
s
is
tin
g
C
lo
u
d
C
o
m
p
u
ti
n
g
p
r
o
v
id
er
s
an
d
u
s
er
s
m
a
k
e
in
telli
g
en
t
d
ec
is
io
n
s
r
eg
ar
d
i
n
g
t
h
eir
i
n
v
est
m
e
n
t
s
o
n
t
h
is
o
u
t
s
ta
n
d
in
g
t
ec
h
n
o
lo
g
y
.
Sev
er
a
l
n
eu
r
al
n
et
w
o
r
k
s
al
g
o
r
ith
m
s
ar
e
av
ailab
le
an
d
co
m
p
ar
in
g
t
h
e
m
in
t
h
e
ai
m
o
f
ch
o
o
s
in
g
t
h
e
b
est
alg
o
r
ith
m
f
o
r
C
P
U
s
c
h
ed
u
li
n
g
i
s
a
co
m
p
lica
ted
m
i
s
s
io
n
g
i
v
in
g
th
e
v
a
s
t
ap
p
licatio
n
f
i
eld
s
o
f
n
e
u
r
al
n
et
wo
r
k
s
.
I
n
t
h
is
p
ap
er
,
a
s
et
o
f
p
r
ac
tical
f
ea
t
u
r
es
h
a
s
b
ee
n
co
n
s
id
er
ed
to
ass
ess
a
n
d
ev
alu
ate
t
h
e
ex
is
ti
n
g
n
e
u
r
al
n
et
w
o
r
k
s
alg
o
r
it
h
m
s
an
d
f
o
r
e
m
o
s
t c
h
o
o
s
i
n
g
t
h
e
m
o
s
t a
p
p
r
o
p
r
iate
alg
o
r
ith
m
f
o
r
C
l
o
u
d
C
o
m
p
u
t
in
g
C
P
U
s
c
h
ed
u
li
n
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
C
o
mp
a
r
a
tive
S
tu
d
y
o
f Neu
r
a
l
N
etw
o
r
k
s
A
lg
o
r
ith
ms fo
r
C
lo
u
d
C
o
mp
u
tin
g
C
P
U
S
ch
e
d
u
lin
g
(
Gib
et
Ta
n
i H
)
3571
2.
CL
O
UD
CO
M
P
UT
I
N
G
P
R
E
S
E
NT
A
T
I
O
N
C
lo
u
d
C
o
m
p
u
ti
n
g
i
s
a
m
o
d
el
f
o
r
en
ab
lin
g
u
b
iq
u
ito
u
s
,
o
n
-
d
e
m
an
d
n
et
w
o
r
k
ac
ce
s
s
to
a
s
h
a
r
ed
p
o
o
l
o
f
co
n
f
i
g
u
r
ab
le
co
m
p
u
t
in
g
r
eso
u
r
ce
s
[
4
]
[
5
]
.
B
y
s
t
u
d
y
i
n
g
t
h
is
n
e
w
s
er
v
ice
d
eli
v
er
y
m
o
d
el
o
r
ig
in
ates
t
h
e
ch
alle
n
g
e
o
f
m
a
n
ag
in
g
h
u
n
d
r
ed
s
o
f
t
h
o
u
s
a
n
d
s
o
f
u
s
er
s
an
d
ap
p
licatio
n
s
r
eq
u
est
s
.
T
h
er
ef
o
r
e,
a
C
lo
u
d
C
o
m
p
u
tin
g
p
r
o
v
id
er
s
h
o
u
ld
co
n
s
id
er
in
te
lli
g
en
t
in
f
r
astr
u
c
tu
r
e
d
ep
lo
y
m
e
n
t
i
n
o
r
d
er
to
estab
lis
h
a
C
lo
u
d
C
o
m
p
u
tin
g
o
f
f
er
,
w
h
ic
h
in
s
u
r
es
tr
an
s
p
ar
e
n
c
y
,
s
ca
lab
ilit
y
,
s
e
cu
r
it
y
a
n
d
f
o
r
e
m
o
s
t
ce
ler
it
y
.
A
C
lo
u
d
C
o
m
p
u
tin
g
o
f
f
er
r
an
g
e
f
r
o
m
o
f
f
er
in
g
a
n
e
n
d
u
s
er
a
s
p
ec
i
f
ic
I
T
in
f
r
astr
u
ctu
r
e
(
s
to
r
ag
e,
s
er
v
er
s
,
n
et
w
o
r
k
…)
,
to
p
r
o
p
o
s
in
g
co
m
p
lica
ted
ap
p
licatio
n
an
d
s
o
f
t
w
ar
e
s
o
l
u
tio
n
s
(
C
R
M,
E
R
P
…)
an
d
all
o
f
th
is
is
o
r
g
an
ized
o
n
a
la
y
er
ed
ar
ch
itect
u
r
e
(
Fig
u
r
e
1
)
.
Fig
u
r
e
1
.
C
lo
u
d
C
o
m
p
u
ti
n
g
L
a
y
er
ed
A
r
ch
itect
u
r
e
an
d
Deli
v
er
y
Mo
d
el
On
e
o
f
th
e
ce
n
tr
al
clo
u
d
p
r
o
v
id
er
s
’
o
b
j
ec
tiv
es
is
th
e
p
r
o
v
is
io
n
in
g
o
f
p
h
y
s
ical
r
eso
u
r
ce
s
f
o
r
u
s
er
s
o
r
a
s
p
ec
if
ic
ap
p
licatio
n
.
T
h
u
s
,
a
clo
u
d
p
r
o
v
id
er
s
h
o
u
ld
s
e
lect
an
d
co
n
tr
o
l
th
e
a
llo
ca
tio
n
o
f
th
e
co
r
r
ec
t
r
eso
u
r
ce
w
h
et
h
er
a
clo
u
d
u
s
er
r
eq
u
est
it
as
a
s
er
v
ice
(
I
aa
S)
o
r
a
clo
u
d
ap
p
licatio
n
o
f
th
e
h
i
g
h
er
la
y
e
r
s
n
ee
d
s
it
(
P
aa
S
o
r
SaaS)
.
3.
NE
URA
L
N
E
T
WO
RK
S AN
D
ARTI
F
I
CI
AL
I
NT
E
L
L
I
G
E
NC
E
3
.
1
.
O
v
er
v
ie
w
A
r
ti
f
icial
Neu
r
al
Net
w
o
r
k
s
(
ANN)
is
an
i
n
f
o
r
m
atio
n
-
p
r
o
ce
s
s
in
g
p
ar
ad
ig
m
th
a
t
s
i
m
u
lates
t
h
e
h
u
m
a
n
b
r
ain
.
I
t
w
as
d
esi
g
n
ed
to
m
i
m
ic
t
h
e
w
a
y
t
h
e
h
u
m
an
b
r
ain
ex
ec
u
te
s
a
s
p
ec
if
ic
ta
s
k
o
r
f
u
n
ct
io
n
[
6
]
[
7
]
.
T
h
is
k
in
d
o
f
n
et
w
o
r
k
s
“Fi
g
u
r
e
2
”
i
s
co
m
p
o
s
ed
o
f
s
ev
er
al
ca
lc
u
la
tio
n
s
u
n
ites
ca
lled
n
e
u
r
o
n
s
,
wh
ich
ar
e
co
m
b
in
e
d
in
la
y
er
s
an
d
o
p
er
atin
g
i
n
p
ar
allel.
T
h
e
i
n
f
o
r
m
atio
n
w
ill
b
e
p
r
o
p
ag
ated
la
y
er
to
la
y
er
,
f
r
o
m
t
h
e
in
p
u
t
la
y
er
to
th
e
o
u
tp
u
t
la
y
er
.
T
h
e
A
NNs
h
av
e
t
h
e
ab
ilit
y
to
s
to
r
e
e
m
p
ir
ic
al
k
n
o
w
led
g
e
a
n
d
m
a
k
e
i
t
av
a
i
lab
le
f
o
r
th
e
u
s
er
s
.
T
h
e
k
n
o
w
led
g
e
o
f
t
h
e
n
et
w
o
r
k
w
il
l
b
e
s
to
r
ed
in
s
y
n
ap
tic
w
ei
g
h
ts
,
o
b
tai
n
ed
b
y
t
h
e
p
r
o
ce
s
s
o
f
ad
ap
tatio
n
o
r
lear
n
in
g
.
Fig
u
r
e
2
.
A
r
ti
f
icial
n
eu
r
al
n
et
w
o
r
k
B
ased
o
n
th
e
w
e
ig
h
t
s
an
d
tr
a
n
s
f
er
f
u
n
ctio
n
s
[
7
]
,
th
e
ac
t
iv
a
tio
n
v
a
lu
e
is
p
as
s
ed
f
r
o
m
n
o
d
e
to
n
o
d
e
.
E
ac
h
n
o
d
e
s
u
m
s
t
h
e
ac
tiv
at
io
n
v
al
u
es
it
r
ec
eiv
e
s
,
an
d
th
e
n
m
o
d
i
f
ie
s
th
e
v
a
lu
e
b
ased
o
n
it
s
tr
an
s
f
er
f
u
n
ct
io
n
.
