I
nte
rna
t
io
na
l J
o
urna
l o
f
I
nfo
rm
a
t
ics a
nd
Co
m
m
un
ica
t
io
n T
ec
hn
o
lo
g
y
(
I
J
-
I
CT
)
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
,
p
p
.
1
1
6
7
~
1
1
7
8
I
SS
N:
2252
-
8
7
7
6
,
DOI
:
1
0
.
1
1
5
9
1
/iji
ct
.
v
1
5
i
3
.
pp
1
1
6
7
-
1
1
7
8
1167
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ict.
ia
esco
r
e.
co
m
A pilot r
ev
iew of
t
he
G
o
o
g
le
clus
ter wo
rklo
a
d t
ra
ce
2
0
1
9
,
metho
do
lo
g
y
and its alterna
tives
:
a
na
ly
sis
of worklo
a
d in
la
rg
e sca
le data
c
entres
Ak
a
s
h P
a
t
el
1
,
Am
it
Na
y
a
k
2
,
K
hu
s
hi P
a
t
el
3
,
Ana
nd
P
a
t
el
4
1
D
e
p
a
r
t
me
n
t
o
f
I
n
f
o
r
mat
i
o
n
Te
c
h
n
o
l
o
g
y
,
D
e
v
a
n
g
P
a
t
e
l
I
n
st
i
t
u
t
e
o
f
A
d
v
a
n
c
e
Te
c
h
n
o
l
o
g
y
a
n
d
R
e
s
e
a
r
c
h
(
D
EPS
TA
R
)
,
F
a
c
u
l
t
y
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
E
n
g
i
n
e
e
r
i
n
g
(
F
TE),
C
h
a
r
o
t
a
r
U
n
i
v
e
r
si
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
(
C
H
A
R
U
S
A
T)
,
A
n
a
n
d
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
a
n
d
E
n
g
i
n
e
e
r
i
n
g
,
D
e
v
a
n
g
P
a
t
e
l
I
n
st
i
t
u
t
e
o
f
A
d
v
a
n
c
e
T
e
c
h
n
o
l
o
g
y
a
n
d
R
e
s
e
a
r
c
h
(
D
EPS
TA
R
)
,
F
a
c
u
l
t
y
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
E
n
g
i
n
e
e
r
i
n
g
(
F
TE),
C
h
a
r
o
t
a
r
U
n
i
v
e
r
si
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
(
C
H
A
R
U
S
A
T)
,
A
n
a
n
d
,
I
n
d
i
a
3
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
D
e
v
a
n
g
P
a
t
e
l
I
n
st
i
t
u
t
e
o
f
A
d
v
a
n
c
e
Te
c
h
n
o
l
o
g
y
a
n
d
R
e
se
a
r
c
h
(
D
EPS
TA
R
)
,
F
a
c
u
l
t
y
o
f
T
e
c
h
n
o
l
o
g
y
a
n
d
E
n
g
i
n
e
e
r
i
n
g
(
F
TE),
C
h
a
r
o
t
a
r
U
n
i
v
e
r
si
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
(
C
H
A
R
U
S
A
T)
,
A
n
a
n
d
,
I
n
d
i
a
4
I
n
f
o
r
mat
i
o
n
T
e
c
h
n
o
l
o
g
y
D
e
p
a
r
t
me
n
t
,
D
h
a
r
ms
i
n
h
D
e
sa
i
I
n
st
i
t
u
t
e
o
f
Te
c
h
n
o
l
o
g
y
(
D
D
I
T)
,
F
a
c
u
l
t
y
o
f
T
e
c
h
n
o
l
o
g
y
,
D
h
a
r
msi
n
h
D
e
s
a
i
U
n
i
v
e
r
si
t
y
(
D
D
U
)
,
N
a
d
i
a
d
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
5
,
2
0
2
5
R
ev
is
ed
Ap
r
1
4
,
2
0
2
6
Acc
ep
ted
Ma
y
1
7
,
2
0
2
6
Clo
u
d
d
a
ta
c
e
n
tres
re
q
u
ire
sh
a
re
d
se
rv
ice
s
th
a
t
a
re
h
ig
h
ly
a
v
a
il
a
b
le,
e
las
ti
c
c
a
p
a
c
it
y
,
m
a
n
a
g
e
d
o
p
e
ra
ti
o
n
s,
a
n
d
ro
b
u
st
re
c
o
v
e
r
y
c
a
p
a
b
il
it
ies
t
o
fa
c
il
it
a
te
th
e
n
e
x
t
g
e
n
e
ra
ti
o
n
o
f
e
fficie
n
t,
re
li
a
b
le,
a
n
d
d
i
v
e
rse
c
o
n
n
e
c
ted
c
o
m
p
u
ti
n
g
e
n
v
iro
n
m
e
n
ts.
Ne
v
e
rth
e
les
s,
lar
g
e
-
sc
a
le
c
lo
u
d
in
fra
stru
c
t
u
re
s
c
o
n
ti
n
u
e
to
fa
il
re
g
u
larly
d
e
s
p
it
e
t
h
e
ir
a
v
a
il
a
b
il
i
t
y
,
sc
a
lab
il
it
y
,
a
n
d
c
o
st
e
fficie
n
c
y
,
p
r
ima
ril
y
d
u
e
t
o
l
o
w
re
so
u
rc
e
u
ti
li
sa
ti
o
n
a
n
d
i
n
a
d
e
q
u
a
te
e
a
rly
-
sta
g
e
fa
il
u
re
m
a
n
a
g
e
m
e
n
t.
Th
e
k
e
y
t
o
e
ffe
c
ti
v
e
re
so
u
rc
e
m
a
n
a
g
e
m
e
n
t
a
n
d
m
in
imiz
in
g
fa
il
u
re
s
in
s
u
c
h
se
tt
i
n
g
s
is
u
n
d
e
rs
tan
d
i
n
g
t
h
e
n
a
t
u
re
o
f
th
e
w
o
rk
l
o
a
d
a
n
d
it
s
fa
il
u
re
m
o
d
e
s.
T
h
e
c
u
rre
n
t
re
v
ie
w
c
o
n
sid
e
rs
t
h
e
G
o
o
g
le
c
l
u
ste
r
wo
rk
l
o
a
d
trac
e
2
0
1
9
to
in
v
e
stig
a
te
wo
r
k
lo
a
d
a
n
d
fa
il
u
re
p
a
tt
e
rn
s
a
n
d
to
g
e
n
e
ra
li
se
th
e
re
su
lt
s
o
f
2
4
a
rti
c
les
.
Th
e
a
n
a
l
y
sis
is
a
lso
c
o
m
p
a
re
d
wit
h
o
t
h
e
r
m
a
jo
r
d
a
tas
e
ts,
su
c
h
a
s
M
icr
o
so
ft
Az
u
r
e
Trac
e
,
Ten
c
e
n
t
Trac
e
,
a
n
d
Ali
b
a
b
a
Trac
e
.
Th
e
p
a
p
e
r
e
sta
b
li
sh
e
s
th
e
re
lev
a
n
c
e
o
f
G
o
o
g
le
c
lu
ste
r
trac
e
s,
d
e
s
c
rib
e
s
th
e
k
e
y
c
o
n
ten
ts o
f
t
h
e
2
0
1
9
d
a
tas
e
t,
a
n
d
c
o
n
tras
ts p
ri
o
r
l
it
e
ra
tu
re
wit
h
re
sp
e
c
t
to
re
se
a
rc
h
o
b
jec
ti
v
e
s,
trac
e
d
a
tas
e
ts,
sig
n
ifi
c
a
n
t
re
su
lt
s,
a
n
d
li
m
it
a
ti
o
n
s.
M
o
re
o
v
e
r,
it
b
riefl
y
d
e
sc
rib
e
s
m
e
th
o
d
s
fo
r
a
n
a
ly
z
i
n
g
a
n
d
m
o
d
e
li
n
g
c
l
u
ste
r
trac
e
s
a
n
d
id
e
n
ti
fies
g
a
p
s
i
n
t
h
e
r
e
se
a
rc
h
th
a
t
sh
o
u
l
d
b
e
a
d
d
re
ss
e
d
t
o
a
d
v
a
n
c
e
th
e
stu
d
y
o
f
c
lu
ste
r
trac
e
s.
K
ey
w
o
r
d
s
:
C
lo
u
d
co
m
p
u
tin
g
Data
ce
n
ter
s
Dee
p
lear
n
in
g
Go
o
g
le
clu
s
ter
tr
ac
e
Ma
ch
in
e
lear
n
in
g
W
o
r
k
lo
ad
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Am
it Na
y
ak
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
Scie
n
ce
an
d
E
n
g
in
ee
r
in
g
Dev
an
g
Patel
I
n
s
titu
te
o
f
Ad
v
an
ce
T
ec
h
n
o
lo
g
y
an
d
R
esear
c
h
(
DE
PS
T
AR
)
Facu
lty
o
f
T
ec
h
n
o
lo
g
y
an
d
E
n
g
in
ee
r
in
g
(
FTE
)
C
h
ar
o
tar
Un
iv
er
s
ity
o
f
Scien
c
e
an
d
T
ec
h
n
o
lo
g
y
(
C
HARUS
AT
)
C
h
an
g
a
3
8
8
4
2
1
,
An
an
d
,
I
n
d
ia
E
m
ail: A
m
itn
ay
ak
.
it@
ch
ar
u
s
at.
ac
.
in
1.
I
NT
RO
D
UCT
I
O
N
I
n
co
m
m
er
cial
s
ettin
g
s
,
la
r
g
e
-
s
ca
le
d
ata
ce
n
ter
s
h
av
e
b
ec
o
m
e
th
e
s
tan
d
ar
d
s
o
lu
tio
n
f
o
r
a
v
ast
r
an
g
e
o
f
ap
p
licatio
n
s
.
T
h
is
is
m
ain
ly
attr
ib
u
tab
le
to
th
e
f
ac
t
t
h
at
th
ese
ce
n
ter
s
o
f
f
er
s
ig
n
if
ican
t
p
o
ten
tial
f
o
r
av
ailab
ilit
y
,
s
ca
lab
ilit
y
,
an
d
c
o
s
t
-
ef
f
ec
tiv
e
o
p
e
r
atio
n
al
ex
p
e
n
s
es.
C
lo
u
d
co
m
p
u
tin
g
s
y
s
tem
s
f
ail
d
u
e
to
th
eir
co
m
p
lex
ity
,
h
eter
o
g
en
eity
,
o
p
en
n
ess
,
an
d
d
is
p
er
s
io
n
.
T
h
ese
f
ailu
r
es
ca
n
d
eg
r
ad
e
clo
u
d
s
y
s
tem
d
ep
en
d
a
b
ilit
y
an
d
co
s
t
u
s
er
s
an
d
p
r
o
v
id
er
s
a
lo
t
o
f
m
o
n
ey
.
C
lo
u
d
p
r
o
v
i
d
er
s
m
ay
in
cu
r
s
u
b
s
tan
tial
lo
s
s
es
f
r
o
m
t
h
ese
er
r
o
r
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
1
1
6
7
-
1
1
7
8
1168
T
h
u
s
,
d
ata
a
n
aly
s
is
o
n
r
ea
l
-
w
o
r
ld
s
o
u
r
ce
s
is
ess
en
tial
to
s
o
lv
in
g
th
e
p
r
o
b
lem
s
f
ac
in
g
lar
g
e
-
s
ca
le
clo
u
d
d
ata
ce
n
ter
s
.
T
h
is
p
ap
er
p
r
esen
ts
an
an
aly
s
is
an
d
s
u
r
v
ey
o
f
th
e
2
0
1
9
Go
o
g
le
C
lu
s
ter
tr
ac
e
s
d
atase
t
[
1
]
,
wh
ich
Go
o
g
le
p
r
o
v
id
es.
I
t
in
clu
d
es
2
.
4
ter
ab
y
tes
o
f
wo
r
k
lo
ad
tr
a
ce
s
co
llected
f
r
o
m
eig
h
t
clu
s
ter
s
th
r
o
u
g
h
o
u
t
Ma
y
2
0
1
9
.
T
h
is
d
ataset
h
as
e
n
ab
le
d
a
co
m
p
a
r
is
o
n
o
f
th
e
d
ev
elo
p
m
en
t
o
f
s
ch
ed
u
lin
g
d
ec
is
io
n
s
m
ad
e
b
y
clu
s
ter
m
an
ag
er
s
,
o
p
e
n
in
g
u
p
n
o
v
el
o
p
p
o
r
tu
n
ities
.
T
h
e
n
ew
d
at
aset
co
v
er
s
allo
c
s
ets,
b
est
-
ef
f
o
r
t
b
atch
s
ch
ed
u
ler
,
v
er
tical
s
ca
lab
ilit
y
,
an
d
task
d
ep
en
d
en
cies.
T
h
is
in
f
o
r
m
atio
n
aid
s
in
jo
b
s
ch
ed
u
lin
g
e
x
p
lan
atio
n
s
.
Un
lik
e
th
e
2
0
1
1
tr
ac
e
d
ataset
[
2
]
,
th
e
2
0
1
9
d
ataset
allo
ws
u
s
to
ex
te
n
d
o
u
r
h
y
p
o
th
esis
to
eig
h
t
cl
u
s
ter
s
ac
r
o
s
s
th
r
ee
co
n
tin
en
ts
an
d
t
o
ex
am
in
e
n
o
n
co
n
f
o
r
m
is
t te
n
d
en
cies [
3
]
.
A
wid
e
r
an
g
e
o
f
r
esear
c
h
ar
t
icles
h
av
e
b
ee
n
p
u
b
lis
h
ed
t
o
d
ate
o
n
th
e
Go
o
g
le
c
lo
u
d
t
r
ac
es
2
0
1
9
d
ataset.
Ad
d
itio
n
ally
,
s
o
m
e
o
f
th
ese
ar
ticles
h
av
e
co
m
p
ar
e
d
th
is
d
ataset
with
o
th
er
p
u
b
licly
av
ailab
le
clo
u
d
d
atasets
.
