I
AE
S
I
nte
rna
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
7
,
No
.
4
,
Dec
em
b
er
201
8
,
p
p
.
185
~
1
8
9
I
SS
N:
2252
-
8938
,
DOI
: 1
0
.
1
1
5
9
1
/i
j
ai.
v
7
.i
4
.
p
p
1
85
-
1
89
185
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
co
r
e.
co
m/jo
u
r
n
a
ls
/in
d
ex
.
p
h
p
/
I
JA
I
Co
m
pa
riso
n of
N
eura
l
Netw
o
rk
Tr
a
ining
Algo
rith
ms
f
o
r
Cla
ss
ificatio
n of
H
ea
rt
Disea
ses
H
esa
m
K
a
ri
m
,
S
ha
ra
re
h R.
Nia
k
a
n
,
Rez
a
Sa
f
da
ri
De
p
a
rtme
n
t
o
f
He
a
lt
h
In
f
o
rm
a
ti
o
n
m
a
n
a
g
e
m
e
n
t,
T
e
h
ra
n
Un
iv
e
rsity
o
f
M
e
d
ica
l
S
c
ien
c
e
s,
Ira
n
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l
1
1
,
2
0
1
8
R
ev
i
s
ed
Oct
10
,
2
0
1
8
A
cc
ep
ted
Oct
2
,
2
0
1
8
He
a
rt
d
ise
a
se
is t
h
e
f
irst
c
a
u
se
o
f
d
e
a
th
i
n
d
if
f
e
r
e
n
t
c
o
u
n
tri
e
s.
A
rti
f
icia
l
n
e
u
ra
l
n
e
tw
o
rk
(
A
NN
)
tec
h
n
iq
u
e
c
a
n
b
e
u
se
d
to
p
re
d
ict
o
r
c
las
sif
ic
a
ti
o
n
p
a
ti
e
n
ts
g
e
tt
in
g
a
h
e
a
rt
d
ise
a
se
.
T
h
e
re
a
re
d
if
fe
re
n
t
train
in
g
a
lg
o
rit
h
m
s
f
o
r
A
N
N.
We
c
o
m
p
a
re
d
e
ig
h
t
n
e
u
ra
l
n
e
tw
o
rk
train
in
g
a
lg
o
rit
h
m
s
f
o
r
c
las
si
f
ic
a
ti
o
n
o
f
h
e
a
rt
d
ise
a
se
d
a
ta
f
ro
m
UCI
re
p
o
sit
o
ry
c
o
n
tain
i
n
g
3
0
3
sa
m
p
les
.
P
e
rf
o
r
m
a
n
c
e
m
e
a
su
re
s
o
f
e
a
c
h
a
lg
o
rit
h
m
c
o
n
tain
in
g
th
e
sp
e
e
d
o
f
train
in
g
,
t
h
e
n
u
m
b
e
r
o
f
e
p
o
c
h
s,
a
c
c
u
ra
c
y
,
a
n
d
m
e
a
n
sq
u
a
re
e
rro
r
(M
S
E)
w
e
re
o
b
tain
e
d
a
n
d
a
n
a
ly
z
e
d
.
Ou
r
re
su
lt
s
s
h
o
w
e
d
th
a
t
train
i
n
g
ti
m
e
f
o
r
g
ra
d
ien
t
d
e
sc
e
n
t
a
lg
o
r
it
h
m
s
wa
s
lo
n
g
e
r
th
a
n
o
t
h
e
r
train
i
n
g
a
lg
o
rit
h
m
s
(8
-
1
0
se
c
o
n
d
s).
I
n
c
o
n
tras
t,
Qu
a
si
-
Ne
w
to
n
a
lg
o
rit
h
m
s
w
e
r
e
f
a
st
e
r
th
a
n
o
th
e
rs
(<
=
0
se
c
o
n
d
).
M
S
E
f
o
r
a
l
l
a
lg
o
rit
h
m
s
w
a
s
b
e
twe
e
n
0
.
1
1
7
a
n
d
0
.
2
2
8
.
W
h
il
e
t
h
e
re
wa
s
a
sig
n
if
ica
n
t
a
ss
o
c
iatio
n
b
e
tw
e
e
n
train
in
g
a
l
g
o
rit
h
m
s
a
n
d
train
in
g
ti
m
e
(p
<
0
.
0
5
),
th
e
n
u
m
b
e
r
o
f
n
e
u
ro
n
s
in
h
i
d
d
e
n
lay
e
r
h
a
d
n
o
t
a
n
y
sig
n
if
ic
a
n
t
e
ff
e
c
t
o
n
th
e
M
S
E
a
n
d
/o
r
a
c
c
u
ra
c
y
o
f
th
e
m
o
d
e
ls
(p
>
0
.
0
5
).
Ba
se
d
o
n
o
u
r
f
in
d
in
g
s,
f
o
r
d
e
v
e
lo
p
m
e
n
t
a
n
A
NN
c
las
sif
ic
a
ti
o
n
m
o
d
e
l
f
o
r
h
e
a
rt
d
ise
a
se
s,
it
is
b
e
st
t
o
u
se
Qu
a
si
-
Ne
w
to
n
train
in
g
a
lg
o
rit
h
m
s b
e
c
a
u
se
o
f
th
e
b
e
st sp
e
e
d
a
n
d
a
c
c
u
ra
c
y
.
K
ey
w
o
r
d
:
Hea
r
t D
is
ea
s
e
Ma
ch
i
n
L
ea
r
n
i
n
g
Me
d
ical
I
n
f
o
r
m
at
ics
Neu
r
al
Net
w
o
r
k
T
r
ain
in
g
A
l
g
o
r
ith
m
s
Co
p
y
rig
h
t
©
2
0
1
8
In
stit
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
:
R
ez
a
Saf
d
ar
i,
Dep
ar
t
m
en
t o
f
Hea
lt
h
I
n
f
o
r
m
a
tio
n
Ma
n
a
g
e
m
en
t,
T
eh
r
an
Un
iv
er
s
it
y
o
f
Me
d
ical
Scien
ce
s
,
E
n
g
h
elab
A
v
e,
G
h
o
d
s
A
v
e,
Fa
r
r
ed
an
esh
A
lle
y
,
No
17
,
T
eh
r
an
,
I
r
an
.
E
m
ail: r
s
a
f
d
ar
i@
t
u
m
s
.
ac
.
ir
1.
I
NT
RO
D
UCT
I
O
N
I
n
r
ec
en
t
d
ec
ad
es,
a
lar
g
e
am
o
u
n
t
o
f
d
ata
is
p
r
o
d
u
ce
d
in
h
ea
lth
ca
r
e
in
d
u
s
tr
y
ab
o
u
t
p
atien
ts
.
T
h
ese
d
ata
ar
e
a
g
o
o
d
r
eso
u
r
ce
s
to
b
e
an
al
y
ze
d
f
o
r
k
n
o
w
led
g
e
e
x
tr
ac
tio
n
th
at
e
n
ab
les
b
est
d
ec
i
s
io
n
m
a
k
in
g
[
1
,
2
]
.
I
n
o
r
d
er
to
co
n
d
u
ct
d
ata
an
aly
zi
n
g
in
th
e
m
ed
ical
d
o
m
ai
n
,
th
er
e
ar
e
v
ar
io
u
s
ap
p
r
o
ac
h
es
co
n
tain
in
g
s
t
atis
tics
,
d
ata
m
i
n
in
g
an
d
m
ac
h
i
n
e
lear
n
i
n
g
m
et
h
o
d
s
.
On
e
p
o
p
u
lar
m
eth
o
d
o
f
th
ese
ap
p
r
o
ac
h
es
is
th
e
ar
tif
icial
n
eu
r
al
n
et
w
o
r
k
(
A
NN)
.
A
N
Ns
p
r
o
v
id
e
a
p
o
w
er
f
u
l
to
o
l
to
an
aly
ze
an
d
m
o
d
el
th
e
d
ata
ac
r
o
s
s
a
b
r
o
ad
r
an
g
e
o
f
m
ed
ica
l
ap
p
licatio
n
s
.
Mo
s
t
ap
p
licatio
n
s
o
f
A
N
Ns
i
n
m
ed
ici
n
e
ar
e
class
i
f
icatio
n
p
r
o
b
lem
s
w
h
ic
h
ass
ig
n
a
n
in
p
u
t
d
ata
to
o
n
e
o
f
a
s
et
o
f
clas
s
es
in
o
u
tp
u
t
lev
el
[
3
,
4
]
.
A
n
eu
r
al
n
et
w
o
r
k
h
as
to
b
e
co
n
f
i
g
u
r
ed
s
u
ch
th
at
th
e
ap
p
licatio
n
o
f
a
s
et
o
f
i
n
p
u
t
s
p
r
o
d
u
ce
s
t
h
e
d
e
s
ir
ed
s
et
o
f
o
u
tp
u
ts
[
5
,
6
]
.
T
h
e
u
s
e
o
f
A
N
N
h
a
s
t
h
r
ee
i
m
p
o
r
tan
t
s
tep
s
f
o
r
a
n
y
p
u
r
p
o
s
es
in
cl
u
d
i
n
g
tr
ain
i
n
g
,
t
esti
n
g
a
n
d
v
a
lid
atio
n
[
7
]
.
Fo
r
co
n
f
i
g
u
r
in
g
t
h
e
ANN,
it
m
u
s
t
tr
ai
n
t
h
e
n
e
u
r
al
n
et
w
o
r
k
b
y
teac
h
i
n
g
p
a
tter
n
s
t
h
r
o
u
g
h
c
h
an
g
i
n
g
th
eir
w
ei
g
h
ts
ac
co
r
d
in
g
to
s
o
m
e
lear
n
in
g
r
u
les.