T
h
e
ac
tiv
atio
n
p
r
o
ce
d
u
r
e
f
o
ll
o
w
s
a
f
ee
d
f
o
r
w
ar
d
p
r
o
ce
s
s
a
n
d
th
e
d
i
f
f
er
en
ce
b
et
w
ee
n
t
h
e
p
r
ed
icted
v
alu
e
an
d
th
e
ac
tu
al
v
al
u
e
(
er
r
o
r
)
w
ill
b
e
p
r
o
p
ag
ated
b
ac
k
w
ar
d
b
y
a
p
p
o
r
tio
n
in
g
t
h
e
m
to
ea
ch
n
o
d
e's
w
ei
g
h
ts
ac
co
r
d
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
5
7
0
–
3
5
7
7
3572
to
th
e
a
m
o
u
n
t
o
f
t
h
e
er
r
o
r
th
e
n
o
d
e
i
s
r
esp
o
n
s
ib
le
f
o
r
(
e.
g
.
,
g
r
ad
ie
n
t
d
esce
n
t
al
g
o
r
ith
m
[
8
]
)
,
as
s
h
o
w
n
i
n
Fig
u
r
e
3
.
Fig
u
r
e
3
.
Feed
f
o
r
w
ar
d
in
p
u
t d
ata
an
d
b
ac
k
w
ar
d
er
r
o
r
p
r
o
p
ag
atio
n
3
.
2
.
Act
iv
a
t
io
n
F
un
ct
io
n
T
h
e
A
cti
v
atio
n
f
u
n
ct
io
n
[
8
]
tr
an
s
late
s
t
h
e
in
p
u
t
s
i
g
n
al
s
to
o
u
tp
u
t
s
i
g
n
al.
T
h
er
e
ar
e
s
e
v
er
a
l
k
i
n
d
s
o
f
ac
tiv
atio
n
f
u
n
ctio
n
s
: U
n
it st
ep
,
Sig
m
o
id
,
Ga
u
s
s
ian
,
etc.
(
Fi
g
u
r
e
4
)
.
Fig
u
r
e
4
.
A
cti
v
atio
n
f
u
n
c
tio
n
s
Un
it st
ep
,
Si
g
m
o
id
,
an
d
Gau
s
s
ian
3
.
3
.
T
y
pes
o
f
Art
if
icia
l N
eur
a
l N
et
wo
rk
s
A
r
ti
f
icial
Neu
r
al
Net
w
o
r
k
s
[
7
]
[
8
]
ar
e
g
en
er
all
y
clas
s
if
ied
in
to
f
ee
d
-
f
o
r
w
ar
d
a
n
d
f
ee
d
b
ac
k
n
et
w
o
r
k
s
.
T
h
e
Feed
-
f
o
r
w
ar
d
[
7
]
n
et
w
o
r
k
is
a
n
o
n
-
r
ec
u
r
r
en
t
n
et
w
o
r
k
,
wh
ich
co
n
tai
n
s
i
n
p
u
t
s
,
o
u
tp
u
ts
,
an
d
h
id
d
en
la
y
er
s
;
th
e
s
i
g
n
a
ls
ca
n
o
n
l
y
tr
a
v
el
i
n
o
n
e
d
ir
ec
tio
n
.
I
n
p
u
t
d
ata
i
s
p
a
s
s
ed
o
n
to
a
la
y
er
o
f
p
r
o
ce
s
s
i
n
g
ele
m
en
t
s
w
h
er
e
it
p
er
f
o
r
m
s
ca
lc
u
latio
n
s
.
I
t in
cl
u
d
es P
er
ce
p
tr
o
n
an
d
R
ad
ial
B
as
is
Fu
n
ctio
n
n
e
t
w
o
r
k
s
.
Feed
-
f
o
r
w
ar
d
n
et
w
o
r
k
s
ar
e
u
s
ed
o
f
te
n
in
d
ata
m
in
in
g
.
M
u
lti
-
la
y
er
[
7
]
P
e
r
ce
p
tr
o
n
“
Fi
g
u
r
e
5
”
is
o
n
e
o
f
th
e
f
ee
d
-
f
o
r
w
ar
d
n
et
w
o
r
k
s
;
it
h
a
s
th
e
s
a
m
e
s
tr
u
ctu
r
e
o
f
a
s
in
g
le
la
y
er
P
er
ce
p
tr
o
n
w
i
th
o
n
e
o
r
m
o
r
e
h
id
d
en
la
y
er
s
.
T
h
e
lear
n
in
g
al
g
o
r
ith
m
u
s
ed
in
t
h
is
n
et
w
o
r
k
is
t
h
e
b
ac
k
p
r
o
p
ag
atio
n
[
9
]
.
I
t
co
n
s
is
t
s
o
f
t
wo
p
h
ases
:
t
h
e
f
o
r
w
ar
d
p
h
ase
wh
er
e
th
e
ac
t
iv
at
io
n
s
ar
e
p
r
o
p
ag
ated
f
r
o
m
t
h
e
i
n
p
u
t
to
t
h
e
o
u
tp
u
t
la
y
er
,
a
n
d
t
h
e
b
ac
k
w
ar
d
p
h
ase,
w
h
er
e
th
e
er
r
o
r
b
etw
ee
n
t
h
e
o
b
s
er
v
ed
ac
tu
al
a
n
d
t
h
e
r
eq
u
ested
n
o
m
i
n
al
v
al
u
e
i
n
t
h
e
o
u
tp
u
t
la
y
er
is
p
r
o
p
ag
ated
b
ac
k
w
ar
d
s
i
n
o
r
d
er
to
m
o
d
i
f
y
th
e
w
e
ig
h
ts
a
n
d
b
ias v
alu
es.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
C
o
mp
a
r
a
tive
S
tu
d
y
o
f Neu
r
a
l
N
etw
o
r
k
s
A
lg
o
r
ith
ms fo
r
C
lo
u
d
C
o
mp
u
tin
g
C
P
U
S
ch
e
d
u
lin
g
(
Gib
et
Ta
n
i H
)
3573
Fig
u
r
e
5
.
Mu
lti
–
la
y
er
P
er
ce
p
tr
o
n
B
ac
k
w
ar
d
p
r
o
p
ag
atio
n
:
P
r
o
p
a
g
ates
t
h
e
er
r
o
r
s
b
ac
k
w
ar
d
b
y
ap
p
o
r
tio
n
in
g
th
e
m
to
ea
ch
u
n
i
t
ac
co
r
d
in
g
to
th
e
a
m
o
u
n
t o
f
th
e
er
r
o
r
ea
ch
u
n
it i
s
r
esp
o
n
s
ib
le
f
o
r
,
s
ee
Fig
u
r
e
6
.
Fig
u
r
e
6
.
E
r
r
o
r
p
r
o
p
ag
atio
n
T
h
e
Feed
-
b
ac
k
[
1
0
]
n
e
t
w
o
r
k
h
as
f
ee
d
-
b
ac
k
p
at
h
s
,
m
ea
n
i
n
g
th
e
y
ca
n
h
a
v
e
s
ig
n
al
s
tr
a
v
eli
n
g
i
n
b
o
th
d
ir
ec
tio
n
s
u
s
i
n
g
lo
o
p
s
.
A
l
l
p
o
s
s
ib
le
co
n
n
ec
t
io
n
s
b
et
w
ee
n
n
e
u
r
o
n
s
ar
e
allo
w
ed
.
S
in
ce
lo
o
p
s
ar
e
p
r
ese
n
t
i
n
th
is
t
y
p
e
o
f
n
et
w
o
r
k
s
,
it
b
ec
o
m
es
a
n
o
n
-
lin
ea
r
d
y
n
a
m
ic
s
y
s
te
m
,
w
h
ic
h
c
h
an
g
es
co
n
ti
n
u
o
u
s
l
y
u
n
til
i
t
r
ea
ch
es
a
s
tate
o
f
eq
u
ilib
r
i
u
m
.
Feed
-
b
ac
k
n
et
w
o
r
k
s
ar
e
o
f
te
n
u
s
ed
i
n
ass
o
ciati
v
e
m
e
m
o
r
ies
an
d
o
p
ti
m
izatio
n
p
r
o
b
le
m
s
w
h
er
e
th
e
n
et
w
o
r
k
lo
o
k
s
f
o
r
t
h
e
b
est ar
r
an
g
e
m
e
n
t o
f
i
n
ter
co
n
n
ec
ted
f
ac
to
r
s
.
3
.
4
.
T
ra
ini
ng
T
ec
hn
iq
ues
T
r
ain
in
g
tech
n
iq
u
e
s
o
r
lear
n
i
n
g
al
g
o
r
ith
m
s
h
a
v
e
a
s
ig
n
i
f
i
ca
n
t
i
m
p
ac
t
o
n
th
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
n
eu
r
al
n
et
w
o
r
k
.
T
h
e
ch
o
ice
o
f
a
s
u
itab
le
lear
n
i
n
g
alg
o
r
i
th
m
i
s
th
er
e
f
o
r
e
ap
p
licatio
n
an
d
in
f
r
as
tr
u
ct
u
r
e
d
ep
en
d
en
t.
T
h
er
e
ar
e
v
ar
ieties
o
f
lear
n
i
n
g
a
lg
o
r
it
h
m
s
t
h
at
ca
n
b
e
u
s
ed
to
tr
ai
n
a
n
e
u
r
al
n
et
w
o
r
k
,
b
elo
w
is
t
h
e
d
escr
ip
tio
n
o
f
s
o
m
e
al
g
o
r
ith
m
s
th
at
w
i
ll b
e
u
s
ed
in
t
h
i
s
co
m
p
ar
ativ
e
s
tu
d
y
.