W
h
en
it c
o
m
es to
r
esear
ch
p
u
b
licatio
n
s
,
h
o
we
v
er
,
t
h
er
e
is
a
lack
o
f
h
ig
h
-
q
u
ality
a
r
ticles.
I
n
clo
u
d
c
o
m
p
u
tin
g
r
esear
ch
,
s
ev
er
al
r
ea
l
-
wo
r
ld
d
ata
ce
n
te
r
tr
ac
e
d
atasets
h
av
e
b
ee
n
r
elea
s
ed
,
alo
n
g
with
th
eir
r
esp
ec
tiv
e
p
er
f
o
r
m
a
n
ce
s
ta
n
d
ar
d
s
.
2
0
1
1
f
ea
tu
r
ed
t
h
e
r
elea
s
e
o
f
Go
o
g
le
clu
s
ter
tr
ac
e
[
2
]
,
wh
ic
h
was
co
n
s
id
er
ed
t
o
b
e
o
n
e
o
f
th
e
e
ar
lies
t
an
d
m
o
s
t
s
ig
n
if
ican
t
d
atasets
.
Fu
r
th
er
m
o
r
e,
th
e
Alib
ab
a
C
lu
s
ter
Data
s
et
[
4
]
is
an
o
th
er
d
ataset
r
ec
en
t
ly
m
ad
e
av
ailab
le.
T
h
e
p
u
r
p
o
s
e
o
f
t
h
is
r
eso
u
r
ce
-
c
o
n
s
u
m
p
tio
n
d
ataset
is
to
en
h
an
ce
u
tili
za
tio
n
an
d
r
ed
u
c
e
in
f
r
astru
ctu
r
e
c
o
s
ts
.
I
t
co
m
e
s
f
r
o
m
th
e
p
r
o
d
u
ctio
n
cl
u
s
ter
o
f
Alib
ab
a,
w
h
ich
o
p
er
ates
b
o
th
o
n
lin
e
s
er
v
ices
an
d
b
atch
p
r
o
ce
s
s
es.
On
t
h
e
o
th
er
h
a
n
d
,
th
e
d
iv
er
s
e
r
eq
u
ir
e
m
en
ts
o
f
wo
r
k
l
o
ad
s
n
ec
ess
itate
th
e
d
ev
elo
p
m
en
t
o
f
n
o
v
el
s
ch
e
d
u
lin
g
alg
o
r
ith
m
s
to
m
an
ag
e
c
o
-
lo
ca
ted
h
eter
o
g
en
eo
u
s
ap
p
licatio
n
g
r
o
u
p
s
ef
f
ec
tiv
ely
[
5
]
.
T
h
is
p
ap
er
co
n
s
is
ts
o
f
th
e
f
o
llo
win
g
s
ec
tio
n
s
.
Sectio
n
1
is
an
I
n
tr
o
d
u
ctio
n
.
Sectio
n
2
d
e
s
cr
ib
es
th
e
s
ig
n
if
ican
ce
o
f
Go
o
g
le
clu
s
ter
tr
ac
e
s
.
Sectio
n
3
d
is
cu
s
s
es
th
e
Go
o
g
le
clu
s
ter
tr
ac
e
2
0
1
9
d
ataset.
Sectio
n
4
p
r
esen
ts
a
co
m
p
ar
ativ
e
an
al
y
s
is
u
s
in
g
alter
n
ativ
e
d
atasets
.
Sectio
n
5
d
escr
ib
es
m
eth
o
d
s
f
o
r
an
aly
zi
n
g
an
d
m
o
d
elin
g
clu
s
ter
tr
ac
e
s
.
Sectio
n
6
d
escr
ib
es
r
esear
ch
c
h
allen
g
es
an
d
o
p
en
ar
ea
s
.
T
h
e
f
in
al
s
ec
tio
n
,
7
,
p
r
esen
ts
th
e
co
n
clu
s
io
n
.
R
esear
ch
o
b
jectiv
es:
−
Pre
s
en
ts
a
s
y
s
tem
atic
liter
atu
r
e
r
ev
iew
o
f
29
ar
ticles
th
at
d
e
p
lo
y
th
e
Go
o
g
le
c
lu
s
ter
wo
r
k
l
o
ad
tr
ac
e
2
0
1
9
d
ataset,
wh
ich
s
u
m
m
a
r
izes
e
v
id
en
ce
o
n
wo
r
k
lo
a
d
as
wel
l
as
f
ailu
r
e
b
eh
a
v
io
u
r
in
r
e
al
-
wo
r
ld
clo
u
d
clu
s
ter
s
.
−
Giv
es
a
co
m
p
ar
ativ
e
s
tu
d
y
o
f
th
e
Go
o
g
le
cl
u
s
ter
tr
ac
e
s
with
th
e
MS
Azu
r
e,
T
e
n
ce
n
t
a
n
d
Alib
ab
a
tr
ac
es
in
clu
d
in
g
t
h
eir
s
im
ilar
ities
,
d
if
f
er
en
ce
s
an
d
t
h
eir
co
m
p
lem
en
tar
y
ad
v
a
n
tag
es to
clo
u
d
r
esea
r
ch
.
−
E
x
p
lain
s
th
e
m
ea
n
in
g
a
n
d
co
n
ten
t
s
o
f
th
e
Go
o
g
le
c
lu
s
ter
d
at
aset
o
n
2
0
1
9
s
u
ch
as
wh
at
k
in
d
o
f
wo
r
k
l
o
ad
s
,
r
eso
u
r
ce
s
,
an
d
e
v
en
ts
ar
e
b
ein
g
r
ec
o
r
d
ed
,
a
n
d
h
o
w
th
ese
h
av
e
b
ee
n
ex
p
lo
ited
in
p
r
ev
io
u
s
liter
atu
r
e.
−
Dete
r
m
in
es
s
o
lu
tio
n
s
to
u
n
r
eso
lv
ed
is
s
u
es
an
d
u
n
d
e
r
in
v
esti
g
ated
f
i
n
d
in
g
s
in
th
e
c
u
r
r
e
n
t
b
o
d
y
o
f
liter
atu
r
e
on
clu
s
ter
tr
ac
e
an
d
o
u
tlin
es
f
u
tu
r
e
r
esear
ch
q
u
esti
o
n
s
in
clu
s
ter
tr
ac
e
a
n
aly
s
is
s
p
ec
if
ically
with
in
th
e
co
n
tex
t o
f
AI
-
an
d
GPU
-
b
ased
wo
r
k
lo
ad
co
n
s
tr
ain
t p
r
ed
ictio
n
.
2.
SI
G
NIF
I
CANC
E
O
F
G
O
O
G
L
E
C
L
US
T
E
R
T
RAC
E
S
T
h
ey
co
n
d
u
cte
d
an
in
-
d
ep
t
h
an
aly
s
is
o
f
th
e
2
0
1
9
Go
o
g
le
clu
s
ter
tr
ac
e
d
ataset,
co
m
p
r
is
in
g
wo
r
k
lo
ad
tr
ac
es
to
talin
g
2
.
4
g
ig
ab
y
tes
f
r
o
m
eig
h
t
clu
s
ter
s
lo
ca
ted
in
d
if
f
er
en
t
p
ar
ts
o
f
t
h
e
wo
r
ld
.
Go
o
g
le
’
s
p
r
o
d
u
ctio
n
clo
u
d
h
as
b
ee
n
task
ed
with
ex
am
i
n
in
g
th
e
ch
ar
a
cter
is
tics
o
f
ab
o
r
ted
o
r
f
ailed
j
o
b
s
an
d
c
o
r
r
elatin
g
th
em
with
s
ig
n
i
f
ican
t
f
ac
t
o
r
s
.
T
h
ese
f
ac
to
r
s
i
n
clu
d
e
r
eso
u
r
c
e
u
tili
za
tio
n
,
w
o
r
k
p
r
i
o
r
ity
,
s
ch
ed
u
lin
g
class
,
jo
b
d
u
r
atio
n
,
an
d
t
h
e
n
u
m
b
er
o
f
t
im
es
th
e
task
was
r
esu
b
m
itted
.
Acc
o
r
d
i
n
g
to
o
u
r
f
in
d
in
g
s
,
s
ev
er
al
s
ig
n
if
ican
t
f
ea
tu
r
es
o
f
f
ailed
task
s
co
n
tr
ib
u
ted
to
th
e
wo
r
k
’
s
f
ailu
r
e
an
d
co
u
ld
b
e
u
s
ed
to
d
ev
elo
p
an
ea
r
ly
f
ailu
r
e
p
r
ed
ictio
n
s
y
s
tem
[
1
]
.
Sev
er
al
r
esear
ch
er
s
h
av
e
u
s
ed
GAR
C
H
an
d
AR
I
MA
m
o
d
els
to
d
ev
el
o
p
tec
h
n
iq
u
es
f
o
r
p
r
e
d
ictin
g
th
e
r
eliab
ilit
y
a
n
d
r
esp
o
n
s
e
tim
es
o
f
o
n
lin
e
s
er
v
ices.
I
n
ad
d
itio
n
,
s
ev
er
al
p
r
io
r
s
tu
d
ies
f
o
u
n
d
th
at
v
ir
tu
al
m
ac
h
in
es
ex
h
ib
ited
r
elativ
ely
co
n
s
is
ten
t
wo
r
k
lo
ad
p
atter
n
s
.
Usi
n
g
Hid
d
en
Ma
r
k
o
v
Mo
d
elin
g
,
th
ey
d
ev
elo
p
ed
a
m
eth
o
d
to
d
escr
ib
e
an
d
p
r
e
d
ict
v
ir
tu
al
m
ac
h
in
e
wo
r
k
lo
a
d
p
atter
n
s
.
T
h
ey
d
ev
elo
p
e
d
s
ev
er
al
a
p
p
r
o
ac
h
es,
an
d
th
is
s
tr
ateg
y
was
am
o
n
g
th
em
.
T
h
e
u
s
e
o
f
s
tatis
tical
tech
n
iq
u
es
b
ased
o
n
Go
o
g
le
C
lu
s
ter
tr
ac
e
s
h
as
led
to
co
m
p
ar
is
o
n
s
o
f
Go
o
g
le
d
at
a
ce
n
ter
s
with
g
r
id
o
r
h
ig
h
-
p
er
f
o
r
m
a
n
ce
co
m
p
u
tin
g
(
HPC
)
s
y
s
tem
s
[
6
]
.
T
h
is
co
m
p
ar
is
o
n
is
m
ad
e
ev
en
th
o
u
g
h
Go
o
g
le
d
ata
ce
n
ter
s
ar
e
class
if
ied
as c
lo
u
d
s
tag
es.
A
th
o
r
o
u
g
h
m
u
lti
-
v
iew
ass
es
s
m
en
t
s
tr
ateg
y
f
o
r
d
u
al
-
clo
u
d
wo
r
k
lo
ad
s
was
d
escr
ib
ed
,
an
d
a
ca
s
e
s
tu
d
y
o
n
Go
o
g
le
C
lu
s
ter
T
r
ac
i
n
g
was
u
s
ed
to
illu
s
tr
ate
th
e
c
o
n
ce
p
t
[
4
]
.
Sev
er
al
s
t
u
d
ies
u
s
ed
th
e
wo
r
k
l
o
ad
o
f
th
e
Go
o
g
le
clu
s
ter
to
d
eter
m
in
e
th
e
d
is
tr
ib
u
tio
n
o
f
clo
u
d
s
er
v
er
s
’
m
ain
ten
an
ce
tim
e
an
d
tim
e
-
to
-
f
ailu
r
e
[
6
]
.
T
h
ese
an
aly
s
es we
r
e
co
n
d
u
cted
to
d
eter
m
i
n
e
th
e
lik
elih
o
o
d
o
f
f
ailu
r
e.
T
h
e
Go
o
g
le
C
lu
s
ter
tr
ac
e
Data
s
et
co
m
p
r
is
es
eig
h
t
B
o
r
g
ce
l
ls
,
ea
ch
lo
ca
ted
in
a
d
if
f
er
en
t
tim
e
zo
n
e,
s
u
ch
as
New
Yo
r
k
,
Helsin
k
i,
an
d
Sin
g
a
p
o
r
e.
Ma
ch
in
e
tab
le
s
,
C
o
llectio
n
tab
les,
an
d
I
n
s
ta
n
ce
tab
les
ar
e
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
A
p
ilo
t revie
w
o
f th
e
G
o
o
g
le
cl
u
s
ter w
o
r
klo
a
d
tr
a
ce
2
0
1
9
,
me
th
o
d
o
lo
g
y
a
n
d
its
a
lter
n
a
tives
…
(
A
ka
s
h
P
a
tel
)
1169
th
r
ee
p
r
im
ar
y
ty
p
es
o
f
tab
le
s
th
at
ar
e
in
clu
d
ed
in
ea
ch
B
o
r
g
ce
ll,
wh
ich
is
co
m
p
r
is
ed
o
f
f
iv
e
tab
les
.
T
o
b
e
m
o
r
e
ex
p
licit,
Ma
ch
i
n
es
ar
e
d
i
v
id
ed
in
to
two
ta
b
les:
th
e
Ma
ch
in
eE
v
en
ts
tab
le
d
escr
ib
es
ev
en
ts
ass
o
ciate
d
with
m
ac
h
in
es,
an
d
th
e
Ma
ch
in
eAttr
ib
u
tes
ta
b
le
p
r
o
v
id
es
in
f
o
r
m
atio
n
o
n
m
ac
h
in
e
f
ea
tu
r
es.
C
o
llectio
n
s
ar
e
wh
at
Go
o
g
le
r
ef
er
s
to
as
jo
b
s
an
d
all
o
c
s
ets,
in
s
tan
ce
s
ar
e
wh
at
th
ey
c
all
task
s
an
d
allo
c
in
s
tan
ce
s
,
an
d
item
s
ar
e
wh
at
th
ey
u
s
e
to
r
ef
er
to
co
llectio
n
s
o
r
in
s
tan
ce
s
.
