I
n
d
on
e
s
i
a
n
Jo
u
r
n
al
of
E
l
e
c
t
r
i
c
al
E
n
g
i
n
e
e
r
i
n
g a
n
d
C
om
p
u
t
e
r
S
c
i
e
n
c
e
V
ol
.
40
, N
o.
2
,
N
ov
e
m
b
e
r
2025
, pp.
1
121
~
1
128
I
S
S
N
:
2502
-
4752, D
O
I
:
10.11591/
i
j
e
e
c
s
.
v
40
.i
2
.
pp1
121
-
1
128
1121
Jou
r
n
al
h
om
e
page
:
ht
t
p:
/
/
i
j
e
e
c
s
.i
ae
s
c
or
e
.c
om
O
p
t
i
m
i
z
i
n
g c
l
u
st
e
r
i
n
g
e
f
f
i
c
i
e
n
c
y
w
i
t
h
w
e
i
g
h
t
e
d
k
-
m
e
a
n
s:
a m
ac
h
i
n
e
l
e
ar
n
i
n
g
-
d
r
i
ve
n
a
p
p
r
oac
h
f
or
e
n
h
a
n
c
e
d
ac
c
u
r
ac
y
a
n
d
sc
a
l
a
b
i
l
i
t
y
V
i
s
h
al
K
au
s
h
i
k
, A
b
d
u
l
A
l
e
e
m
S
c
h
o
ol
of
C
o
m
put
e
r
S
c
i
e
n
c
e
a
n
d
E
ng
i
n
e
e
r
i
ng,
G
a
l
g
ot
i
a
s
U
ni
ve
r
s
i
t
y
,
G
r
e
a
t
e
r
N
oi
da
,
I
ndi
a
A
r
t
i
c
l
e
I
n
f
o
A
B
S
T
R
A
C
T
A
r
t
i
c
l
e
h
i
s
t
o
r
y
:
R
e
c
e
i
v
e
d
D
e
c
3
,
2024
R
e
v
i
s
e
d
J
ul
22
,
2025
A
c
c
e
pt
e
d
O
c
t
15
,
2025
Da
ta
a
n
a
l
y
s
i
s
u
n
l
o
c
k
s
th
e
h
i
d
d
e
n
,
l
a
t
e
n
t
p
a
tt
e
r
n
s
a
n
d
s
tr
u
c
tu
r
e
s
wi
th
i
n
d
a
tas
e
ts
.
C
l
u
s
ter
i
n
g
a
lg
o
r
i
th
m
s
,
th
e
c
o
r
n
e
r
s
t
o
n
e
o
f
a
n
y
d
a
ta
a
n
a
l
y
s
i
s
,
a
r
e
u
s
u
a
l
l
y
c
h
a
ll
e
n
g
e
d
b
y
h
i
g
h
-
d
i
m
e
n
s
i
o
n
a
l
i
ty
,
c
o
m
p
l
e
x
i
ty
,
o
r
l
a
r
g
e
-
s
c
a
l
e
d
a
ta.
T
h
i
s
r
e
s
e
a
r
c
h
p
r
o
p
o
s
e
s
a
h
y
b
r
i
d
m
o
d
e
l
th
a
t
m
e
r
g
e
s
n
e
u
r
a
l
n
e
two
r
k
s
a
n
d
c
l
u
s
ter
i
n
g
tec
h
n
i
q
u
e
s
to
h
a
n
d
l
e
th
e
s
e
p
r
o
b
l
e
m
s
.
Ne
u
r
a
l
n
e
two
r
k
s
a
r
e
u
s
e
d
f
o
r
f
e
a
tu
r
e
e
x
tr
a
c
ti
o
n
a
n
d
d
i
m
e
n
s
i
o
n
a
l
i
ty
r
e
d
u
c
ti
o
n
;
r
a
w
d
a
ta
wil
l
b
e
tr
a
n
s
f
o
r
m
e
d
i
n
to
a
r
o
b
u
s
t,
l
o
w
-
d
i
m
e
n
s
i
o
n
a
l
r
e
p
r
e
s
e
n
tati
o
n
.
W
i
th
th
e
s
e
r
e
f
i
n
e
d
f
e
a
tu
r
e
s
,
th
e
p
e
r
f
o
r
m
a
n
c
e
o
f
c
l
u
s
ter
i
n
g
a
l
g
o
r
i
th
m
s
i
m
p
r
o
v
e
s
i
n
ter
m
s
o
f
s
c
a
l
a
b
il
i
ty
,
e
f
f
i
c
i
e
n
c
y
,
a
n
d
a
c
c
u
r
a
c
y
.
T
h
e
p
r
o
p
o
s
e
d
m
o
d
e
l
i
s
tes
ted
o
n
d
i
v
e
r
s
i
f
i
e
d
d
a
tas
e
ts
s
u
c
h
a
s
th
e
w
i
s
c
o
n
s
i
n
b
r
e
a
s
t
c
a
n
c
e
r
d
a
tas
e
t
(
W
B
C
D)
,
GE
O
Da
tas
e
t,
a
n
d
i
m
a
g
e
a
n
d
tex
t
d
a
ta
b
e
n
c
h
m
a
r
k
s
f
o
r
wh
i
c
h
s
u
b
s
tan
ti
a
l
i
m
p
r
o
v
e
m
e
n
ts
i
n
c
l
u
s
ter
i
n
g
m
e
tr
i
c
s
s
u
c
h
a
s
s
il
h
o
u
e
tt
e
s
c
o
r
e
,
p
u
r
i
ty
,
a
n
d
c
o
m
p
u
tati
o
n
a
l
e
f
f
i
c
i
e
n
c
y
a
r
e
r
e
p
o
r
ted
.
T
h
e
r
e
s
u
l
ts
d
e
m
o
n
s
tr
a
te
t
h
e
e
f
f
i
c
a
c
y
o
f
th
e
h
y
b
r
i
d
a
p
p
r
o
a
c
h
i
n
o
p
ti
m
i
z
i
n
g
c
l
u
s
ter
i
n
g
a
p
p
li
c
a
t
i
o
n
s
a
c
r
o
s
s
d
o
m
a
i
n
s
,
s
u
c
h
a
s
b
i
o
i
n
f
o
r
m
a
ti
c
s
,
h
e
a
l
th
c
a
r
e
,
a
n
d
i
m
a
g
e
a
n
a
l
y
s
i
s
.
K
e
y
w
or
d
s
:
C
l
us
t
e
r
i
ng
a
l
g
or
i
t
h
m
s
C
l
us
t
e
r
i
ng
opt
i
m
i
z
a
t
i
on
D
e
e
p l
e
a
r
ni
n
g
H
y
br
i
d
m
ode
l
s
M
a
c
hi
ne
l
e
a
r
ni
ng
T
h
i
s
i
s
a
n
o
p
e
n
a
c
c
e
s
s
a
r
ti
c
l
e
u
n
d
e
r
th
e
C
C
B
Y
-
SA
l
i
c
e
n
s
e
.
C
or
r
e
s
pon
di
n
g A
u
t
h
or
:
A
bdul
A
l
e
e
m
S
c
hool
of
C
om
put
e
r
S
c
i
e
nc
e
a
nd E
ng
i
ne
e
r
i
ng
,
G
a
l
g
ot
i
a
s
U
ni
v
e
r
s
i
t
y
G
r
e
a
t
e
r
N
oi
da
,
I
ndi
a
E
m
a
i
l
:
e
r
.a
l
e
e
m
@
gm
a
i
l
.c
o
m
1.
I
N
T
R
O
D
U
C
T
I
O
N
C
l
us
t
e
r
i
ng
i
s
a
s
i
g
ni
f
i
c
a
nt
da
t
a
a
n
a
l
y
s
i
s
a
c
t
i
v
i
t
y
.
I
t
i
s
f
r
e
que
nt
l
y
ut
i
l
i
s
e
d
t
o
di
s
c
ov
e
r
pa
t
t
e
r
ns
a
nd
s
t
r
uc
t
ur
e
s
i
n
unm
a
r
ke
d
s
e
t
s
of
i
nf
or
m
a
t
i
on
i
n
bi
oi
nf
or
m
a
t
i
c
s
,
i
m
a
g
e
p
r
oc
e
s
s
i
ng
,
na
t
ur
a
l
l
a
ng
ua
g
e
p
r
oc
e
s
s
i
ng,
a
nd
m
a
r
ke
t
s
e
gm
e
nt
a
t
i
on.
C
onv
e
nt
i
ona
l
c
l
us
t
e
r
i
ng
a
l
g
or
i
t
h
m
s
s
uc
h
a
s
K
-
M
e
a
ns
,
D
B
S
C
A
N
a
nd
hi
e
r
a
r
c
hi
c
a
l
c
l
us
t
e
r
i
ng
us
e
di
s
t
a
nc
e
s
i
m
i
l
a
r
i
t
y
m
e
a
s
ur
e
s
. S
t
i
l
l
, t
he
y
do not
pe
r
f
or
m
w
e
l
l
w
i
t
h hi
g
h
-
di
m
e
n
s
i
ona
l
, non
-
l
i
ne
a
r
,
a
nd
l
a
r
ge
-
s
c
a
l
e
d
a
t
a
be
c
a
us
e
of
t
he
i
r
s
hor
t
c
o
m
i
n
g
s
,
s
uc
h
a
s
s
e
ns
i
t
i
v
i
t
y
t
o
i
ni
t
i
a
l
i
s
a
t
i
on,
f
a
i
l
ur
e
t
o
c
l
us
t
e
r
non
-
c
on
v
e
x
s
t
r
uc
t
ur
e
s
a
nd
i
na
bi
l
i
t
y
t
o
s
c
a
l
e
up
[
1]
,
[
2]
.
T
o
a
v
oi
d
s
uc
h
pr
obl
e
m
s
,
s
c
i
e
nt
i
s
t
s
ha
v
e
c
o
m
bi
ne
d
m
a
c
hi
ne
l
e
a
r
ni
ng
m
e
t
hods
,
s
pe
c
i
f
i
c
a
l
l
y
ne
ur
a
l
ne
t
w
or
ks
,
w
i
t
h
c
l
us
t
e
r
i
ng
,
di
m
e
ns
i
ona
l
i
t
y
r
e
duc
t
i
on
a
n
d
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on.
A
ut
o
e
nc
ode
r
s
a
nd
C
N
N
s
ha
v
e
e
f
f
e
c
t
i
v
e
l
y
r
e
duc
e
d
di
m
e
ns
i
ons
a
nd
e
xt
r
a
c
t
e
d
f
e
a
t
ur
e
s
i
n
hi
g
h
-
di
m
e
ns
i
ona
l
da
t
a
[
3]
,
m
a
ki
n
g
c
l
us
t
e
r
i
ng
of
t
he
da
t
a
m
or
e
f
e
a
s
i
bl
e
t
o
r
un
a
l
g
or
i
t
h
m
s
on
[
4]
M
or
e
ov
e
r
,
m
e
t
a
he
ur
i
s
t
i
c
a
l
g
or
i
t
h
m
s
a
nd
r
e
i
nf
or
c
e
m
e
nt
l
e
a
r
ni
ng
ha
v
e
b
e
e
n
ut
i
l
i
s
e
d
t
o
r
e
f
i
ne
c
l
us
t
e
r
i
n
g
pa
r
a
m
e
t
e
r
s
on
a
dy
na
m
i
c
ba
s
i
s
ba
s
e
d
on
t
he
c
o
m
pl
e
xi
t
y
of
t
he
da
t
a
s
e
t
[
5]
,
[
6
]
.
T
he
pr
e
s
e
nt
pa
pe
r
s
u
gg
e
s
t
s
us
i
ng
n
e
ur
a
l
ne
t
w
or
ks
t
o
l
e
a
r
n
l
a
t
e
nt
f
e
a
t
ur
e
s
a
nd
s
ubs
e
que
nt
l
y
us
i
n
g
c
l
us
t
e
r
i
ng
a
l
g
o
r
i
t
h
m
s
t
o
e
nha
nc
e
t
he
qua
l
i
t
y
o
f
c
l
us
t
e
r
i
ng
out
c
o
m
e
s
.