T
r
ain
in
g
o
f
th
e
n
eu
r
al
n
et
w
o
r
k
s
ca
n
b
e
d
o
n
e
b
y
v
ar
io
u
s
s
u
g
g
ested
alg
o
r
it
h
m
s
[
4
,
8
]
.
Dif
f
er
en
t
t
y
p
es
o
f
tr
ain
i
n
g
al
g
o
r
ith
m
s
w
er
e
co
m
p
ar
ed
in
v
ar
io
u
s
f
ie
l
d
s
an
d
th
eir
p
r
o
s
an
d
co
n
s
h
av
e
b
ee
n
an
al
y
ze
d
[
9
-
12
]
.
Ho
w
ev
er
,
n
o
s
tu
d
ies h
a
v
e
b
ee
n
co
n
d
u
cted
in
th
e
ca
r
d
io
v
ascu
lar
d
o
m
a
in
.
On
e
o
f
t
h
e
ar
ea
s
o
f
h
ea
lt
h
ca
r
e
w
h
er
e
t
h
e
d
a
ta
ar
e
g
r
o
w
i
n
g
u
p
i
s
th
e
ca
r
d
io
v
ascu
lar
f
ie
ld
.
Hea
r
t
d
is
ea
s
e
is
th
e
f
ir
s
t
ca
u
s
e
o
f
d
ea
th
in
d
if
f
er
e
n
t
co
u
n
tr
ie
s
an
d
ac
co
u
n
ts
f
o
r
ap
p
r
o
x
im
a
tel
y
8
0
%
o
f
all
d
ea
th
s
.
B
ased
o
n
W
HO
r
ep
o
r
t,
ab
o
u
t
1
2
m
ill
io
n
d
ea
t
h
s
p
er
y
ea
r
o
cc
u
r
in
th
e
w
o
r
l
d
d
u
e
to
th
e
h
ea
r
t
d
is
ea
s
e
s
.
T
h
e
ter
m
h
ea
r
t
d
is
ea
s
e
co
m
p
r
i
s
es
t
h
e
v
ar
io
u
s
d
is
ea
s
es
t
h
at
af
f
ec
t
t
h
e
h
ea
r
t
[
1
,
13
,
14
]
.
E
f
f
o
r
ts
to
i
m
p
r
o
v
e
lif
e
s
t
y
les
an
d
co
n
tr
o
l
r
is
k
f
ac
to
r
s
w
ill
d
ef
in
i
tel
y
co
n
tr
ib
u
te
to
h
ea
r
t
d
is
ea
s
e
p
r
ev
e
n
tio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
7
,
No
.
4
,
Dec
em
b
er
201
8
:
1
8
5
–
1
89
186
I
n
d
ee
d
,
th
e
p
r
e
d
ictiv
e
an
d
d
iag
n
o
s
is
o
f
h
ea
r
t
d
is
ea
s
e
s
in
th
e
ea
r
ly
s
ta
g
e
s
h
o
u
ld
b
e
d
o
n
e
to
r
ed
u
ce
th
e
r
is
k
o
f
h
ea
r
t d
is
ea
s
e
a
n
d
is
v
ita
l f
o
r
th
e
p
r
ev
en
tio
n
o
f
p
atie
n
t
’
s
d
ea
t
h
s
[
1
,
13
,
14
]
.
I
n
o
r
d
er
to
d
iag
n
o
s
e
h
ea
r
t
d
is
ea
s
es,
th
er
e
ar
e
v
ar
io
u
s
w
a
y
s
i
n
cl
u
d
in
g
p
h
y
s
ical
e
x
a
m
i
n
atio
n
,
ec
h
o
ca
r
d
io
g
r
a
m
,
ca
r
d
iac
n
u
cl
ea
r
s
ca
n
,
an
d
an
g
io
g
r
ap
h
y
.
Ho
w
e
v
er
,
p
h
y
s
ician
s
d
iag
n
o
s
e
h
ea
r
t
d
is
ea
s
e
b
y
lear
n
in
g
a
n
d
ex
p
er
ien
ce
.
B
ec
a
u
s
e
o
f
h
u
m
an
m
i
s
ta
k
es,
d
ia
g
n
o
s
t
ic
m
et
h
o
d
s
m
i
g
h
t
b
e
le
s
s
a
cc
u
r
ate
an
d
lead
to
er
r
o
r
s
,
f
alse
p
r
esu
m
p
tio
n
s
a
n
d
u
n
p
r
ed
ictab
le
ef
f
ec
ts
[
1
]
.
T
h
u
s
m
a
th
e
m
atica
l
al
g
o
r
it
h
m
s
s
u
c
h
as
A
NN
s
h
a
v
e
b
ee
n
u
s
ed
to
clas
s
i
f
y
h
ea
r
t
d
is
ea
s
es
[
15
]
.
Am
o
n
g
all
ap
p
lied
d
ata
m
i
n
i
n
g
m
et
h
o
d
s
,
A
N
Ns
h
av
e
h
ad
an
ac
ce
p
tab
le
p
er
f
o
r
m
an
ce
a
n
d
k
n
o
w
n
as
a
v
alu
ab
le
al
g
o
r
ith
m
f
o
r
h
ea
r
t
d
is
ea
s
e
class
i
f
icatio
n
[
16
]
.
I
n
th
e
lear
n
i
n
g
p
r
o
ce
s
s
,
u
n
d
er
s
ta
n
d
in
g
t
h
e
b
es
t
s
tr
u
ct
u
r
e
a
n
d
f
u
n
c
tio
n
to
o
b
ta
in
t
h
e
b
est
r
es
u
lt
is
cr
u
cial;
o
t
h
er
w
i
s
e,
th
er
e
w
o
u
ld
b
e
tim
e
an
d
co
s
t
co
n
s
u
m
i
n
g
if
th
e
y
ar
e
f
o
u
n
d
b
y
tr
y
an
d
er
r
o
r
.
Fo
r
A
NNs
alg
o
r
ith
m
ap
p
licatio
n
in
th
e
ar
ea
o
f
h
ea
r
t
d
is
ea
s
e,
th
e
b
est
m
et
h
o
d
an
d
s
tr
u
ctu
r
e
is
n
o
t
k
n
o
w
n
y
et.
T
h
is
s
tu
d
y
is
ai
m
ed
to
co
m
p
ar
e
s
o
m
e
A
NN
tr
ain
i
n
g
al
g
o
r
ith
m
s
a
n
d
f
i
n
d
o
u
t th
e
b
es
t
m
et
h
o
d
f
o
r
class
i
f
i
ca
tio
n
o
f
h
ea
r
t d
is
ea
s
es.
2.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
w
as
a
p
r
o
s
p
ec
ti
v
e
cr
o
s
s
-
s
ec
tio
n
al
s
tu
d
y
t
h
at
m
ea
s
u
r
ed
an
d
co
m
p
ar
ed
p
er
f
o
r
m
an
ce
a
n
d
f
u
n
ctio
n
alit
y
o
f
ar
ti
f
icial
n
e
u
r
al
n
et
w
o
r
k
tr
ai
n
in
g
a
lg
o
r
it
h
m
s
f
o
r
clas
s
i
f
icatio
n
o
f
h
ea
r
t
d
i
s
e
ases
.
Data
s
et
ta
k
e
n
f
r
o
m
U
C
I
m
ac
h
in
e
lear
n
i
n
g
r
ep
o
s
ito
r
y
[
17
]
w
as
u
s
ed
to
d
ev
e
lo
p
th
e
A
NN
-
b
ased
m
o
d
el
s
.
T
h
e
d
atab
ase
co
n
ta
in
s
3
0
3
s
am
p
les
w
it
h
7
6
attr
ib
u
tes.
Ho
w
e
v
er
,
w
e
u
s
ed
o
n
l
y
1
3
m
o
s
t
i
m
p
o
r
tan
t
attr
ib
u
tes
li
s
t
ed
in
T
ab
le
1
.
T
h
e
p
r
ed
ict
attr
ib
u
te
w
a
s
d
iag
n
o
s
i
s
o
f
h
ea
r
t
d
is
ea
s
e
in
w
h
ic
h
it
s
v
alu
e
is
‘
0
’
if
d
ia
m
eter
n
ar
r
o
w
i
n
g
=<
5
0
%
(
n
o
h
ea
r
t
di
s
ea
s
e)
an
d
is
‘
1
’
i
f
th
i
s
p
ar
am
eter
is
>5
0
%
(
p
o
s
itiv
e
h
ea
r
t
d
is
ea
s
e)
.
Fo
r
A
NNs
lear
n
in
g
p
r
o
ce
s
s
,
d
ata
w
a
s
d
iv
id
ed
in
to
th
r
ee
s
et
s
f
o
r
tr
ain
in
g
(
6
0
%),
v
alid
atio
n
(
2
0
%)
an
d
test
i
n
g
(
2
0
%).
T
o
av
o
id
p
o
s
s
ib
le
b
ias
in
t
h
e
p
r
esen
tatio
n
o
r
d
er
o
f
th
e
s
a
m
p
le
p
atter
n
s
to
th
e
A
NN,
t
h
ese
s
a
m
p
le
s
ets
w
er
e
r
an
d
o
m
ized
.
T
ab
le
1
.