T
h
e
C
o
l
lectio
n
E
v
en
ts
an
d
I
n
s
tan
ce
E
v
en
ts
tab
les,
r
esp
ec
tiv
ely
,
r
ec
o
r
d
th
e
life
cy
cle
o
f
co
llectio
n
s
an
d
in
s
tan
ce
s
[
7
]
.
T
h
e
2
0
1
9
tr
ac
es
i
n
clu
d
e
o
n
l
y
r
eso
u
r
ce
r
eq
u
ests
an
d
u
tili
za
tio
n
;
th
e
y
d
o
n
o
t
ca
p
tu
r
e
e
n
d
u
s
er
s
,
d
ata,
o
r
ac
ce
s
s
p
atter
n
s
to
s
to
r
ag
e
s
y
s
tem
s
an
d
o
th
e
r
s
er
v
ices.
E
ac
h
r
ec
o
r
d
c
o
n
tain
s
a
m
icr
o
s
ec
o
n
d
tim
estam
p
r
elativ
e
to
a
r
ef
er
e
n
ce
p
o
in
t
6
0
0
s
ec
o
n
d
s
b
e
f
o
r
e
th
e
s
tar
t
o
f
t
h
e
tr
ac
e
p
er
io
d
.
T
h
e
tim
estam
p
s
ar
e
6
4
-
b
it
in
teg
er
s
.
Fo
r
e
x
am
p
le,
a
2
0
-
s
ec
o
n
d
e
v
en
t
o
cc
u
r
r
i
n
g
6
2
0
s
ec
o
n
d
s
af
ter
th
e
tr
ac
e
b
e
g
in
s
wo
u
ld
h
av
e
a
tim
estam
p
o
f
6
2
0
s
ec
o
n
d
s
.
T
wo
ex
ce
p
tio
n
al
tim
estam
p
v
alu
es
ar
e
av
ailab
le
f
o
r
ev
en
ts
b
ey
o
n
d
th
e
t
r
ac
e
win
d
o
w:
0
f
o
r
p
r
e
-
tr
ac
e
ev
e
n
ts
an
d
2
6
3
−
1
(
MA
XI
NT
)
f
o
r
p
o
s
t
-
tr
ac
e
e
v
en
ts
.
L
im
itin
g
u
s
e
m
ea
s
u
r
em
en
t
tim
es
to
3
0
0
s
ec
o
n
d
s
,
o
f
f
s
et
r
u
les
ar
e
u
s
ed
to
d
etec
t
o
cc
u
r
r
en
ce
s
r
elativ
e
to
th
e
tr
ac
e
p
er
io
d
.
Me
asu
r
em
e
n
ts
ar
e
ac
cu
r
ate
to
th
e
n
ea
r
est
s
ec
o
n
d
b
u
t
p
r
esen
ted
in
m
icr
o
s
ec
o
n
d
s
f
o
r
clar
ity
.
T
h
ese
tim
estam
p
s
ar
e
g
en
er
ally
ac
cu
r
ate,
alth
o
u
g
h
s
m
all
v
ar
ia
tio
n
s
m
ay
o
cc
u
r
d
u
e
to
cl
o
ck
d
r
if
t
ac
r
o
s
s
clu
s
ter
n
o
d
es,
r
ath
er
th
an
s
ch
ed
u
lin
g
o
r
u
s
ag
e
co
n
ce
r
n
s
[
8
]
.
W
h
en
co
n
d
u
ctin
g
r
esear
ch
o
n
th
e
clo
u
d
,
it
is
ess
en
tial
to
en
s
u
r
e
th
e
av
ailab
ilit
y
o
f
em
p
ir
ic
al
d
atasets
th
at
d
em
o
n
s
tr
ate
th
e
e
x
ten
t
to
wh
ich
th
e
clo
u
d
m
ee
ts
cu
s
to
m
er
r
e
q
u
ir
em
en
ts
.
Fo
r
e
x
am
p
l
e,
n
u
m
er
o
u
s
p
u
b
lic
clo
u
d
p
latf
o
r
m
tr
ac
es
f
o
r
clu
s
t
er
tr
ac
e
an
aly
s
is
h
av
e
b
ee
n
m
ad
e
av
ailab
le.
T
h
ese
in
clu
d
e
t
h
e
Face
b
o
o
k
tr
ac
e,
th
e
T
ao
b
a
o
d
ataset,
th
e
Go
o
g
le
clu
s
ter
tr
ac
e
,
an
d
th
e
Al
ib
ab
a
tr
ac
e
2
0
1
8
.
I
n
No
v
e
m
b
er
2
0
1
1
,
Go
o
g
le
r
elea
s
ed
a
“
clu
s
ter
u
s
ag
e
tr
ac
e
d
ataset
”
.
T
h
er
e
is
a
r
em
ar
k
ab
l
e
am
o
u
n
t o
f
d
etail
in
th
is
tr
ac
e,
as
well
as
a
wid
e
v
ar
iety
o
f
task
s
.
T
h
e
p
o
s
itio
n
s
en
co
m
p
ass
a
wid
er
r
an
g
e
o
f
r
esp
o
n
s
ib
ilit
ies,
in
clu
d
in
g
p
r
o
v
id
i
n
g
o
n
lin
e
s
er
v
ices
s
u
ch
as
s
ea
r
ch
q
u
er
ies
an
d
o
th
er
u
s
es
o
f
Ma
p
R
ed
u
ce
m
eth
o
d
s
.
T
h
e
ter
m
“
wo
r
k
lo
ad
”
r
ef
e
r
s
to
th
e
n
u
m
b
er
o
f
jo
b
s
b
ein
g
r
ec
ei
v
e
d
,
th
e
task
s
ass
o
ciate
d
with
th
o
s
e
jo
b
s
th
at
u
s
er
s
h
av
e
s
u
b
m
itted
,
an
d
t
h
e
lo
a
d
a
p
ar
ticu
lar
s
y
s
tem
ex
p
er
ien
ce
s
wh
en
co
n
d
u
ctin
g
task
s
.
W
i
t
h
in
th
e
s
co
p
e
o
f
th
is
s
tu
d
y
,
th
e
tr
ac
e
d
ataset
is
ch
ar
ac
ter
ized
with
r
esp
ec
t
to
h
o
w
task
s
ar
e
p
r
o
ce
s
s
ed
an
d
t
er
m
in
ated
with
in
t
h
e
clu
s
ter
,
h
o
w
ter
m
in
ated
jo
b
s
ar
e
d
is
tr
ib
u
ted
,
a
n
d
h
o
w
s
ch
ed
u
lin
g
class
an
d
p
r
i
o
r
ity
in
f
l
u
en
ce
ter
m
in
ated
jo
b
s
[
9
]
,
[
1
0
]
.
T
o
ca
p
tu
r
e
lo
n
g
-
r
an
g
e
d
ep
e
n
d
en
cies
in
clo
u
d
wo
r
k
lo
a
d
s
,
th
is
an
aly
s
is
is
co
n
d
u
cted
u
s
in
g
wo
r
k
lo
ad
s
d
er
iv
ed
f
r
o
m
a
s
tan
d
ar
d
r
ea
l
-
wo
r
ld
d
ataset,
Go
o
g
le
C
lu
s
ter
tr
ac
e
.
T
h
e
r
esear
ch
s
h
o
ws
th
at
th
e
m
et
r
ics
u
n
d
er
an
aly
s
is
ar
e
h
ea
v
y
-
tailed
;
m
o
r
eo
v
er
,
a
m
ath
em
atica
l
f
o
r
m
is
d
er
iv
ed
t
o
ch
ar
ac
ter
ize
lo
n
g
-
r
an
g
e
d
e
p
en
d
e
n
cies
in
ag
g
r
eg
ate
wo
r
k
lo
ad
s
.
T
h
e
wo
r
k
lo
ad
s
in
th
e
Go
o
g
le
clu
s
ter
tr
ac
e
ar
e
an
aly
ze
d
in
p
r
io
r
wo
r
k
,
d
ep
e
n
d
in
g
o
n
th
e
ch
ar
ac
ter
is
tics
o
f
th
e
tas
k
s
an
d
t
h
e
r
eso
u
r
ce
s
u
s
ed
.
T
h
ey
o
b
s
er
v
ed
th
at
o
cc
u
p
atio
n
s
in
th
e
f
o
r
est
in
d
u
s
tr
y
ca
n
b
e
class
if
ied
in
to
t
h
r
ee
c
ateg
o
r
ies:
s
h
o
r
t,
m
e
d
iu
m
,
a
n
d
lo
n
g
task
s
.
T
h
is
class
if
icati
o
n
is
b
ased
o
n
th
e
r
eso
u
r
ce
-
in
ten
s
iv
e
n
atu
r
e
o
f
t
h
e
jo
b
s
.
Ad
d
itio
n
ally
,
th
e
y
wo
r
k
ed
o
n
clu
s
ter
in
g
wo
r
k
l
o
ad
p
atter
n
s
.
On
t
h
e
o
th
er
h
an
d
,
th
eir
r
esear
ch
d
o
e
s
n
o
t
f
o
cu
s
o
n
u
n
d
er
s
tan
d
in
g
th
e
n
atu
r
e
o
f
clo
u
d
wo
r
k
lo
ad
s
th
at
ar
e
d
ep
en
d
e
n
t
o
n
an
e
x
ten
d
ed
p
er
io
d
o
f
ti
m
e.
T
h
e
v
a
r
ied
n
atu
r
e
o
f
w
o
r
k
lo
ad
s
in
Go
o
g
le
clu
s
ter
tr
a
ce
is
th
e
s
u
b
ject
o
f
d
is
cu
s
s
io
n
in
an
o
th
er
p
iece
o
f
r
esear
ch
.
T
h
e
r
ep
o
r
t
o
b
s
er
v
es
th
at
th
e
wo
r
k
lo
ad
is
h
ig
h
l
y
d
y
n
am
ic,
m
ea
n
i
n
g
it
ch
an
g
es
o
v
er
tim
e
an
d
is
d
r
i
v
en
b
y
n
u
m
er
o
u
s
b
r
ief
ass
ig
n
m
en
ts
th
at
r
eq
u
ir
e
r
ap
id
s
c
h
ed
u
lin
g
d
ec
is
io
n
s
.
T
h
e
r
esear
ch
t
h
at
th
ey
h
av
e
d
o
n
e,
o
n
t
h
e
o
th
e
r
h
a
n
d
,
d
o
e
s
n
o
t
an
aly
ze
t
h
e
lo
n
g
-
r
an
g
e
r
elian
ce
th
at
clo
u
d
wo
r
k
lo
ad
s
h
a
v
e
[
1
1
]
.
A
b
ase
p
r
e
d
icto
r
is
a
ter
m
t
h
at
is
wid
ely
u
s
ed
to
r
ef
er
to
ea
c
h
o
f
th
e
n
u
m
e
r
o
u
s
p
r
ed
ictio
n
m
o
d
els
th
at
ar
e
u
tili
ze
d
in
an
en
s
em
b
le
m
eth
o
d
.
T
h
is
in
clu
d
es
th
e
u
s
e
o
f
m
u
ltip
le
p
r
ed
ic
tio
n
m
o
d
els
to
f
o
r
ec
ast
f
u
tu
r
e
o
u
tco
m
es.
I
n
Go
o
g
le
C
lu
s
ter
tr
ac
e
2
0
1
1
,
th
e
wo
r
k
lo
ad
is
an
aly
ze
d
b
y
p
ar
titi
o
n
in
g
it
in
to
th
r
ee
d
o
m
ai
n
s
:
p
o
s
itiv
e
d
o
m
ain
(
POS),
n
eg
a
tiv
e
d
o
m
ain
(
NE
G)
,
an
d
b
o
r
d
er
d
o
m
ain
.
A
th
r
ee
-
way
d
e
cisi
o
n
is
th
e
to
o
l
em
p
lo
y
ed
f
o
r
th
is
ty
p
e
o
f
wo
r
k
lo
ad
an
aly
s
is
.
E
MA
p
r
ed
ict
io
n
m
eth
o
d
o
lo
g
y
is
u
tili
ze
d
f
o
r
th
e
g
iv
en
s
tab
le
p
er
io
d
.
Fo
r
th
e
v
o
latile
p
er
io
d
,
th
e
Ho
lt
-
W
in
ter
s
p
r
ed
ictio
n
m
o
d
el
is
ap
p
lied
,
an
d
th
is
is
s
en
s
itiv
e
to
ju
m
p
p
atter
n
s
in
th
e
d
ata.
Fo
r
th
e
v
o
latilit
y
p
er
io
d
,
a
weig
h
ted
p
r
ed
i
ctio
n
tech
n
iq
u
e
f
o
r
m
s
t
h
e
co
r
e
p
r
ed
ictio
n
m
o
d
el
with
in
th
e
p
r
ed
ictiv
e
f
r
am
ewo
r
k
,
as
th
e
m
o
d
el
r
esp
o
n
s
ib
le
f
o
r
d
eter
m
in
in
g
th
e
r
elev
an
t
o
u
tc
o
m
es.
Ou
tco
m
e
weig
h
ti
n
g
is
u
s
ed
to
esti
m
ate
th
e
f
in
al
p
r
o
ject
ed
wo
r
k
lo
ad
b
ased
o
n
weig
h
ts
o
b
tain
ed
u
s
in
g
r
ec
ip
r
o
ca
l e
r
r
o
r
tech
n
iq
u
es.
Hen
ce
,
m
o
d
els with
h
ig
h
er
ac
c
u
r
ac
y
ar
e
g
iv
e
n
g
r
ea
ter
weig
h
t,
wh
er
ea
s
th
o
s
e
with
lo
w
ac
cu
r
ac
y
r
ec
eiv
e
m
in
im
al
weig
h
t [
1
2
]
.
3.
O
VE
RVI
E
W
O
F
G
O
O
G
L
E
CL
US
T
E
R
T
RA
CE
S 2
0
1
9
D
AT
AS
E
T
An
ex
ten
s
iv
e
d
ataset
th
at
is
m
ad
e
ac
ce
s
s
ib
le
to
th
e
g
en
e
r
al
p
u
b
lic
i
s
k
n
o
wn
as
Go
o
g
le
clu
s
ter
tr
ac
e
s
.