B
a
ck
-
pro
pa
g
a
t
io
n
:
an
ab
b
r
ev
iatio
n
o
f
b
ac
k
w
ar
d
p
r
o
p
ag
atio
n
o
f
er
r
o
r
alg
o
r
it
h
m
[
1
2
]
w
as
o
r
ig
in
all
y
in
tr
o
d
u
ce
d
in
th
e
1
9
7
0
s
.
I
t
is
a
m
et
h
o
d
o
f
tr
ain
in
g
ar
tific
ia
l
n
eu
r
al
n
et
w
o
r
k
s
b
ased
o
n
th
e
g
r
ad
ien
t
d
escen
t
[
1
3
]
,
o
n
e
o
f
t
h
e
o
p
ti
m
izatio
n
m
et
h
o
d
s
.
I
t
ca
lcu
la
tes
th
e
g
r
ad
ien
t
o
f
a
lo
s
s
f
u
n
ctio
n
w
it
h
r
esp
ec
t
to
all
th
e
w
ei
g
h
ts
i
n
t
h
e
cu
r
r
en
t
n
et
w
o
r
k
.
T
h
e
alg
o
r
ith
m
i
s
d
escr
ib
ed
b
elo
w
:
T
ab
le
1
.
B
ac
k
-
P
r
o
p
ag
atio
n
T
r
ain
i
n
g
al
g
o
r
ith
m
1.
I
n
i
t
i
a
l
i
z
e
w
e
i
g
h
t
s t
o
smal
l
r
a
n
d
o
m v
a
l
u
e
s
2.
C
h
o
o
se
i
n
p
u
t
p
a
t
t
e
r
n
3.
P
r
o
p
a
g
a
t
e
si
g
n
a
l
f
o
r
w
a
r
d
t
h
r
o
u
g
h
n
e
t
w
o
r
k
4.
D
e
t
e
r
mi
n
e
Er
r
o
r
(
E)
a
n
d
p
r
o
p
a
g
a
t
e
i
t
b
a
c
k
w
a
r
d
s
t
h
r
o
u
g
h
n
e
t
w
o
r
k
t
o
a
ss
i
g
n
c
r
e
d
i
t
t
o
e
a
c
h
u
n
i
t
5.
U
p
d
a
t
e
w
e
i
g
h
t
b
y
me
a
n
s g
r
a
d
i
e
n
t
d
e
sce
n
t
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
5
7
0
–
3
5
7
7
3574
B
ac
k
p
r
o
p
ag
atio
n
ac
tio
n
ca
n
c
au
s
e
c
h
a
n
g
e
s
i
n
th
e
w
ei
g
h
t
o
f
th
e
p
r
es
y
n
ap
tic
co
n
n
ec
t
io
n
s
,
th
er
e
is
n
o
s
i
m
p
le
m
ec
h
a
n
is
m
f
o
r
a
n
er
r
o
r
s
ig
n
al
to
p
r
o
p
ag
ate
t
h
r
o
u
g
h
m
u
ltip
le
la
y
er
s
n
et
w
o
r
k
,
a
n
d
it
i
s
a
m
o
n
g
th
e
d
is
ad
v
an
ta
g
es o
f
th
is
lear
n
in
g
m
et
h
o
d
.
Resili
ent
P
ro
pa
g
a
t
io
n
:
Hein
r
ich
B
r
au
n
cr
ea
ted
r
esil
ie
n
t
p
r
o
p
ag
atio
n
“Rp
r
o
p
”,
an
ab
b
r
e
v
iatio
n
o
f
r
esil
ien
t
b
ac
k
-
p
r
o
p
ag
atio
n
,
i
n
1
9
9
2
[
1
4
]
.
I
t
is
a
lear
n
i
n
g
h
eu
r
is
tic
f
o
r
s
u
p
er
v
is
ed
lear
n
i
n
g
i
n
f
ee
d
-
f
o
r
w
ar
d
ar
tif
icial
n
e
u
r
al
n
et
w
o
r
k
s
.
“Rp
r
o
p
”
i
s
co
n
s
id
er
ed
th
e
b
est al
g
o
r
ith
m
,
m
ea
s
u
r
ed
in
ter
m
s
o
f
co
n
v
er
g
e
n
ce
s
p
ee
d
,
ac
cu
r
ac
y
a
n
d
r
o
b
u
s
tn
es
s
w
it
h
r
esp
ec
t to
tr
ain
in
g
p
ar
a
m
eter
s
[
1
6
]
.
“Rp
r
o
p
”
is
s
i
m
i
lar
to
t
h
e
b
ac
k
-
p
r
o
p
ag
atio
n
al
g
o
r
ith
m
.
Ho
w
e
v
er
,
it
h
a
s
t
w
o
m
ain
ad
v
a
n
tag
e
s
o
v
er
b
ac
k
p
r
o
p
ag
atio
n
:
Tr
ain
in
g
w
it
h
“
R
p
r
o
p
”
is
o
f
te
n
f
a
s
ter
th
a
n
tr
ain
in
g
w
it
h
b
ac
k
p
r
o
p
ag
atio
n
.
“Rp
r
o
p
”
d
o
es
n
o
t
r
eq
u
ir
e
th
e
s
p
ec
if
icatio
n
o
f
an
y
f
r
ee
p
ar
am
eter
v
al
u
es,
as
o
p
p
o
s
ed
to
b
ac
k
p
r
o
p
ag
atio
n
th
at
n
ee
d
s
v
al
u
es
f
o
r
th
e
lear
n
i
n
g
r
ate.
T
h
e
m
ai
n
d
is
ad
v
a
n
ta
g
e
o
f
“
R
p
r
o
p
”
is
th
at
it
is
a
m
o
r
e
co
m
p
lex
al
g
o
r
ith
m
to
i
m
p
le
m
e
n
t
th
an
b
ac
k
p
r
o
p
ag
atio
n
.
G
enet
ic
a
lg
o
rit
h
m
t
ra
ini
ng
:
T
h
e
Gen
etic
alg
o
r
it
h
m
s
[
1
6
]
ar
e
alg
o
r
ith
m
s
f
o
r
o
p
ti
m
iz
atio
n
an
d
lear
n
in
g
b
ased
o
n
s
e
v
er
al
f
ea
t
u
r
es
o
f
n
a
tu
r
al
s
elec
tio
n
.
T
h
ey
ca
n
also
b
e
u
s
ed
f
o
r
tr
ain
in
g
o
f
ar
tific
ial
n
eu
r
a
l
n
et
w
o
r
k
.
T
h
e
d
esig
n
o
f
t
h
e
alg
o
r
ith
m
w
as
in
s
p
ir
ed
b
y
o
b
s
er
v
atio
n
o
f
n
at
u
r
al
ev
o
l
u
tio
n
p
r
o
ce
s
s
.
T
h
e
g
en
etic
alg
o
r
ith
m
p
er
f
o
r
m
s
s
e
v
er
al
o
p
er
atio
n
s
in
c
lu
d
i
n
g
[
1
7
]
:
T
ab
le
2
.
Gen
etic
tr
ain
in
g
al
g
o
r
ith
m
1.
R
a
n
d
o
m
i
n
i
t
i
a
l
i
z
a
t
i
o
n
o
f
t
h
e
p
r
e
l
i
mi
n
a
r
y
p
o
p
u
l
a
t
i
o
n
.
2.
In
-
l
o
o
p
e
v
a
l
u
a
t
i
o
n
o
f
e
v
e
r
y
c
h
r
o
mo
so
me
b
y
me
a
su
r
i
n
g
i
t
s f
i
t
n
e
ss.
3.
C
o
mp
a
r
i
so
n
w
i
t
h
t
h
e
m
i
n
i
mal
d
e
si
r
e
d
f
i
t
n
e
ss.
4.
S
e
l
e
c
t
i
o
n
o
f
t
h
e
f
i
t
t
e
st
s
u
b
se
t
o
f
c
h
r
o
mo
so
me
s.
5.
P
e
r
f
o
r
m c
r
o
ssi
n
g
-
o
v
e
r
,
w
h
i
c
h
i
s e
x
c
h
a
n
g
e
o
f
f
e
a
t
u
r
e
s fr
o
m t
h
e
se
l
e
c
t
e
d
s
u
b
se
t
o
f
c
h
r
o
mo
so
me
s.
6.
I
n
t
r
o
d
u
c
e
m
u
t
a
t
i
o
n
s,
w
h
i
c
h
a
r
e
r
a
n
d
o
m c
h
a
n
g
e
s a
p
p
l
i
e
d
t
o
r
a
n
d
o
ml
y
c
h
o
se
n
f
e
a
t
u
r
e
s o
f
t
h
e
c
h
r
o
mo
so
me
s.
7.
R
e
t
u
r
n
t
o
t
h
e
2
n
d
p
o
i
n
t
.
Du
r
in
g
tr
ai
n
in
g
p
r
o
ce
s
s
,
ev
er
y
c
h
r
o
m
o
s
o
m
e
o
n
t
h
e
g
e
n
etic
al
g
o
r
ith
m
e
v
o
l
v
es
f
r
o
m
all
t
h
e
co
n
n
ec
tio
n
w
eig
h
t
s
f
r
o
m
th
e
a
r
tif
icial
n
eu
r
al
n
et
w
o
r
k
.