A
ttrib
u
tes o
f
h
ea
r
t d
is
ea
s
es d
ata
u
s
ed
i
n
d
ev
elo
p
in
g
A
N
N
V
a
r
i
a
b
l
e
V
a
r
i
a
b
l
e
D
e
f
i
n
i
t
i
o
n
C
a
t
e
g
o
r
i
e
s o
f
V
a
l
u
e
s
A
g
e
A
g
e
o
f
p
a
t
i
e
n
t
[
2
9
-
7
7
]
S
e
x
G
e
n
d
e
r
o
f
p
a
t
i
e
n
t
(
1
=
mal
e
;
0
=
f
e
mal
e
)
CP
C
h
e
st
p
a
i
n
t
y
p
e
[1
-
4]
RBP
R
e
st
i
n
g
b
l
o
o
d
p
r
e
ssu
r
e
[
9
4
-
2
0
0
]
SC
S
e
r
u
m c
h
o
l
e
st
e
r
o
l
i
n
mg
/
d
l
[
1
2
6
,
5
6
4
]
F
B
S
F
a
st
i
n
g
b
l
o
o
d
su
g
a
r
>
1
2
0
mg
/
d
l
[0
-
1]
R
ER
R
e
st
i
n
g
e
l
e
c
t
r
o
g
r
a
p
h
i
c
r
e
su
l
t
s
[0
-
2]
M
H
R
A
M
a
x
i
m
u
m
h
e
a
r
t
r
a
t
e
a
c
h
i
e
v
e
d
[
7
1
-
2
0
2
]
E
I
A
Ex
e
r
c
i
se
i
n
d
u
c
e
d
a
n
g
i
n
a
[0
-
1]
O
l
d
-
p
e
a
k
S
T
d
e
p
r
e
ssi
o
n
i
n
d
u
c
e
d
b
y
e
x
e
r
c
i
se
r
e
l
a
t
i
v
e
t
o
r
e
st
[0
-
6
.
2
]
S
l
o
p
e
S
l
o
p
e
o
f
t
h
e
p
e
a
k
e
x
e
r
c
i
se
S
T
seg
me
n
t
[1
-
3]
NUM
N
u
mb
e
r
o
f
ma
j
o
r
v
e
sse
l
s c
o
l
o
r
e
d
b
y
f
l
u
o
r
o
sco
p
y
[0
-
3]
D
e
f
-
t
D
e
f
e
c
t
t
y
p
e
(
n
o
r
mal
,
f
i
x
e
d
,
r
e
v
e
r
si
b
l
e
d
e
f
e
c
t
)
[
3
,
6
,
7
]
D
i
a
g
n
o
si
s
C
l
a
ss o
f
h
e
a
r
t
d
i
se
a
se
0
(
n
o
h
e
a
r
t
d
i
se
a
se
)
o
r
1
(
h
a
s
h
e
a
r
t
d
i
s
e
a
se
)
T
ab
le
2
.
A
ll tr
ain
i
n
g
f
u
n
c
tio
n
s
f
o
r
co
n
d
u
ctin
g
A
N
N
T
r
a
i
n
i
n
g
A
l
g
o
r
i
t
h
m
T
r
a
i
n
i
n
g
F
u
n
c
t
i
o
n
D
e
scri
p
t
i
o
n
G
r
a
d
i
e
n
t
D
e
sce
n
t
GD
G
r
a
d
i
e
n
t
d
e
sce
n
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
G
D
M
G
r
a
d
i
e
n
t
d
e
sce
n
t
w
i
t
h
mo
me
n
t
u
m
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
RP
R
e
si
l
i
e
n
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
(
R
p
r
o
p
)
C
o
n
j
u
g
a
t
e
G
r
a
d
i
e
n
t
S
C
G
S
c
a
l
e
d
c
o
n
j
u
g
a
t
e
g
r
a
d
i
e
n
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
C
G
P
C
o
n
j
u
g
a
t
e
G
r
a
d
i
e
n
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
w
i
t
h
P
o
l
a
k
-
R
i
e
b
e
r
U
p
d
a
t
e
s
C
G
F
F
l
e
t
c
h
e
r
-
P
o
w
e
l
l
c
o
n
j
u
g
a
t
e
g
r
a
d
i
e
n
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
Q
u
a
si
-
N
e
w
t
o
n
B
F
G
B
F
G
S
q
u
a
s
i
-
N
e
w
t
o
n
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
LM
L
e
v
e
n
b
e
r
g
-
M
a
r
q
u
a
r
d
t
b
a
c
k
-
p
r
o
p
a
g
a
t
i
o
n
I
n
o
r
d
er
to
d
ev
elo
p
Mu
ltil
a
y
er
P
er
ce
p
tr
o
n
Neu
r
al
Ne
t
w
o
r
k
s
(
ML
P
NN)
,
w
e
u
s
ed
th
r
ee
m
ai
n
tr
ai
n
in
g
alg
o
r
ith
m
s
(
GD
:
Gr
ad
ien
t
D
escen
t,
C
G:
C
o
n
j
u
g
ate
Gr
ad
ien
t,
Qu
a
s
i
-
Ne
w
to
n
)
co
n
tai
n
i
n
g
ei
g
h
t
tr
ain
i
n
g
f
u
n
ctio
n
s
d
escr
ib
ed
in
tab
le
2
.
T
h
e
s
ig
m
o
id
tr
an
s
f
er
f
u
n
c
tio
n
i
s
u
s
ed
f
o
r
th
e
h
id
d
en
la
y
er
.
B
asic
s
y
s
te
m
tr
ai
n
i
n
g
p
ar
am
eter
s
ar
e
m
a
x
_
ep
o
ch
s
=1
0
0
0
,
s
h
o
w
=5
,
p
er
f
o
r
m
an
ce
g
o
al=
0
,
tim
e=
I
n
f
,
m
in
_
g
r
ad
=1
e
-
0
1
0
,
m
a
x
_
f
ail=6
ar
e
f
i
x
ed
f
o
r
ea
ch
tr
ain
in
g
f
u
n
cti
o
n
.
Fin
all
y
,
p
er
f
o
r
m
a
n
ce
ev
al
u
atio
n
o
f
ea
ch
tr
ain
i
n
g
f
u
n
ct
i
o
n
co
n
d
u
cted
w
it
h
m
ea
s
u
r
in
g
a
n
d
co
m
p
ar
in
g
t
h
e
s
p
ee
d
o
f
tr
ain
i
n
g
(
ti
m
e)
,
n
u
m
b
er
o
f
ep
o
ch
at
t
h
e
e
n
d
o
f
tr
ain
i
n
g
,
co
r
r
ec
t
class
i
f
icatio
n
p
er
ce
n
tag
e
(
ac
c
u
r
ac
y
)
,
r
eg
r
ess
io
n
o
n
tr
ain
i
n
g
,
r
eg
r
ess
io
n
o
n
v
al
id
atio
n
an
d
m
ea
n
s
q
u
ar
e
er
r
o
r
(
MSE
)
as
th
e
ev
al
u
atio
n
cr
iter
i
a
o
f
ea
ch
f
u
n
ctio
n
.
A
ll
t
h
e
s
e
p
ar
a
m
eter
s
w
er
e
ch
ec
k
ed
f
o
r
1
0
,
2
0
an
d
3
0
n
u
m
b
er
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
C
o
mp
a
r
is
o
n
o
f Neu
r
a
l Netw
o
r
k
Tr
a
in
in
g
A
lg
o
r
ith
ms fo
r
.
.
.
(
Hesa
m
K
a
r
im
)
187
o
f
n
e
u
r
o
n
s
i
n
t
h
e
h
id
d
en
la
y
er
.
On
e
-
w
a
y
a
n
al
y
s
i
s
o
f
v
ar
ia
n
c
e
(
A
NOV
A
)
w
a
s
u
s
ed
to
d
eter
m
i
n
e
w
h
e
th
er
t
h
er
e
ar
e
an
y
s
tati
s
ticall
y
s
i
g
n
i
f
i
ca
n
t
d
if
f
er
en
ce
s
b
et
w
ee
n
t
h
e
m
ea
n
s
o
f
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es
f
o
r
all
tr
ain
i
n
g
al
g
o
r
ith
m
s
.
A
NN
to
o
lb
o
x
in
MA
T
L
A
B
2
0
1
0
w
as u
s
ed
to
c
o
n
s
tr
u
ct
n
e
u
r
al
n
et
w
o
r
k
s
f
o
r
d
iag
n
o
s
i
n
g
o
f
th
e
h
ea
r
t
d
is
ea
s
e.
SP
SS
(
v
er
s
io
n
2
0
1
5
)
also
u
s
ed
f
o
r
s
ta
tis
tica
l
d
ata
an
al
y
s
i
s
.
A
ll
t
h
ese
e
x
p
er
i
m
en
ts
w
er
e
ca
r
r
ied
o
u
t o
n
W
in
d
o
w
s
7
(
3
2
-
bi
t)
o
p
er
atin
g
s
y
s
te
m
w
it
h
I
n
tel(
R
)
C
o
r
e(
T
M)
i5
2
.
5
0
GHz
p
r
o
ce
s
s
o
r
an
d
6
GB
R
A
M.
3.
RE
SU
L
T
S
A
ND
D
I
SCU
SS
I
O
N
I
n
t
h
is
s
t
u
d
y
,
w
e
co
m
p
ar
ed
th
e
p
er
f
o
r
m
an
ce
o
f
ei
g
h
t
ANN
tr
ain
in
g
f
u
n
ct
io
n
f
o
r
h
e
ar
t
d
is
ea
s
e
class
i
f
icatio
n
.