Mo
r
e
th
an
1
2
,
5
0
0
n
o
d
es
wer
e
u
s
ed
to
g
en
e
r
ate
th
e
d
ata
f
o
r
th
is
lar
g
e
d
ataset.
Du
r
i
n
g
th
e
p
er
i
o
d
f
r
o
m
Feb
r
u
ar
y
8
to
Ma
r
c
h
2
8
,
Go
o
g
le
’
s
clu
s
ter
tr
ac
e
s
co
m
p
r
is
e
a
p
p
r
o
x
im
ately
2
8
m
illi
o
n
task
s
an
d
6
7
2
,
0
7
4
j
o
b
s
.
E
ac
h
jo
b
h
as
u
n
iq
u
e
c
h
ar
ac
te
r
is
tics
b
ased
o
n
its
s
ch
ed
u
lin
g
ca
teg
o
r
y
,
p
r
io
r
ity
,
r
eso
u
r
ce
r
eq
u
ir
e
m
en
ts
,
an
d
r
eso
u
r
ce
u
tili
za
tio
n
.
T
h
e
f
o
ll
o
win
g
s
u
m
m
a
r
y
c
o
v
er
s
G
o
o
g
l
e
T
r
ac
es
’
k
e
y
s
tatis
tics
.
T
h
e
li
n
k
s
b
etwe
en
f
ailed
task
s
,
s
ch
ed
u
led
co
u
r
s
es,
an
d
r
eso
u
r
ce
d
em
an
d
s
ar
e
b
r
o
k
e
n
to
elim
in
ate
th
em
.
B
asic
in
f
o
r
m
atio
n
was d
er
iv
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
1
1
6
7
-
1
1
7
8
1170
b
y
a
n
aly
zin
g
f
ailu
r
e
p
atter
n
s
o
v
e
r
a
2
9
-
d
ay
p
e
r
io
d
.
T
h
e
Go
o
g
le
T
r
ac
e
o
f
2
9
d
ay
s
r
ev
ea
ls
s
ig
n
if
ican
t
d
if
f
er
en
ce
s
i
n
p
atter
n
s
o
f
f
ai
lu
r
es
o
f
jo
b
s
a
n
d
task
s
.
I
n
co
m
p
lete
ac
tiv
ities
in
Go
o
g
le
T
r
ac
e
wen
t
h
ig
h
ly
b
etwe
en
d
ay
s
2
a
n
d
1
0
,
wh
ic
h
s
h
o
ws s
ig
n
if
ican
t d
estru
ctio
n
o
f
p
ar
ticu
lar
d
ata
ce
n
tr
e
r
eso
u
r
ce
s
[
6
]
.
Go
o
g
le
h
as
o
n
ly
r
ec
en
tly
r
elea
s
ed
n
ew
tr
ac
e
d
ata
ap
p
lica
b
le
at
a
lar
g
e
s
ca
le.
C
o
m
p
ar
e
d
with
th
e
p
r
io
r
v
er
s
io
n
,
th
e
Ma
y
2
0
1
9
d
ataset
co
n
tain
s
eig
h
t
d
is
tin
ct
Go
o
g
le
co
m
p
u
ter
clu
s
ter
s
,
r
ef
er
r
ed
to
as
ce
lls
.
I
n
co
n
tr
ast
to
ea
r
lier
r
esear
ch
,
th
is
ap
p
r
o
ac
h
en
ab
les
in
ter
-
clu
s
ter
an
aly
s
is
,
wh
ich
was
p
r
ev
io
u
s
ly
d
if
f
icu
lt.
J
o
b
s
,
m
ac
h
in
e
ev
en
ts
,
an
d
o
t
h
er
u
tili
za
tio
n
d
ata
ar
e
in
clu
d
ed
in
th
e
Go
o
g
le
T
r
ac
e
2
0
1
9
d
ataset.
J
o
b
s
an
d
m
ac
h
in
e
ev
en
ts
co
n
tain
in
f
o
r
m
atio
n
o
n
j
o
b
an
d
allo
ca
tio
n
s
ch
ed
u
lin
g
,
as
well
as
o
n
m
ac
h
in
e
life
-
cy
cle
o
p
er
atin
g
p
r
o
ce
s
s
es.
Use
th
e
p
r
o
v
id
e
d
d
ata
to
co
m
p
u
te
th
e
n
o
r
m
alize
d
C
PU
an
d
m
em
o
r
y
u
s
ag
e
f
o
r
ea
ch
ac
tiv
ity
,
r
an
g
in
g
f
r
o
m
0
to
1
.
T
h
e
c
o
m
p
u
tatio
n
d
iv
id
es
ac
tu
al
u
s
ag
e
b
y
th
e
m
ax
im
u
m
n
u
m
b
er
o
f
Go
o
g
le
c
o
m
p
u
te
u
n
its
(
GC
Us)
o
r
m
em
o
r
y
ca
p
ac
ity
o
f
all
co
m
p
u
ter
s
in
a
ce
ll [
1
3
]
.
Go
o
g
le
r
elea
s
ed
its
th
ir
d
an
d
latest
d
ataset.
I
t
wa
s
r
ele
ased
in
ea
r
ly
2
0
2
0
an
d
r
ec
o
r
d
s
clo
u
d
r
eso
u
r
ce
u
s
ag
e
f
r
o
m
eig
h
t
clu
s
ter
s
,
o
n
e
o
f
wh
ich
h
ad
m
o
r
e
th
an
1
2
,
0
0
0
co
m
p
u
ter
s
in
Ma
y
2
0
1
9
.
T
h
is
d
ataset
f
o
cu
s
es
o
n
r
eso
u
r
ce
r
eq
u
ests
an
d
co
n
s
u
m
p
tio
n
an
d
d
o
es
n
o
t
in
clu
d
e
en
d
u
s
er
s
,
u
n
lik
e
th
e
2
0
1
1
d
ataset.
T
h
e
2
0
1
9
d
ataset
ad
d
s
task
-
r
eser
v
ed
s
h
ar
ed
r
eso
u
r
ce
s
,
C
PU
-
u
s
e
h
is
to
g
r
am
s
f
o
r
e
ac
h
f
iv
e
m
in
u
tes,
an
d
jo
b
-
p
ar
en
t
r
elatio
n
s
h
ip
s
f
o
r
m
aster
-
wo
r
k
er
r
elatio
n
s
h
ip
s
,
s
u
ch
as
Ma
p
R
ed
u
ce
jo
b
s
[
1
4
]
.
T
a
b
le
1
s
h
o
ws
a
co
m
p
ar
is
o
n
b
etwe
en
th
e
Go
o
g
le
clu
s
ter
tr
ac
e
2
0
1
1
an
d
th
e
G
o
o
g
le
clu
s
ter
tr
ac
e
2
0
1
9
d
atas
ets.
T
h
e
d
ataset
o
f
Go
o
g
le
c
lu
s
ter
2
0
1
9
ca
n
b
e
s
u
cc
ess
f
u
lly
a
p
p
lied
in
th
e
d
ev
el
o
p
m
en
t
o
f
th
e
p
r
ed
ictiv
e
m
o
d
els to
o
p
tim
ize
clu
s
ter
p
er
f
o
r
m
an
ce
a
n
d
r
eso
u
r
ce
u
tili
za
tio
n
s
in
ce
it in
clu
d
es a
b
u
n
d
a
n
t
m
etr
ics th
at
ca
n
b
e
u
s
ed
to
an
aly
ze
wo
r
k
l
o
ad
p
atter
n
s
an
d
r
eso
u
r
ce
r
eq
u
i
r
em
en
t
with
in
th
e
clo
u
d
e
n
v
ir
o
n
m
en
t.
T
h
ese
u
n
d
er
s
tan
d
i
n
g
s
f
ac
ilit
ate
m
o
r
e
s
o
p
h
is
ticated
m
eth
o
d
s
lik
e
r
e
in
f
o
r
ce
m
e
n
t
lear
n
in
g
(
R
L
)
,
in
wh
ich
f
r
am
ewo
r
k
s
lik
e
Q
-
lear
n
in
g
an
d
d
ee
p
Q
-
n
etwo
r
k
s
m
ay
o
p
tim
ize
th
e
r
eso
u
r
ce
allo
ca
tio
n
d
y
n
am
icall
y
b
y
b
alan
ci
n
g
t
h
e
wo
r
k
lo
ad
s
an
d
en
h
an
c
in
g
th
e
ef
f
icien
cy
o
f
th
e
s
y
s
tem
s
.
Als
o
,
p
r
ed
ictiv
e
s
y
s
tem
s
c
an
u
s
e
f
o
r
ec
asti
n
g
tech
n
iq
u
es
lik
e
NARX
an
d
AR
I
MA
to
p
r
ed
ict
f
u
tu
r
e
wo
r
k
lo
ad
s
u
s
in
g
p
ast
d
ata
to
e
n
ab
le
th
e
p
r
o
ac
ti
v
e
allo
ca
tio
n
o
f
r
eso
u
r
ce
s
,
m
in
i
m
ize
co
s
ts
,
an
d
ac
h
iev
e
h
ig
h
p
er
f
o
r
m
an
ce
.
Ma
ch
i
n
e
lear
n
in
g
s
o
lu
tio
n
s
ar
e
also
in
s
tr
u
m
en
tal
in
task
ass
ig
n
m
en
t
wh
e
r
e
ar
tific
ial
n
e
u
r
al
n
etwo
r
k
s
,
as
well
as,
en
s
em
b
le
alg
o
r
ith
m
s
ca
n
b
e
u
s
ed
to
d
eter
m
in
e
t
h
e
b
est
n
o
d
e
-
task
ass
ig
n
m
e
n
ts
r
esu
ltin
g
in
an
ef
f
icien
t
u
s
e
o
f
r
eso
u
r
ce
s
.
Nev
er
th
eless
,
r
eg
ar
d
less
o
f
i
ts
p
o
ten
tial,
th
e
m
o
d
el
ac
cu
r
ac
y
an
d
p
er
f
o
r
m
a
n
ce
m
a
y
b
e
in
f
lu
e
n
ce
d
b
y
ch
allen
g
es,
in
clu
d
i
n
g
d
ata
q
u
ality
p
r
o
b
lem
s
,
an
d
th
e
d
y
n
a
m
ic
an
d
u
n
p
r
ed
ictab
le
n
at
u
r
e
o
f
wo
r
k
lo
ad
s
,
a
n
d
to
m
ax
im
ize
th
e
b
e
n
ef
its
o
f
p
r
ed
ictiv
e
an
aly
tics
in
clu
s
ter
m
an
ag
em
en
t,
it
is
i
m
p
o
r
tan
t
to
m
itig
ate
th
ese
lim
itatio
n
s
[
1
5
]
-
[
1
7
]
.
T
ab
le
1
.
C
o
m
p
a
r
is
o
n
b
etwe
en
Go
o
g
le
C
lu
s
ter
tr
ac
e
2
0
1
1
(
v
2
)
an
d
2
0
1
9
(
v
3
)
d
atasets
V
e
r
si
o
n
2
(
c
e
l
l
)
V
e
r
si
o
n
3
(
t
o
t
a
l
)
V
e
r
si
o
n
3
(
c
e
l
l
a
v
g
)
To
t
a
l
e
v
e
n
t
s
3
7
7
8
0
3
0
9
8
3
3
3
8
7
2
9
I
n
i
t
i
a
l
m
a
c
h
i
n
e
s
1
2
4
7
7
9
1
9
0
2
1
1
4
8
7
N
e
w
A
D
D
1
0
6
4
6
5
3
5
8
2
R
E
M
O
V
E
8
9
5
7
1
5
3
7
7
9
1
9
2
2
2
U
P
D
A
TE
7
3
8
0
3
2
4
40
R
EJO
I
N
8
8
6
0
1
5
1
0
7
7
1
8
8
8
4
R
E
M
O
V
E
(
p
e
r
ma
n
e
n
t
l
y
)
98
2
7
0
2
3
3
8
R
e
p
a
i
r
p
r
o
b
a
b
i
l
i
t
y
0
.
9
8
9
1
7
0
.
9
8
2
4
3
-
4.
CO
M
P
ARA
T
I
V
E
ANA
L
YS
I
S WI
T
H
AL
T
E
RNA
T
I
V
E
DATAS
E
T
S
R
ec
en
t
s
tu
d
ies
o
n
clo
u
d
c
o
m
p
u
tin
g
an
d
wo
r
k
l
o
ad
tr
ac
e
a
n
al
y
s
is
h
av
e
f
o
cu
s
ed
o
n
im
p
r
o
v
i
n
g
r
eso
u
r
ce
u
tili
za
tio
n
,
s
ch
ed
u
lin
g
ef
f
ici
en
cy
,
wo
r
k
lo
ad
p
r
e
d
ictio
n
,
an
d
s
y
s
tem
r
eliab
ilit
y
in
lar
g
e
-
s
ca
le
d
is
tr
ib
u
ted
en
v
ir
o
n
m
en
ts
.
T
h
ese
s
tu
d
ies
u
tili
ze
r
ea
l
-
wo
r
ld
clu
s
t
er
tr
ac
e
d
atasets
f
r
o
m
m
ajo
r
clo
u
d
p
r
o
v
id
er
s
s
u
ch
as
Alib
ab
a,
Go
o
g
le,
T
e
n
ce
n
t,
a
n
d
Mic
r
o
s
o
f
t
Azu
r
e
t
o
ev
alu
ate
d
if
f
er
en
t
m
o
d
els,
s
ch
ed
u
l
in
g
tech
n
iq
u
es,
an
d
wo
r
k
lo
ad
g
e
n
er
atio
n
f
r
am
ewo
r
k
s
.
T
ab
le
2
in
th
e
Ap
p
en
d
ix
[
1
8
]
-
[
2
2
]
.