T
he
m
ode
l
o
v
e
r
c
o
m
e
s
t
he
c
ur
s
e
of
di
m
e
ns
i
ona
l
i
t
y
by
di
m
i
ni
s
hi
ng
di
m
e
ns
i
ona
l
i
t
y
a
nd
i
nc
r
e
a
s
i
ng
f
e
a
t
ur
e
r
e
p
r
e
s
e
nt
a
t
i
on,
w
hi
c
h
i
m
pr
o
v
e
s
t
he
c
l
us
t
e
r
i
ng
a
c
c
u
r
a
c
y
i
n
a
ddi
t
i
on
t
o
c
ont
r
i
but
i
ng
t
o
c
o
m
put
a
t
i
ona
l
e
f
f
i
c
i
e
nc
y
[
7]
-
[
9]
.
I
t
i
s
t
e
s
t
e
d
on
v
a
r
i
ous
da
t
a
ba
s
e
s
l
i
ke
,
W
B
C
D
(
l
ow
-
di
m
e
n
s
i
ona
l
)
,
G
E
O
g
e
n
e
e
xpr
e
s
s
i
on
da
t
a
s
e
t
(
hi
g
h
-
di
m
e
ns
i
ona
l
)
,
M
N
I
S
T
(
i
m
a
g
e
)
,
a
nd
20
N
e
w
s
g
r
oups
(
t
e
xt
)
a
n
d
i
t
s
m
e
t
r
i
c
s
a
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
, V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
112
1
-
1
128
1122
s
i
l
houe
t
t
e
s
c
or
e
,
a
dj
us
t
e
d
r
a
nd
i
nde
x
(
A
R
I
)
,
a
nd
c
l
us
t
e
r
i
n
g
pur
i
t
y
,
a
nd
r
unt
i
m
e
[
10]
-
[
1
2]
.
T
hi
s
r
e
s
e
a
r
c
h
g
e
n
e
r
a
l
l
y
br
i
n
g
s
a
s
t
r
ong
h
y
br
i
d
t
e
c
hni
que
,
a
hi
g
h
l
e
v
e
l
of
be
nc
h
m
a
r
ki
n
g
on
do
m
a
i
ns
,
a
nd
pr
o
v
a
bl
y
b
e
t
t
e
r
r
e
s
ul
t
s
t
ha
n c
l
a
s
s
i
c
a
nd e
xi
s
t
i
ng
hy
br
i
d c
l
us
t
e
r
i
ng
t
e
c
hni
que
s
.
a)
N
e
ur
a
l
ne
t
w
o
r
k
-
ba
s
e
d c
l
us
t
e
r
i
ng
N
e
ur
a
l
ne
t
w
or
ks
a
r
e
popul
a
r
i
n
c
l
us
t
e
r
i
n
g
a
ppl
i
c
a
t
i
ons
pr
i
m
a
r
i
l
y
due
t
o
t
he
i
r
a
bi
l
i
t
y
t
o
l
e
a
r
n
non
-
l
i
ne
a
r
a
nd
hi
g
h
-
di
m
e
ns
i
ona
l
r
e
pr
e
s
e
nt
a
t
i
ons
.
A
ut
oe
nc
ode
r
s
,
f
or
i
ns
t
a
nc
e
,
l
ow
e
r
t
he
da
t
a
di
m
e
ns
i
on
but
r
e
t
a
i
n
m
e
a
ni
n
g
f
ul
f
e
a
t
ur
e
s
[
13
]
.
H
i
nt
on
a
nd
S
a
l
a
khut
di
nov
ha
v
e
s
ho
w
n
t
ha
t
de
e
p
a
ut
oe
nc
ode
r
s
m
a
y
s
i
g
ni
f
i
c
a
nt
l
y
e
nha
nc
e
c
l
us
t
e
r
i
ng
pe
r
f
or
m
a
nc
e
by
pr
oj
e
c
t
i
n
g
hi
g
h
-
di
m
e
ns
i
o
na
l
da
t
a
i
nt
o
a
l
ow
e
r
s
pa
c
e
.
A
na
l
og
ous
l
y
,
i
m
a
ge
c
l
us
t
e
r
i
ng
t
a
s
ks
c
a
n
be
c
a
r
r
i
e
d
out
us
i
ng
C
N
N
s
t
o
l
e
a
r
n
s
pa
t
i
a
l
f
e
a
t
ur
e
s
t
ha
t
t
r
a
di
t
i
ona
l
c
l
u
s
t
e
r
i
ng
a
l
g
or
i
t
h
m
s
c
a
nnot
[
14]
, [
15
].
b)
R
e
i
nf
or
c
e
m
e
nt
l
e
a
r
n
i
ng
i
n c
l
us
t
e
r
i
ng
R
e
i
nf
or
c
e
m
e
nt
l
e
a
r
ni
ng
,
w
hi
c
h
ha
s
be
e
n
f
ound
t
o
pl
a
y
a
s
i
g
ni
f
i
c
a
nt
r
ol
e
i
n
opt
i
m
i
z
i
ng
c
l
us
t
e
r
i
ng
a
l
g
or
i
t
h
m
s
,
l
e
a
r
ns
how
t
o
d
y
na
m
i
c
a
l
l
y
a
dj
us
t
t
he
pa
r
a
m
e
t
e
r
s
of
c
l
us
t
e
r
s
ba
s
e
d
on
f
e
e
dba
c
k
f
r
o
m
t
h
e
e
n
v
i
r
on
m
e
nt
. T
hr
ou
g
h t
hi
s
a
ppr
oa
c
h,
m
or
e
e
f
f
e
c
t
i
v
e
c
l
us
t
e
r
i
ng
s
ol
ut
i
ons
a
r
e
di
s
c
o
v
e
r
e
d
[
16
]
, [
17
].
c)
C
l
us
t
e
r
i
ng
us
i
ng
m
e
t
a
he
ur
i
s
t
i
c
a
l
g
or
i
t
h
m
s
M
e
t
a
he
ur
i
s
t
i
c
a
l
g
or
i
t
h
m
s
,
s
uc
h
a
s
pa
r
t
i
c
l
e
s
w
a
r
m
opt
i
m
i
z
a
t
i
on
(
P
S
O
)
,
g
e
ne
t
i
c
a
l
g
or
i
t
h
m
s
a
nd
a
nt
c
ol
ony
opt
i
m
i
z
a
t
i
on,
a
r
e
pr
o
m
i
s
i
ng
f
or
s
ol
v
i
n
g
c
l
us
t
e
r
i
ng
i
s
s
ue
s
.
I
n
ge
ne
r
a
l
,
m
e
t
a
he
ur
i
s
t
i
c
a
l
g
or
i
t
hm
s
a
r
e
be
t
t
e
r
t
ha
n
e
xha
us
t
i
v
e
s
e
a
r
c
h
e
s
f
or
opt
i
m
a
l
or
ne
a
r
-
opt
i
m
a
l
s
ol
ut
i
ons
i
n
l
a
r
g
e
he
a
l
t
h
s
pa
s
.
F
or
e
xa
m
pl
e
,
a
PSO
-
ba
s
e
d
a
ppr
oa
c
h
f
or
c
l
us
t
e
r
i
ng
,
kno
w
n
a
s
S
w
a
r
m
C
l
us
t
,
a
l
l
ow
s
f
or
h
a
ndl
i
ng
hi
g
h
-
di
m
e
ns
i
ona
l
da
t
a
a
nd
o
bt
a
i
ns
r
e
l
a
t
i
v
e
l
y
r
obus
t
c
l
us
t
e
r
i
ng
c
o
m
p
a
r
e
d
t
o
t
r
a
di
t
i
ona
l
a
ppr
oa
c
he
s
[
18
]
,
[
19
]
.
E
v
ol
ut
i
ona
r
y
m
e
t
hods
a
r
e
c
o
m
bi
ne
d
w
i
t
h ne
ur
a
l
ne
t
w
or
ks
t
o f
or
m
h
y
br
i
d
m
od
e
l
s
t
ha
t
a
da
pt
i
v
e
l
y
opt
i
m
i
z
e
c
l
us
t
e
r
i
ng
p
e
r
f
or
m
a
nc
e
[
20
].
d)
H
y
br
i
d
c
l
us
t
e
r
i
ng
m
ode
l
s
T
he
i
nt
e
g
r
a
t
i
on
of
de
e
p
l
e
a
r
ni
n
g
a
nd
c
l
us
t
e
r
i
ng
(
e
.
g
.
ge
ne
t
i
c
a
l
g
or
i
t
h
m
s
+
C
N
N
s
)
he
l
ps
t
o
i
m
pr
o
ve
t
he
pe
r
f
or
m
a
n
c
e
a
c
r
os
s
a
r
a
nge
of
da
t
a
s
e
t
s
[
21
]
,
[
22
]
.
T
he
de
e
p
be
l
i
e
f
ne
t
w
or
k
pe
r
f
or
m
e
d
b
e
t
t
e
r
o
n
e
xt
e
ns
i
v
e
di
m
e
ns
i
ona
l
da
t
a
[
23
]
,
[
24
]
.
F
e
a
t
ur
e
l
e
a
r
ni
ng
a
nd
c
l
us
t
e
r
i
ng
a
r
e
r
e
c
e
nt
a
d
v
a
nc
e
s
a
nd
a
r
e
t
hus
uni
f
i
e
d
by
ne
ur
a
l
f
r
a
m
e
w
or
ks
, but
s
t
i
l
l
f
a
c
e
i
s
s
ue
s
of
s
c
a
l
a
bi
l
i
t
y
a
nd c
om
pl
e
xi
t
y
[
25
]
-
[
27
].
e)
L
i
m
i
t
a
t
i
ons
a
nd
r
e
s
e
a
r
c
h ga
ps
T
he
h
y
br
i
d
a
ppr
oa
c
h
ha
s
s
e
v
e
r
a
l
l
i
m
i
t
a
t
i
ons
:
m
os
t
r
e
qui
r
e
he
a
v
y
c
o
m
put
a
t
i
ona
l
r
e
s
our
c
e
s
,
m
a
ki
ng
t
he
m
i
m
p
r
a
c
t
i
c
a
l
f
or
r
e
a
l
-
t
i
m
e
a
ppl
i
c
a
t
i
ons
;
t
he
a
ppr
oa
c
he
s
he
a
v
i
l
y
r
e
l
y
on
pa
r
a
m
e
t
e
r
t
uni
ng
a
nd
a
r
e
not
i
nt
e
r
pr
e
t
a
bl
e
i
n pr
a
c
t
i
c
e
.
T
hi
s
w
or
k a
i
m
s
t
o f
i
l
l
s
uc
h g
a
ps
b
y
de
v
e
l
opi
ng
a
s
c
a
l
a
bl
e
a
nd e
f
f
i
c
i
e
nt
hy
br
i
d ne
ur
a
l
ne
t
w
or
k
-
ba
s
e
d
m
ode
l
w
h
e
r
e
n
e
ur
a
l
ne
t
w
or
ks
a
r
e
i
nt
e
g
r
a
t
e
d
w
i
t
h
c
l
us
t
e
r
i
ng
a
l
g
or
i
t
hm
s
t
o
i
m
pr
o
ve
pe
r
f
or
m
a
nc
e
f
or
a
w
i
de
v
a
r
i
e
t
y
of
da
t
a
s
e
t
s
a
s
s
ho
w
n i
n T
a
bl
e
1
.