T
h
e
r
esu
lt
o
f
t
h
is
ev
al
u
atio
n
is
s
h
o
w
n
in
ta
b
le
3
.
A
s
s
h
o
w
n
in
tab
le
3
,
t
r
ain
in
g
t
i
m
e
r
an
g
es
b
et
w
ee
n
8
an
d
1
0
s
ec
o
n
d
s
f
o
r
GD
an
d
GDM
(
g
r
ad
ien
t
d
esce
n
t
w
it
h
m
o
m
e
n
tu
m
)
r
esp
ec
tiv
el
y
.
T
im
e
m
ea
s
u
r
e
m
e
n
t
f
o
r
r
e
m
a
in
alg
o
r
ith
m
s
w
as
i
n
a
r
ag
e
o
f
0
-
2
s
e
co
n
d
s
.
T
r
ain
in
g
p
r
o
ce
s
s
e
n
d
ed
in
ep
o
ch
1
0
0
0
f
o
r
GD
an
d
GD
M
al
g
o
r
ith
m
s
.
All
o
th
er
alg
o
r
it
h
m
s
e
n
d
ed
in
ep
o
ch
2
-
2
2
.
A
v
er
a
g
e
o
f
ac
c
u
r
ac
y
f
o
r
Qu
a
s
i
-
Ne
w
to
n
alg
o
r
ith
m
s
(
8
6
.
0
6
%),
GD
(
8
3
.
1
3
)
an
d
C
G
(
8
3
.
1
4
)
w
er
e
o
b
tain
ed
.
Ma
x
i
m
u
m
an
d
m
i
n
i
m
u
m
r
eg
r
ess
io
n
v
al
u
e
o
n
tr
ain
i
n
g
w
er
e
0
.
9
9
9
(
L
M:
L
ev
e
n
b
er
g
-
Ma
r
q
u
ar
d
t
b
ac
k
-
p
r
o
p
ag
atio
n
)
an
d
0
.
1
7
3
(
C
GF:
C
o
n
j
u
g
ate
Gr
ad
ie
n
t
b
ac
k
-
p
r
o
p
ag
atio
n
w
it
h
Fletc
h
er
-
R
e
ev
es
Up
d
ates)
,
r
esp
ec
tiv
el
y
.
MSE
f
o
r
all
alg
o
r
ith
m
s
w
as
b
et
w
ee
n
0
.
1
1
7
an
d
0
.
2
2
8
.
B
ased
o
n
r
esu
lts
o
f
v
ar
i
an
ce
an
al
y
s
i
s
s
h
o
w
ed
in
tab
le
4
,
s
tatis
ticall
y
,
th
er
e
w
as n
o
s
i
g
n
i
f
ica
n
t
d
i
f
f
er
e
n
ce
b
et
w
ee
n
MSE
/
A
cc
u
r
ac
y
i
n
g
r
o
u
p
s
o
f
alg
o
r
ith
m
s
an
d
n
u
m
b
er
o
f
h
id
d
en
lay
er
s
(
p
>0
.
0
5
)
.
B
etw
ee
n
tr
ain
i
n
g
al
g
o
r
ith
m
s
a
n
d
tr
ain
in
g
ti
m
e,
t
h
er
e
w
as a
s
ig
n
i
f
ica
n
t a
s
s
o
ciat
io
n
(
p
<0
.
0
5
)
.
T
h
e
m
ea
n
tr
ai
n
i
n
g
t
i
m
e
f
o
r
GD
a
n
d
GDM
w
as
9
.
3
an
d
8
.
3
s
ec
o
n
d
s
r
esp
ec
tiv
el
y
.
I
n
r
etu
r
n
,
th
e
m
ea
n
tr
ain
i
n
g
ti
m
e
f
o
r
R
P
(
r
esil
ien
t
b
ac
k
-
p
r
o
p
ag
atio
n
)
(
0
s
ec
.
)
,
L
M
an
d
C
GF (
0
.
3
3
s
ec
.
)
w
er
e
o
b
tain
ed
an
d
r
ep
o
r
ted
in
T
ab
le
3
.
T
r
ain
in
g
o
f
th
e
n
eu
r
al
n
et
w
o
r
k
s
ca
n
b
e
d
o
n
e
b
y
d
if
f
er
en
t
o
p
tim
izatio
n
alg
o
r
ith
m
s
[
7
,
8
]
.
I
n
t
h
is
s
t
u
d
y
,
w
e
co
m
p
ar
ed
t
h
r
ee
m
ai
n
c
lass
es
o
f
tr
ai
n
i
n
g
al
g
o
r
ith
m
s
co
n
tai
n
i
n
g
e
ig
h
t
m
et
h
o
d
s
f
o
r
class
i
f
ica
tio
n
o
f
h
ea
r
t
d
i
s
ea
s
es.
On
e
o
f
t
h
e
m
ai
n
m
ea
s
u
r
e
m
e
n
t
s
f
o
r
ev
al
u
a
tio
n
o
f
ea
c
h
al
g
o
r
it
h
m
w
as
ac
c
u
r
ac
y
.
B
ased
o
n
o
u
r
r
esu
lt
s
th
e
m
a
x
i
m
u
m
ac
cu
r
ac
y
w
as
f
o
r
Qu
asi
-
Ne
w
to
n
al
g
o
r
ith
m
s
(
9
1
.
7
5
%)
.
Qu
asi
-
Ne
w
t
o
n
m
et
h
o
d
s
ex
p
lo
it
g
r
ad
ien
t i
n
f
o
r
m
atio
n
to
ap
p
r
o
x
i
m
ate
th
e
Hess
ian
m
atr
i
x
o
f
t
h
e
er
r
o
r
f
u
n
c
tio
n
w
it
h
r
esp
ec
t
to
th
e
p
ar
a
m
eter
s
o
f
th
e
n
et
w
o
r
k
.
T
h
is
ap
p
r
o
x
im
at
io
n
m
atr
i
x
is
s
u
b
s
eq
u
e
n
tl
y
u
s
ed
to
d
eter
m
i
n
e
an
e
f
f
ec
tiv
e
s
ea
r
ch
d
ir
ec
tio
n
an
d
u
p
d
ate
th
e
v
al
u
es
o
f
th
e
p
ar
am
eter
s
[
18
]
.
T
h
e
ef
f
ec
ti
v
e
n
es
s
o
f
tr
ai
n
i
n
g
al
g
o
r
it
h
m
s
w
as
m
ea
s
u
r
ed
b
y
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
MSE
)
.
A
lt
h
o
u
g
h
s
o
m
e
s
tu
d
ies
b
eliev
e
th
at
n
e
t
w
o
r
k
s
ar
e
s
e
n
s
iti
v
e
to
t
h
e
n
u
m
b
er
o
f
n
eu
r
o
n
s
i
n
th
eir
h
id
d
en
la
y
er
s
[
19
]
,
w
e
d
id
n
o
t
f
in
d
a
n
y
s
i
g
n
if
ican
t
as
s
o
ciatio
n
b
et
w
ee
n
t
h
e
n
u
m
b
er
o
f
n
e
u
r
o
n
s
i
n
h
id
d
e
n
la
y
er
s
an
d
m
o
d
els
ac
cu
r
ac
y
,
an
d
MSE
.
W
e
u
s
ed
r
eg
r
ess
io
n
an
al
y
s
is
f
u
n
ctio
n
in
o
r
d
er
to
c
o
m
p
ar
e
th
e
ac
tu
a
l
o
u
tp
u
ts
t
h
e
alg
o
r
ith
m
s
w
i
th
t
h
e
d
esire
d
o
u
tp
u
ts
.
Ma
x
i
m
u
m
r
eg
r
ess
io
n
v
al
u
e
o
n
tr
ain
i
n
g
w
a
s
f
o
r
L
M
alg
o
r
ith
m
.
Ou
r
r
esu
lt
s
ab
o
u
t
r
eg
r
ess
io
n
v
alu
es
is
s
i
m
i
lar
to
th
e
r
esu
lt
o
f
Sh
ar
m
a
’
s
s
t
u
d
y
[
9
]
.
I
t
s
h
o
w
s
th
at
th
e
co
r
r
elatio
n
co
ef
f
icie
n
t
(
R
)
b
et
w
ee
n
ac
t
u
al
an
d
d
esire
d
o
u
tp
u
t
i
n
L
M
al
g
o
r
ith
m
is
ac
ce
p
tab
le,
s
o
,
t
h
is
a
lg
o
r
ith
m
i
s
p
r
o
p
er
to
class
i
f
icatio
n
tas
k
.
An
o
th
er
p
er
f
o
r
m
a
n
ce
m
ea
s
u
r
e
ev
alu
ated
in
th
i
s
s
tu
d
y
w
a
s
co
m
p
u
tatio
n
ti
m
e
o
f
tr
ain
i
n
g
alg
o
r
ith
m
s
.
B
ased
o
n
o
u
r
f
i
n
d
in
g
s
,
s
i
m
p
le
GD
a
n
d
GDM
al
g
o
r
ith
m
s
r
u
n
s
lo
w
er
t
h
an
o
t
h
er
s
.
GD
al
g
o
r
ith
m
i
s
k
n
o
w
n
as
s
teep
est
d
escen
t
s
tar
t
w
it
h
a
r
an
d
o
m
w
e
ig
h
t
v
ec
to
r
.