O
t
her
t
ra
ini
ng
m
et
ho
d
s
:
T
h
e
r
e
ar
e
o
th
er
tr
ai
n
i
n
g
m
e
th
o
d
s
t
h
at
ca
n
b
e
u
s
ed
to
tr
ai
n
s
e
v
er
a
l a
r
tific
ial
n
eu
r
al
n
e
t
w
o
r
k
s
,
e.
g
.
“
Scaled
C
o
n
j
u
g
ate
Gr
ad
ien
t
[
1
8
]
,
C
o
m
p
eti
tiv
e
L
ea
r
n
i
n
g
[
1
9
]
,
L
ev
en
b
er
g
-
Ma
r
q
u
ar
d
t
[
2
0
]
,
Ho
p
f
ield
lear
n
i
n
g
[
2
1
]
,
etc.
”,
m
o
s
t
o
f
t
h
o
s
e
al
g
o
r
ith
m
s
b
elo
n
g
to
t
h
e
s
u
p
er
v
is
ed
l
ea
r
n
in
g
f
a
m
il
y
,
an
d
ea
ch
o
f
t
h
e
m
h
as
s
p
ec
if
ic
f
e
atu
r
es
,
ad
v
a
n
ta
g
es,
a
n
d
d
is
ad
v
an
ta
g
e
s
th
a
t
m
o
s
tl
y
ca
n
’
t
b
e
ad
ap
ted
to
C
P
U
s
ch
ed
u
lin
g
p
r
o
b
le
m
atic.
4.
NE
URA
L
N
E
T
WO
RK
S AN
D
CL
O
UD
CO
M
P
UT
I
N
G
C
P
U
SCH
E
DU
L
I
N
G
C
P
U
s
ch
ed
u
li
n
g
is
in
v
o
lv
ed
i
n
ea
c
h
o
f
th
e
C
lo
u
d
C
o
m
p
u
ti
n
g
la
y
er
s
(
Fi
g
u
r
e
1
)
,
w
h
er
ea
s
it
w
ill
a
f
f
ec
t
s
ig
n
i
f
ica
n
tl
y
th
e
p
lat
f
o
r
m
s
p
er
f
o
r
m
an
ce
(
Op
er
ati
n
g
S
y
s
te
m
)
,
m
id
d
le
w
ar
e
an
d
s
o
f
t
w
ar
e
r
esp
o
n
s
es.
Hen
ce
,
ch
o
o
s
in
g
t
h
e
ac
c
u
r
ate
alg
o
r
it
h
m
f
o
r
C
P
U
s
c
h
ed
u
lin
g
w
i
ll
h
av
e
a
m
a
s
s
i
v
e
i
m
p
ac
t
o
n
t
h
e
C
lo
u
d
d
eliv
er
y
r
esp
o
n
s
e
ti
m
e
an
d
p
r
esen
t
s
a
f
i
n
er
alter
n
ati
v
e
to
ex
p
a
n
d
in
g
th
e
in
f
r
as
tr
u
ct
u
r
es
i
n
o
r
d
er
to
p
r
o
m
o
te
ce
ler
it
y
,
th
u
s
r
ed
u
ci
n
g
co
s
ts
r
elat
iv
e
to
ac
q
u
ir
in
g
t
h
e
n
e
w
i
n
f
r
as
tr
u
ct
u
r
es,
m
an
a
g
e
m
en
t,
p
r
o
v
i
s
io
n
i
n
g
,
m
o
n
ito
r
in
g
an
d
tr
o
u
b
lesh
o
o
tin
g
.
T
h
e
f
in
e
s
t
C
P
U
s
ch
ed
u
li
n
g
al
g
o
r
ith
m
o
n
a
C
lo
u
d
C
o
m
p
u
tin
g
m
o
d
el
s
h
o
u
ld
p
r
ed
ict
th
e
a
m
o
u
n
t
o
f
ti
m
e
(
T
im
e
Qu
a
n
t
u
m
)
th
a
t
i
s
es
s
en
tial
f
o
r
ea
ch
tas
k
s
u
b
m
itted
f
o
r
ex
ec
u
tio
n
in
r
esp
ec
t
to
t
h
e
f
o
llo
w
in
g
d
ir
ec
tio
n
s
:
R
ed
u
ce
t
h
e
n
u
m
b
er
o
f
co
n
te
x
t
s
w
itch
e
s
(
th
e
a
m
o
u
n
t o
f
ti
m
e
s
th
e
C
P
U
s
w
itc
h
es
f
r
o
m
a
ta
s
k
to
an
o
th
er
)
R
ed
u
ce
t
h
e
av
er
ag
e
a
m
o
u
n
t o
f
ti
m
e
t
h
at
a
tas
k
s
p
en
t o
n
th
e
w
ait
in
g
li
s
t.
R
ed
u
ce
t
h
e
av
er
ag
e
a
m
o
u
n
t o
f
ti
m
e
n
ec
e
s
s
ar
y
to
ca
r
r
y
o
u
t th
e
ex
ec
u
tio
n
o
f
a
tas
k
.
B
y
s
t
u
d
y
in
g
th
e
s
e
g
u
id
elin
e
s
an
d
th
e
ex
i
s
ti
n
g
C
P
U
Sc
h
ed
u
li
n
g
alg
o
r
ith
m
s
,
w
e
w
e
r
e
ab
le
to
e
m
p
h
a
s
ize
th
e
f
o
llo
w
i
n
g
A
N
N
k
e
y
cr
iter
ia
t
h
at
w
ill a
f
f
ec
t th
e
C
lo
u
d
C
o
m
p
u
ti
n
g
s
er
v
ice
d
e
liv
er
y
m
o
d
el:
R
esp
o
n
s
e
T
i
m
e
(
S
1
)
: T
h
e
a
m
o
u
n
t o
f
ti
m
e
n
ec
e
s
s
ar
y
to
p
r
o
d
u
ce
a
r
esu
lt
.
T
r
ain
in
g
m
eth
o
d
s
(
S
2
)
: S
u
p
p
o
r
t o
f
A
N
N
ex
i
s
ti
n
g
tr
ai
n
i
n
g
m
eth
o
d
s
T
r
ain
in
g
d
u
r
atio
n
(
S
3
)
:
T
h
e
am
o
u
n
t
o
f
ti
m
e
r
eq
u
ir
ed
to
co
ac
h
th
e
al
g
o
r
ith
m
b
ef
o
r
e
it
c
an
s
tar
t
ta
k
in
g
d
ec
is
io
n
.
I
n
teg
r
atio
n
(
S
4
)
:
Si
m
p
l
icit
y
o
f
co
d
in
g
an
d
i
n
teg
r
atio
n
with
ex
is
ti
n
g
p
latf
o
r
m
s
(
Op
er
atio
n
s
y
s
te
m
s
,
H
y
p
er
v
i
s
o
r
s
,
C
lo
u
d
p
r
o
v
is
io
n
i
n
g
p
lat
f
o
r
m
s
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
C
o
mp
a
r
a
tive
S
tu
d
y
o
f Neu
r
a
l
N
etw
o
r
k
s
A
lg
o
r
ith
ms fo
r
C
lo
u
d
C
o
mp
u
tin
g
C
P
U
S
ch
e
d
u
lin
g
(
Gib
et
Ta
n
i H
)
3575
A
th
eo
r
et
ic
w
eig
h
t
th
at
v
ar
ie
s
f
r
o
m
0
to
1
h
a
s
b
ee
n
g
i
v
e
n
to
ea
ch
o
n
e
o
f
t
h
e
cr
i
ter
ia
m
en
t
io
n
ed
ab
o
v
e
th
at
r
ep
r
esen
t it
s
i
m
p
o
r
ta
n
ce
to
s
o
lv
in
g
t
h
e
s
c
h
ed
u
li
n
g
p
r
o
b
le
m
atic:
R
esp
o
n
s
e
T
i
m
e:
w
1
=
0
.
3
5
,
T
r
ain
i
n
g
m
et
h
o
d
s
:
w
2
=
0
.
2
5
,
T
r
ain
i
n
g
d
u
r
at
io
n
:
w
3
=
0
.
3
,
I
n
teg
r
atio
n
:
w
4
=
0
.
1
∑
=
1
4
=
1
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
ev
alu
at
io
n
co
n
s
id
er
ed
in
th
is
p
ap
er
co
n
s
is
ts
o
f
e
v
alu
a
ti
n
g
th
e
t
y
p
e
o
f
ar
ti
f
icial
n
e
u
r
al
n
et
w
o
r
k
s
b
ased
o
n
th
e
cr
iter
ia
d
escr
ib
e
d
o
n
th
e
p
r
ev
io
u
s
s
ec
tio
n
.
A
c
co
r
d
in
g
to
liter
at
u
r
e,
th
er
e
ar
e
a
v
ar
iet
y
o
f
ANN
T
y
p
es
a
n
d
ea
ch
o
n
e
o
f
t
h
e
m
h
as
p
r
o
v
e
n
it
s
ca
p
ac
it
y
i
n
o
n
e
o
r
m
u
ltip
le
f
ield
s
.
T
h
e
c
h
al
len
g
e
is
to
f
in
d
th
e
A
N
N
t
y
p
e
th
a
t
ca
n
b
e
ad
ap
ted
th
e
m
o
s
t
to
C
P
U
s
ch
ed
u
lin
g
f
o
r
cl
o
u
d
co
m
p
u
ti
n
g
an
d
th
is
b
y
r
ev
ie
w
i
n
g
t
h
e
A
r
ti
f
icial
Ne
u
r
al
Net
w
o
r
k
al
g
o
r
ith
m
s
ap
p
licatio
n
s
o
n
t
h
e
f
ie
ld
:
T
ab
le
3
.