T
a
bl
e
1. C
o
m
bi
ne
d
t
a
bl
e
i
nt
e
g
r
a
t
i
ng
bot
h
r
e
s
e
a
r
c
h ga
ps
i
n
m
ode
l
s
/
pa
pe
r
s
a
nd
l
i
m
i
t
a
t
i
ons
a
nd pr
opos
e
d
s
ol
ut
i
on
P
a
pe
r
/
m
ode
l
F
oc
us
/
c
ont
r
i
but
i
o
n
R
e
s
e
a
r
c
h
g
a
p/
l
i
m
i
t
a
t
i
o
n
P
r
opos
e
d
s
ol
ut
i
o
n
T
r
a
di
t
i
o
n
a
l
C
l
us
t
e
r
i
ng
(K
-
M
e
a
n
s
,
D
B
S
C
A
N
,
H
i
e
r
a
r
c
h
i
c
a
l
)
E
f
f
i
c
i
e
n
t
gr
oupi
ng
o
f
da
t
a
ba
s
e
d
on
di
s
t
a
n
c
e
m
e
a
s
ur
e
s
S
t
r
u
gg
l
e
s
w
i
t
h
h
i
gh
-
di
m
e
n
s
i
o
n
a
l
,
n
o
n
-
l
i
n
e
a
r
,
a
n
d
s
pa
r
s
e
da
t
a
;
s
e
n
s
i
t
i
ve
t
o
i
ni
t
i
a
l
i
z
a
t
i
o
n;
l
a
c
ks
s
c
a
l
a
bi
l
i
t
y
[
12
]
D
e
ve
l
op
a
s
c
a
l
a
bl
e
h
ybr
i
d
m
ode
l
l
e
ve
r
a
g
i
n
g
n
e
ur
a
l
n
e
t
w
or
ks
f
or
di
m
e
n
s
i
on
a
l
i
t
y
r
e
du
c
t
i
on
a
n
d
i
m
pr
ove
d
h
a
n
dl
i
ng
of
n
on
-
l
i
n
e
a
r
da
t
a
.
A
ut
oe
n
c
ode
r
-
B
a
s
e
d
C
l
us
t
e
r
i
ng
(
H
i
n
t
o
n
a
n
d
S
a
l
a
khut
di
n
ov
)
D
i
m
e
n
s
i
o
n
a
l
i
t
y
r
e
duc
t
i
o
n
us
i
ng
de
e
p
a
ut
oe
n
c
o
d
e
r
s
f
or
c
l
us
t
e
r
i
ng
t
a
s
ks
L
i
m
i
t
e
d
t
o
c
e
r
t
a
i
n
da
t
a
t
ype
s
(
e
.
g
.
,
i
m
a
g
e
s
)
;
c
o
m
p
ut
a
t
i
o
n
a
l
l
y
i
n
t
e
n
s
i
ve
;
s
e
n
s
i
t
i
ve
t
o
h
ype
r
pa
r
a
m
e
t
e
r
s
e
l
e
c
t
i
o
n
.
U
s
e
a
da
pt
i
ve
a
r
c
h
i
t
e
c
t
ur
e
s
a
n
d
i
n
t
e
gr
a
t
e
pa
r
a
m
e
t
e
r
t
u
n
i
ng
t
e
c
hn
i
que
s
t
o
i
m
p
r
ove
g
e
n
e
r
a
l
i
z
a
bi
l
i
t
y
a
n
d
e
f
f
i
c
i
e
n
c
y.
D
e
e
pC
l
us
t
C
om
bi
ni
ng
de
e
p
l
e
a
r
ni
ng
a
n
d
c
l
us
t
e
r
i
ng
a
l
g
o
r
i
t
hm
s
f
o
r
f
e
a
t
ur
e
e
x
t
r
a
c
t
i
o
n
D
oe
s
n
ot
a
ddr
e
s
s
r
e
a
l
-
t
i
m
e
pe
r
f
or
m
a
n
c
e
;
r
e
qu
i
r
e
s
pr
e
-
t
r
a
i
ni
ng
on
l
a
r
g
e
da
t
a
s
e
t
s
,
l
i
m
i
t
i
ng
a
da
pt
a
bi
l
i
t
y
[
14
].
O
pt
i
m
i
z
e
t
h
e
m
ode
l
f
or
r
e
a
l
-
t
i
m
e
c
l
us
t
e
r
i
ng
by
r
e
duc
i
ng
c
om
put
a
t
i
o
n
a
l
ove
r
h
e
a
d
a
n
d
l
e
ve
r
a
g
i
ng
l
i
ght
w
e
i
gh
t
n
e
ur
a
l
n
e
t
w
o
r
k
de
s
i
gn
s
.
R
e
i
n
f
o
r
c
e
m
e
n
t
L
e
a
r
n
i
ng
i
n
C
l
us
t
e
r
i
ng
O
pt
i
m
i
z
e
s
c
l
us
t
e
r
i
ng
pa
r
a
m
e
t
e
r
s
dyn
a
m
i
c
a
l
l
y
us
i
ng
r
e
i
n
f
or
c
e
m
e
nt
l
e
a
r
n
i
ng
H
i
gh
c
o
m
pu
t
a
t
i
o
n
a
l
ove
r
h
e
a
d
f
o
r
pa
r
a
m
e
t
e
r
opt
i
m
i
z
a
t
i
o
n;
n
ot
va
l
i
da
t
e
d
on
di
ve
r
s
e
da
t
a
s
e
t
s
l
i
ke
i
m
a
g
e
s
or
t
e
x
t
[
15
]
.
I
n
c
o
r
por
a
t
e
a
da
pt
i
ve
R
L
-
ba
s
e
d
f
r
a
m
e
w
or
ks
t
h
a
t
dy
n
a
m
i
c
a
l
l
y
t
u
n
e
c
l
us
t
e
r
i
ng
pa
r
a
m
e
t
e
r
s
a
c
r
os
s
di
f
f
e
r
e
n
t
da
t
a
s
e
t
s
.
S
w
a
r
m
C
l
us
t
P
a
r
t
i
c
l
e
s
w
a
r
m
op
t
i
m
i
z
a
t
i
o
n
f
or
c
l
us
t
e
r
i
ng
F
oc
us
e
s
on
opt
i
m
i
z
a
t
i
o
n
f
o
r
hi
gh
-
di
m
e
n
s
i
o
n
a
l
da
t
a
but
l
a
c
ks
i
n
t
e
gr
a
t
i
o
n
w
i
t
h
n
e
ur
a
l
n
e
t
w
o
r
k
f
e
a
t
ur
e
e
x
t
r
a
c
t
i
o
n
[
17
].
C
om
bi
n
e
P
S
O
w
i
t
h
n
e
u
r
a
l
n
e
t
w
or
ks
f
or
e
nh
a
n
c
e
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
a
nd
c
l
us
t
e
r
i
ng
a
c
c
ur
a
c
y.
G
e
n
e
r
a
l
H
yb
r
i
d
A
ppr
oa
c
h
e
s
N
e
ur
a
l
n
e
t
w
o
r
ks
a
n
d
c
l
us
t
e
r
i
ng
i
nt
e
g
r
a
t
i
o
n
f
o
r
f
e
a
t
ur
e
e
x
t
r
a
c
t
i
o
n
a
n
d
gr
oupi
ng
R
e
qui
r
e
s
e
x
t
e
n
s
i
ve
c
om
pu
t
a
t
i
o
n
a
l
r
e
s
our
c
e
s
;
r
e
l
i
e
s
h
e
a
vi
l
y
on
pa
r
a
m
e
t
e
r
t
u
n
i
ng;
l
a
c
ks
i
n
t
e
r
pr
e
t
a
bi
l
i
t
y
[
21
].
D
e
s
i
gn
a
n
e
f
f
i
c
i
e
n
t
f
r
a
m
e
w
or
k
w
i
t
h
a
ut
o
m
a
t
e
d
pa
r
a
m
e
t
e
r
t
u
n
i
ng
a
n
d
i
n
t
e
r
pr
e
t
a
bl
e
A
I
m
e
c
h
a
n
i
s
m
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
O
pt
i
m
i
z
i
n
g c
l
us
t
e
r
i
n
g e
f
f
i
c
i
e
nc
y
w
i
t
h
w
e
i
ght
e
d
k
-
m
e
ans
:
a m
ac
hi
ne
l
e
ar
ni
ng
-
dr
i
v
e
n
…
(
V
i
s
hal
K
aus
hi
k
)
1123
2.
P
R
O
P
O
S
E
D
M
E
T
H
O
D
T
he
h
y
br
i
d
m
ode
l
pr
opos
e
d
i
nt
e
g
r
a
t
e
s
ne
ur
a
l
ne
t
w
or
ks
f
or
t
he
f
e
a
t
ur
e
-
e
xt
r
a
c
t
i
ng
m
odu
l
e
a
nd
a
c
l
us
t
e
r
i
ng
a
l
g
or
i
t
h
m
f
or
gr
oupi
ng
da
t
a
i
nt
o
m
e
a
ni
n
g
f
ul
c
l
us
t
e
r
s
.
I
t
dr
a
w
s
upon
bot
h
s
t
r
e
ng
t
hs
s
i
nc
e
t
he
ne
ur
a
l
ne
t
w
or
k
r
e
duc
e
s
di
m
e
ns
i
ona
l
i
t
y
f
r
o
m
t
he
da
t
a
a
nd
l
e
a
r
ns
l
a
t
e
nt
f
e
a
t
ur
e
s
.
T
h
e
c
l
us
t
e
r
i
ng
a
l
g
or
i
t
h
m
pr
oc
e
e
ds
w
i
t
h t
he
s
e
r
e
f
i
ne
d f
e
a
t
ur
e
s
t
o a
c
hi
e
v
e
be
t
t
e
r
a
c
c
ur
a
c
y
a
nd e
f
f
i
c
i
e
nc
y
i
n c
l
us
t
e
r
i
ng
.
2.1
.
N
e
u
r
al
n
e
t
w
or
k
c
o
m
p
on
e
n
t
T
he
ne
ur
a
l
n
e
t
w
o
r
k
c
o
m
pr
i
s
e
s
pr
e
-
pr
oc
e
s
s
i
ng
a
nd
c
on
v
e
r
t
i
ng
r
a
w
i
nput
da
t
a
i
nt
o
a
c
o
m
p
a
c
t
a
nd
i
nf
or
m
a
t
i
v
e
f
e
a
t
ur
e
s
pa
c
e
.
W
e
us
e
a
n
a
ut
oe
nc
ode
r
a
r
c
hi
t
e
c
t
ur
e
be
c
a
us
e
i
t
c
a
n
e
f
f
i
c
i
e
nt
l
y
l
e
a
r
n
c
o
m
pl
e
x,
non
-
l
i
ne
a
r
r
e
pr
e
s
e
nt
a
t
i
ons
.
A
n
a
ut
oe
nc
ode
r
ha
s
t
w
o
m
a
i
n
c
o
m
pone
nt
s
:
t
he
e
nc
ode
r
a
nd
t
he
de
c
ode
r
.
T
h
e
e
nc
ode
r
c
o
m
p
r
e
s
s
e
s
t
he
i
nf
or
m
a
t
i
on
f
r
o
m
t
he
i
nput
da
t
a
i
nt
o
a
l
a
t
e
nt
r
e
pr
e
s
e
nt
a
t
i
on,
w
hi
l
e
t
he
de
c
ode
r
t
a
k
e
s
t
hi
s
c
o
m
pr
e
s
s
e
d
r
e
pr
e
s
e
nt
a
t
i
on
a
nd
r
e
c
ons
t
r
uc
t
s
t
he
or
i
g
i
na
l
da
t
a
.
B
y
m
i
ni
m
i
z
i
ng
t
he
r
e
c
ons
t
r
uc
t
i
on
l
os
s
,
t
he
e
nc
ode
r
l
e
a
r
ns
m
e
a
ni
n
g
f
ul
f
e
a
t
ur
e
s
t
ha
t
de
f
i
ne
t
he
unde
r
l
y
i
n
g
s
t
r
uc
t
ur
e
of
t
he
da
t
a
.
T
he
r
e
f
or
e
,
t
hi
s
l
a
t
e
nt
r
e
pr
e
s
e
nt
a
t
i
on
w
i
l
l
be
us
e
d a
s
t
he
out
put
of
t
he
ne
ur
a
l
ne
t
w
or
k t
o s
e
r
v
e
a
s
a
m
e
a
ns
t
o a
c
c
e
s
s
t
he
c
l
us
t
e
r
i
ng
c
o
m
pone
nt
[
27
].
A
r
c
hi
t
e
c
t
ur
e
T
he
a
ut
oe
nc
od
e
r
c
ons
i
s
t
s
of
t
w
o
m
a
i
n c
o
m
pone
nt
s
:
E
nc
ode
r
:
C
om
pr
e
s
s
e
s
t
he
hi
g
h
-
di
m
e
ns
i
ona
l
i
nput
∈
i
nt
o a
l
a
t
e
nt
r
e
pr
e
s
e
nt
a
t
i
on
∈
w
he
r
e
≪
.