T
h
e
w
ei
g
h
t
v
ec
to
r
w
ill
b
e
m
o
d
if
i
ed
iter
ativ
el
y
u
n
t
il
a
m
i
n
i
m
u
m
in
t
h
e
er
r
o
r
s
u
r
f
ac
e
is
f
o
u
n
d
[
20
-
22
]
.
GD
tak
es
m
a
n
y
s
m
all
s
tep
s
to
r
ea
ch
t
h
e
m
i
n
i
m
u
m
er
r
o
r
;
th
er
ef
o
r
e,
its
r
elativ
el
y
s
lo
w
a
n
d
in
ef
f
icie
n
t
[
22
]
.
A
lth
o
u
g
h
s
o
m
e
alg
o
r
ith
m
s
s
u
c
h
as
th
e
GDM
an
d
R
P
h
av
e
b
ee
n
p
r
o
p
o
s
ed
f
o
r
im
p
r
o
v
i
n
g
t
h
e
s
p
ee
d
o
f
co
n
v
er
g
e
n
ce
o
f
G
D
alg
o
r
ith
m
s
,
o
u
r
r
esu
lts
s
h
o
wed
a
lo
w
er
ex
ec
u
tio
n
ti
m
e
f
o
r
GDM
.
T
h
e
m
o
m
e
n
t
u
m
v
ar
iat
io
n
is
u
s
u
all
y
f
aster
t
h
an
s
i
m
p
le
GD
b
ec
au
s
e
it
all
o
w
s
h
i
g
h
er
lear
n
in
g
r
ates
[
19
]
.
Ho
w
ev
er
,
R
P
ex
ec
u
tio
n
ti
m
e
w
a
s
f
aster
th
a
n
G
D
an
d
G
DM
(
n
ea
r
0
s
ec
)
.
R
P
tr
ain
i
n
g
al
g
o
r
ith
m
k
n
o
w
n
a
s
R
p
r
o
p
ch
an
g
e
s
th
e
w
ei
g
h
t
v
ec
to
r
ac
co
r
d
in
g
to
s
ep
ar
ate
u
p
d
ate
v
alu
e.
T
h
is
alg
o
r
ith
m
is
ea
s
y
to
co
m
p
u
te
lo
ca
l
lear
n
i
n
g
s
c
h
e
m
e
an
d
ea
s
y
to
i
m
p
le
m
en
t
;
it
is
d
u
e
to
n
o
ch
o
ice
o
f
p
ar
am
eter
s
r
eq
u
ir
e
m
e
n
t
at
all
p
r
o
ce
s
s
to
o
b
tain
o
p
tim
al
co
n
v
er
g
en
ce
ti
m
e
s
.
T
h
e
n
u
m
b
e
r
o
f
lear
n
in
g
s
tep
s
is
s
i
g
n
i
f
i
ca
n
tl
y
r
ed
u
ce
d
in
co
m
p
ar
is
o
n
to
th
e
o
r
ig
i
n
al
g
r
ad
ien
t
-
d
esce
n
t p
r
o
ce
d
u
r
e
[
23
]
t
h
u
s
R
P
is
f
aster
t
h
an
GD
a
n
d
GDM
.
Ou
r
f
i
n
d
in
g
s
h
o
w
ed
lo
w
e
x
e
cu
tio
n
ti
m
e
f
o
r
SC
G
(
Scaled
co
n
j
u
g
ate
g
r
ad
ien
t)
,
C
GP
(
C
o
n
j
u
g
a
te
Gr
ad
ien
t
b
ac
k
-
p
r
o
p
ag
atio
n
w
it
h
P
o
lak
-
R
ieb
er
Up
d
ates)
,
an
d
C
GF
a
s
C
G
al
g
o
r
ith
m
s
.
C
G
alg
o
r
it
h
m
i
m
p
le
m
en
ted
as
an
iter
ativ
e
alg
o
r
ith
m
.
I
t
s
tar
ts
o
u
t
b
y
s
e
ar
ch
in
g
in
th
e
n
e
g
ati
v
e
o
f
th
e
g
r
ad
ien
t
an
d
th
en
p
er
f
o
r
m
s
a
li
n
e
s
ea
r
ch
to
d
eter
m
i
n
e
t
h
e
o
p
ti
m
al
d
is
ta
n
ce
to
m
o
v
e
alo
n
g
th
e
cu
r
r
en
t
s
ea
r
c
h
d
ir
ec
tio
n
.
Sear
ch
i
n
g
alo
n
g
w
it
h
co
n
j
u
g
ate
d
ir
ec
tio
n
s
lead
s
to
f
aster
co
n
v
er
g
e
n
ce
th
an
s
teep
es
t
d
escen
t
d
ir
ec
tio
n
s
[
24
,
25
]
.
T
h
e
SC
G
m
et
h
o
d
w
a
s
d
esi
g
n
ed
to
av
o
id
th
e
ti
m
e
-
co
n
s
u
m
i
n
g
lin
e
s
ea
r
c
h
i
n
C
G
al
g
o
r
it
h
m
s
.
T
h
is
al
g
o
r
ith
m
r
eq
u
ir
es
m
o
r
e
iter
atio
n
s
to
co
n
v
er
g
e
r
at
h
er
t
h
an
th
e
o
th
er
C
G
a
lg
o
r
it
h
m
s
;
h
o
w
ev
er
,
t
h
e
n
u
m
b
er
o
f
co
m
p
u
tatio
n
s
i
n
ea
c
h
s
tep
is
s
i
g
n
if
ic
an
tl
y
r
ed
u
ce
d
as n
o
l
in
e
s
ea
r
c
h
is
p
er
f
o
r
m
ed
[
19
]
.
C
GF
i
s
an
u
p
d
ated
v
er
s
io
n
o
f
C
G
w
h
ich
co
m
p
u
te
s
n
e
w
s
ea
r
c
h
d
ir
ec
tio
n
a
s
t
h
e
r
ati
o
o
f
th
e
n
o
r
m
s
q
u
ar
ed
o
f
th
e
c
u
r
r
en
t
g
r
ad
ien
t
to
t
h
e
n
o
r
m
s
q
u
a
r
ed
o
f
th
e
p
r
ev
io
u
s
g
r
ad
ien
t
[
26
-
28
]
.
C
GP
ca
lcu
lat
es
n
e
w
s
ea
r
ch
d
ir
ec
tio
n
as
th
e
r
a
tio
n
o
f
t
h
e
i
n
n
er
p
r
o
d
u
ct
o
f
t
h
e
p
r
ev
io
u
s
c
h
an
g
e
in
t
h
e
g
r
ad
ien
t
w
i
th
t
h
e
c
u
r
r
en
t
g
r
ad
ien
t
d
i
v
id
ed
b
y
t
h
e
n
o
r
m
s
q
u
ar
ed
o
f
th
e
p
r
ev
io
u
s
g
r
ad
ien
t
[
9
,
28
]
.
Gen
er
all
y
,
th
e
e
x
ec
u
tio
n
t
i
m
e
o
f
Q
u
a
s
i
-
N
e
w
to
n
al
g
o
r
ith
m
s
w
as
s
i
m
ilar
t
o
C
G
al
g
o
r
ith
m
s
.
I
n
Ne
w
to
n
m
et
h
o
d
s
,
a
q
u
ad
r
atic
ap
p
r
o
x
im
a
tio
n
i
s
u
s
ed
in
s
tea
d
o
f
a
lin
ea
r
ap
p
r
o
x
i
m
at
io
n
o
f
th
e
er
r
o
r
f
u
n
ct
io
n
.
T
h
e
m
a
in
ad
v
a
n
ta
g
e
o
f
t
h
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8938
IJ
-
AI
Vo
l.
7
,
No
.
4
,
Dec
em
b
er
201
8
:
1
8
5
–
1
89
188
Ne
w
to
n
m
eth
o
d
s
i
s
th
at
it
h
as
a
q
u
ad
r
atic
co
n
v
er
g
e
n
ce
r
ate
w
h
ile
t
h
e
s
teep
est
d
esce
n
t
h
as
a
m
u
c
h
s
lo
w
er
li
n
ea
r
co
n
v
er
g
e
n
ce
r
ate.
Ho
w
ev
er
,
e
ac
h
s
tep
o
f
t
h
i
s
m
et
h
o
d
r
eq
u
i
r
es
a
lar
g
e
a
m
o
u
n
t
o
f
co
m
p
u
tat
io
n
[
29
]
.
A
v
ar
iet
y
o
f
al
g
o
r
ith
m
s
w
er
e
d
esi
g
n
e
d
b
ase
o
n
Ne
w
to
n
m
et
h
o
d
s
.
B
FGS
(
B
r
o
y
d
en
–
Fletc
h
er
–
Go
ld
f
ar
b
–
Sh
an
n
o
)
alg
o
r
ith
m
is
a
n
iter
ativ
e
m
et
h
o
d
f
o
r
s
o
lv
in
g
u
n
co
n
s
tr
ai
n
e
d
n
o
n
lin
ea
r
o
p
ti
m
iza
tio
n
p
r
o
b
le
m
s
t
h
at
u
s
es
an
ap
p
r
o
x
im
a
te
Hess
ia
n
m
atr
i
x
in
co
m
p
u
ti
n
g
th
e
s
ea
r
ch
d
ir
ec
tio
n
[
29
]
.
L
M
alg
o
r
ith
m
w
as
d
esig
n
ed
to
ap
p
r
o
ac
h
s
ec
o
n
d
-
o
r
d
er
tr
ain
in
g
s
p
ee
d
w
ith
o
u
t
h
av
in
g
to
co
m
p
u
te
th
e
Hess
ia
n
m
atr
i
x
.