A
NN
A
p
p
licatio
n
s
T
y
p
e
OF
A
N
N
A
p
p
l
i
c
a
t
i
o
n
A
d
a
p
t
e
d
f
o
r
C
PU
S
c
h
e
d
u
l
i
n
g
/
S
y
st
e
m
r
e
s
o
u
r
c
e
s m
a
n
a
g
e
m
e
n
t
M
u
l
t
i
-
l
a
y
e
r
P
e
r
c
e
p
t
r
o
n
[
2
2
]
S
u
p
e
r
v
i
se
d
l
e
a
r
n
i
n
g
[
2
3
]
P
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
[
2
4
]
S
p
e
e
c
h
r
e
c
o
g
n
i
t
i
o
n
[
2
4
]
I
mag
e
r
e
c
o
g
n
i
t
i
o
n
[
2
4
]
M
a
c
h
i
n
e
t
r
a
n
sl
a
t
i
o
n
[
2
4
]
“
M
u
l
t
i
-
l
a
y
e
r
P
e
r
c
e
p
t
r
o
n
”
h
a
s
b
e
e
n
u
se
d
t
o
o
p
t
i
m
i
z
e
j
o
b
s
c
h
e
d
u
l
i
n
g
r
e
su
l
t
s
[
3
]
.
R
B
F
n
e
t
w
o
r
k
[
2
5
]
M
a
c
-
K
e
y
G
l
a
ss C
h
a
o
t
i
c
t
i
me
se
r
i
e
s [2
6
]
L
o
g
i
st
i
c
M
a
p
[
2
7
]
P
r
e
d
i
c
t
i
o
n
N
o
n
L
i
n
e
a
r
sy
st
e
m [
2
6
]
[
2
7
]
F
o
r
e
c
a
st
i
n
g
[
2
8
]
R
B
F
n
e
u
r
a
l
n
e
t
w
o
r
k
i
s u
se
d
i
n
t
h
e
p
r
e
d
i
c
t
i
o
n
o
f
t
h
e
t
i
me
a
n
d
r
e
so
u
r
c
e
s c
o
n
s
u
me
d
b
y
a
p
p
l
i
c
a
t
i
o
n
s
[
4
0
]
K
o
h
o
n
e
n
se
l
f
-
o
r
g
a
n
i
z
i
n
g
n
e
t
w
o
r
k
[
2
9
]
M
e
t
e
o
r
o
l
o
g
y
,
O
c
e
a
n
o
g
r
a
p
h
y
[
3
0
]
P
r
o
j
e
c
t
p
r
i
o
r
i
t
i
z
a
t
i
o
n
a
n
d
se
l
e
c
t
i
o
n
[
3
1
]
--
R
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
[
3
2
]
H
a
n
d
w
r
i
t
i
n
g
a
n
d
sp
e
e
c
h
R
e
c
o
g
n
i
t
i
o
n
[
3
3
]
C
o
mp
u
t
e
r
V
i
s
i
o
n
[
3
4
]
L
a
n
g
u
a
g
e
P
r
o
c
e
ssi
n
g
[
3
5
]
R
e
c
u
r
r
e
n
t
N
e
u
r
a
l
N
e
t
w
o
r
k
h
a
s
b
e
e
n
u
se
d
t
o
o
p
t
i
m
i
z
e
t
h
e
n
u
m
b
e
r
o
f
q
u
e
u
e
s a
n
d
q
u
a
n
t
u
m
t
o
d
e
c
r
e
a
se
t
h
e
r
e
sp
o
n
se
t
i
me
o
f
p
r
o
c
e
sse
s a
n
d
i
n
c
r
e
a
se
t
h
e
p
e
r
f
o
r
man
c
e
o
f
sch
e
d
u
l
i
n
g
.
[
4
1
]
.
M
o
d
u
l
a
r
n
e
u
r
a
l
n
e
t
w
o
r
k
s
[
3
6
]
P
r
e
d
i
c
a
t
i
o
n
[
3
7
]
P
a
t
t
e
r
n
r
e
c
o
g
n
i
t
i
o
n
[
3
8
]
C
l
a
ssi
f
i
c
a
t
i
o
n
[
3
9
]
--
T
ab
le
4
.
A
NN
Sco
r
in
g
T
y
p
e
OF
A
N
N
R
e
sp
o
n
se
T
i
m
e
T
r
a
i
n
i
n
g
m
e
t
h
o
d
s
T
r
a
i
n
i
n
g
d
u
r
a
t
i
o
n
In
t
e
g
r
a
t
i
o
n
M
u
l
t
i
-
l
a
y
e
r
P
e
r
c
e
p
t
r
o
n
0
.
8
-
B
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
-
R
e
si
l
i
e
n
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
-
G
e
n
e
t
i
c
a
l
g
o
r
i
t
h
mi
c
0
.
3
0
.
6
0
.
8
R
B
F
n
e
t
w
o
r
k
0
.
7
-
G
r
a
d
i
e
n
t
D
e
sce
n
t
-
K
a
l
man
F
i
l
t
e
r
i
n
g
-
G
e
n
e
t
i
c
A
l
g
o
r
i
t
h
mi
c
0
.
3
0
.
5
0
.
7
K
o
h
o
n
e
n
se
l
f
-
o
r
g
a
n
i
z
i
n
g
n
e
t
w
o
r
k
0
.
1
-
S
e
l
f
-
O
r
g
a
n
i
z
i
n
g
M
a
p
0
.
1
0
.
1
0
.
1
R
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
0
.
5
-
R
e
c
u
r
r
e
n
t
l
e
a
r
n
i
n
g
-
Ex
t
e
n
d
e
d
K
a
l
man
-
G
r
a
d
i
e
n
t
d
e
sce
n
t
-
G
l
o
b
a
l
o
p
t
i
m
i
z
a
t
i
o
n
0
.
4
0
.
4
0
.
5
M
o
d
u
l
a
r
n
e
u
r
a
l
n
e
t
w
o
r
k
s
0
.
1
-
M
o
d
u
l
a
r
n
e
u
r
a
l
n
e
t
w
o
r
k
t
r
a
i
n
i
n
g
a
l
g
o
r
i
t
h
m
0
.
1
0
.
1
0
.
1
T
h
e
o
v
er
all
s
co
r
e
f
o
r
ea
ch
alg
o
r
ith
m
i
s
ca
lcu
la
ted
as f
o
llo
w
:
S =
∑
∗
4
=
1
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2088
-
8708
I
J
E
C
E
Vo
l.
7
,
No
.
6
,
Dec
em
b
er
2
0
1
7
:
3
5
7
0
–
3
5
7
7
3576
Fig
u
r
e
6
.
A
NN
al
g
o
r
ith
m
s
Ov
er
all
Sco
r
e
A
cc
o
r
d
in
g
to
f
i
g
u
r
e
6
,
t
h
e
M
u
lti
-
la
y
er
P
er
ce
p
tr
o
n
A
NN
att
ain
ed
t
h
e
f
in
e
s
t
s
co
r
e,
f
o
llo
wed
b
y
R
B
F
n
et
w
o
r
k
a
n
d
R
ec
u
r
r
en
t
Neu
r
al
Net
w
o
r
k
r
esp
ec
ti
v
el
y
.
T
h
er
ef
o
r
e,
Mu
lti
-
la
y
er
P
er
ce
p
tr
o
n
is
th
e
A
N
N
t
y
p
e
th
a
t
ca
n
b
etter
an
s
w
er
to
th
e
p
r
o
b
le
m
atic
o
f
C
P
U
s
ch
ed
u
li
n
g
o
n
C
lo
u
d
C
o
m
p
u
ti
n
g
.
6.
CO
NCLU
SI
O
N
T
h
e
s
tu
d
y
en
g
a
g
ed
o
n
th
is
p
a
p
er
is
a
th
eo
r
etica
l
ev
alu
atio
n
o
f
A
r
ti
f
icial
Ne
u
r
al
Net
w
o
r
k
s
an
d
th
eir
ab
ilit
ies
to
s
o
lv
e
t
h
e
p
r
o
b
le
m
r
elate
d
to
C
P
U
s
c
h
ed
u
li
n
g
o
n
C
lo
u
d
C
o
m
p
u
t
in
g
.
A
s
et
o
f
co
n
ce
p
tu
al
m
etr
ic
s
h
av
e
b
ee
n
co
n
s
id
er
ed
to
s
co
r
e
ea
ch
A
N
N
t
y
p
e
a
n
d
tr
a
i
n
i
n
g
tech
n
iq
u
e
s
a
n
d
th
a
t
is
in
r
eg
ar
d
s
to
s
p
ec
if
ic
cr
iter
ia
u
s
ed
to
e
v
al
u
ate
t
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
s
c
h
ed
u
li
n
g
alg
o
r
ith
m
s
i
n
t
h
e
C
lo
u
d
t
h
at
ca
n
b
e
r
es
u
m
ed
o
n
r
ed
u
cin
g
th
e
av
er
ag
e
w
ait
in
g
t
i
m
e
o
f
tas
k
s
o
n
th
e
ex
ec
u
tio
n
q
u
eu
e
an
d
s
ti
m
u
lati
n
g
t
h
e
r
esp
o
n
s
e
ti
m
e.