C
o
m
p
ar
ativ
e
an
al
y
s
is
p
r
esen
ted
b
elo
w
h
ig
h
lig
h
ts
th
e
o
b
jectiv
es,
d
atasets
,
m
ajo
r
f
in
d
in
g
s
,
an
d
id
en
tifie
d
lim
itatio
n
s
o
f
ex
is
tin
g
r
esear
ch
wo
r
k
s
,
th
er
eb
y
p
r
o
v
id
in
g
a
clea
r
u
n
d
er
s
tan
d
in
g
o
f
c
u
r
r
e
n
t
ad
v
an
ce
m
en
ts
an
d
p
o
ten
tial
f
u
tu
r
e
r
es
ea
r
ch
d
i
r
ec
tio
n
s
in
clo
u
d
wo
r
k
lo
ad
t
r
ac
e
an
a
ly
s
is
.
5.
M
E
T
H
O
DS F
O
R
ANA
L
Y
Z
I
NG
AND
M
O
D
E
L
I
NG
C
L
UST
E
R
T
RACE
S
A
co
m
p
r
eh
en
s
iv
e
co
llectio
n
o
f
wo
r
k
lo
ad
d
ata
th
at
was
co
llected
f
r
o
m
eig
h
t
Go
o
g
le
B
o
r
g
co
m
p
u
te
clu
s
ter
s
o
v
er
th
e
m
o
n
th
o
f
Ma
y
2
0
1
9
is
p
r
o
v
i
d
ed
b
y
th
e
Go
o
g
le
clu
s
ter
wo
r
k
l
o
ad
t
r
ac
es
2
0
1
9
d
ataset,
wh
ich
s
er
v
es
as
a
lan
d
m
ar
k
to
th
e
d
ev
elo
p
m
en
t
o
f
clo
u
d
co
m
p
u
tin
g
r
esear
ch
.
I
t
is
a
co
r
n
er
s
to
n
e
f
o
r
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
A
p
ilo
t revie
w
o
f th
e
G
o
o
g
le
cl
u
s
ter w
o
r
klo
a
d
tr
a
ce
2
0
1
9
,
me
th
o
d
o
lo
g
y
a
n
d
its
a
lter
n
a
tives
…
(
A
ka
s
h
P
a
tel
)
1171
u
n
d
er
s
tan
d
i
n
g
t
h
e
d
y
n
am
ics
o
f
lar
g
e
-
s
ca
le
clu
s
ter
m
an
ag
em
en
t
an
d
wo
r
k
lo
ad
s
ch
ed
u
lin
g
in
s
id
e
a
clo
u
d
en
v
ir
o
n
m
en
t,
a
n
d
th
e
d
ataset
o
b
tain
ed
f
r
o
m
Kag
g
le
is
th
e
s
o
u
r
ce
o
f
th
e
in
f
o
r
m
ati
o
n
.
Giv
en
th
at
t
h
e
p
r
o
ject
is
p
r
im
ar
ily
co
n
ce
r
n
ed
with
th
e
m
an
ag
em
en
t
o
f
h
ea
v
y
tr
af
f
ic,
th
e
Go
o
g
le
clu
s
ter
wo
r
k
l
o
ad
t
r
ac
es
2
0
1
9
d
ataset
h
as
em
er
g
ed
as
an
im
p
o
r
tan
t
r
eso
u
r
ce
th
at
ca
n
b
e
u
tili
ze
d
f
o
r
jo
b
s
u
b
m
is
s
io
n
,
s
ch
e
d
u
lin
g
ch
o
ices,
an
d
r
eso
u
r
ce
u
tili
za
tio
n
ac
r
o
s
s
a
v
ar
iety
o
f
clu
s
ter
s
.
T
h
e
p
u
r
p
o
s
e
o
f
th
is
co
m
p
o
n
en
t
is
to
c
o
n
d
u
ct
an
an
aly
s
is
in
o
r
d
er
t
o
ev
alu
ate
th
e
ef
f
icac
y
an
d
o
u
tco
m
e
o
f
o
u
r
alg
o
r
ith
m
.
A
co
m
p
ar
is
o
n
o
f
th
e
m
eth
o
d
with
m
o
r
e
co
n
v
en
tio
n
al
lo
a
d
b
ala
n
cin
g
tech
n
iq
u
es,
s
u
c
h
as
least
co
n
n
e
ctio
n
s
an
d
least
r
esp
o
n
s
e
tim
e
,
is
u
s
ed
to
an
aly
ze
th
e
ef
f
ec
tiv
e
n
ess
o
f
th
e
alg
o
r
i
t
h
m
.
I
n
o
r
d
er
to
d
o
th
e
co
m
p
ar
is
o
n
,
cr
itical
v
a
r
iab
les
s
u
ch
as
r
eso
u
r
ce
u
s
ag
e,
laten
cy
,
th
r
o
u
g
h
p
u
t,
an
d
er
r
o
r
r
ate
wer
e
tak
en
i
n
to
co
n
s
id
er
atio
n
.
T
h
e
ef
f
icie
n
cy
an
d
d
ep
en
d
ab
ilit
y
o
f
lo
ad
-
b
alan
cin
g
alg
o
r
ith
m
s
in
r
ea
l
-
wo
r
ld
cir
cu
m
s
tan
ce
s
m
ay
b
e
an
aly
ze
d
u
s
in
g
th
ese
m
ea
s
u
r
es,
wh
ich
ar
e
ac
ce
p
tab
le
f
o
r
t
h
e
p
u
r
p
o
s
e.
T
h
is
an
aly
s
is
is
ca
r
r
ied
o
u
t
b
y
d
is
tr
ib
u
tin
g
5
0
0
task
s
to
ea
ch
o
f
th
e
alg
o
r
ith
m
s
,
b
eg
in
n
in
g
with
1
0
0
task
s
an
d
in
cr
ea
s
in
g
th
e
n
u
m
b
er
o
f
task
s
b
y
1
0
0
in
ea
c
h
s
u
cc
es
s
iv
e
s
tag
e.
I
n
th
is
,
th
e
m
etr
ics
wer
e
watc
h
ed
an
d
k
ep
t
clo
s
ely
in
a
m
an
n
e
r
f
o
r
an
aly
s
is
to
d
eter
m
in
e
wh
ich
im
p
lem
en
tatio
n
h
ad
an
ef
f
ec
tiv
e
o
u
tco
m
e
f
r
o
m
all
o
f
th
em
[
7
]
.
T
h
is
r
esear
ch
an
aly
ze
d
th
e
G
o
o
g
le
T
r
ac
e
Data
s
et.
T
h
is
d
at
aset
is
ex
tr
em
ely
v
alu
ab
le
d
u
e
to
th
e
f
ac
t
th
at
it
is
cu
r
r
en
tly
b
e
in
g
wo
r
k
ed
o
n
b
y
a
lar
g
e
n
u
m
b
er
o
f
a
ca
d
em
ics,
an
d
th
e
f
in
d
in
g
s
o
f
th
is
r
esear
ch
h
av
e
th
e
p
o
ten
tial
to
in
f
lu
en
ce
th
e
n
ex
t
g
en
er
atio
n
o
f
clo
u
d
co
m
p
u
tin
g
an
d
d
ata
ce
n
ter
s
.
Fo
u
r
p
er
s
p
ec
tiv
es
wer
e
u
s
ed
to
an
aly
ze
th
e
Go
o
g
le
T
r
ac
e
to
en
h
an
ce
cl
u
s
ter
p
er
f
o
r
m
an
ce
.
Me
m
o
r
y
af
f
ec
ts
wo
r
k
p
er
f
o
r
m
an
ce
m
o
r
e
th
an
th
e
C
PU,
ac
co
r
d
i
n
g
to
o
u
r
in
itial
d
ata
r
esear
ch
.
Des
p
ite
m
em
o
r
y
b
ein
g
th
e
b
ar
r
ie
r
,
Go
o
g
le
r
ec
en
tly
u
p
d
ated
s
er
v
er
s
with
f
u
ll
C
PUs
to
h
av
e
f
u
ll
m
em
o
r
y
.
B
o
r
g
p
er
f
o
r
m
a
n
ce
will
im
p
r
o
v
e
b
y
r
ed
u
cin
g
R
AM
b
y
2
5
%
f
r
o
m
2
2
1
8
f
u
ll
C
PU
an
d
m
em
o
r
y
s
y
s
tem
s
to
1
0
,
1
8
8
with
0
.
5
C
PU
an
d
0
.
2
4
9
3
m
e
m
o
r
y
.
I
t
ca
n
b
e
s
ee
n
th
at
m
o
r
e
r
estrictiv
e
p
r
o
f
ess
io
n
s
ten
d
to
d
elay
less
in
th
e
s
ec
o
n
d
ex
p
e
r
im
en
t,
alth
o
u
g
h
t
h
e
t
h
ir
d
tr
ial
h
as g
iv
en
r
esu
lts
s
h
o
win
g
th
at
lo
wer
-
p
r
io
r
ity
jo
b
s
co
n
t
r
ib
u
te
a
lo
t
in
ter
m
s
o
f
r
esch
ed
u
lin
g
a
n
d
au
g
m
en
t
clo
u
d
o
v
er
h
ea
d
,
wh
ich
wer
e
n
o
t
ex
p
ec
ted
.
I
t
was
p
r
o
p
o
s
ed
to
m
in
im
ize
th
e
tim
e
s
p
en
t
o
n
ea
ch
jo
b
a
n
d
let
th
e
lo
wer
-
p
r
io
r
it
y
jo
b
s
f
in
is
h
f
o
r
o
p
tim
al
u
s
e
o
f
r
eso
u
r
ce
s
.
Fin
ally
,
it
was
o
b
s
er
v
ed
th
at
th
e
clo
u
d
m
an
ag
em
e
n
t
s
y
s
tem
p
u
ts
th
e
m
ax
im
u
m
c
o
n
s
tr
ain
ts
to
KI
L
L
E
D
jo
b
s
,
an
d
th
is
is
th
e
m
ain
o
b
s
er
v
a
tio
n
o
f
t
h
is
s
tu
d
y
.
I
t
s
h
o
u
ld
b
e
b
r
o
k
en
d
o
wn
in
t
o
jo
b
s
with
s
u
b
task
s
in
o
r
d
er
to
m
in
im
ize
th
e
n
u
m
b
e
r
o
f
jo
b
s
elim
in
ated
an
d
im
p
r
o
v
e
p
er
f
o
r
m
an
ce
[
9
]
.
Alth
o
u
g
h
G
o
o
g
le
cl
u
s
ter
d
at
a
s
et
o
f
2
0
1
9
h
as
b
ee
n
wid
ely
e
x
p
lo
r
ed
,
th
er
e
ar
e
s
till
m
ajo
r
o
p
er
atio
n
al
in
s
ig
h
ts
to
b
e
m
ad
e
.
I
n
th
is
p
ap
er
,
th
e
au
t
h
o
r
s
s
u
g
g
est
ar
tif
icial
in
tellig
en
ce
m
o
d
els
th
at
ar
e
r
u
n
o
n
g
r
ap
h
ics
p
r
o
ce
s
s
in
g
u
n
its
to
f
o
r
ec
ast
th
e
wo
r
k
l
o
ad
lim
itatio
n
s
,
wh
ich
is
o
n
e
o
f
th
e
p
r
im
a
r
y
b
o
ttlen
ec
k
s
in
cl
o
u
d
en
v
ir
o
n
m
en
ts
at
th
e
s
ca
le.
T
h
e
ap
p
r
o
ac
h
wo
u
l
d
h
elp
i
n
b
etter
p
r
o
v
is
io
n
in
g
o
f
its
r
eso
u
r
ce
s
,
less
en
in
g
p
er
f
o
r
m
an
ce
d
eg
r
a
d
atio
n
,
an
d
ex
tr
ac
tin
g
m
o
r
e
v
al
u
e
o
u
t
o
f
th
e
ex
is
tin
g
clu
s
ter
t
r
a
ce
in
f
o
r
m
atio
n
b
y
an
ticip
atin
g
s
u
ch
co
n
s
tr
ain
ts
in
th
e
f
u
tu
r
e.
R
esear
ch
s
u
g
g
e
s
ts
co
m
b
in
in
g
m
ac
h
in
e
lear
n
i
n
g
alg
o
r
ith
m
s
lik
e
XGBo
o
s
t
with
h
y
p
er
p
ar
am
ete
r
tu
n
in
g
ap
p
r
o
ac
h
es
lik
e
s
war
m
-
b
ased
ar
tific
ial
b
ee
co
lo
n
y
o
p
tim
izatio
n
to
s
av
e
en
er
g
y
an
d
tim
e.
Acc
u
r
ate
co
n
s
tr
ain
t
n
u
m
b
er
f
o
r
ec
asti
n
g
s
av
es
tim
e
an
d
en
er
g
y
.
Ou
r
r
e
s
ea
r
ch
en
d
ea
v
o
r
s
to
p
u
s
h
th
e
lim
its
o
f
t
h
e
ca
p
a
b
ilit
y
o
f
th
is
d
ataset,
with
th
e
g
o
al
o
f
d
is
co
v
er
in
g
s
ig
n
if
ica
n
t
k
n
o
wled
g
e
th
at
h
as
an
in
f
lu
en
ce
in
th
e
ac
tu
al
wo
r
ld
.
Als
o
,
th
e
B
o
r
g
s
y
s
tem
s
h
o
u
ld
o
p
tim
ize
its
ef
f
icien
cy
b
y
an
aly
zin
g
an
o
p
tio
n
th
at
d
is
r
eg
ar
d
s
VM
s
th
at
h
av
e
an
u
n
u
s
u
ally
lo
w
n
u
m
b
er
o
f
r
eso
u
r
ce
s
:
eith
e
r
C
PU
o
r
R
AM
.
T
h
er
ef
o
r
e,
t
h
e
f
u
tu
r
e
r
esear
ch
a
r
ea
is
p
r
o
v
i
d
in
g
s
o
p
h
is
ticated
id
ea
s
to
B
o
r
g
ab
o
u
t
r
eso
u
r
ce
allo
ca
tio
n
te
ch
n
iq
u
es,
with
th
e
aim
to
elim
in
ate
p
o
te
n
tial
b
o
ttlen
ec
k
ef
f
ec
ts
s
u
ch
u
n
d
er
-
p
r
o
v
is
io
n
ed
v
ir
tu
al
m
ac
h
in
es
cr
ea
te.