T
h
is
alg
o
r
ith
m
ap
p
ea
r
s
to
b
e
th
e
f
aste
s
t
m
eth
o
d
f
o
r
tr
ain
in
g
m
o
d
er
ate
-
s
ized
f
ee
d
-
f
o
r
w
ar
d
n
e
u
r
al
n
et
w
o
r
k
s
[
19
,
30
]
b
u
t is n
o
t su
i
tab
le
f
o
r
a
lar
g
e
n
u
m
b
er
o
f
d
ata
[
31
]
.
T
h
e
m
ai
n
d
r
aw
b
ac
k
o
f
t
h
e
L
M
is
t
h
at
it
r
eq
u
ir
es th
e
s
to
r
ag
e
o
f
s
o
m
e
m
atr
ices t
h
at
ca
n
b
e
q
u
ite
lar
g
e
f
o
r
ce
r
tain
p
r
o
b
lem
s
[
19
]
.
C
G
alg
o
r
it
h
m
s
ar
e
ch
ar
ac
ter
ized
b
y
lo
w
m
e
m
o
r
y
r
e
q
u
ir
e
m
e
n
ts
,
f
ast
a
n
d
s
tr
o
n
g
lo
ca
l a
n
d
g
lo
b
al
co
n
v
er
g
en
ce
p
r
o
p
er
ties
[
3
2
]
.
T
h
u
s
,
it
ca
n
b
e
u
s
ed
to
s
p
ar
s
e
s
y
s
te
m
s
th
at
d
i
m
en
s
io
n
ar
e
to
o
l
ar
g
e
an
d
to
s
o
lv
e
u
n
co
n
s
t
r
ain
ed
o
p
ti
m
izatio
n
p
r
o
b
le
m
s
[
25
,
33
]
.
T
h
e
s
to
r
ag
e
r
eq
u
ir
e
m
en
ts
f
o
r
C
GP
(
f
o
u
r
v
ec
to
r
s
)
ar
e
s
lig
h
tl
y
lar
g
er
t
h
a
n
f
o
r
C
G
F
[
9
,
28
]
.
So
m
e
i
m
p
o
r
tan
t
f
ac
to
r
s
s
u
c
h
as
tr
ain
in
g
ti
m
e,
m
e
m
o
r
y
n
ee
d
an
d
ac
cu
r
ac
y
m
u
s
t
b
e
co
n
s
id
er
ed
in
o
r
d
e
r
to
ch
o
o
s
e
th
e
b
est tr
ain
in
g
al
g
o
r
ith
m
.
A
cc
o
r
d
in
g
to
th
e
f
in
d
i
n
g
o
f
th
i
s
s
t
u
d
y
,
GD
a
n
d
GD
M
alg
o
r
ith
m
s
ar
e
to
o
s
lo
w
;
in
co
n
tr
ast,
tr
ain
i
n
g
alg
o
r
ith
m
s
b
ased
o
n
Ne
w
to
n
m
e
th
o
d
co
n
v
er
g
e
i
n
les
s
iter
atio
n
an
d
ar
e
f
aster
a
n
d
m
o
r
e
ac
c
u
r
ate
.
I
n
ad
d
itio
n
,
t
h
e
C
G
alg
o
r
it
h
m
s
r
eq
u
ir
e
m
o
r
e
s
to
r
ag
e
th
a
n
th
e
o
t
h
er
alg
o
r
it
h
m
s
.
I
t
is
b
etter
to
u
s
e
L
M
tr
ain
in
g
f
o
r
s
m
all
an
d
m
e
d
iu
m
-
s
ize
n
et
w
o
r
k
s
if
t
h
er
e
is
en
o
u
g
h
m
e
m
o
r
y
.
Fo
r
lar
g
e
n
et
w
o
r
k
s
,
SC
G
o
r
R
P
alg
o
r
ith
m
s
ar
e
a
s
u
itab
le
c
h
o
ice
[
19
,
24
,
33
]
.
Fin
all
y
,
Qu
a
s
i
-
Ne
w
to
n
m
eth
o
d
s
ar
e
g
en
er
al
l
y
co
n
s
id
er
ed
m
o
r
e
p
o
w
er
f
u
l c
o
m
p
ar
ed
to
o
th
er
tr
ain
i
n
g
al
g
o
r
ith
m
s
[
18
]
.
T
ab
le
3
.
C
o
m
p
ar
is
o
n
o
f
A
NN
T
r
ain
in
g
F
u
n
c
tio
n
s
b
ased
o
n
th
e
v
al
u
es o
f
A
cc
u
r
ac
y
,
ti
m
e
an
d
n
e
u
r
o
n
n
u
m
b
er
in
h
id
d
en
la
y
er
T
r
a
i
n
i
n
g
A
l
g
o
r
i
t
h
m
T
r
a
i
n
i
n
g
F
u
n
c
t
i
o
n
H
M
S
E
Ep
o
c
h
R
T
r
a
i
n
R
V
a
l
i
d
a
t
i
o
n
A
c
c
u
r
a
c
y
Ex
e
c
u
t
i
o
n
T
i
me
(
S
e
c
)
G
r
a
d
i
e
n
t
D
e
sce
n
t
GD
10
0
.
1
1
7
1
0
0
0
0
.
6
0
6
0
.
7
3
5
8
3
.
5
0
9
20
0
.
1
7
3
1
0
0
0
0
.
7
1
7
0
.
5
7
3
8
1
.
8
5
9
30
0
.
1
3
1
1
0
0
0
0
.
6
9
2
0
.
6
8
1
8
1
.
8
5
10
G
D
M
10
0
.
1
7
5
1
0
0
0
0
.
7
5
5
0
.
5
7
8
8
4
.
4
9
9
20
0
.
1
7
5
1
0
0
0
0
.
7
3
4
0
.
6
0
8
8
1
.
1
9
8
30
0
.
2
0
0
1
0
0
0
0
.
6
9
5
0
.
5
6
4
8
1
.
5
2
8
RP
10
0
.
1
3
8
6
0
.
7
8
7
0
.
6
3
2
8
4
.
1
6
0
20
0
.
1
5
4
14
0
.
8
1
5
0
.
6
3
8
8
4
.
8
2
0
30
0
.
2
0
2
19
0
.
8
6
2
0
.
4
8
9
8
4
.
8
2
0
C
o
n
j
u
g
a
t
e
G
r
a
d
i
e
n
t
S
C
G
10
0
.
1
2
2
17
0
.
8
3
8
0
.
6
7
7
8
6
.
9
0
1
20
0
.
1
2
1
14
0
.
8
1
7
0
.
6
5
7
8
4
.
4
9
1
30
0
.
1
9
1
13
0
.
8
3
7
0
.
5
2
6
8
3
.
5
0
0
C
G
P
10
0
.
1
8
9
4
0
.
1
9
1
0
.
3
2
4
8
0
.
8
6
1
20
0
.
1
4
8
5
0
.
3
4
6
0
.
6
3
4
8
1
.
8
5
0
30
0
.
1
5
5
16
0
.
3
5
9
0
.
5
8
3
8
7
.
0
9
1
C
G
F
10
0
.
1
1
2
10
0
.
5
3
2
0
.
7
0
5
8
3
.
5
0
1
20
0
.
1
3
6
11
0
.
5
0
7
0
.
6
5
4
8
4
.
1
6
0
30
0
.
2
2
8
3
0
.
1
7
3
0
.
2
3
1
7
5
.
9
1
0
Q
u
a
si
-
N
e
w
t
o
n
B
F
G
10
0
.
1
1
1
22
0
.
4
0
1
0
.
5
0
4
8
6
.
4
7
1
20
0
.
1
6
0
15
0
.
3
3
3
0
.
4
0
5
8
5
.
1
5
2
30
0
.
1
2
4
8
0
.
3
7
1
0
.
6
6
5
8
6
.
8
0
2
LM
10
0
.
1
3
9
5
0
.
8
8
8
0
.
7
8
8
8
2
.
5
1
0
20
0
.
1
6
5
2
0
.
9
5
2
0
.
8
7
5
8
3
.
8
3
0
30
0
.
1
5
3
8
0
.
9
9
9
0
.
8
5
8
9
1
.
7
5
1
H
:
N
u
mb
e
r
o
f
n
e
u
r
o
n
s
i
n
h
i
d
d
e
n
l
a
y
e
r
.
M
S
E:
M
e
a
n
o
f
S
q
u
a
r
e
Er
r
o
r
.
R:
R
e
g
r
e
ssi
o
n
.
T
ab
le
4
.
On
e
-
w
a
y
ANOV
A
r
e
s
u
lt
f
o
r
co
m
p
ar
in
g
m
ea
n
s
o
f
p
er
f
o
r
m
an
ce
m
ea
s
u
r
es i
n
an
y
tr
ain
i
n
g
al
g
o
r
ith
m
s
P
e
r
f
o
r
man
c
e
me
a
su
r
e
s
T
r
a
i
n
i
n
g
a
l
g
o
r
i
t
h
m
N
u
mb
e
r
o
f
h
i
d
d
e
n
l
a
y
e
r
F
P
-
v
a
l
u
e
F
P
-
v
a
l
u
e
M
S
E
0
.
7
2
5
0
.
6
5
4
2
.
8
4
0
0
.
0
8
1
A
c
c
u
r
a
c
y
1
.
2
2
8
0
.
3
4
4
0
.
1
3
5
0
.
8
7
5
T
i
me
1
1
.