I
n
s
p
ite
o
f
th
e
d
if
f
ic
u
lt
ies
en
co
u
n
ter
ed
in
o
r
d
er
to
s
p
o
t
th
e
ac
cu
r
ate
A
N
N
t
y
p
e
s
u
ited
f
o
r
th
e
C
P
U
s
ch
ed
u
l
in
g
ch
alle
n
g
e
o
n
t
h
e
C
lo
u
d
,
th
e
Mu
lti
-
la
y
er
P
er
ce
p
tr
o
n
A
NN
r
ad
iates
as
th
e
b
est
ca
n
d
id
ate
to
an
s
w
er
to
ea
ch
o
f
th
e
cr
iter
ia
co
n
s
id
er
ed
d
u
r
in
g
t
h
e
ev
al
u
atio
n
an
d
ass
e
s
s
m
e
n
t.
T
h
is
ac
co
m
p
lis
h
m
e
n
t
w
i
ll
b
e
ex
p
an
d
ed
b
y
co
n
d
u
cti
n
g
m
o
r
e
s
t
u
d
ies
a
n
d
tes
tin
g
o
n
t
h
e
Mu
lti
-
la
y
er
P
er
ce
p
tr
o
n
A
NN
a
lg
o
r
it
h
m
u
s
in
g
s
p
ec
if
ic
s
i
m
u
lato
r
s
.
Fu
r
t
h
er
m
o
r
e,
a
n
i
m
p
le
m
en
ta
tio
n
o
f
t
h
e
al
g
o
r
ith
m
in
o
n
e
o
f
clo
u
d
co
m
p
u
t
in
g
p
la
tf
o
r
m
s
in
o
r
d
er
to
ass
ess
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
al
g
o
r
ith
m
o
n
r
ea
l
b
ased
s
itu
a
tio
n
s
.
RE
F
E
R
E
NC
E
S
[1
]
F
.
Da
río
Ba
p
t
ista,
S
.
Ro
d
rig
u
e
s,
F
.
M
o
rg
a
d
o
-
Dia
s.
“
P
e
rf
o
rm
a
n
c
e
c
o
m
p
a
riso
n
o
f
A
NN
train
in
g
a
lg
o
rit
h
m
s
f
o
r
c
las
si
f
ica
ti
o
n
”
,
In
telli
g
e
n
t
S
i
g
n
a
l
Pro
c
e
ss
in
g
(
W
IS
P)
,
2
0
1
3
IEE
E
8
t
h
In
ter
n
a
ti
o
n
a
l
S
y
m
p
o
siu
m
.
[2
]
R.
Ca
ru
a
n
a
,
A
.
Nic
u
les
c
u
-
M
izil,
“
A
n
e
m
p
iri
c
a
l
c
o
m
p
a
riso
n
o
f
su
p
e
rv
ise
d
lea
rn
in
g
a
lg
o
rit
h
m
s”
,
Pro
c
e
e
d
in
g
s
o
f
th
e
2
3
rd
in
ter
n
a
ti
o
n
a
l
c
o
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
lea
r
n
in
g
.
ACM
,
2
0
0
6
.
[3
]
M
.
M
a
q
a
b
leh
,
H.
Ka
ra
jeh
,
R.
M
a
sa
’d
e
h
,
“
Jo
b
S
c
h
e
d
u
l
in
g
f
o
r
Clo
u
d
Co
m
p
u
ti
n
g
Us
in
g
Ne
u
ra
l
Ne
tw
o
rk
s
”
,
Co
mm
u
n
ica
ti
o
n
s a
n
d
Ne
two
rk
,
2
0
1
4
,
6
,
1
9
1
-
2
0
0
.
[4
]
C.
El
Am
ra
n
i,
K.B.
F
il
a
li
,
K.
Be
n
A
h
m
e
d
,
A
.
T
.
Dia
ll
o
,
S
.
T
e
lo
lah
y
,
“
A
Co
m
p
a
rti
v
e
S
tu
d
y
o
f
Clo
u
d
Co
m
p
u
ti
n
g
m
id
d
lew
a
re
”
,
IEE
E/
ACM
In
ter
n
a
ti
o
n
a
l
S
y
mp
o
si
u
m o
n
Cl
u
ste
r,
Clo
u
d
a
n
d
Gr
id
Co
mp
u
ti
n
g
,
2
0
1
2
.
[5
]
G
.
T
.
Hic
h
a
m
,
E.
Ch
a
k
e
r,
“
Clo
u
d
Co
m
p
u
ti
n
g
C
P
U
A
ll
o
c
a
ti
o
n
a
n
d
S
c
h
e
d
u
li
n
g
A
lg
o
rit
h
m
s
U
sin
g
Clo
u
d
S
im
S
im
u
lato
r”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
El
e
c
trica
l
a
n
d
C
o
mp
u
ter
En
g
i
n
e
e
rin
g
,
Vo
l.
6
,
N
o
.
4
,
A
u
g
u
st 2
0
1
6
.
[6
]
M
.
A
b
d
e
ll
a
,
T
.
M
a
r
wa
la
,
“
T
h
e
U
se
o
f
G
e
n
e
ti
c
A
l
g
o
rit
h
m
s
a
n
d
Ne
u
ra
l
Ne
tw
o
rk
s
to
A
p
p
ro
x
im
a
te
M
issin
g
Da
ta
in
Da
tab
a
se
”
,
Co
mp
u
ti
n
g
a
n
d
In
fo
r
ma
ti
c
s
,
V
o
l
2
4
,
2
0
0
5
,
5
7
7
–
5
8
9
.
[7
]
S
a
e
d
S
a
y
a
d
:
h
tt
p
:/
/w
ww
.
s
a
e
d
sa
y
a
d
.
c
o
m
/artif
icia
l_
n
e
u
ra
l_
n
e
tw
o
rk
.
h
tm
.
[8
]
J.
S
k
o
rin
-
Ka
p
o
v
,
K.W
.
T
a
n
g
,
“
T
r
a
in
in
g
A
rti
f
icia
l
Ne
u
ra
l
N
e
t
w
o
rk
s
:
Ba
c
k
p
ro
p
a
g
a
ti
o
n
v
ia
No
n
li
n
e
a
r
Op
ti
m
iza
ti
o
n
”
,
J
o
u
rn
a
l
o
f
Co
m
p
u
ti
n
g
a
n
d
I
n
f
o
rm
a
ti
o
n
T
e
c
h
n
o
l
o
g
y
,
CIT
9
,
2
0
0
1
,
1
,
0
0
1
–
0
1
4
.
[9
]
D.Ka
u
l,
N.
A
n
a
m
,
S
.
G
a
ik
wa
d
,
S
.
T
i
w
a
ri,
“
Do
m
a
in
Ba
se
d
C
a
teg
o
risa
ti
o
n
Us
in
g
A
d
a
p
ti
v
e
P
re
p
ro
c
e
ss
in
g
”
,
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
f
o
r R
e
se
a
rc
h
in
Ap
p
li
e
d
S
c
ien
c
e
&
En
g
in
e
e
rin
g
T
e
c
h
n
o
lo
g
y
,
V
o
l
u
m
e
4
,
F
e
b
ru
a
r
y
2
0
1
6
.
[1
0
]
S
.
E.
F
a
h
lm
a
n
,
“
A
n
Em
p
iri
c
a
l
S
tu
d
y
o
f
L
e
a
rn
in
g
S
p
e
e
d
in
Ba
c
k
p
ro
p
a
g
a
ti
o
n
Ne
tw
o
rk
s”
,
Ca
rn
e
g
ie
M
e
ll
o
n
Rep
o
rt
,
No
CM
U
-
Cs,
p
p
.
88
-
1
6
2
.
[1
1
]
S
.
P
a
d
m
a
p
riy
a
,
S
.
A
n
d
a
v
a
n
,
“
A
S
tu
d
y
o
n
A
lg
o
rit
h
m
ic
A
p
p
ro
a
c
h
e
s
a
n
d
M
in
i
n
g
M
e
th
o
d
o
l
o
g
ies
In
Da
ta
M
in
in
g
”
,
In
ter
n
a
t
io
n
a
l
J
o
u
rn
a
l
o
f
C
o
mp
u
ter
S
c
ien
c
e
T
re
n
d
s
a
n
d
T
e
c
h
n
o
l
o
g
y
,
Vo
lu
m
e
3
,
Iss
u
e
1
,
Ja
n
-
F
e
b
2
0
1
5
.
0
.
6
2
5
0
.
5
5
0
.
1
0
.
4
7
5
0
.
1
0
0
.
1
0
.
2
0
.
3
0
.
4
0
.
5
0
.
6
0
.
7
M
u
lt
i-l
a
y
e
r
P
e
rc
e
p
tro
n
RBF
n
e
two
rk
Ko
h
o
n
e
n
se
lf-
o
r
g
a
n
izin
g
n
e
two
rk
Re
c
u
rre
n
t
n
e
u
ra
l
n
e
two
rk
M
o
d
u
lar
n
e
u
ra
l
n
e
two
rk
s
Sco
r
e
A
NN
A
lg
o
r
ith
m
s
Ov
era
l
l
Sco
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
C
o
mp
a
r
a
tive
S
tu
d
y
o
f Neu
r
a
l
N
etw
o
r
k
s
A
lg
o
r
ith
ms fo
r
C
lo
u
d
C
o
mp
u
tin
g
C
P
U
S
ch
e
d
u
lin
g
(
Gib
et
Ta
n
i H
)
3577
[1
2
]
E.