T
he
t
r
a
ns
f
or
m
a
t
i
on i
s
de
f
i
ne
d a
s
:
=
(
;
,
)
=
(
+
)
w
he
r
e
W
e
a
nd
b
e
a
r
e
t
he
w
e
i
g
ht
s
a
nd bi
a
s
e
s
of
t
he
e
nc
ode
r
, a
nd
σ
\
s
i
gm
a
σ
i
s
a
n a
c
t
i
v
a
t
i
on f
unc
t
i
on (
e
.g
.,
R
e
L
U
)
.
D
e
c
ode
r
:
R
e
c
ons
t
r
uc
t
s
t
he
i
npu
t
f
r
om
t
he
l
a
t
e
nt
r
e
pr
e
s
e
nt
a
t
i
on
=
(
;
,
)
.
T
he
r
e
c
ons
t
r
uc
t
i
on pr
oc
e
s
s
a
i
m
s
t
o
m
i
ni
m
i
z
e
t
he
r
e
c
ons
t
r
uc
t
i
on l
os
s
:
=
|
−
|
2
̌
A
t
r
a
i
ne
d
a
ut
oe
nc
ode
r
i
s
w
ha
t
r
e
duc
e
s
r
e
c
ons
t
r
uc
t
i
on
l
os
s
i
n
or
de
r
t
o
t
r
a
i
n
t
he
e
nc
ode
r
s
o
t
ha
t
i
t
l
e
a
r
ns
t
o
r
e
pr
e
s
e
nt
da
t
a
c
o
m
pa
c
t
l
y
a
nd
i
n
a
m
e
a
ni
ng
f
ul
w
a
y
.
T
hi
s
l
a
t
e
nt
r
e
p
r
e
s
e
nt
a
t
i
on
Z
w
i
l
l
be
us
e
d
by
t
he
c
l
us
t
e
r
i
ng
c
o
m
pone
nt
a
s
i
nput
.
2.2
.
C
l
u
s
t
e
r
i
n
g
c
om
p
on
e
n
t
T
he
ne
ur
a
l
ne
t
w
or
k
out
put
s
t
he
l
ow
-
di
m
e
ns
i
ona
l
f
e
a
t
ur
e
s
pa
c
e
of
t
he
K
-
M
e
a
ns
a
l
g
or
i
t
hm
.
T
hi
s
c
r
e
a
t
e
s
a
n
i
m
pr
o
v
e
d
i
ni
t
i
a
l
i
s
a
t
i
on,
be
t
t
e
r
s
c
a
l
a
bi
l
i
t
y
a
nd
hi
g
he
r
a
c
c
ur
a
c
y
,
s
i
nc
e
t
he
c
o
m
p
a
c
t
f
e
a
t
ur
e
s
be
t
t
e
r
c
a
pt
ur
e
t
he
unde
r
l
y
i
n
g
s
t
r
uc
t
ur
e
of
t
he
da
t
a
, t
he
r
e
by
pr
o
m
ot
i
ng
c
l
us
t
e
r
i
ng
.
K
-
M
e
a
ns
C
l
us
t
e
r
i
ng
K
-
M
e
a
ns
i
s
a
pa
r
t
i
t
i
oni
ng
-
ba
s
e
d a
l
g
or
i
t
h
m
t
ha
t
m
i
ni
m
i
z
e
s
t
he
w
i
t
hi
n
-
c
l
us
t
e
r
v
a
r
i
a
nc
e
.
G
i
v
e
n
k
c
l
us
t
e
r
s
, t
he
a
l
g
or
i
t
h
m
a
i
m
s
t
o:
:
∑
∑
|
−
|
2
∈
=
1
w
he
r
e
μ
i
i
s
t
he
c
e
nt
r
oi
d of
c
l
us
t
e
r
C
i
a
nd Z
j
i
s
a
da
t
a
poi
nt
i
n t
he
l
a
t
e
nt
f
e
a
t
ur
e
s
pa
c
e
.
T
he
a
l
g
or
i
t
h
m
f
ol
l
o
w
s
t
he
s
e
s
t
e
ps
:
a)
I
ni
t
i
a
l
i
z
e
k c
l
us
t
e
r
c
e
nt
r
oi
ds
r
a
ndo
m
l
y
.
b)
A
s
s
i
g
n e
a
c
h da
t
a
poi
nt
Z
j
t
o t
he
ne
a
r
e
s
t
c
e
nt
r
oi
d:
:
=
(
:
|
−
|
≤
|
−
|
,
∀
}
.
c)
U
pda
t
e
t
he
c
e
nt
r
oi
ds
:
=
1
|
|
∑
∈
d)
R
e
pe
a
t
s
t
e
ps
2 a
nd 3 unt
i
l
c
onv
e
r
g
e
nc
e
(
no c
ha
n
g
e
i
n c
l
us
t
e
r
a
s
s
i
g
n
m
e
nt
s
or
c
e
nt
r
oi
ds
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
, V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
112
1
-
1
128
1124
3.
M
E
T
H
O
D
T
hi
s
s
e
c
t
i
on
de
s
c
r
i
be
s
t
he
e
xpe
r
i
m
e
nt
a
l
de
s
i
g
n
a
i
m
e
d
a
t
v
a
l
i
da
t
i
ng
t
he
pr
opos
e
d
hy
br
i
d
m
ode
l
.
T
h
e
e
xpe
r
i
m
e
nt
w
a
s
c
onf
i
g
ur
e
d
by
a
c
hoi
c
e
of
da
t
a
s
e
t
,
c
r
i
t
e
r
i
a
,
ba
s
e
l
i
ne
s
a
nd
c
a
l
c
ul
a
t
i
ng
c
o
m
put
a
t
i
ona
l
e
f
f
i
c
i
e
nc
y
.
3.1.
D
at
as
e
t
s
W
e
w
i
l
l
s
how
t
he
w
i
de
a
ppl
i
c
a
bi
l
i
t
y
of
t
he
h
y
br
i
d
m
ode
l
us
i
ng
t
he
f
ol
l
ow
i
n
g
v
a
r
i
e
d da
t
a
s
e
t
s
.
a)
W
i
s
c
ons
i
n
br
e
a
s
t
c
a
nc
e
r
da
t
a
s
e
t
(
W
B
C
D
)
D
e
s
c
r
i
pt
i
on:
M
e
di
c
a
l
da
t
a
s
e
t
w
h
e
r
e
t
he
di
a
g
nos
t
i
c
i
nf
or
m
a
t
i
on
i
s
a
v
a
i
l
a
bl
e
f
or
t
he
pr
e
s
e
nc
e
of
br
e
a
s
t
c
a
n
c
e
r
,
be
ni
g
n or
m
a
l
i
g
na
nt
t
u
m
or
.
F
e
a
t
ur
e
s
:
30 num
e
r
i
c
f
e
a
t
ur
e
s
(
r
a
di
us
, t
e
xt
ur
e
, a
nd s
m
oot
hne
s
s
)
O
bj
e
c
t
i
v
e
:
C
he
c
ki
ng
c
l
us
t
e
r
i
n
g
pe
r
f
or
m
a
nc
e
f
or
s
m
a
l
l
l
ow
-
di
m
e
ns
i
ona
l
da
t
a
.
b)
G
e
n
e
e
xpr
e
s
s
i
on o
m
ni
bus
(
G
E
O
)
d
a
t
a
s
e
t
:
D
e
s
c
r
i
pt
i
on:
H
i
g
h
-
di
m
e
ns
i
ona
l
g
e
ne
e
xpr
e
s
s
i
on da
t
a
f
r
o
m
m
i
c
r
oa
r
r
a
y
e
xpe
r
i
m
e
nt
s
.
F
e
a
t
ur
e
s
:
H
undr
e
ds
t
o t
hous
a
nds
of
num
e
r
i
c
a
t
t
r
i
but
e
s
.
T
a
s
k:
A
bi
l
i
t
y
t
o m
od
e
l
hi
g
h
-
di
m
e
ns
i
ona
l
bi
ol
og
i
c
a
l
da
t
a
.
3.2.
E
val
u
at
i
o
n
m
e
t
r
i
c
s
T
o pr
ope
r
l
y
e
v
a
l
ua
t
e
t
he
pe
r
f
or
m
a
nc
e
o
f
t
he
hy
br
i
d
m
od
e
l
, w
e
e
m
pl
o
y
t
he
f
ol
l
ow
i
n
g
m
e
t
r
i
c
s
:
a)
I
nt
e
r
n
a
l
M
e
t
r
i
c
s
(
e
v
a
l
ua
t
e
c
l
us
t
e
r
i
ng
w
i
t
hout
g
r
ound t
r
ut
h)
S
i
l
houe
t
t
e
S
c
or
e
:
I
t
m
e
a
s
ur
e
s
t
he
de
g
r
e
e
of
s
i
m
i
l
a
r
i
t
y
of
a
n obj
e
c
t
t
o i
t
s
c
l
us
t
e
r
c
om
pa
r
e
d t
o ot
he
r
c
l
us
t
e
r
s
.
F
or
m
ul
a
:
ℎ
=
−
m
a
x
(
,
)
w
he
r
e
a
i
s
t
he
a
v
e
r
a
ge
i
nt
r
a
-
c
l
us
t
e
r
di
s
t
a
nc
e
a
nd b i
s
t
he
a
v
e
r
a
g
e
ne
a
r
e
s
t
-
c
l
us
t
e
r
di
s
t
a
nc
e
.
D
a
v
i
e
s
-
boul
di
n i
nde
x
(
D
B
I
)
:
M
e
a
s
ur
e
s
t
he
a
v
e
r
a
ge
s
i
m
i
l
a
r
i
t
y
r
a
t
i
o of
c
l
us
t
e
r
s
t
o t
he
i
r
s
e
pa
r
a
t
i
on.
L
o
w
e
r
v
a
l
u
e
s
i
ndi
c
a
t
e
be
t
t
e
r
-
de
f
i
ne
d c
l
us
t
e
r
s
.
b)
E
xt
e
r
na
l
M
e
t
r
i
c
s
(
r
e
qui
r
e
g
r
ound t
r
ut
h l
a
be
l
s
)
:
P
ur
i
t
y
:
M
e
a
s
ur
e
s
t
he
f
r
a
c
t
i
on of
da
t
a
poi
nt
s
c
or
r
e
c
t
l
y
c
l
a
s
s
i
f
i
e
d t
o
t
he
i
r
g
r
ound t
r
ut
h c
l
us
t
e
r
.
F
or
m
ul
a
:
=
1
∑
|
∩
|
=
1
w
he
r
e
C
k
i
s
a
pr
e
di
c
t
e
d c
l
us
t
e
r
a
nd
T
j
i
s
a
g
r
ound t
r
ut
h c
l
a
s
s
.
A
dj
us
t
e
d
r
a
nd i
nde
x
(
A
R
I
)
:
C
om
pa
r
e
s
c
l
us
t
e
r
i
ng
w
i
t
h t
he
g
r
ound t
r
ut
h w
hi
l
e
a
dj
us
t
i
ng
f
or
c
h
a
nc
e
.
V
a
l
ue
s
r
a
nge
be
t
w
e
e
n
-
1 (
poor
c
l
us
t
e
r
i
ng
)
t
o 1 (
pe
r
f
e
c
t
c
l
us
t
e
r
i
ng
)
.
N
or
m
a
l
i
z
e
d
m
ut
ua
l
i
nf
or
m
a
t
i
on
(
N
M
I
)
:
M
e
a
s
ur
e
s
t
he
a
m
ount
of
i
nf
or
m
a
t
i
on s
ha
r
e
d be
t
w
e
e
n pr
e
di
c
t
e
d c
l
us
t
e
r
s
a
nd
g
r
ound t
r
ut
h c
l
a
s
s
e
s
.
c)
C
om
put
a
t
i
ona
l
E
f
f
i
c
i
e
nc
y
:
R
unt
i
m
e
:
M
e
a
s
ur
e
s
t
he
t
i
m
e
t
a
ke
n t
o
c
o
m
pl
e
t
e
c
l
us
t
e
r
i
ng
.
M
e
m
or
y
U
s
a
g
e
:
E
v
a
l
ua
t
e
s
t
he
m
e
m
or
y
c
ons
u
m
pt
i
on dur
i
ng
t
he
t
r
a
i
ni
ng
a
nd c
l
us
t
e
r
i
n
g
pr
oc
e
s
s
.