I
t
is
f
u
r
th
e
r
h
ig
h
ly
r
ec
o
m
m
e
n
d
ed
th
at
o
t
h
er
m
ajo
r
clo
u
d
s
er
v
ice
p
r
o
v
id
er
s
in
clu
d
i
n
g
Am
az
o
n
W
eb
Ser
v
ices
(
AW
S),
Mic
r
o
s
o
f
t
Azu
r
e,
an
d
Alib
ab
a
b
e
c
o
n
s
id
er
ed
in
s
u
b
s
eq
u
en
t
r
esear
ch
ef
f
o
r
ts
.
C
o
m
p
ar
ativ
e
s
tu
d
ies
ac
r
o
s
s
th
ese
p
latf
o
r
m
s
ca
n
th
e
r
ef
o
r
e
p
r
o
v
id
e
th
e
r
esear
ch
co
m
m
u
n
ity
as a
wh
o
le
with
v
alu
a
b
le
f
in
d
in
g
s
[
9
]
.
W
ith
in
th
e
r
ea
lm
o
f
cu
r
r
e
n
t
clu
s
ter
m
an
ag
em
en
t
s
y
s
tem
s
,
tr
ac
e
an
aly
s
is
h
as
b
ee
n
p
er
f
o
r
m
in
g
a
s
ig
n
if
ican
t
r
o
le
in
p
r
o
v
id
in
g
an
u
n
d
er
s
tan
d
in
g
o
f
t
h
e
b
e
h
a
v
io
r
s
th
at
t
h
ey
ex
h
ib
it.
T
h
is
p
ap
er
p
r
esen
ts
th
e
f
in
d
in
g
s
o
f
o
u
r
in
ter
-
clu
s
ter
an
aly
s
is
o
f
th
e
r
ec
en
tly
d
is
clo
s
ed
Go
o
g
le
tr
ac
e
2
0
1
9
d
ata
s
et.
Ou
r
a
n
aly
s
is
f
o
cu
s
ed
o
n
th
e
d
is
tr
ib
u
tio
n
o
f
u
n
d
er
ly
i
n
g
p
h
y
s
ical
m
ac
h
in
es
,
s
u
ch
as
th
e
n
u
m
b
e
r
o
f
m
ac
h
i
n
es
an
d
r
eso
u
r
ce
s
,
an
d
co
m
p
u
tatio
n
al
task
r
eq
u
est
m
etr
ics,
s
u
ch
as
wo
r
k
d
u
r
atio
n
,
tim
e
u
s
ed
,
a
n
d
c
y
cles
p
er
in
s
tr
u
ctio
n
.
T
h
is
f
u
r
th
er
r
ev
ea
l
th
at
th
e
ce
lls
p
o
s
s
ess
in
g
alm
o
s
t
th
e
s
am
e
co
m
p
o
s
itio
n
o
f
m
ac
h
in
e
r
eso
u
r
ce
co
u
ld
co
n
s
u
m
e
th
o
s
e
r
eso
u
r
ce
s
in
an
en
tire
ly
d
i
f
f
er
e
n
t
m
an
n
er
b
y
c
o
n
s
id
er
in
g
th
e
p
atter
n
o
f
r
eq
u
est
o
f
t
ask
s
.
A
s
h
ap
e
h
as
b
ee
n
f
ig
u
r
ed
o
u
t
th
r
o
u
g
h
an
u
n
s
u
p
er
v
is
ed
lear
n
in
g
alg
o
r
ith
m
f
r
o
m
v
ar
io
u
s
clu
s
ter
s
;
s
u
ch
s
h
ap
es
ap
p
ea
r
to
b
e
two
ty
p
es
-
co
n
ca
v
e
a
n
d
co
n
v
ex
f
o
r
m
s
.
I
n
ca
s
e
o
f
th
e
co
n
v
ex
s
h
ap
e,
it
wo
u
ld
b
e
th
e
p
r
esen
ce
o
f
co
h
er
en
t
s
ca
le
v
ar
iatio
n
f
o
r
all
d
im
en
s
i
o
n
,
b
u
t c
o
n
ca
v
e
d
e
p
icts
m
is
s
in
g
v
ar
ian
ce
al
o
n
g
with
a
n
y
d
im
en
s
io
n
.
Pro
v
id
e
a
ch
a
r
ac
ter
izatio
n
an
aly
s
is
o
f
a
wo
r
k
lo
ad
tr
ac
e
th
at
was
o
b
tain
ed
o
v
e
r
a
p
e
r
io
d
o
f
two
m
o
n
th
s
f
r
o
m
a
p
r
o
d
u
ctio
n
m
a
ch
in
e
lear
n
in
g
as
a
s
er
v
ice
clu
s
ter
in
Alib
ab
a
th
at
h
ad
m
o
r
e
th
an
6
,
0
0
0
GPUs
.
T
h
e
ex
p
lan
atio
n
o
f
clu
s
ter
s
c
h
ed
u
lin
g
ch
allen
g
es
is
g
iv
en
.
Su
ch
is
s
u
es
in
clu
d
e
lo
w
u
tili
za
tio
n
o
f
g
r
ap
h
ical
p
r
o
ce
s
s
in
g
u
n
its
,
lo
n
g
tim
es
to
wait
in
th
e
q
u
eu
es,
th
e
e
x
is
ten
ce
o
f
ce
r
tai
n
h
ar
d
-
to
-
s
ch
ed
u
le
ap
p
licatio
n
s
r
eq
u
ir
in
g
h
ig
h
-
e
n
d
GPUs
b
u
t
h
av
e
s
p
ec
if
ic
s
ch
ed
u
lin
g
r
e
q
u
ir
em
en
ts
an
d
m
ay
ca
u
s
e
i
m
b
alan
ce
d
lo
ad
in
g
ac
r
o
s
s
h
eter
o
g
en
e
o
u
s
m
ac
h
i
n
es,
ca
u
s
in
g
b
o
ttlen
ec
k
s
o
n
C
PUs
.
T
h
ese
cu
r
r
en
t
s
o
lu
tio
n
s
in
clu
d
e
a
s
ch
ed
u
ler
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
1
1
6
7
-
1
1
7
8
1172
wh
ich
allo
ws
f
o
r
GPU
s
h
ar
in
g
an
d
u
s
es
a
r
eser
v
in
g
-
an
d
-
p
a
ck
in
g
s
tr
ateg
y
in
s
ep
ar
atin
g
h
ig
h
-
GPU
jo
b
s
f
r
o
m
lo
w
-
GPU
task
s
to
b
etter
m
an
ag
e
th
e
s
ch
ed
u
lin
g
o
f
PAI
wo
r
k
lo
ad
s
.
T
h
e
o
b
s
er
v
atio
n
is
th
at
m
o
s
t
o
f
th
e
jo
b
s
ar
e
u
s
u
ally
r
u
n
n
in
g
an
d
s
ch
ed
u
led
b
y
g
r
o
u
p
s
,
with
less
th
an
o
n
e
GPU
p
er
i
n
s
tan
ce
.
On
ly
v
e
r
y
f
ew
b
u
s
in
ess
-
cr
itical
jo
b
s
u
s
e
th
e
h
ig
h
-
en
d
GPUs
in
o
n
e
s
y
s
tem
co
n
n
ec
ted
v
ia
NVL
in
k
s
.
T
h
e
C
PU
s
ee
m
s
to
b
e
a
b
o
ttlen
ec
k
f
o
r
s
u
ch
lo
w
GP
U
r
eso
u
r
ce
jo
b
s
;
in
f
ac
t,
it
also
h
o
s
ts
d
ata
p
r
e
-
p
r
o
ce
s
s
in
g
an
d
s
im
u
latio
n
.
So
m
e
p
r
o
b
lem
s
r
em
ai
n
,
in
clu
d
in
g
p
o
s
s
ib
le
C
PU
b
o
ttlen
ec
k
an
d
lo
a
d
im
b
alan
ce
f
o
r
h
eter
o
g
en
eo
u
s
d
ev
ices.
T
h
e
tr
ac
e
-
th
e
m
o
s
t
d
etailed
i
n
ter
m
s
o
f
w
o
r
k
lo
a
d
s
an
d
clu
s
ter
s
ca
le
-
is
p
u
b
lis
h
ed
to
p
r
o
m
o
te
f
u
r
t
h
er
r
esear
c
h
o
n
b
etter
GPU
s
ch
ed
u
lin
g
[
1
7
]
.
W
ith
in
th
e
s
co
p
e
o
f
th
is
r
esea
r
ch
,
A
co
m
p
r
e
h
en
s
iv
e
s
tr
ateg
y
f
o
r
th
e
task
f
ailu
r
e
d
etec
tio
n
m
o
d
el
is
p
r
o
p
o
s
ed
.
Fiv
e
tr
ad
itio
n
al
m
o
d
els
o
f
m
ac
h
in
e
lear
n
in
g
a
n
d
t
h
r
ee
d
ee
p
lea
r
n
in
g
m
o
d
els
we
r
e
cr
ea
ted
u
s
in
g
th
e
GC
T
d
ataset
to
ass
e
s
s
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
ese
d
if
f
er
e
n
t
m
o
d
els
in
jo
b
a
n
d
task
f
ailu
r
e
p
r
ed
ictio
n
.
T
h
e
b
est
m
o
d
el
is
th
e
XGBo
o
s
t
class
i
f
ier
,
wh
ich
g
iv
es
a
n
ac
c
u
r
ac
y
s
co
r
e
o
f
9
4
.
3
5
%
a
n
d
a
n
F
-
s
co
r
e
o
f
0
.
9
3
1
0
as
o
b
tain
ed
b
y
th
e
p
er
f
o
r
m
an
ce
an
aly
s
is
.
T
wo
m
o
d
els
wer
e
f
o
u
n
d
at
th
e
task
lev
el
t
h
at
ar
e
at
ex
ce
llen
t
le
v
els.
T
h
ese
m
o
d
els
ar
e
b
a
s
ed
o
n
d
ec
is
io
n
tr
ee
a
n
d
r
an
d
o
m
f
o
r
es
t
.
A
s
co
r
e
o
f
8
9
.
7
5
%
f
o
r
ac
c
u
r
ac
y
a
n
d
a
s
co
r
e
o
f
0
.
9
1
5
4
f
o
r
th
e
F
-
s
co
r
e
h
av
e
b
ee
n
ca
lcu
lated
f
o
r
b
o
th
m
o
d
els.
T
ak
in
g
e
v
er
y
th
i
n
g
in
t
o
co
n
s
id
er
atio
n
,
th
e
f
in
d
in
g
s
h
a
v
e
d
em
o
n
s
tr
ated
t
h
at
th
e
t
r
ad
itio
n
al
ML
m
o
d
els
p
er
f
o
r
m
s
ig
n
if
ican
tly
b
etter
t
h
an
th
e
DL
m
o
d
els
wh
en
it
co
m
es
to
class
if
y
in
g
t
h
e
ter
m
in
atio
n
s
tatu
s
o
f
jo
b
s
an
d
task
s
.
Fu
r
th
er
m
o
r
e,
ac
co
r
d
in
g
to
th
e
f
i
n
d
in
g
s
o
f
o
u
r
r
esear
ch
o
f
th
e
s
ig
n
i
f
ican
ce
o
f
f
ea
t
u
r
es,
th
e
s
ch
ed
u
lin
g
class
an
d
th
e
C
PU
r
eq
u
est
ar
e
th
e
m
o
s
t
im
p
o
r
tan
t
ch
ar
ac
ter
is
tics
f
o
r
T
r
ad
itio
n
al
ML
,
wh
er
ea
s
th
e
d
is
k
s
p
ac
e
r
eq
u
est
an
d
th
e
m
em
o
r
y
r
eq
u
est
ar
e
th
e
m
o
s
t
im
p
o
r
tan
t
f
ea
tu
r
es
f
o
r
D
L
in
ter
m
s
o
f
th
e
p
r
e
d
ictio
n
o
f
task
f
ailu
r
e.
On
th
e
o
th
er
h
an
d
,
wh
en
it
co
m
es
to
task
-
lev
el
p
r
ed
ictio
n
,
th
e
m
o
s
t
ess
en
tial
e
lem
en
t
f
o
r
t
r
ad
itio
n
al
ML
is
r
eso
u
r
ce
d
em
an
d
s
,
wh
er
ea
s
p
r
io
r
ity
is
th
e
m
o
s
t
cr
itical
f
ac
to
r
f
o
r
D
L
m
o
d
els.
I
n
co
n
cl
u
s
io
n
,
o
u
r
s
ca
lab
ilit
y
an
aly
s
is
r
ev
ea
led
t
h
at
T
r
ad
itio
n
al
ML
m
o
d
els
ar
e
ca
p
ab
le
o
f
m
ak
in
g
p
r
ed
ictio
n
s
in
a
r
ea
s
o
n
ab
le
am
o
u
n
t
o
f
tim
e,
d
esp
i
te
th
e
f
ac
t
th
at
th
ey
ar
e
im
p
lem
en
ted
o
n
co
n
s
u
m
er
-
lev
el
h
ar
d
war
e
[
1
4
]
.
I
n
th
e
f
ield
o
f
clo
u
d
co
m
p
u
ti
n
g
,
s
ev
er
al
s
tu
d
ies
h
av
e
m
ad
e
u
s
e
o
f
clu
s
ter
in
g
t
o
an
aly
ze
clo
u
d
tr
ac
e
d
ata
in
o
r
d
er
to
i
m
p
r
o
v
e
r
es
o
u
r
ce
m
an
ag
em
e
n
t.
Mu
c
h
o
f
th
is
r
esear
ch
h
av
e
b
ee
n
s
u
c
ce
s
s
f
u
l.
Du
r
in
g
th
e
clu
s
ter
in
g
p
r
o
ce
s
s
,
th
e
d
im
en
s
io
n
s
an
d
t
h
e
ap
p
r
o
ac
h
th
at
a
r
e
u
tili
ze
d
ca
n
h
av
e
a
s
ig
n
if
i
ca
n
t
im
p
ac
t
o
n
th
e
q
u
ality
o
f
th
e
clo
u
d
t
r
ac
es
th
at
ar
e
clu
s
ter
ed
.