3
7
8
<
=
0
.
0
0
1
0
.
2
1
4
0
.
8
0
9
Ep
o
c
h
2
.
9
8
5
E
4
<
=
0
.
0
0
1
0
.
0
0
0
1
.
0
0
0
4.
CO
NCLU
SI
O
N
I
n
co
n
cl
u
s
io
n
,
f
o
r
A
N
N
clas
s
i
f
icatio
n
m
o
d
el
d
e
v
elo
p
m
en
t
f
o
r
h
ea
r
t
d
is
ea
s
es,
it
is
b
est
to
u
s
e
Q
u
a
s
i
-
Ne
w
to
n
tr
ain
i
n
g
alg
o
r
it
h
m
s
b
ec
au
s
e
o
f
b
est
s
p
ee
d
an
d
ac
cu
r
ac
y
.
A
l
s
o
,
th
e
n
u
m
b
er
o
f
n
e
u
r
o
n
s
in
h
id
d
en
la
y
er
h
as
n
o
s
ig
n
i
f
ica
n
t e
f
f
ec
t o
n
th
e
p
er
f
o
r
m
a
n
ce
m
o
d
el.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
I
SS
N:
2252
-
8938
C
o
mp
a
r
is
o
n
o
f Neu
r
a
l Netw
o
r
k
Tr
a
in
in
g
A
lg
o
r
ith
ms fo
r
.
.
.
(
Hesa
m
K
a
r
im
)
189
RE
F
E
R
E
NC
E
S
[1
]
Da
n
g
a
re
CS
,
e
t
a
l
.
"
A
d
a
ta
m
in
in
g
a
p
p
r
o
a
c
h
f
o
r
p
re
d
ictio
n
o
f
h
e
a
rt
d
ise
a
se
u
si
n
g
n
e
u
ra
l
n
e
tw
o
rk
s
,
"
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
m
p
u
ter
E
n
g
i
n
e
e
rin
g
&
T
e
c
h
n
o
lo
g
y
(
IJ
CET
),
v
o
l.
3
,
p
p
.
3
0
-
4
0
p
,
2
0
1
2
.
[2
]
S
p
a
th
is
D,
e
t
a
l.
"
Dia
g
n
o
sin
g
a
sth
m
a
a
n
d
c
h
ro
n
ic
o
b
str
u
c
ti
v
e
p
u
l
m
o
n
a
r
y
d
ise
a
se
w
it
h
m
a
c
h
in
e
le
a
rn
in
g
,
"
He
a
lt
h
In
fo
rm
a
t
ics
J
o
u
rn
a
l,
v
o
l.
0
,
p
p
.
1
4
6
.
[3
]
Kh
a
n
IY,
e
t
a
l.
"
Im
p
o
rtan
c
e
o
f
Artif
icia
l
Ne
u
ra
l
Ne
t
w
o
rk
in
M
e
d
ica
l
Dia
g
n
o
sis
d
ise
a
se
li
k
e
a
c
u
te
n
e
p
h
rit
is
d
ise
a
se
a
n
d
h
e
a
rt
d
ise
a
se
,
"
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
E
n
g
i
n
e
e
rin
g
S
c
ien
c
e
a
n
d
I
n
n
o
v
a
ti
v
e
T
e
c
h
n
o
lo
g
y
(
IJ
ES
IT
),
v
o
l.
2
,
p
p
.
210
-
7
,
2
0
1
3
.
[4
]
Krö
se
B,
e
t
a
l.
"
A
n
in
tro
d
u
c
ti
o
n
t
o
n
e
u
ra
l
n
e
tw
o
rk
s,"
v
o
l.
p
p
.
1
9
9
3
.
[5
]
Al
-
Ba
h
ra
n
i
R,
e
t
a
l.
"
S
u
rv
iv
a
b
il
it
y
p
re
d
ictio
n
o
f
c
o
lo
n
c
a
n
c
e
r
p
a
ti
e
n
ts
u
si
n
g
n
e
u
ra
l
n
e
tw
o
rk
s,"
He
a
lt
h
I
n
f
o
rm
a
ti
c
s
J
o
u
rn
a
l,
v
o
l
.
0
,
p
p
.
1
4
6
.
[6
]
S
e
lm
a
B,
e
t
a
l.
"
Ne
u
ra
l
n
e
tw
o
rk
n
a
v
ig
a
ti
o
n
tec
h
n
i
q
u
e
f
o
r
u
n
m
a
n
n
e
d
v
e
h
icle
,
"
T
re
n
d
s
i
n
Ap
p
li
e
d
S
c
i
e
n
c
e
s
Res
e
a
rc
h
,
v
o
l.
9
,
p
p
.
2
4
6
,
2
0
1
4
.
[7
]
Ha
sh
im
M
N,
e
t
a
l.
"
A
Co
m
p
a
riso
n
S
t
u
d
y
o
f
L
e
a
rn
in
g
A
l
g
o
rit
h
m
s
fo
r
Esti
m
a
ti
n
g
F
a
u
lt
L
o
c
a
ti
o
n
,
"
I
n
d
o
n
e
sia
n
J
o
u
rn
a
l
o
f
El
e
c
trica
l
En
g
in
e
e
rin
g
a
n
d
Co
mp
u
ter
S
c
ien
c
e
,
v
o
l
.
6
,
p
p
.
2
0
1
7
.
[8
]
Ya
d
a
v
N,
e
t
a
l.
A
n
In
tro
d
u
c
ti
o
n
t
o
Ne
u
ra
l
Ne
tw
o
rk
M
e
th
o
d
s f
o
r
Di
ff
e
r
e
n
ti
a
l
Eq
u
a
ti
o
n
s:
S
p
rin
g
e
r
;
2
0
1
5
.
[9
]
S
h
a
rm
a
B.
"
Co
m
p
a
riso
n
o
f
n
e
u
ra
l
n
e
tw
o
rk
train
in
g
f
u
n
c
ti
o
n
s
f
o
r
He
m
a
to
m
a
c
las
si
f
ica
ti
o
n
in
b
ra
in
CT
im
a
g
e
s,
"
IOS
R
J
o
u
rn
a
ls (
IOS
R
J
o
u
r
n
a
l
o
f
C
o
mp
u
ter
En
g
in
e
e
ri
n
g
),
v
o
l
.
1
,
p
p
.
3
1
-
5
,
2
0
1
4
.
[1
0
]
Ag
g
a
r
wa
l
K,
e
t
a
l.
"
E
v
a
lu
a
ti
o
n
o
f
v
a
rio
u
s
train
in
g
a
lg
o
rit
h
m
s
in
a
n
e
u
ra
l
n
e
tw
o
rk
m
o
d
e
l
f
o
r
so
f
t
wa
re
e
n
g
in
e
e
rin
g
a
p
p
li
c
a
ti
o
n
s,"
ACM
S
IGS
OF
T
S
o
f
twa
re
En
g
in
e
e
rin
g
No
tes
,
v
o
l.
3
0
,
p
p
.
1
-
4
,
2
0
0
5
.
[1
1
]
Zh
o
u
L
,
e
t
a
l.
"
T
ra
in
in
g
a
lg
o
rit
h
m
p
e
rf
o
r
m
a
n
c
e
f
o
r
i
m
a
g
e
c
las
sifica
ti
o
n
b
y
n
e
u
ra
l
n
e
tw
o
rk
s,"
Ph
o
to
g
ra
mm
e
tric
En
g
i
n
e
e
rin
g
&
Rem
o
te S
e
n
sin
g
,
v
o
l.
7
6
,
p
p
.
9
4
5
-
5
1
,
2
0
1
0
.
[1
2
]
Al
-
Ha
sa
n
a
t
A
,
e
t
a
l.
"
Ex
p
e
ri
m
e
n
t
a
l
In
v
e
siti
g
a
ti
o
n
o
f
T
ra
in
in
g
A
l
g
o
rit
h
m
s u
se
d
in
Ba
c
k
p
ro
p
a
g
a
ti
o
n
A
rti
f
icia
l
Ne
u
ra
l
Ne
tw
o
rk
s t
o
A
p
p
ly
Cu
rv
e
F
it
ti
n
g
.
"
[1
3
]
S
o
n
i
J,
e
t
a
l.
"
P
re
d
ictiv
e
d
a
ta
m
in
in
g
f
o
r
m
e
d
ica
l
d
iag
n
o
sis:
A
n
o
v
e
rv
ie
w
o
f
h
e
a
rt
d
ise
a
se
p
re
d
ictio
n
,
"
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
Co
m
p
u
ter
A
p
p
l
ica
ti
o
n
s,
v
o
l.
1
7
,
p
p
.
4
3
-
8
,
2
0
1
1
.
[1
4
]
S
a
n
tu
ll
i
G
.
"
Ep
id
e
m
io
lo
g
y
o
f
c
a
r
d
io
v
a
sc
u
lar
d
ise
a
se
in
th
e
2
1
st
c
e
n
tu
ry
:
u
p
d
a
ted
n
u
m
b
e
rs
a
n
d
u
p
d
a
t
e
d
f
a
c
ts,"
J
Cv
D,
v
o
l.
1
,
p
p
.
1
-
2
,
2
0
1
3
.
[1
5
]
A
b
d
a
r
M
,
e
t
a
l.
"
Co
m
p
a
rin
g
P
e
rf
o
rm
a
n
c
e
o
f
Da
ta
M
in
i
n
g
A
lg
o
rit
h
m
s
in
P
re
d
ictio
n
He
a
rt
Dise
a
se
s
,
"
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
(
IJ
ECE
),
v
o
l.