3.3.
B
as
e
l
i
n
e
s
T
he
g
i
v
e
n h
y
br
i
d
m
od
e
l
i
s
c
o
m
pa
r
e
d
w
i
t
h t
hr
e
e
ba
s
e
l
i
ne
c
a
t
e
g
or
i
e
s
:
t
r
a
di
t
i
ona
l
c
l
us
t
e
r
i
ng
(
K
-
M
e
a
ns
,
D
B
S
C
A
N
)
,
di
m
e
ns
i
on
r
e
duc
t
i
on
w
i
t
h
c
l
us
t
e
r
i
ng
(
P
C
A
+
K
-
M
e
a
ns
,
t
-
S
N
E
+
D
B
S
C
A
N
)
,
a
nd
hy
br
i
d
m
ode
l
s
(
D
e
e
pC
l
us
t
,
A
ut
oe
nc
ode
r
+
K
-
M
e
a
ns
)
.
S
uc
h
c
o
m
pa
r
i
s
ons
de
t
e
r
m
i
ne
t
he
c
l
us
t
e
r
i
ng
pe
r
f
or
m
a
nc
e
w
i
t
h
a
nd
w
i
t
hout
pr
e
pr
oc
e
s
s
i
ng
a
nd
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on.
T
h
e
a
r
r
a
n
g
e
m
e
nt
f
a
c
i
l
i
t
a
t
e
s
a
t
hor
oug
h
e
v
a
l
ua
t
i
on
of
t
he
e
f
f
e
c
t
i
v
e
ne
s
s
of
t
he
m
od
e
l
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
O
pt
i
m
i
z
i
n
g c
l
us
t
e
r
i
n
g e
f
f
i
c
i
e
nc
y
w
i
t
h
w
e
i
ght
e
d
k
-
m
e
ans
:
a m
ac
hi
ne
l
e
ar
ni
ng
-
dr
i
v
e
n
…
(
V
i
s
hal
K
aus
hi
k
)
1125
3.4.
E
x
p
e
r
i
m
e
n
t
r
e
s
u
l
t
T
he
c
l
a
i
m
e
d
e
f
f
i
c
i
e
nc
y
of
t
he
pr
e
s
e
nt
e
d
hy
br
i
d
m
od
e
l
i
s
c
onf
i
r
m
e
d
b
y
t
he
r
e
s
ul
t
s
of
e
xpe
r
i
m
e
nt
a
l
w
or
k
on
t
w
o
da
t
a
(
W
i
s
c
ons
i
n
a
nd
G
E
O
)
s
e
t
s
.
I
t
ha
s
be
e
n
s
how
n
t
o
be
s
upe
r
i
or
t
o
s
u
c
h
ba
s
e
l
i
ne
s
a
nd
a
l
t
e
r
na
t
i
v
e
s
a
s
P
C
A
+
K
-
M
e
a
ns
a
nd
A
ut
oe
nc
ode
r
+
K
-
M
e
a
ns
i
n
t
e
r
m
s
of
a
c
c
ur
a
c
y
a
nd
s
c
a
l
a
bi
l
i
t
y
,
a
nd
c
he
a
pe
r
i
n
t
e
r
m
s
of
c
o
m
put
a
t
i
ona
l
c
os
t
.
T
he
pa
r
a
m
e
t
e
r
s
of
D
B
I
a
nd
N
M
I
r
e
c
or
de
d
t
he
pot
e
nt
i
a
l
of
t
he
m
ode
l
t
o r
e
pr
e
s
e
nt
a
c
t
ua
l
pa
t
t
e
r
ns
of
da
t
a
pa
r
t
i
c
ul
a
r
l
y
on t
he
hi
g
h di
m
e
ns
i
ona
l
G
E
O
da
t
a
F
i
g
u
r
e
1.
F
i
g
ur
e
1. S
i
l
houe
t
t
e
s
c
or
e
c
om
pa
r
i
s
on
a
nd
pur
i
t
y
s
c
or
e
c
o
m
p
a
r
i
s
on
4.
R
E
S
U
L
T
S
A
N
D
D
I
S
C
U
S
S
I
O
N
T
he
r
e
s
ul
t
s
of
t
he
e
xpe
r
i
m
e
nt
s
v
a
l
i
da
t
e
t
he
pr
opos
e
d
hy
br
i
d
m
ode
l
'
s
e
f
f
e
c
t
i
v
e
n
e
s
s
a
c
r
os
s
t
he
t
w
o
da
t
a
s
e
t
s
, na
m
e
l
y
t
he
W
B
C
D
a
nd t
he
G
E
O
D
a
t
a
s
e
t
.
4.1.
Wi
s
c
on
s
i
n
d
at
as
e
t
T
he
hy
br
i
d
m
ode
l
e
xpl
oi
t
e
d
t
he
ne
ur
a
l
ne
t
w
or
k
t
o
c
a
r
r
y
out
e
f
f
e
c
t
i
v
e
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on
a
nd
di
m
e
ns
i
ona
l
i
t
y
r
e
duc
t
i
on
on
t
he
W
B
C
D
.
I
t
w
a
s
be
t
t
e
r
t
ha
n
c
on
v
e
nt
i
ona
l
a
ppr
oa
c
he
s
r
e
g
a
r
d
i
ng
t
he
S
i
l
houe
t
t
e
S
c
or
e
0.85
(
a
s
c
o
m
pa
r
e
d
t
o
0.76
a
nd
0.79
of
K
-
M
e
a
ns
a
nd
P
C
A
+
K
-
M
e
a
ns
,
r
e
s
pe
c
t
i
ve
l
y
)
,
w
hi
c
h
m
e
a
nt
m
o
r
e
di
s
t
i
nc
t
a
nd
c
ohe
r
e
nt
c
l
us
t
e
r
s
.
T
he
pur
i
t
y
s
c
or
e
w
a
s
0.91,
c
om
pa
r
e
d
t
o
K
-
M
e
a
ns
(
0.
85)
a
nd
P
C
A
+
K
-
M
e
a
ns
(
0.87)
. B
ot
h w
e
r
e
c
o
m
p
a
r
e
d t
o t
he
i
r
gr
ound t
r
ut
h, i
ndi
c
a
t
i
ng
a
be
t
t
e
r
m
a
t
c
h.
T
he
A
R
I
w
a
s
0.88,
w
hi
c
h
m
e
a
ns
a
s
t
r
on
g
c
on
g
r
ue
nc
y
w
i
t
h t
he
da
t
a
s
e
t
'
s
s
t
r
uc
t
ur
e
a
s
s
how
n i
n
F
i
g
ur
e
2.
F
i
g
ur
e
2
. P
e
r
f
or
m
a
nc
e
c
o
m
pa
r
i
s
on on w
i
s
c
ons
i
n da
t
a
s
e
t
a
nd
c
o
m
pa
r
a
t
i
v
e
pe
r
f
or
m
a
n
c
e
of
c
l
us
t
e
r
i
ng
m
e
t
hods
4.2.
G
E
O
d
at
a
s
e
t
T
hi
s
w
a
s
m
or
e
of
t
he
c
a
s
e
w
i
t
h
t
he
G
E
O
D
a
t
a
s
e
t
,
a
s
i
t
i
s
bi
ol
og
i
c
a
l
l
y
c
o
m
pl
e
x
a
nd
of
hi
gh
di
m
e
ns
i
ons
.
S
t
i
l
l
,
f
or
t
hi
s
r
e
a
s
on,
t
he
h
y
br
i
d
m
ode
l
w
a
s
a
bl
e
t
o
e
xt
r
a
c
t
l
a
t
e
nt
pa
t
t
e
r
n
s
us
i
ng
i
t
s
ne
ur
a
l
ne
t
w
or
k.
I
t
r
e
a
c
h
e
d
a
S
i
l
houe
t
t
e
S
c
or
e
of
0.78,
w
hi
c
h
w
a
s
hi
g
h
e
r
t
ha
n
K
-
M
e
a
ns
(
0.68)
a
nd
P
C
A
+
K
-
M
e
a
ns
(
0.72)
,
i
m
pl
y
i
ng
t
ha
t
i
t
i
s
be
t
t
e
r
a
t
ha
ndl
i
ng
di
m
e
ns
i
ona
l
i
t
y
.
I
t
a
l
s
o
ha
d
m
o
r
e
bi
ol
og
i
c
a
l
l
y
r
e
l
e
v
a
nt
c
l
us
t
e
r
s
w
i
t
h
a
P
ur
i
t
y
S
c
or
e
of
0.82
(
i
n
c
o
m
pa
r
i
s
on
t
o
0.78
of
K
-
M
e
a
ns
)
.
T
he
hi
g
h
N
M
I
of
0.80
on
t
he
m
ode
l
a
l
s
o
s
how
e
d a
hi
g
h de
gr
e
e
of
c
o
m
pa
t
i
bi
l
i
t
y
w
i
t
h t
he
da
t
a
s
t
r
uc
t
ur
e
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
, V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
112
1
-
1
128
1126
4.3
.
C
om
p
ar
at
i
ve
p
e
r
f
or
m
an
c
e
T
he
h
y
br
i
d
m
e
t
hod
w
a
s
,
i
n
one
w
a
y
,
be
t
t
e
r
t
ha
n
t
he
e
xi
s
t
i
ng
a
nd
t
r
a
di
t
i
ona
l
a
ppr
oa
c
he
s
.
T
h
e
c
l
a
s
s
i
c
a
l
c
l
us
t
e
r
i
ng
(
K
-
M
e
a
ns
,
D
B
S
C
A
N
)
c
oul
d
not
ha
ndl
e
hi
g
h
-
di
m
e
ns
i
ona
l
noi
s
e
,
a
nd
t
he
P
C
A
-
ba
s
e
d
m
e
t
hods
di
d
not
pe
r
f
or
m
m
uc
h
be
t
t
e
r
a
nd
f
a
i
l
e
d
t
o
c
a
pt
ur
e
non
-
l
i
ne
a
r
t
r
e
nds
.
T
he
m
ode
l
s
of
A
ut
oe
nc
ode
r
+
K
-
M
e
a
ns
w
e
r
e
s
i
m
i
l
a
r
but
di
d
not
pe
r
f
or
m
w
e
l
l
e
noug
h
a
s
t
he
y
unf
ol
de
d
due
t
o
a
l
a
c
k
of
a
r
c
hi
t
e
c
t
ur
a
l
opt
i
m
i
s
a
t
i
on a
s
s
how
n i
n F
i
g
ur
e
3
.
F
i
g
ur
e
3. C
om
pa
r
a
t
i
v
e
pe
r
f
or
m
a
n
c
e
of
c
l
us
t
e
r
i
ng
m
e
t
hods
4.4
.
I
n
s
i
gh
t
s
f
r
om
vi
s
u
al
i
z
at
i
on
V
i
s
ua
l
i
z
a
t
i
on
of
t
he
l
a
t
e
nt
f
e
a
t
ur
e
s
pa
c
e
c
l
us
t
e
r
t
o
g
a
i
n
de
e
pe
r
i
ns
i
g
ht
s
i
n
t
o
t
he
f
e
a
t
ur
e
e
xt
r
a
c
t
i
on
c
a
pa
bi
l
i
t
y
of
t
he
m
o
de
l
f
or
t
he
W
i
s
c
ons
i
n
da
t
a
s
e
t
a
s
s
how
n
i
n
F
i
g
ur
e
4
,
t
he
hy
br
i
d
m
ode
l
y
i
e
l
de
d
c
om
pa
c
t
a
nd
w
e
l
l
-
s
e
pa
r
a
t
e
d
c
l
us
t
e
r
,
d
e
m
ons
t
r
a
t
i
ng
e
f
f
e
c
t
i
v
e
di
m
e
ns
i
ona
l
i
t
y
r
e
d
uc
t
i
on
a
nd
g
ood
c
l
a
s
s
di
s
c
r
i
m
i
na
t
i
on.