T
h
e
p
ap
er
p
r
esen
ted
two
co
n
tr
ib
u
tio
n
s
:
an
in
-
d
ep
th
a
n
aly
s
is
,
co
n
s
id
er
in
g
th
e
p
o
s
s
ib
il
ity
o
f
im
p
r
o
v
in
g
clu
s
ter
in
g
q
u
ality
u
s
in
g
d
im
en
s
io
n
s
th
at
ar
e
m
o
r
e
ef
f
icien
t
th
a
n
tr
ad
itio
n
al
o
n
es
an
d
p
r
o
ce
d
u
r
e
s
d
if
f
er
en
t
f
r
o
m
th
e
u
s
u
al
o
n
e
s
u
s
ed
.
T
h
e
ch
allen
g
e
o
f
p
r
e
d
ictin
g
wh
ich
o
f
th
e
two
elem
en
ts
wil
l
o
cc
u
r
is
s
o
m
eth
in
g
th
at
h
as
n
o
t
y
et
b
ee
n
an
tici
p
ated
,
s
o
a
d
etec
tio
n
m
eth
o
d
,
E
Fectio
n
,
was
in
v
en
ted
.
A
co
m
p
ar
is
o
n
o
f
th
e
p
er
f
o
r
m
an
ce
o
f
t
h
r
ee
in
ter
n
a
l
v
alid
atio
n
m
ea
s
u
r
es
-
Dav
ies
-
B
o
u
ld
in
,
Sil
h
o
u
ette
C
o
ef
f
icien
t,
an
d
C
alin
s
k
i
–
Har
ab
asz
—
was
co
n
d
u
cted
to
f
i
n
d
th
e
m
o
s
t
s
u
itab
le
v
alid
ati
o
n
m
et
h
o
d
f
o
r
th
is
tec
h
n
iq
u
e
wh
ile
d
e
v
elo
p
in
g
t
h
e
E
Fectio
n
ap
p
r
o
ac
h
.
T
h
is
m
eth
o
d
allo
ws
o
n
e
to
f
in
d
w
h
ich
d
im
en
s
io
n
a
n
d
m
eth
o
d
will
p
r
o
d
u
ce
th
e
h
ig
h
e
s
t
q
u
ality
clu
s
ter
in
g
o
f
a
p
ar
tic
u
lar
clo
u
d
tr
ac
e
with
o
u
t
an
an
aly
s
is
o
f
th
e
en
tire
d
ataset.
I
n
d
o
in
g
s
o
,
th
e
Dav
ies
–
B
o
u
ld
en
in
d
e
x
as
well
as
th
e
Pear
s
o
n
co
r
r
elatio
n
co
ef
f
icien
t
is
u
s
ed
.
Per
f
o
r
m
an
ce
m
etr
ics:
I
n
itially
,
th
e
test
ab
ilit
y
o
f
E
Fectio
n
s
ta
r
ted
with
a
v
e
r
if
icatio
n
o
f
ac
c
u
r
ac
y
an
d
p
r
ec
is
io
n
to
id
en
tify
in
s
tan
ce
s
o
v
er
v
ar
i
o
u
s
co
m
b
in
atio
n
s
o
f
attr
ib
u
tes
with
d
if
f
er
en
t
m
eth
o
d
s
.
T
h
en
its
u
ti
lity
is
s
h
o
wn
b
y
m
ak
i
n
g
two
a
p
p
licatio
n
s
in
r
elate
d
an
aly
s
is
.
A
co
m
p
ar
is
o
n
with
th
e
m
o
s
t
r
ec
en
t
d
o
cu
m
en
ted
ef
f
o
r
ts
in
th
e
liter
atu
r
e
r
ev
ea
led
th
at
E
Fectio
n
was
p
o
ten
tially
ap
p
licab
l
e
in
alm
o
s
t
8
3
%
o
f
ca
s
es
an
d
h
ad
th
e
h
ig
h
est
ac
cu
r
ac
y
p
r
ef
e
r
e
n
ce
.
A
d
d
itio
n
ally
,
it
s
u
r
p
ass
ed
th
e
m
o
s
t
r
ec
en
t
attem
p
t
b
y
elev
e
n
p
er
ce
n
t
ag
e
p
o
i
n
ts
an
d
h
ad
h
ig
h
er
s
tab
ilit
y
th
an
th
e
p
r
ev
io
u
s
ap
p
r
o
ac
h
[
2
3
]
,
[2
4
].
Usi
n
g
th
e
Go
o
g
le
clu
s
ter
tr
ac
e
d
ata
s
et
v
er
s
io
n
3
,
wh
ich
was
ac
q
u
ir
ed
f
r
o
m
ab
o
u
t
9
6
,
0
0
0
c
o
m
p
u
ter
s
,
Failu
r
e
s
tatis
tic
s
an
d
m
ac
h
in
e
life
tim
es
o
v
er
tim
e
wer
e
s
tu
d
ied
.
I
n
less
th
an
a
m
in
u
te
f
r
o
m
th
e
p
r
e
v
io
u
s
f
ailu
r
e
o
f
th
e
m
ac
h
in
e,
it
was
ca
lcu
lated
th
at
1
3
%
th
at
an
o
t
h
er
m
ac
h
in
e
o
n
th
e
s
am
e
n
etwo
r
k
s
witch
will
f
ail.
A
Ma
r
k
o
v
c
h
ain
m
o
d
el
was d
esig
n
ed
in
o
r
d
er
to
p
r
ed
ict
wh
i
ch
m
ac
h
in
es a
r
e
in
wh
ich
s
tates a
t a
n
y
tim
e.
W
ith
th
is
m
o
d
el
an
d
co
m
p
u
ted
p
r
o
b
ab
ilit
y
esti
m
ates,
v
er
y
ac
cu
r
ate
p
r
ed
ictio
n
s
ab
o
u
t
m
ac
h
in
e
s
t
atu
s
co
u
ld
b
e
d
o
n
e
f
o
r
s
ev
er
al
d
ay
s
.
Usi
n
g
th
e
esti
m
ated
m
ac
h
in
e
s
tates,
th
e
tr
en
d
o
f
th
e
ac
tiv
e
m
ac
h
in
e
was
g
en
er
ated
an
d
co
m
p
ar
ed
to
th
e
tr
e
n
d
p
r
o
v
id
e
d
in
th
e
Go
o
g
le
cl
u
s
ter
wo
r
k
lo
ad
tr
ac
es
d
ataset.
T
h
e
er
r
o
r
o
b
tain
ed
was
1
.
7
6
%.
W
ith
in
th
e
co
m
p
r
ess
ed
f
o
r
m
a
t,
th
e
d
ata
s
et
v
er
s
io
n
3
h
as
a
s
ize
o
f
2
.
7
ter
ab
y
tes.
T
h
e
d
ata
co
llectio
n
was
m
ad
e
av
ailab
le
b
y
Go
o
g
le
i
n
b
o
th
th
e
J
SON
f
o
r
m
at
an
d
as
B
ig
Qu
er
y
tab
les.
R
eg
ar
d
in
g
J
SON,
th
e
s
ize
o
f
th
e
d
ata
co
llectio
n
is
g
r
ea
ter
th
a
n
7
ter
ab
y
tes.
W
ith
s
u
ch
a
l
ar
g
e
d
ata
s
et
co
m
es
a
s
et
o
f
ch
allen
g
es
th
at
ar
e
en
tire
ly
u
n
iq
u
e
.
T
h
e
m
ajo
r
ity
o
f
th
e
ch
allen
g
es
th
at
ar
e
r
el
ated
with
th
e
J
SON
f
o
r
m
at
m
ay
b
e
r
eso
lv
ed
b
y
u
s
in
g
th
e
v
er
s
io
n
o
f
th
e
d
ata
s
et
th
at
is
h
o
s
ted
o
n
Go
o
g
le
B
i
g
Qu
er
y
.
B
ig
Qu
er
y
is
q
u
ite
ex
p
en
s
iv
e,
b
u
t
d
u
e
t
o
th
e
J
SON
f
o
r
m
at,
it
ca
n
r
u
n
o
n
co
m
m
o
d
ity
h
ar
d
wa
r
e.
Fo
r
t
h
is
r
esear
ch
wo
r
k
,
t
h
e
d
ata
s
et
was
r
etr
iev
ed
f
r
o
m
th
e
Go
o
g
le
Sto
r
ag
e
b
u
ck
et,
a
n
d
th
e
n
it
was
d
o
wn
l
o
ad
ed
c
o
m
p
r
ess
ed
in
J
SON
f
o
r
m
at.
T
o
in
cr
ea
s
e
m
em
o
r
y
ef
f
icien
cy
,
th
e
co
m
p
r
ess
ed
J
SON
was
im
p
o
r
ted
in
to
th
e
Sp
ar
k
C
lu
s
ter
an
d
s
av
e
d
i
n
Ap
a
ch
e
Par
q
u
et
f
o
r
m
at
b
ec
au
s
e
u
n
co
m
p
r
ess
ed
J
SON
o
cc
u
p
ies
to
o
m
u
c
h
o
f
th
e
s
y
s
tem
’
s
m
em
o
r
y
.
On
ly
th
e
n
e
ed
ed
d
ata
f
r
o
m
th
e
Ap
ac
h
e
Sp
ar
k
Par
q
u
et
f
ile
was
r
ea
d
,
co
n
v
er
ted
,
a
n
d
s
av
ed
to
a
C
SV
f
ile
f
o
r
f
u
r
th
er
p
r
o
ce
s
s
in
g
with
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
A
p
ilo
t revie
w
o
f th
e
G
o
o
g
le
cl
u
s
ter w
o
r
klo
a
d
tr
a
ce
2
0
1
9
,
me
th
o
d
o
lo
g
y
a
n
d
its
a
lter
n
a
tives
…
(
A
ka
s
h
P
a
tel
)
1173
Py
th
o
n
[
2
2
]
.
I
n
o
r
d
er
to
in
cr
e
ase
th
e
ac
cu
r
ac
y
o
f
th
e
f
o
r
ec
a
s
t,
First,
p
r
esen
t
a
th
r
ee
-
way
e
n
s
em
b
le
d
ata
ce
n
te
r
wo
r
k
lo
ad
p
r
ed
ictio
n
.
I
t
was
s
tated
th
at
wo
r
k
lo
ad
s
wer
e
co
n
tin
u
o
u
s
,
v
ar
y
in
g
,
an
d
d
ela
y
s
.
T
h
u
s
,
u
s
in
g
th
e
an
n
ea
lin
g
s
ch
em
e,
w
o
r
k
lo
a
d
-
d
iv
is
io
n
th
r
esh
o
ld
in
g
was
d
is
co
v
er
ed
.
Su
b
s
eq
u
en
tly
,
b
y
m
ak
in
g
a
th
r
ee
-
way
d
ec
is
io
n
p
r
in
cip
le
to
ass
ig
n
o
n
e
o
f
th
r
ee
ty
p
es
o
f
f
o
r
e
ca
s
tin
g
m
o
d
els
d
ep
e
n
d
in
g
u
p
o
n
wo
r
k
l
o
ad
ty
p
e
an
d
p
r
io
r
a
p
r
io
r
i
er
r
o
r
p
r
ed
ictio
n
,
ac
c
u
r
ac
y
ca
n
b
e
im
p
r
o
v
ed
u
p
o
n
.
L
astl
y
,
T
W
D
-
R
C
PM
en
h
an
ce
d
wo
r
k
lo
a
d
f
o
r
ec
ast
ac
cu
r
ac
y
b
y
6
9
.
0
%,
6
8
.
6
%,
an
d
7
2
.
6
%
wh
en
test
ed
o
n
Go
o
g
le
clu
s
ter
tr
ac
e
C
PU
lo
ad
m
o
n
ito
r
in
g
lo
g
s
co
m
p
ar
ed
to
AR
I
MA
,
NN,
an
d
DM
ASVR
-
3
W
D
[
1
2
]
.
T
h
e
an
aly
s
is
o
f
Go
o
g
le
clu
s
t
er
tr
ac
e
s
n
ee
d
s
to
u
tili
ze
s
o
p
h
is
ticated
clu
s
ter
in
g
tech
n
iq
u
es
th
at
ca
n
ef
f
icien
tly
o
p
e
r
ate
with
h
ig
h
-
d
im
en
s
io
n
al
an
d
d
y
n
a
m
ically
ch
an
g
in
g
d
ata.
I
n
co
r
p
o
r
atin
g
cu
r
r
en
t
alg
o
r
ith
m
s
an
d
m
o
d
el
s
,
r
esear
ch
er
s
will
b
e
ab
le
to
en
h
an
ce
th
e
ac
c
u
r
a
cy
an
d
in
ter
p
r
etab
ilit
y
o
f
th
e
r
esu
lts
o
f
clu
s
ter
in
g
,
wh
ich
is
n
ec
ess
ar
y
to
a
n
aly
ze
d
ata
m
ea
n
i
n
g
f
u
lly
.
T
r
im
m
ed
an
d
s
p
ar
s
e
cl
u
s
ter
in
g
ar
e
au
t
o
m
ated
m
eth
o
d
s
o
f
clu
s
ter
in
g
th
at
em
p
h
asize
im
p
o
r
tan
t
f
ea
tu
r
es
an
d
d
ee
m
p
h
a
s
ize
n
o
is
e
s
o
th
at
th
e
p
r
o
ce
s
s
is
le
s
s
s
en
s
itiv
e
to
o
u
tlier
s
an
d
th
e
p
a
r
am
eter
s
ar
e
m
o
r
e
ea
s
ily
ch
o
s
en
b
y
u
s
er
s
.
Fra
m
ewo
r
k
s
s
u
ch
as
B
UR
ST,
to
o
,
ar
e
tailo
r
e
d
to
s
tr
ea
m
in
g
tim
e
-
s
er
ies
d
ata,
allo
win
g
r
ea
l
-
tim
e
a
n
aly
s
is
th
r
o
u
g
h
d
y
n
am
ically
a
d
ap
tin
g
to
c
h
an
g
es
an
d
esti
m
atin
g
th
e
b
est
n
u
m
b
er
o
f
clu
s
ter
s
.