R.
Da
v
id
,
E.
H.
G
e
o
ff
re
y
,
J.
W
.
Ro
n
a
ld
.
“
L
e
a
rn
in
g
re
p
re
se
n
tatio
n
s
b
y
b
a
c
k
-
p
ro
p
a
g
a
ti
n
g
e
rro
rs”
,
Na
t
u
re
In
ter
n
a
t
io
n
a
l
W
e
e
k
ly Jo
u
rn
a
l
o
f
S
c
ien
c
e
,
1
9
8
6
.
[1
3
]
A
.
S
n
y
m
a
n
,
“
P
ra
c
ti
c
a
l
M
a
th
e
m
a
ti
c
a
l
Op
ti
m
iza
ti
o
n
:
A
n
In
tr
o
d
u
c
ti
o
n
t
o
Ba
sic
Op
ti
m
iza
ti
o
n
T
h
e
o
ry
a
n
d
C
las
sic
a
l
a
n
d
Ne
w
G
ra
d
ien
t
-
Ba
se
d
A
lg
o
rit
h
m
s
”
,
S
p
ri
n
g
e
r
P
u
b
li
s
h
in
g
,
I
S
BN 0
-
3
8
7
-
2
4
3
4
8
-
8
,
2
0
0
5
[1
4
]
M
.
Ried
m
il
ler,
H.
Bra
u
n
,
“
Rp
r
o
p
-
A
F
a
st
A
d
a
p
ti
v
e
L
e
a
rn
in
g
A
lg
o
rit
h
m
”
,
Pro
c
e
e
d
in
g
s
o
f
t
h
e
In
ter
n
a
ti
o
n
a
l
S
y
mp
o
si
u
m o
n
Co
m
p
u
ter
a
n
d
In
f
o
rm
a
ti
o
n
S
c
ien
c
e
VII
,
1
9
9
2
.
[1
5
]
M
.
Ried
m
il
ler,
H.
Bra
u
n
,
“
A
d
irec
t
a
d
a
p
ti
v
e
m
e
th
o
d
f
o
r
f
a
ste
r
b
a
c
k
p
ro
p
a
g
a
ti
o
n
lea
rn
in
g
:
T
h
e
R
P
R
OP
a
lg
o
rit
h
m
”
,
Pro
c
.
IEE
E
In
t.
Co
n
f.
O
n
Ne
u
ra
l
Ne
two
rk
,
p
p
.
5
8
6
-
5
9
1
,
1
9
9
3
.
[1
6
]
A
.
E.
Ei
b
e
n
,
“
G
e
n
e
ti
c
a
l
g
o
rit
h
m
s
w
it
h
m
u
lt
i
-
p
a
re
n
t
re
c
o
m
b
in
a
ti
o
n
”
,
P
PS
N
III:
Pro
c
e
e
d
in
g
s
o
f
t
h
e
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Evo
l
u
ti
o
n
a
ry
Co
m
p
u
t
a
ti
o
n
,
7
8
–
8
7
,
I
S
BN 3
-
5
4
0
-
5
8
4
8
4
-
6
.
[1
7
]
B.
P
.
Jia
n
g
,
“
Ne
u
ra
l
Ne
tw
o
rk
Alg
o
rit
h
m
Op
ti
m
ize
d
b
y
Ge
n
e
ti
c
A
l
g
o
rit
h
m
a
n
d
Its
S
im
u
latio
n
”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
m
p
u
ter
S
c
ien
c
e
,
Iss
u
e
s 1
0
,
2
0
1
3
,
5
1
6
-
5
1
9
.
[1
8
]
M
.
F
o
d
sle
tt
e
,
“
A
s
c
a
led
c
o
n
ju
g
a
te
g
r
a
d
ien
t
a
lg
o
rit
h
m
f
o
r
f
a
st
-
su
p
e
rv
ise
d
lea
rn
in
g
.
Ne
u
ra
l
Ne
t
w
o
rk
s”
,
DAI
M
I
P
B
339
,
1
9
9
3
.
[1
9
]
R.
Da
v
id
,
D.
Zi
p
se
r,
J.L
.
M
c
Clell
a
n
d
,
“
P
a
ra
ll
e
l
Distri
b
u
te
d
P
r
o
c
e
ss
in
g
”
,
M
IT
Pre
ss
,
Vo
l.
1
,
pp.
1
5
1
–
1
9
3
.
[2
0
]
D.
M
a
rq
u
a
rd
t,
“
A
n
A
l
g
o
rit
h
m
f
o
r
L
e
a
st
-
S
q
u
a
re
s
Esti
m
a
ti
o
n
o
f
N
o
n
li
n
e
a
r
P
a
ra
m
e
ters
”
,
S
IAM
J
o
u
rn
a
l
o
n
Ap
p
li
e
d
M
a
th
e
ma
ti
c
s
,
V
o
l.
1
1
,
No
.
2
,
J
u
n
e
1
9
6
3
,
p
p
.
4
3
1
–
4
4
1
.
[2
1
]
M
a
c
Ka
y
,
J.C.
Da
v
id
,
“
Ho
p
f
ield
Ne
tw
o
rk
s”
,
In
fo
rm
a
ti
o
n
T
h
e
o
ry
,
In
fer
e
n
c
e
a
n
d
L
e
a
rn
i
n
g
Al
g
o
rith
ms
.
C
a
mb
ri
d
g
e
Un
ive
rs
it
y
Pre
ss
.
p
.
5
0
8
.
IS
BN
0
5
2
1
6
4
2
9
8
1
.
[2
2
]
Ro
se
n
b
latt,
F
ra
n
k
,
“
P
r
in
c
ip
les
o
f
Ne
u
ro
d
y
n
a
m
ic
s:
P
e
rc
e
p
tr
o
n
s
a
n
d
th
e
T
h
e
o
ry
o
f
Bra
in
M
e
c
h
a
n
ism
s”
.
S
p
a
rt
a
n
Bo
o
k
s,
W
a
s
h
in
g
to
n
DC
,
1
9
6
1
[2
3
]
G
.
C
y
b
e
n
k
o
,
“
A
p
p
ro
x
ima
ti
o
n
b
y
su
p
e
rp
o
si
ti
o
n
s
o
f
a
sig
m
o
id
a
l
fu
n
c
ti
o
n
”
,
M
a
t
h
e
ma
ti
c
s
o
f
Co
n
tro
l,
S
ig
n
a
ls,
a
n
d
S
y
ste
ms
,
2
(4
),
3
0
3
–
3
1
4
[2
4
]
P
.
D
W
a
ss
e
r
m
a
n
,
T
.
S
c
h
w
a
rtz,
“
Ne
u
ra
l
n
e
tw
o
rk
s.
II.
W
h
a
t
a
re
th
e
y
a
n
d
w
h
y
is
e
v
e
r
y
b
o
d
y
so
in
t
e
re
ste
d
in
t
h
e
m
n
o
w
?
”
,
IEE
E
Exp
e
rt
,
1
9
8
8
,
Vo
l
u
m
e
3
,
Iss
u
e
1
[2
5
]
H.
Ha
v
il
u
d
d
i
n
,
I.
T
a
h
y
u
d
in
,
“
T
ime
S
e
ries
P
re
d
ictio
n
Us
i
n
g
Ra
d
ial
Ba
sis
F
u
n
c
t
io
n
Ne
u
ra
l
Ne
tw
o
rk
”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
trica
l
a
n
d
C
o
mp
u
t
e
r E
n
g
i
n
e
e
rin
g
,
Vo
l.
5
,
N
o
.
4
,
A
u
g
u
st 2
0
1
5
.
[2
6
]
E.
S
.
Ch
o
n
g
,
S
.
C
h
e
n
,
B.
M
u
lg
re
w
,
“
G
r
a
d
ien
t
Ra
d
ial
Ba
sis
F
u
n
c
t
io
n
Ne
tw
o
rk
s
f
o
r
No
n
li
n
e
a
r
a
n
d
No
n
sta
ti
o
n
a
ry
T
i
m
e
S
e
ries
P
re
d
ictio
n
”
,
IEE
E
T
r
a
n
sa
c
ti
o
n
s
o
n
Ne
u
ra
l
Ne
two
rk
s
,
V
o
l
.
7
,
No
.
1
[2
7
]
R.
Zam
o
ra
,
D.Ra
c
o
c
e
a
n
u
,
N.Z
e
rh
o
u
n
i,
“
Re
c
u
rre
n
t
ra
d
ial
b
a
sis
f
u
n
c
ti
o
n
n
e
tw
o
rk
f
o
r
ti
m
e
-
s
e
ri
e
s
p
re
d
ictio
n
”
,
En
g
i
n
e
e
rin
g
Ap
p
li
c
a
ti
o
n
s
o
f
Arti
fi
c
ia
l
In
tell
ig
e
n
c
e
,
E
lse
v
ier,
2
0
0
3
.
[2
8
]
L
.
Yu
,
K.
Ke
u
n
g
L
a
i,
S
.
W
a
n
g
,
“
M
u
lt
istag
e
RBF
n
e
u
ra
l
n
e
tw
o
rk
e
n
se
m
b
le l
e
a
rn
in
g
f
o
r
e
x
c
h
a
n
g
e
r
a
t
e
s f
o
re
c
a
stin
g
”
,
Ne
u
ro
c
o
m
p
u
ti
n
g
.