I
n
t
he
G
E
O
da
t
a
s
e
t
(
F
i
g
ur
e
4)
,
t
he
m
ode
l
s
uc
c
e
s
s
f
ul
l
y
c
a
pt
ur
e
d
bi
ol
og
i
c
a
l
l
y
m
e
a
ni
ng
f
ul
pa
t
t
e
r
ns
,
a
s
e
v
i
de
nc
e
d
b
y
t
he
n
a
t
ur
a
l
s
e
pa
r
a
t
i
on
c
l
us
t
e
r
s
oc
c
ur
de
s
pi
t
e
t
he
da
t
a
'
s
hi
g
h
di
m
e
ns
i
ona
l
i
t
y
a
nd
c
o
m
pl
e
xi
t
y
o
f
t
he
da
t
a
.
F
i
g
ur
e
4. t
-
S
N
E
v
i
s
ua
l
i
z
a
t
i
on of
l
a
t
e
nt
f
e
a
t
ur
e
s
pa
c
e
f
or
w
i
s
c
ons
i
n da
t
a
s
e
t
a
nd
t
-
S
N
E
v
i
s
ua
l
i
z
a
t
i
on of
l
a
t
e
nt
f
e
a
t
ur
e
s
pa
c
e
f
o
r
G
E
O
da
t
a
s
e
t
4.5.
D
i
s
c
u
s
s
i
on
T
he
p
r
opos
e
d
hy
br
i
d
s
c
h
e
m
e
c
ons
i
s
t
i
ng
of
ne
ur
a
l
ne
t
w
or
k
-
ba
s
e
d
f
e
a
t
ur
e
l
e
a
r
ni
n
g
a
nd
c
l
us
t
e
r
i
n
g
a
l
g
or
i
t
h
m
s
ho
w
e
d i
m
p
r
e
s
s
i
v
e
pe
r
f
or
m
a
n
c
e
boos
t
s
a
c
r
os
s
m
ul
t
i
pl
e
da
t
a
s
e
t
s
.
T
he
W
i
s
c
ons
i
n da
t
a
s
e
t
pr
oduc
e
d
a
S
i
l
houe
t
t
e
S
c
or
e
of
0.85 a
nd
a
P
ur
i
t
y
S
c
or
e
of
0.91;
t
he
c
o
m
pl
e
x
G
E
O
d
a
t
a
s
e
t
ha
d
a
S
i
l
hou
e
t
t
e
S
c
or
e
of
0.78
,
a
n
N
M
I
of
0.80,
a
nd
out
pe
r
f
or
m
i
n
g
t
r
a
di
t
i
ona
l
c
l
us
t
e
r
i
ng
m
e
t
hods
a
nd
P
C
A
-
ba
s
e
d
c
l
us
t
e
r
i
ng
m
e
t
hods
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
I
S
S
N
:
2502
-
4752
O
pt
i
m
i
z
i
n
g c
l
us
t
e
r
i
n
g e
f
f
i
c
i
e
nc
y
w
i
t
h
w
e
i
ght
e
d
k
-
m
e
ans
:
a m
ac
hi
ne
l
e
ar
ni
ng
-
dr
i
v
e
n
…
(
V
i
s
hal
K
aus
hi
k
)
1127
C
om
pa
r
e
d
w
i
t
h
a
d
v
a
n
c
e
d
m
ode
l
s
l
i
ke
D
E
C
,
I
D
E
C
,
a
nd
D
e
e
pC
l
us
t
e
r
,
our
m
e
t
hod
hol
ds
c
om
p
e
t
i
t
i
v
e
c
l
us
t
e
r
i
ng
q
ua
l
i
t
y
w
i
t
hout
r
e
qui
r
i
ng
e
xt
e
ns
i
v
e
p
r
e
-
t
r
a
i
ni
n
g
a
nd
o
f
f
e
r
i
n
g
a
n
e
f
f
e
c
t
i
v
e
ba
l
a
nc
e
be
t
w
e
e
n
pe
r
f
or
m
a
nc
e
a
nd e
f
f
i
c
i
e
nc
y
i
n c
o
m
put
a
t
i
on.
4.6.
S
c
al
a
b
i
l
i
t
y an
d
c
om
p
u
t
at
i
on
a
l
e
f
f
i
c
i
e
n
c
y
A
s
c
a
l
a
bi
l
i
t
y
a
s
s
e
s
s
m
e
nt
of
t
he
h
y
br
i
d
m
ode
l
oc
c
ur
r
e
d
us
i
ng
e
qui
p
m
e
nt
w
hi
c
h
i
nc
l
ude
d
a
n
I
nt
e
l
C
or
e
i
7
pr
oc
e
s
s
or
w
i
t
h
32G
B
R
A
M
a
nd
a
n
N
V
I
D
I
A
R
T
X
3060
G
P
U
.
I
m
pl
e
m
e
nt
i
ng
ne
ur
a
l
ne
t
w
or
k
t
r
a
i
ni
ng
a
dde
d
a
bout
20
–
30%
e
xt
r
a
t
i
m
e
pr
oc
e
s
s
i
ng
dur
i
ng
t
he
t
r
a
i
ni
n
g
s
t
a
g
e
but
pr
o
v
i
de
d
35
–
40%
a
c
c
e
l
e
r
a
t
e
d
c
l
us
t
e
r
i
ng
pe
r
f
or
m
a
nc
e
f
r
o
m
P
C
A
-
ba
s
e
d t
e
c
hni
que
s
.
5.
C
O
N
C
L
U
S
I
O
N
A
N
D
F
U
T
U
R
E
WO
R
K
T
he
r
e
s
e
a
r
c
h
de
s
i
g
n
us
e
d
hy
br
i
d
c
l
us
t
e
r
i
ng
m
e
t
hods
,
w
hi
c
h
i
nt
e
g
r
a
t
e
d
ne
ur
a
l
ne
t
w
or
k
f
e
a
t
ur
e
s
w
i
t
h
s
t
a
nda
r
d
c
l
us
t
e
r
i
ng
a
l
g
or
i
t
hm
s
t
o
c
r
e
a
t
e
m
or
e
s
t
a
bl
e
,
pr
e
c
i
s
e
,
a
nd
r
e
s
i
l
i
e
nt
s
y
s
t
e
m
s
.
T
he
W
i
s
c
ons
i
n
B
r
e
a
s
t
C
a
nc
e
r
a
nd
G
E
O
da
t
a
s
e
t
s
pr
oduc
e
d
e
xp
e
r
i
m
e
nt
a
l
out
c
o
m
e
s
s
ho
w
i
ng
s
ubs
t
a
nt
i
a
l
e
nha
nc
e
m
e
nt
s
of
S
i
l
houe
t
t
e
S
c
or
e
,
P
ur
i
t
y
,
A
R
I
a
nd
N
M
I
c
o
m
p
a
r
e
d
t
o
s
t
a
nda
r
d
t
e
c
hni
que
s
.
F
ut
ur
e
de
v
e
l
opm
e
nt
of
t
hi
s
m
od
e
l
w
i
l
l
a
i
m
t
o
i
m
pr
o
v
e
c
o
m
put
a
t
i
ona
l
e
f
f
i
c
i
e
nc
y
t
hr
oug
h
l
i
g
ht
w
e
i
g
ht
de
s
i
g
ns
us
i
ng
t
r
a
ns
f
e
r
l
e
a
r
ni
ng
t
o
de
c
r
e
a
s
e
t
r
a
i
ni
ng
e
xpe
ns
e
s
a
nd
br
oa
de
n
i
t
s
a
ppl
i
c
a
bi
l
i
t
y
t
o
s
e
m
i
-
s
upe
r
v
i
s
e
d
c
l
us
t
e
r
i
ng
w
i
t
h
e
xpl
a
i
na
bl
e
A
r
t
i
f
i
c
i
a
l
I
nt
e
l
l
i
g
e
nc
e
m
e
t
hods
t
o
e
nha
nc
e
i
nt
e
r
pr
e
t
a
bi
l
i
t
y
.
R
e
a
l
-
w
or
l
d
v
a
l
i
da
t
i
on
of
t
he
m
ode
l
w
i
l
l
be
c
om
e
pos
s
i
bl
e
t
hr
oug
h t
e
s
t
i
ng
a
c
r
os
s
m
ul
t
i
pl
e
m
ode
s
a
nd e
xt
e
ns
i
v
e
da
t
a
s
e
t
s
a
m
pl
e
s
.
F
U
N
D
I
N
G
I
N
F
O
R
M
A
T
I
O
N
A
ut
hor
s
s
t
a
t
e
no f
undi
ng
i
n
v
ol
v
e
d.
C
O
N
F
L
I
C
T
O
F
I
N
T
E
R
E
S
T
S
T
A
T
E
M
E
N
T
A
ut
hor
s
s
t
a
t
e
no c
onf
l
i
c
t
of
i
nt
e
r
e
s
t
.
D
A
T
A
A
V
A
I
L
A
B
I
L
I
T
Y
D
a
t
a
a
v
a
i
l
a
bi
l
i
t
y
i
s
not
a
ppl
i
c
a
bl
e
t
o
t
hi
s
pa
pe
r
a
s
no
ne
w
da
t
a
w
e
r
e
c
r
e
a
t
e
d
o
r
a
na
l
y
z
e
d
i
n
t
hi
s
s
t
udy
.
R
E
F
E
R
E
N
C
E
S
[
1]
A
.
K
.
J
a
i
n
a
n
d
R
.
C
.
D
ube
s
,
“
A
l
g
o
r
i
t
hm
s
f
o
r
c
l
us
t
e
r
i
ng
da
t
a
,
”
P
r
e
nt
i
c
e
-
H
al
l
,
I
n
c
.
,
1988.
[
2]
M
.
E
s
t
e
r
,
H
.
-
P
.
K
r
i
e
g
e
l
,
J
.
S
a
n
de
r
,
a
nd
X
.
X
u
,
“
A
de
n
s
i
t
y
-
ba
s
e
d
a
l
g
o
r
i
t
hm
f
o
r
di
s
c
ove
r
i
ng
c
l
us
t
e
r
s
i
n
l
a
r
g
e
s
pa
t
i
a
l
da
t
a
ba
s
e
s
w
i
t
h
n
oi
s
e
,
”
I
n
k
dd
,
vol
.
96,
n
o.
34,
pp.
226
–
231,
1996.
[
3]
Y
.
B
e
ng
i
o
,
A
.
C
our
vi
l
l
e
,
a
n
d
P
.
V
i
n
c
e
n
t
,
“
R
e
p
r
e
s
e
n
t
a
t
i
o
n
l
e
a
r
n
i
ng
:
A
r
e
vi
e
w
a
nd
n
e
w
pe
r
s
pe
c
t
i
ve
s
,
”
I
E
E
E
T
r
ans
ac
t
i
ons
o
n
P
at
t
e
r
n
A
nal
y
s
i
s
and
M
ac
hi
ne
I
nt
e
l
l
i
ge
n
c
e
,
vol
.
35,
n
o.
8,
pp.
1798
–
1828,
A
u
g
.
2013,
doi
:
10.
1109/
T
P
A
M
I
.
201
3.
50.
[
4]
G
.
E
.
H
i
nt
o
n
a
n
d
R
.
R
.
S
a
l
a
k
h
ut
di
n
o
v,
“
R
e
d
uc
i
ng
t
h
e
di
m
e
n
s
i
o
n
a
l
i
t
y
of
da
t
a
w
i
t
h
n
e
ur
a
l
n
e
t
w
o
r
ks
,
”
S
c
i
e
n
c
e
,
vol
.
313,
n
o.
5786
,
pp.
504
–
507,
J
ul
.
2006,
doi
:
10.
1126/
s
c
i
e
n
c
e
.
1127647.
[
5]
R
.
S
.
S
ut
t
on
a
n
d
A
.
G
.
B
a
r
t
o,
R
e
i
nf
or
c
e
m
e
nt
l
e
ar
ni
ng:
A
n
i
nt
r
odu
c
t
i
on
,
vol
.
1,
n
o.
1.
1998.
[
6]
Y
.
L
e
c
u
n,
Y
.
B
e
ng
i
o
,
a
n
d
G
.
H
i
n
t
o
n,
“
D
e
e
p
l
e
a
r
n
i
ng,
”
N
at
ur
e
,
vol
.
521,
n
o.
7553
,
pp.
436
–
444,
M
a
y
2015
,
doi
:
10.
1038/
n
a
t
ur
e
14539.
[
7]
Z
.