5
,
p
p
.
1
5
6
9
-
7
6
,
2
0
1
5
.
[1
6
]
Da
s
R,
e
t
a
l.
"
Eff
e
c
ti
v
e
d
iag
n
o
sis
o
f
h
e
a
rt
d
ise
a
se
th
ro
u
g
h
n
e
u
ra
l
n
e
tw
o
rk
s
e
n
se
m
b
les
,
"
Exp
e
rt
sy
ste
ms
wit
h
a
p
p
li
c
a
ti
o
n
s,
v
o
l.
3
6
,
p
p
.
7
6
7
5
-
8
0
,
2
0
0
9
.
[1
7
]
A
h
a
D
W
.
UCI M
a
c
h
in
e
L
e
a
rn
in
g
Re
p
o
sit
o
ry
.
2
0
1
6
.
[1
8
]
L
ik
a
s
A
,
e
t
a
l.
"
T
r
a
in
in
g
th
e
ra
n
d
o
m
n
e
u
ra
l
n
e
tw
o
rk
u
sin
g
q
u
a
si
-
Ne
w
to
n
m
e
th
o
d
s,"
Eu
ro
p
e
a
n
J
o
u
rn
a
l
o
f
Op
e
ra
ti
o
n
a
l
Res
e
a
rc
h
,
v
o
l.
1
2
6
,
p
p
.
3
3
1
-
9
,
2
0
0
0
.
[1
9
]
Be
a
le M
H,
e
t
a
l.
"
Ne
u
ra
l
n
e
tw
o
rk
to
o
lb
o
x
7
,
"
Us
e
r’s
Gu
id
e
,
M
a
t
h
W
o
rk
s,
v
o
l.
p
p
.
2
0
1
0
.
[2
0
]
W
il
so
n
DR,
e
t
a
l.
"
T
h
e
g
e
n
e
ra
l
i
n
e
ff
icie
n
c
y
o
f
b
a
tch
train
in
g
f
o
r
g
ra
d
ien
t
d
e
sc
e
n
t
lea
rn
in
g
,
"
Ne
u
ra
l
Ne
two
rk
s,
v
o
l.
1
6
,
p
p
.
1
4
2
9
-
5
1
,
2
0
0
3
.
[2
1
]
Qia
n
N.
"
On
th
e
m
o
m
e
n
tu
m
ter
m
in
g
ra
d
ien
t
d
e
sc
e
n
t
lea
rn
in
g
a
l
g
o
rit
h
m
s,"
Ne
u
ra
l
n
e
two
rk
s,
v
o
l.
1
2
,
p
p
.
1
4
5
-
5
1
,
1
9
9
9
.
[2
2
]
Bish
o
p
CM
.
Ne
u
ra
l
n
e
tw
o
rk
s
f
o
r
p
a
tt
e
rn
re
c
o
g
n
it
i
o
n
:
Ox
f
o
rd
u
n
iv
e
rsity
p
re
ss
;
1
9
9
5
.
[2
3
]
Ried
m
il
ler
M
,
e
t
a
l.
,
"
A
d
ire
c
t
a
d
a
p
ti
v
e
me
th
o
d
f
o
r
f
a
ste
r
b
a
c
k
p
r
o
p
a
g
a
ti
o
n
le
a
rn
i
n
g
:
T
h
e
RP
ROP
a
l
g
o
rit
h
m
,
"
in
1
9
9
3
.
Ne
u
ra
l
Ne
tw
o
rk
s,
1
9
9
3
,
IE
EE
In
ter
n
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
,
1
9
9
3
,
p
p
.
5
8
6
-
91
.
[2
4
]
Birg
in
EG
,
e
t
a
l.
"
A
sp
e
c
tr
a
l
c
o
n
ju
g
a
te
g
ra
d
ien
t
m
e
th
o
d
f
o
r
u
n
c
o
n
s
train
e
d
o
p
ti
m
iza
ti
o
n
,
"
Ap
p
li
e
d
M
a
th
e
ma
t
ics
a
n
d
o
p
ti
miz
a
ti
o
n
,
v
o
l.
4
3
,
p
p
.
1
1
7
-
2
8
,
2
0
0
1
.
[2
5
]
Da
i
Y
-
H,
e
t
a
l.
"
Ne
w
c
o
n
ju
g
a
c
y
c
o
n
d
it
io
n
s
a
n
d
re
late
d
n
o
n
li
n
e
a
r
c
o
n
ju
g
a
te
g
ra
d
ien
t
m
e
th
o
d
s,"
A
p
p
l
ied
M
a
t
h
e
ma
ti
c
s
a
n
d
Op
ti
miz
a
ti
o
n
,
v
o
l.
4
3
,
p
p
.
8
7
-
1
0
1
,
2
0
0
1
.
[2
6
]
F
letc
h
e
r
R,
e
t
a
l.
"
F
u
n
c
ti
o
n
m
in
i
m
iz
a
ti
o
n
b
y
c
o
n
ju
g
a
te g
ra
d
ien
ts,"
T
h
e
c
o
m
p
u
ter
j
o
u
rn
a
l,
v
o
l.
7
,
p
p
.
1
4
9
-
5
4
,
1
9
6
4
.
[2
7
]
M
a
so
o
d
S
,
e
t
a
l.
,
e
d
i
to
rs.
An
a
lys
i
s
o
f
we
i
g
h
t
i
n
it
i
a
li
za
t
io
n
ro
u
ti
n
e
s
fo
r
c
o
n
j
u
g
a
te
g
ra
d
ien
t
t
ra
i
n
in
g
a
lg
o
rith
m
wit
h
Fl
e
tch
e
r
-
Ree
v
e
s
u
p
d
a
tes
.
C
o
m
p
u
t
in
g
,
Co
m
m
u
n
ica
ti
o
n
a
n
d
A
u
to
m
a
t
io
n
(ICCCA
),
2
0
1
6
I
n
tern
a
ti
o
n
a
l
Co
n
f
e
re
n
c
e
o
n
;
2
0
1
6
:
I
EE
E
.
[2
8
]
Al
-
Ba
y
a
ti
A
,
e
t
a
l.
"
Co
n
j
u
g
a
te
Gra
d
ien
t
Ba
c
k
-
p
ro
p
a
g
a
ti
o
n
w
it
h
M
o
d
if
ied
P
o
lac
k
–
Re
b
ier
u
p
d
a
tes
f
o
r
train
i
n
g
f
e
e
d
fo
r
w
a
rd
n
e
u
ra
l
n
e
tw
o
rk
,
"
Ira
q
i
J
S
ta
ti
s
S
c
i,
v
o
l.
2
0
,
p
p
.
1
6
4
-
7
3
,
2
0
1
1
.
[2
9
]
S
a
in
i
L
,
e
t
a
l.
"
Arti
fi
c
ia
l
n
e
u
ra
l
n
e
two
rk
b
a
se
d
p
e
a
k
lo
a
d
fo
re
c
a
stin
g
u
si
n
g
L
e
v
e
n
b
e
rg
–
M
a
rq
u
a
r
d
t
a
n
d
q
u
a
si
-
Ne
wt
o
n
me
th
o
d
s
,
"
IEE
P
ro
c
e
e
d
i
n
g
s
-
Ge
n
e
ra
ti
o
n
,
T
ra
n
s
m
issio
n
a
n
d
Distrib
u
ti
o
n
,
v
o
l
.
1
4
9
,
p
p
.
5
7
8
-
8
4
,
2
0
0
2
.
[3
0
]
Ha
g
a
n
M
T
,
e
t
a
l.
"
T
ra
in
in
g
f
e
e
d
f
o
r
w
a
rd
n
e
tw
o
rk
s
w
it
h
th
e
M
a
r
q
u
a
rd
t
a
lg
o
rit
h
m
,
"
IEE
E
tra
n
sa
c
ti
o
n
s
o
n
Ne
u
ra
l
Ne
two
rk
s,
v
o
l.
5
,
p
p
.
9
8
9
-
9
3
,
1
9
9
4
.
[3
1
]
Ilo
n
e
n
J,
e
t
a
l.
"
Dif
fe
re
n
ti
a
l
e
v
o
lu
ti
o
n
train
i
n
g
a
lg
o
rit
h
m
f
o
r
f
e
e
d
-
f
o
rw
a
rd
n
e
u
ra
l
n
e
tw
o
rk
s,"
Ne
u
ra
l
Pro
c
e
ss
in
g
L
e
tt
e
rs
,
v
o
l.
1
7
,
p
p
.
9
3
-
1
0
5
,
2
0
0
3
.
[3
2
]
Ha
g
e
r
WW
,
e
t
a
l.
"
A
su
rv
e
y
o
f
n
o
n
li
n
e
a
r
c
o
n
j
u
g
a
te
g
ra
d
ien
t
m
e
th
o
d
s,"
Pa
c
if
ic
jo
u
rn
a
l
o
f
O
p
ti
miza
ti
o
n
,
v
o
l.
2
,
p
p
.
35
-
5
8
,
2
0
0
6
.
[3
3
]
F
letc
h
e
r
R.
Co
n
j
u
g
a
te g
ra
d
ien
t
m
e
th
o
d
s f
o
r
i
n
d
e
f
in
it
e
sy
ste
m
s.
Nu
m
e
rica
l
a
n
a
l
y
sis:
S
p
rin
g
e
r
;
1
9
7
6
.
p
.
7
3
-
89
.
Evaluation Warning : The document was created with Spire.PDF for Python.