[2
9
]
Ko
h
o
n
e
n
,
T
e
u
v
o
,
“
S
e
lf
-
Org
a
n
ize
d
F
o
rm
a
ti
o
n
o
f
T
o
p
o
l
o
g
ica
ll
y
Co
rre
c
t
F
e
a
tu
re
M
a
p
s”
,
Bi
o
lo
g
ica
l
Cy
b
e
rn
e
ti
c
s.
4
3
(1
):
5
9
–
6
9
.
Do
i
:
1
0
.
1
0
0
7
/b
f
0
0
3
3
7
2
8
8
.
[3
0
]
Y.L
iu
,
R.
H.
W
e
isb
e
rg
,
“
A
re
v
ie
w
o
f
se
l
f
-
o
rg
a
n
izin
g
m
a
p
a
p
p
li
c
a
ti
o
n
s
i
n
m
e
teo
ro
lo
g
y
a
n
d
o
c
e
a
n
o
g
ra
p
h
y
”
,
S
e
lf
-
Or
g
a
n
izi
n
g
M
a
p
s
-
A
p
p
li
c
a
ti
o
n
s a
n
d
N
o
v
e
l
Al
g
o
rit
h
m De
sig
n
,
2
5
3
-
2
7
2
.
[3
1
]
G.
Zh
e
n
g
,
V
.
V
a
ish
n
a
v
i,
“
A
M
u
lt
id
im
e
n
sio
n
a
l
P
e
rc
e
p
tu
a
l
M
a
p
A
p
p
r
o
a
c
h
to
P
ro
jec
t
P
ri
o
rit
iza
ti
o
n
a
n
d
S
e
lec
ti
o
n
”
,
AIS
T
r
a
n
sa
c
ti
o
n
s
o
n
Hu
ma
n
-
Co
m
p
u
ter
In
ter
a
c
ti
o
n
,
(
3
)
2
,
p
p
.
8
2
-
1
0
3
.
[3
2
]
A
.
G
ra
v
e
s,
M
.
L
iw
ic
k
i,
S
.
F
e
rn
a
n
d
e
z
,
R.
Be
rto
lam
i,
H.
Bu
n
k
e
,
J.
S
c
h
m
id
h
u
b
e
r,
“
A
No
v
e
l
Co
n
n
e
c
ti
o
n
ist
S
y
ste
m
f
o
r
Im
p
ro
v
e
d
Un
c
o
n
stra
in
e
d
Ha
n
d
w
rit
in
g
Re
c
o
g
n
it
io
n
”
,
IEE
E
T
r
a
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
rn
An
a
lys
is
a
n
d
M
a
c
h
in
e
In
telli
g
e
n
c
e
,
v
o
l.
3
1
,
n
o
.
5
,
2
0
0
9
.
[3
3
]
H.
S
a
k
,
A
.
W
.
S
e
n
io
r,
F
.
Be
a
u
f
a
y
s,
“
L
o
n
g
sh
o
rt
-
term
m
e
m
o
r
y
re
c
u
rre
n
t
n
e
u
ra
l
n
e
tw
o
rk
a
rc
h
it
e
c
tu
re
s
f
o
r
larg
e
sc
a
le
a
c
o
u
stic m
o
d
e
li
n
g
”
,
Pro
c
.
I
n
ter
s
p
e
e
c
h
,
p
p
3
3
8
-
3
4
2
,
S
i
n
g
a
p
o
re
,
S
e
p
t.
2
0
1
[3
4
]
T
rip
a
th
i,
S
u
b
a
rn
a
,
“
Co
n
tex
t
M
a
tt
e
rs:
Re
f
in
in
g
Ob
jec
t
De
tec
ti
o
n
in
V
id
e
o
w
it
h
Re
c
u
rre
n
t
Ne
u
ra
l
Ne
tw
o
rk
s”
,
a
rXiv
p
re
p
rin
t
a
rXiv
:
1
6
0
7
.
0
4
6
4
8
(
2
0
1
6
).
[3
5
]
S
o
c
h
e
r,
Rich
a
r
d
,
L
in
,
Ng
.
Cli
f
f
,
Y.
A
n
d
re
w
,
M
a
n
n
in
g
,
D.
Ch
r
i
sto
p
h
e
r,
“
P
a
rsin
g
Na
tu
ra
l
S
c
e
n
e
s
a
n
d
Na
tu
ra
l
L
a
n
g
u
a
g
e
w
it
h
Re
c
u
rsiv
e
N
e
u
ra
l
Ne
t
w
o
rk
”
,
T
h
e
2
8
th
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
M
a
c
h
in
e
L
e
a
rn
in
g
(ICM
L
2
0
1
1
).
[3
6
]
A
.
F
a
ro
o
q
,
“
Bio
l
o
g
ica
ll
y
In
sp
ired
M
o
d
u
lar
Ne
u
r
a
l
Ne
tw
o
rk
s
”
,
Ph
D
Diss
e
rta
ti
o
n
,
Vi
rg
i
n
i
a
T
e
c
h
.
2
0
0
0
[
In
tern
e
t
A
c
e
ss
]
h
tt
p
:/
/sc
h
o
lar.l
i
b
.
v
t.
e
d
u
/t
h
e
se
s/a
v
a
il
a
b
le/e
td
-
0
6
0
9
2
0
0
0
-
1
2
1
5
0
0
2
8
/u
n
re
stricte
d
/etd
.
p
d
[3
7
]
T
.
Ki
m
o
to
,
K.
A
sa
k
a
w
a
,
M
.
Yo
d
a
,
M
.
T
a
k
e
o
k
a
,
“
S
to
c
k
m
a
rk
e
t
p
re
d
ictio
n
sy
ste
m
w
it
h
m
o
d
u
lar
n
e
u
ra
l
n
e
tw
o
rk
s”
,
In
ter
n
a
t
io
n
a
l
J
o
in
t
Co
n
fer
e
n
c
e
o
n
Ne
u
ra
l
Ne
two
rk
s
,
P
a
g
e
s 1
-
6
.
P
isc
a
ta
w
a
y
,
NJ
,
US
A
1
9
9
0
.
[3
8
]
L
.
M
u
i,
A
.
A
g
a
r
w
a
l,
A
.
G
u
p
ta,
P
.
W
.
S
h
e
n
-
P
e
i,
“
A
n
A
d
a
p
ti
v
e
M
o
d
u
lar
Ne
u
ra
l
Ne
tw
o
rk
w
it
h
A
p
p
li
c
a
ti
o
n
t
o
Un
c
o
n
stra
in
e
d
Ch
a
ra
c
ter
Re
c
o
g
n
it
io
n
”
,
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Pa
tt
e
rn
Rec
o
g
n
it
io
n
a
n
d
Art
if
ici
a
l
I
n
telli
g
e
n
c
e
,
V
o
l
.
8
,
No
.
5
,
P
a
g
e
s 1
1
8
9
-
1
2
0
4
.
Oc
to
b
e
r
1
9
9
4
.
[3
9
]
P
.
Blo
n
d
a
,
V
.
L
a
f
o
rg
iv
a
,
G
.
P
a
sq
u
a
riello
,
G
.
S
a
talin
o
,
“
M
u
lt
isp
e
c
tral
c
las
si
f
ica
ti
o
n
b
y
m
o
d
u
lar
n
e
u
ra
l
n
e
tw
o
rk
a
rc
h
it
e
c
tu
re
”
,
In
ter
n
a
ti
o
n
a
l
Ge
o
s
c
ien
c
e
a
n
d
Rem
o
te
S
e
n
sin
g
T
e
c
h
n
o
l
o
g
ies
,
D
a
ta
An
a
lys
is
a
n
d
In
te
rp
re
ta
ti
o
n
,
Vo
l.
4
.
P
a
g
e
s 1
8
7
3
-
1
8
7
6
.
Ne
w
Yo
rk
,
1
9
9
3
.
[4
0
]
M
a
tsu
n
a
g
a
,
A
n
d
ré
a
,
A
.
B.
F
.
Jo
sé
,
“
On
th
e
u
se
o
f
m
a
c
h
i
n
e
lea
rn
in
g
to
p
re
d
ict
th
e
ti
m
e
a
n
d
re
so
u
rc
e
s
c
o
n
su
m
e
d
b
y
a
p
p
li
c
a
ti
o
n
s”
,
Pro
c
e
e
d
in
g
s
o
f
th
e
2
0
1
0
1
0
t
h
IEE
E
/A
CM
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Clu
ste
r,
C
lo
u
d
a
n
d
Gr
i
d
Co
mp
u
t
in
g
.
IEE
E
Co
m
p
u
ter
S
o
c
i
e
ty
,
2
0
1
0
.
[4
1
]
L
.
Be
c
c
h
e
tt
i,
S
.
L
e
o
n
a
rd
i,
S
.
A
M
a
rc
h
e
tt
i,
“
A
v
e
r
a
g
e
-
Ca
se
a
n
d
S
m
o
o
th
e
d
Co
m
p
e
ti
ti
v
e
A
n
a
l
y
sis
o
f
th
e
M
u
lt
il
e
v
e
l
F
e
e
d
b
a
c
k
A
l
g
o
rit
h
m
”
,
M
a
th
e
ma
ti
c
s o
f
Op
e
ra
ti
o
n
Res
e
a
rc
h
,
V
o
l
.
3
1
,
2
0
0
6
.
Evaluation Warning : The document was created with Spire.PDF for Python.