X
u,
Y
.
G
ong,
Y
.
Z
h
ou
,
Q
.
B
a
o,
a
nd
W
.
Q
i
a
n,
“
E
nh
a
n
c
i
ng
K
ube
r
n
e
t
e
s
a
ut
o
m
a
t
e
d
s
c
h
e
dul
i
ng
w
i
t
h
de
e
p
l
e
a
r
n
i
ng
a
nd
r
e
i
n
f
or
c
e
m
e
nt
t
e
c
hn
i
que
s
f
or
l
a
r
g
e
-
s
c
a
l
e
c
l
oud
c
om
pu
t
i
ng
opt
i
m
i
z
a
t
i
o
n
,
”
i
n
N
i
nt
h
I
nt
e
r
nat
i
onal
Sy
m
pos
i
um
on
A
dv
a
nc
e
s
i
n
E
l
e
c
t
r
i
c
a
l
,
E
l
e
c
t
r
oni
c
s
,
and
C
om
put
e
r
E
ngi
ne
e
r
i
ng
(
I
S
A
E
E
C
E
2024)
,
O
c
t
.
2024,
p.
175,
doi
:
10.
1117/
12.
3034052.
[
8]
L.
E
r
i
c
s
s
on,
H
.
G
ouk
,
a
n
d
T
.
M
.
H
os
pe
da
l
e
s
,
“
H
o
w
w
e
l
l
do
s
e
l
f
-
s
upe
r
vi
s
e
d
m
ode
l
s
t
r
a
n
s
f
e
r
?
,
”
i
n
P
r
o
c
e
e
di
n
gs
of
t
he
I
E
E
E
C
om
put
e
r
Soc
i
e
t
y
C
onf
e
r
e
nc
e
on
C
om
put
e
r
V
i
s
i
on
and
P
at
t
e
r
n
R
e
c
og
ni
t
i
on
,
J
un
.
2021
,
pp.
5410
–
5419,
do
i
:
10.
1109/
C
V
P
R
46437.
2021.
00537.
[
9]
X.
A
n
e
t
al
.
,
“
S
T
R
P
-
D
B
S
C
A
N
:
A
pa
r
a
l
l
e
l
D
B
S
C
A
N
a
l
gor
i
t
hm
ba
s
e
d
on
s
pa
t
i
a
l
-
t
e
m
po
r
a
l
r
a
n
do
m
pa
r
t
i
t
i
o
n
i
ng
f
or
c
l
us
t
e
r
i
n
g
t
r
a
j
e
c
t
or
y
da
t
a
,
”
A
pp
l
i
e
d
Sc
i
e
n
c
e
s
(
Sw
i
t
z
e
r
l
and)
,
vol
.
13,
n
o.
20,
p.
11122,
O
c
t
.
2023,
doi
:
10.
3390/
a
pp13201
1122.
[
10]
S
.
Z
h
ou
e
t
al
.
,
“
A
c
o
m
pr
e
h
e
n
s
i
ve
s
ur
ve
y
on
de
e
p
c
l
us
t
e
r
i
ng
:
t
a
x
o
n
o
m
y
,
c
h
a
l
l
e
ng
e
s
,
a
n
d
f
ut
ur
e
di
r
e
c
t
i
o
n
s
,
”
A
C
M
C
om
put
i
ng
Sur
v
e
y
s
,
vol
.
57,
n
o.
3,
pp.
1
–
38,
N
ov.
2024,
doi
:
10.
1145/
3689036.
[
11]
S
.
C
.
J
oh
n
s
o
n
,
“
H
i
e
r
a
r
c
h
i
c
a
l
c
l
us
t
e
r
i
ng
s
c
h
e
m
e
s
,
”
P
s
y
c
hom
e
t
r
i
k
a
,
vol
.
32,
n
o.
3,
pp.
241
–
25
4,
S
e
p.
1967,
doi
:
10.
1007/
B
F
02289588.
[
12]
A
.
K
r
i
z
h
e
vs
ky
,
I
.
S
ut
s
ke
ve
r
,
a
n
d
G
.
E
.
H
i
n
t
o
n,
“
I
m
a
g
e
N
e
t
c
l
a
s
s
i
f
i
c
a
t
i
o
n
w
i
t
h
de
e
p
c
on
vol
ut
i
o
n
a
l
n
e
ur
a
l
n
e
t
w
o
r
ks
,
”
C
om
m
uni
c
a
t
i
ons
of
t
he
A
C
M
,
vol
.
60,
n
o.
6,
pp.
84
–
90,
2017,
doi
:
10.
1145/
306
5386.
[
13]
G
.
L
i
,
W
.
W
a
ng,
W
.
Z
h
a
ng
,
W
.
Y
ou,
F
.
W
u,
a
n
d
H
.
T
u,
“
H
a
n
dl
i
ng
m
ul
t
i
m
oda
l
m
ul
t
i
-
obj
e
c
t
i
ve
pr
obl
e
m
s
t
h
r
ou
gh
s
e
l
f
-
or
g
a
ni
z
i
n
g
qua
n
t
u
m
-
i
n
s
pi
r
e
d
pa
r
t
i
c
l
e
s
w
a
r
m
opt
i
m
i
z
a
t
i
o
n,
”
I
nf
or
m
at
i
on
Sc
i
e
nc
e
s
,
vol
.
577,
pp.
510
–
540,
O
c
t
.
2021,
doi
:
10.
1016/
j
.
i
n
s
.
2021.
07.
011.
[
14]
L
.
L
i
u
a
nd
Z
.
W
a
ng,
“
S
ur
ve
y
o
n
m
ul
t
i
ob
j
e
c
t
i
ve
opt
i
m
i
z
a
t
i
o
n
e
vol
ut
i
o
n
a
r
y
a
l
g
or
i
t
hm
ba
s
e
d
on
de
c
o
m
pos
i
t
i
on,
”
J
our
nal
of
Sof
t
w
ar
e
,
vol
.
10,
n
o.
4743
–
4771,
p.
2023,
34A
D
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2502
-
4752
I
ndone
s
i
a
n J
E
l
e
c
E
n
g
&
C
om
p S
c
i
, V
ol
.
40
, N
o.
2
,
N
ov
e
m
be
r
20
25
:
112
1
-
1
128
1128
[
15]
C
.
G
.
S
oa
r
e
s
a
n
d
T
.
A
.
S
a
n
t
os
,
A
dv
an
c
e
s
i
n
m
ar
i
t
i
m
e
t
e
c
hnol
ogy
and
e
ngi
ne
e
r
i
n
g,
v
ol
um
e
2
.
C
R
C
P
r
e
s
s
,
2024.
[
16]
Q
.
F
e
n
g,
C
.
L
.
P
.
C
h
e
n,
a
n
d
L
.
L
i
u,
“
A
r
e
vi
e
w
of
c
on
ve
x
c
l
us
t
e
r
i
ng
f
r
o
m
m
ul
t
i
pl
e
pe
r
s
pe
c
t
i
ve
s
:
m
ode
l
s
,
opt
i
m
i
z
a
t
i
o
n
s
,
s
t
a
t
i
s
t
i
c
a
l
pr
ope
r
t
i
e
s
,
a
ppl
i
c
a
t
i
o
n
s
,
a
n
d
c
onn
e
c
t
i
o
n
s
,
”
I
E
E
E
T
r
ans
ac
t
i
ons
on
N
e
ur
al
N
e
t
w
or
k
s
and
L
e
ar
ni
ng
Sy
s
t
e
m
s
,
vol
.
35,
n
o.
10,
pp.
13122
–
13142
,
O
c
t
.
2024,
doi
:
10.
1109/
T
N
N
L
S
.
2023.
3276393.
[
17]
A
.
N
g
,
M
.
J
or
da
n,
a
n
d
Y
.
W
e
i
s
s
,
“
O
n
s
pe
c
t
r
a
l
c
l
us
t
e
r
i
ng
:
A
n
a
l
ys
i
s
a
n
d
a
n
a
l
g
or
i
t
hm
,
”
A
dv
an
c
e
s
i
n
ne
ur
al
i
n
f
or
m
at
i
on
pr
oc
e
s
s
i
ng
s
y
s
t
e
m
s
,
vol
.
14,
2001.
[
18]
J
.
X
i
e
,
R
.
G
i
r
s
h
i
c
k
,
a
n
d
A
.
F
a
r
h
a
di
,
“
U
n
s
upe
r
vi
s
e
d
de
e
p
e
m
be
ddi
ng
f
or
c
l
us
t
e
r
i
ng
a
n
a
l
ys
i
s
,
”
I
nt
e
r
nat
i
ona
l
c
o
nf
e
r
e
nc
e
on
m
a
c
hi
n
e
l
e
ar
ni
ng
,
pp.
478
–
487,
2016.
[
19]
E
.
A
l
j
a
l
bout
,
V
.
G
ol
kov,
Y
.
S
i
ddi
qui
,
M
.
S
t
r
obe
l
,
a
n
d
D
.
C
r
e
m
e
r
s
,
“
C
l
us
t
e
r
i
ng
w
i
t
h
de
e
p
l
e
a
r
n
i
ng
:
T
a
x
o
n
o
m
y
a
n
d
n
e
w
m
e
t
hods
,
”
ar
X
i
v
pr
e
pr
i
nt
ar
X
i
v
:
1801.
07648
,
2018.
[
20]
X
.
G
uo,
X
.
L
i
u,
E
.
Z
h
u
,
a
n
d
J
.
Y
i
n,
“
D
e
e
p
c
l
us
t
e
r
i
ng
w
i
t
h
c
o
n
vol
ut
i
o
n
a
l
a
ut
oe
n
c
ode
r
s
,
”
i
n
L
e
c
t
ur
e
N
ot
e
s
i
n
C
o
m
put
e
r
Sc
i
e
nc
e
(
i
n
c
l
udi
ng
s
ubs
e
r
i
e
s
L
e
c
t
ur
e
N
ot
e
s
i
n
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
n
c
e
and
L
e
c
t
ur
e
N
ot
e
s
i
n
B
i
oi
nf
or
m
at
i
c
s
)
,
vol
.
10635
L
N
C
S
,
S
pr
i
ng
e
r
I
nt
e
r
n
a
t
i
o
n
a
l
P
ubl
i
s
h
i
ng,
2017,
pp.
373
–
382.
[
21]
P
.
H
ua
ng,
Y
.
H
ua
ng,
W
.
W
a
ng
,
a
n
d
L
.
W
a
ng,
“
D
e
e
p
e
m
be
ddi
ng
n
e
t
w
or
k
f
or
c
l
us
t
e
r
i
ng
,
”
i
n
P
r
o
c
e
e
di
ngs
-
I
nt
e
r
nat
i
ona
l
C
onf
e
r
e
nc
e
on
P
at
t
e
r
n
R
e
c
ogni
t
i
on
,
A
u
g
.
2014,
pp.
1532
–
1537,
doi
:
10.
1109/
I
C
P
R
.
2014.
272.
[
22]
L
.
V
a
n
de
r
M
a
a
t
e
n
a
n
d
G
.
H
i
nt
o
n
,
“
V
i
s
ua
l
i
z
i
ng
da
t
a
us
i
ng
t
-
S
N
E
,
”
J
our
nal
of
m
ac
hi
ne
l
e
ar
ni
ng
r
e
s
e
ar
c
h
,
vol
.
9,
n
o.
11,
2008.
[
23]
A
.
B
.
L
.
L
a
r
s
e
n
,
S
.
K
.
S
ø
nde
r
by
,
H
.
L
a
r
oc
h
e
l
l
e
,
a
n
d
O
.
W
i
n
t
h
e
r
,
“
A
ut
oe
n
c
od
i
ng
be
yo
n
d
p
i
x
e
l
s
us
i
ng
a
l
e
a
r
n
e
d
s
i
m
i
l
a
r
i
t
y
m
e
t
r
i
c
,
”
i
nt
e
r
nat
i
onal
c
onf
e
r
e
nc
e
on
m
ac
hi
ne
l
e
ar
ni
n
,
pp.
1558
–
1566,
2016.
[
24]
Z
.
J
i
a
ng
,
Y
.
Z
h
e
ng,
H
.
T
a
n,
B
.
T
a
ng
,
a
n
d
H
.
Z
h
ou
,
“
V
a
r
i
a
t
i
o
n
a
l
de
e
p
e
m
be
ddi
ng
:
A
n
u
n
s
upe
r
vi
s
e
d
g
e
n
e
r
a
t
i
ve
a
ppr
oa
c
h
t
o
C
l
us
t
e
r
i
ng,
”
i
n
I
J
C
A
I
I
nt
e
r
nat
i
onal
J
oi
n
t
C
onf
e
r
e
nc
e
on
A
r
t
i
f
i
c
i
al
I
nt
e
l
l
i
ge
n
c
e
,
A
u
g
.
2017,
vol
.
0
,
pp.
1965
–
1972,
d
oi
:
10.
24963/
i
j
c
a
i
.
2017/
273.
[
25]
Y
.
B
e
ng
i
o,
P
.
L
a
m
b
l
i
n
,
D
.
P
opov
i
c
i
,
a
n
d
H
.
L
a
r
oc
h
e
l
l
e
,
“
G
r
e
e
dy
l
a
ye
r
-
w
i
s
e
t
r
a
i
n
i
ng
of
de
e
p
n
e
t
w
o
r
k
s
,
”
i
n
N
I
P
S
2006
:
P
r
oc
e
e
di
ngs
of
t
he
19t
h
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
on
N
e
ur
al
I
nf
o
r
m
at
i
on
P
r
oc
e
s
s
i
ng
Sy
s
t
e
m
s
,
T
h
e
M
I
T
P
r
e
s
s
,
2006,
pp.
153
–
160.
[
26]
L
.
S
o
ng
,
A
.
S
m
o
l
a
,
A
.
G
r
e
t
t
o
n,
K
.
M
.
B
o
r
gw
a
r
d
t
,
a
n
d
J
.
B
e
do
,
“
S
upe
r
vi
s
e
d
f
e
a
t
u
r
e
s
e
l
e
c
t
i
o
n
vi
a
de
pe
n
de
n
c
e
e
s
t
i
m
a
t
i
o
n
,
”
i
n
A
C
M
I
nt
e
r
nat
i
onal
C
onf
e
r
e
nc
e
P
r
oc
e
e
di
ng
Se
r
i
e
s
,
J
un
.
2007,
vol
.
227,
pp.
823
–
830,
doi
:
10.
1145/
1
273496.
1273600.
[
27]
D
.
P
.
K
i
ngm
a
a
n
d
M
.
W
e
l
l
i
ng
,
“
A
ut
o
-
e
n
c
odi
ng
va
r
i
a
t
i
o
n
a
l
ba
ye
s
,
”
ar
X
i
v
pr
e
pr
i
nt
ar
X
i
v
:
1312.
6114
,
2013
.
B
I
O
G
R
A
P
H
I
E
S
O
F
A
U
T
H
O
R
S
Vis
h
a
l
K
a
u
s
h
ik
i
s
a
tec
h
n
o
l
o
g
y
a
n
d
b
u
s
i
n
e
s
s
v
i
s
i
o
n
a
r
y
p
r
o
f
e
s
s
i
o
n
a
l
w
i
th
o
v
e
r
2
8
+
y
e
a
r
s
o
f
e
x
p
e
r
i
e
n
c
e
a
s
a
G
l
o
b
a
l
De
li
v
e
r
y
E
x
e
c
u
ti
v
e
(
VP)
a
t
HC
L
T
e
c
h
n
o
l
o
g
i
e
s
L
td
.
H
i
s
e
x
p
e
r
ti
s
e
i
n
d
r
iv
i
n
g
d
ig
i
ta
l
tr
a
n
s
f
o
r
m
a
ti
o
n
i
n
i
t
i
a
t
i
v
e
s
h
a
s
c
o
n
s
i
s
ten
tl
y
c
o
n
tr
i
b
u
ted
to
t
h
e
s
u
c
c
e
s
s
o
f
i
n
d
u
s
tr
y
-
l
e
a
d
i
n
g
I
o
T
t
r
a
n
s
f
o
r
m
a
ti
o
n
p
r
o
g
r
a
m
s
.
He
h
a
s
wo
r
k
e
d
a
c
r
o
s
s
h
i
s
c
a
r
e
e
r
li
f
e
s
p
a
n
a
s
s
o
f
twa
r
e
d
e
v
e
l
o
p
e
r
,
tes
ter
,
QA
,
D
B
A,
l
e
a
d
e
n
g
i
n
e
e
r
,
p
r
o
jec
t
m
a
n
a
g
e
r
,
d
e
li
v
e
r
y
m
a
n
a
g
e
r
,
p
r
o
g
r
a
m
m
a
n
a
g
e
r
,
d
e
l
iv
e
r
y
e
x
e
c
u
ti
v
e
,
i
n
n
o
v
a
to
r
,
r
e
s
e
a
r
c
h
e
r
,
a
d
m
i
n
i
s
tr
a
to
r
,
a
n
d
i
n
tell
e
c
tu
a
l
p
r
o
f
e
s
s
i
o
n
a
l
.
H
i
s
r
e
s
e
a
r
c
h
i
n
ter
e
s
ts
i
n
c
l
u
d
e
I
n
ter
n
e
t
o
f
th
i
n
g
s
,
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
Ge
n
e
r
a
ti
v
e
a
r
ti
f
i
c
i
a
l
i
n
te
ll
ig
e
n
c
e
.
He
h
a
s
d
o
n
e
h
i
s
M
a
s
ter
’
s
De
g
r
e
e
i
n
C
o
m
p
u
ter
s
f
r
o
m
th
e
S
a
i
n
t
P
e
ter
s
b
u
r
g
S
tate
U
n
iv
e
r
s
i
ty
o
f
I
n
f
o
r
m
a
ti
o
n
T
e
c
h
n
o
l
o
g
i
e
s
,
M
e
c
h
a
n
i
c
s
a
n
d
Op
ti
c
s
,
S
t.
P
e
ter
s
b
u
r
g
,
i
n
1
9
9
6
a
n
d
M
B
A
f
r
o
m
M
a
n
i
p
a
l
Ac
a
d
e
m
y
o
f
Hi
g
h
e
r
E
d
u
c
a
ti
o
n
,
M
a
n
i
p
a
l
,
2
0
0
4
.
He
i
s
c
u
r
r
e
n
t
l
y
p
a
r
t
ti
m
e
P
h
D
S
c
h
o
l
a
r
f
r
o
m
I
n
d
u
s
tr
y
i
n
Ga
lg
o
ti
a
s
U
n
i
v
e
r
s
i
ty
r
e
s
e
a
r
c
h
i
n
g
i
n
a
r
e
a
o
f
i
n
ter
n
e
t
o
f
th
i
n
g
s
a
n
d
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
.
H
e
h
a
s
wr
i
tt
e
n
a
c
o
r
p
o
r
a
te
p
a
p
e
r
o
n
De
li
v
e
r
i
n
g
I
n
ter
n
e
t
o
f
th
i
n
g
s
a
s
a
s
e
r
v
i
c
e
a
n
d
c
u
r
r
e
n
tl
y
r
e
s
e
a
r
c
h
i
n
g
o
n
a
r
e
a
s
f
o
r
o
p
ti
m
i
z
i
n
g
c
l
u
s
ter
i
n
g
a
n
d
d
a
ta
a
g
g
r
e
g
a
ti
o
n
i
n
I
o
T
e
c
o
s
y
s
tem
.
He
h
a
s
b
e
e
n
a
wa
r
d
e
d
f
o
r
h
i
s
i
n
n
o
v
a
ti
o
n
i
d
e
a
s
a
n
d
e
x
e
m
p
l
a
r
y
r
e
s
e
a
r
c
h
th
i
n
k
i
n
g
wh
il
e
wo
r
k
i
n
g
i
n
th
e
c
o
r
p
o
r
a
te
I
n
d
u
s
tr
y
a
s
p
a
r
t
o
f
h
i
s
s
ti
n
t
wi
th
I
n
f
o
s
y
s
T
e
c
h
n
o
l
o
g
i
e
s
li
m
i
ted
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
i
l:
v
k
p
h
d
s
c
h
o
l
a
r
@g
m
a
il
.
c
o
m
.
Dr
.
Abd
u
l
Ale
e
m
i
s
a
P
r
o
f
e
s
s
o
r
a
n
d
He
a
d
o
f
C
o
m
p
u
ti
n
g
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
De
p
a
r
tm
e
n
t,
a
t
Ga
l
g
o
ti
a
s
U
n
iv
e
r
s
i
ty
.
He
h
a
s
d
o
n
e
h
i
s
B
.
T
e
c
h
.
(
C
S
E
)
f
r
o
m
U
.
P
.
T
e
c
h
n
i
c
a
l
U
n
iv
e
r
s
i
ty
,
L
u
c
k
n
o
w.
He
h
a
s
d
o
n
e
h
i
s
M
.
T
e
c
h
.
(
S
o
f
twa
r
e
E
n
g
g
.
)
f
r
o
m
M
NN
I
T
A
ll
a
h
a
b
a
d
.
He
wa
s
b
e
s
to
we
d
Go
l
d
M
e
d
a
l
f
o
r
b
e
i
n
g
th
e
to
p
p
e
r
o
f
h
i
s
b
a
tch
i
n
th
e
m
a
s
ter
'
s
p
r
o
g
r
a
m
.
He
h
a
s
o
b
tai
n
e
d
P
h
.
D.
(
C
S
E
)
f
r
o
m
M
NN
I
T
A
ll
a
h
a
b
a
d
i
n
th
e
a
r
e
a
o
f
Da
ta
M
i
n
i
n
g
,
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
a
n
d
E
d
u
c
a
ti
o
n
a
l
S
y
s
tem
s
.
He
h
a
s
b
e
e
n
t
h
r
i
c
e
a
wa
r
d
e
d
a
s
B
e
s
t/
E
x
e
m
p
l
a
r
y
F
a
c
u
l
ty
,
f
i
v
e
t
i
m
e
s
c
o
n
f
e
r
r
e
d
th
e
B
e
s
t
P
a
p
e
r
a
wa
r
d
,
a
n
d
two
ti
m
e
s
a
wa
r
d
e
d
f
o
r
c
o
d
i
n
g
E
x
c
e
ll
e
n
c
e
.
He
p
o
s
s
e
s
s
e
s
13
+
y
e
a
r
s
o
f
b
l
e
n
d
e
d
e
x
p
e
r
i
e
n
c
e
i
n
c
l
u
d
i
n
g
6
+
y
e
a
r
s
i
n
to
p
s
o
f
twa
r
e
i
n
d
u
s
tr
i
e
s
a
n
d
7
+
y
e
a
r
s
i
n
p
r
e
m
i
u
m
e
d
u
c
a
ti
o
n
a
l
i
n
s
ti
tu
ti
o
n
s
a
s
a
v
a
l
u
e
-
a
d
d
e
d
a
c
a
d
e
m
i
c
i
a
n
,
s
o
f
twa
r
e
d
e
v
e
l
o
p
e
r
,
l
e
a
d
e
n
g
i
n
e
e
r
,
r
e
s
e
a
r
c
h
e
r
,
a
d
m
i
n
i
s
tr
a
to
r
,
a
n
d
i
n
te
ll
e
c
tu
a
l
p
r
o
f
e
s
s
i
o
n
a
l.
Hi
s
r
e
s
e
a
r
c
h
i
n
ter
e
s
ts
i
n
c
l
u
d
e
d
i
g
i
ta
l
i
m
a
g
e
p
r
o
c
e
s
s
i
n
g
,
o
p
ti
m
i
z
a
t
i
o
n
tec
h
n
i
q
u
e
s
,
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
a
r
ti
f
i
c
i
a
l
i
n
telli
g
e
n
c
e
.
He
h
a
s
p
u
b
l
i
s
h
e
d
m
o
r
e
th
a
n
3
0
r
e
s
e
a
r
c
h
p
a
p
e
r
s
i
n
r
e
p
u
ted
c
o
n
f
e
r
e
n
c
e
s
a
n
d
jo
u
r
n
a
l
s
o
f
d
i
s
ti
n
g
u
i
s
h
e
d
i
n
d
e
x
i
n
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
i
l:
e
r
.
a
lee
m
@g
m
a
il
.
c
o
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.