B
esid
es
ac
cu
r
ac
y
,
ex
p
lain
ab
ilit
y
is
an
im
p
o
r
ta
n
t
p
a
r
t
o
f
clu
s
ter
in
g
a
n
d
s
u
ch
to
o
ls
as
C
lu
s
ter
-
E
x
p
lo
r
er
ca
n
b
e
u
s
ed
to
im
p
r
o
v
e
in
ter
p
r
etab
ilit
y
b
y
d
eter
m
i
n
in
g
p
atter
n
s
an
d
ch
ar
ac
ter
is
tics
th
at
ar
e
im
p
o
r
tan
t
in
clu
s
ter
s
an
d
em
p
l
o
y
m
eth
o
d
s
s
u
ch
as
f
r
eq
u
e
n
t
-
item
s
et
m
in
in
g
.
Nev
er
th
eless
,
s
o
m
e
p
r
o
b
lem
s
ar
e
s
till
p
r
esen
t,
esp
ec
ial
ly
in
th
e
p
r
o
b
lem
o
f
g
u
a
r
an
teein
g
ea
s
y
an
d
u
n
d
er
s
tan
d
a
b
le
o
u
tco
m
es
with
co
m
p
lex
d
ata,
s
o
m
o
r
e
r
esear
ch
is
n
ec
ess
ar
y
to
en
h
an
c
e
th
e
ef
f
ic
ien
cy
an
d
v
er
s
atility
o
f
clu
s
ter
in
g
ap
p
r
o
a
ch
es [
2
5
]
-
[
2
8
].
6.
RE
S
E
ARCH
CH
A
L
L
E
NG
E
S IN
CL
US
T
E
R
T
RAC
E
A
NALYS
I
S
T
h
er
e
ar
e
s
ev
er
al
ch
allen
g
es in
th
e
Go
o
g
le
clu
s
ter
wo
r
k
lo
ad
tr
ac
es
d
ataset
an
d
its
alter
n
ati
v
e
d
atasets
n
am
ely
Alib
ab
a
T
r
ac
e,
MS
Azu
r
e
T
r
ac
e,
a
n
d
T
e
n
c
en
t T
r
ac
e
.
−
E
d
g
e
clo
u
d
AI
in
f
e
r
en
ce
o
n
I
o
T
d
ata
p
r
esen
ts
n
o
v
el
ch
all
en
g
es.
E
d
g
e
clo
u
d
s
,
lik
e
co
n
v
en
tio
n
al
clo
u
d
p
latf
o
r
m
s
,
will
o
p
er
ate
n
u
m
e
r
o
u
s
ten
a
n
t
ap
p
s
o
n
ea
c
h
s
e
r
v
er
.
T
h
ese
a
p
p
licatio
n
s
u
tili
ze
ed
g
e
s
er
v
er
h
ar
d
war
e,
s
u
ch
as
ac
ce
ler
ato
r
s
.
T
r
ad
itio
n
al
r
eso
u
r
ce
s
li
k
e
C
PUs
an
d
s
er
v
e
r
GPUs
ac
ce
p
t
v
i
r
tu
aliza
tio
n
f
o
r
ap
p
licatio
n
m
u
ltip
lex
in
g
,
wh
i
le
ed
g
e
ac
ce
ler
ato
r
s
d
o
n
o
t.
Usi
n
g
a
DNN
ac
ce
ler
ato
r
f
o
r
s
ev
er
al
ten
an
t
ap
p
licatio
n
s
m
ig
h
t
ca
u
s
e
p
e
r
f
o
r
m
a
n
ce
in
ter
f
er
e
n
ce
o
win
g
to
a
lack
o
f
is
o
latio
n
m
e
th
o
d
s
,
s
u
ch
as
v
ir
tu
aliz
atio
n
.
T
h
is
ca
n
n
eg
ativ
ely
im
p
ac
t
r
esp
o
n
s
e
tim
es
f
o
r
laten
cy
-
s
en
s
itiv
e
I
o
T
ap
p
licatio
n
s
.
Dev
elo
p
in
g
n
o
v
el
clu
s
ter
r
eso
u
r
ce
m
an
ag
e
m
en
t
s
tr
ateg
ies
is
n
ec
ess
ar
y
to
ef
f
icien
tly
m
u
ltip
lex
s
h
ar
e
d
ed
g
e
clo
u
d
r
eso
u
r
ce
s
f
o
r
laten
cy
-
s
en
s
itiv
e
ap
p
licatio
n
s
[2
3
]
.
−
Sev
e
r
al
r
esear
ch
er
s
f
o
u
n
d
th
at
an
aly
zin
g
clu
s
ter
wo
r
k
l
o
ad
tr
ac
es
o
n
a
s
in
g
le
p
latf
o
r
m
m
ig
h
t
b
e
ch
allen
g
in
g
d
u
e
to
v
ar
iab
ilit
y
,
lim
itin
g
th
e
ap
p
licab
ilit
y
o
f
an
ap
p
r
o
ac
h
.
Mo
s
t
m
icr
o
s
er
v
ices
in
th
e
Alib
ab
a
clu
s
ter
ar
e
m
o
r
e
s
en
s
itiv
e
to
C
PU
th
an
m
em
o
r
y
,
ac
co
r
d
in
g
t
o
an
an
aly
s
is
o
f
Alib
ab
a
tr
ac
e.
Su
ch
ex
p
lo
r
atio
n
is
en
co
u
r
a
g
ed
b
y
t
r
ac
e
cr
o
s
s
-
p
latf
o
r
m
co
m
p
a
r
is
o
n
[
2
5
].
−
Alth
o
u
g
h
th
e
Go
o
g
le
clu
s
ter
d
ataset
2
0
1
9
h
as
b
ee
n
an
al
y
z
ed
,
in
s
ig
h
ts
r
em
ain
u
n
ex
p
lo
r
e
d
.
Ou
r
s
o
lu
tio
n
in
v
o
lv
es u
s
in
g
AI
an
d
GPU
p
o
wer
to
esti
m
ate
jo
b
lim
its
,
a
s
ig
n
if
ican
t c
h
allen
g
e
in
clo
u
d
s
y
s
tem
s
[
9
]
.
−
T
h
e
s
ea
r
ch
f
o
r
n
ew
clu
s
ter
s
was
attem
p
ted
in
th
e
2
0
1
9
Go
o
g
le
C
lu
s
ter
d
ataset,
m
ea
s
u
r
ed
at
2
.
4
ter
ab
y
tes,
h
u
g
e
i
n
co
m
p
ar
is
o
n
with
its
p
r
ed
ec
ess
o
r
th
at
m
ea
s
u
r
ed
o
n
ly
at
4
4
g
ig
a
b
y
tes.
T
h
e
r
ef
o
r
e,
p
r
o
p
er
ef
f
o
r
t
h
as
to
b
e
p
u
t
b
eh
i
n
d
tim
e
a
n
d
r
eso
u
r
ce
s
to
m
an
e
u
v
er
t
h
e
a
m
o
u
n
t
o
f
d
ata
p
r
o
ce
s
s
ed
.
T
o
o
v
er
co
m
e
th
ese
ch
allen
g
es,
a
G
o
o
g
le
c
lo
u
d
to
o
l
ca
lled
B
ig
Qu
er
y
was
u
s
ed
.
I
t
en
a
b
les
co
m
p
lex
s
ea
r
ch
es
o
n
th
e
Go
o
g
le
C
lu
s
ter
d
ataset
with
o
u
t h
av
in
g
to
d
o
wn
l
o
a
d
th
e
e
n
tire
tr
ac
e
[
9
]
.
−
Go
o
g
le
C
lu
s
ter
tr
ac
e
is
an
im
a
g
e
o
f
th
eir
d
ata
ce
n
ter
o
p
er
atio
n
s
.
As in
f
r
astru
ctu
r
e
d
ev
elo
p
s
,
th
is
tr
ac
e
m
ay
b
ec
o
m
e
o
u
td
ated
an
d
lim
ited
.
W
ith
o
u
t
s
tan
d
ar
d
s
o
r
p
r
o
ce
d
u
r
es,
d
ec
is
io
n
-
m
ak
i
n
g
m
ig
h
t
b
e
ch
allen
g
in
g
to
u
n
d
er
s
tan
d
.
Priv
ac
y
an
d
co
n
f
id
en
tiality
co
n
ce
r
n
s
m
ak
e
Go
o
g
le
T
r
ac
e
ch
allen
g
i
n
g
.
I
t
is
ch
allen
g
in
g
to
m
an
ag
e
an
d
an
aly
ze
h
u
g
e
am
o
u
n
ts
o
f
d
ata
in
lar
g
e
-
s
ca
le
s
y
s
tem
s
,
d
etec
t
v
alu
ab
le
p
atter
n
s
,
an
d
g
en
e
r
ate
tr
u
s
two
r
th
y
r
esu
lts
.
Gen
er
aliza
b
ilit
y
m
ay
b
e
r
estricte
d
b
y
Go
o
g
le
’
s
r
u
l
es,
w
h
ich
m
a
y
d
i
f
f
er
s
ig
n
if
ican
tl
y
f
r
o
m
th
o
s
e
o
f
o
th
e
r
b
u
s
in
ess
es.
An
aly
zin
g
an
d
ca
teg
o
r
izin
g
th
e
Go
o
g
le
C
lu
s
ter
tr
ac
e
p
r
o
v
id
es
v
alu
ab
le
in
s
ig
h
ts
in
to
m
an
ag
in
g
lar
g
e
-
s
ca
le
d
is
tr
ib
u
ted
s
y
s
tem
s
,
s
u
ch
as c
lo
u
d
co
m
p
u
tin
g
a
n
d
d
ata
c
en
ter
s
[
1
0
]
.
−
R
esear
ch
er
s
wan
t
to
u
s
e
u
n
s
u
p
er
v
is
ed
lear
n
in
g
tech
n
iq
u
es
to
an
aly
ze
wo
r
k
ca
teg
o
r
ies.
A
ch
allen
g
e
ar
is
es
wh
en
d
ea
lin
g
with
m
illi
o
n
s
o
f
m
u
lti
-
d
im
en
s
io
n
al
d
ata
item
s
.
Du
e
to
th
e
cu
r
s
e
o
f
d
im
en
s
io
n
ality
,
with
lar
g
er
d
ata
s
ets,
ea
ch
d
im
en
s
i
o
n
g
r
ea
tly
i
n
cr
ea
s
es
p
r
o
ce
s
s
in
g
co
s
ts
.
Ad
d
itio
n
ally
,
d
ata
wit
h
s
ev
er
e
o
u
tlier
s
an
d
h
ea
v
y
b
ias
m
ig
h
t
b
ias
t
h
e
clu
s
ter
in
g
alg
o
r
ith
m
.
T
h
e
s
e
ch
allen
g
es
ar
e
o
v
er
co
m
e
b
y
r
e
d
u
cin
g
th
e
m
etr
ics
to
th
r
ee
b
ased
o
n
th
e
o
u
tco
m
e
o
f
co
r
r
elatio
n
an
aly
s
is
.
Oth
er
s
tr
ateg
ies
in
clu
d
e
s
c
alin
g
o
f
o
u
tlier
s
u
s
in
g
B
o
x
-
C
o
x
tr
an
s
f
o
r
m
atio
n
an
d
ce
n
ter
i
n
g
th
e
d
ata
b
y
a
s
tan
d
ar
d
s
ca
ler
wh
ich
also
elim
in
ates
s
tan
d
ar
d
d
ev
iatio
n
[
1
3
].
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
1
1
6
7
-
1
1
7
8
1174
−
Ou
r
s
ch
ed
u
ler
p
o
licy
d
esig
n
h
as
ch
allen
g
es,
m
a
n
y
o
f
w
h
ich
ar
e
y
et
u
n
ad
d
r
ess
ed
.
N
ex
t,
s
o
m
e
o
p
en
ch
allen
g
es
th
at
m
ay
ar
is
e
in
GPU
clu
s
ter
s
with
h
e
ter
o
g
en
eo
u
s
m
ac
h
in
es
ar
e
g
o
i
n
g
to
b
e
d
ea
lt
with
:
m
is
m
atch
ed
s
p
ec
if
icat
io
n
s
an
d
r
eq
u
est
f
o
r
in
s
ta
n
ce
s
,
o
v
er
cr
o
w
d
in
g
lo
w
-
p
e
r
f
o
r
m
an
ce
GPU
co
m
p
u
ter
s
,
a
n
d
im
b
ala
n
ce
d
lo
ad
s
in
h
ig
h
-
en
d
m
ac
h
in
es [
1
7
].
−
T
h
is
ch
allen
g
e
h
elp
s
clo
u
d
s
er
v
ice
co
m
p
an
ies
s
av
e
ex
p
en
s
es.
T
h
e
s
u
s
tain
ab
le
d
ev
elo
p
m
en
t
g
o
als
ar
e
also
s
u
p
p
o
r
ted
.
T
o
tr
ain
th
is
m
o
d
e
l,
co
n
s
id
er
u
s
in
g
alter
n
ativ
e
d
atasets
.
T
r
an
s
f
er
lear
n
in
g
tech
n
iq
u
es
m
ay
b
e
u
s
ed
to
cr
ea
te
h
i
g
h
-
q
u
ality
p
r
e
d
ictio
n
m
o
d
els f
o
r
clo
u
d
r
eso
u
r
ce
m
an
ag
em
e
n
t
[
1
4
]
.
−
Stra
g
g
ler
task
s
,
wh
ich
ar
e
s
o
m
etim
es
r
elate
d
to
th
e
lo
n
g
tail
p
r
o
b
lem
,
is
s
till
a
b
u
r
n
in
g
is
s
u
e
in
clu
s
ter
co
m
p
u
tin
g
s
in
ce
o
n
ly
a
f
ew
s
l
o
w
r
u
n
n
in
g
task
s
ca
n
g
r
ea
tly
p
o
s
tp
o
n
e
th
e
ex
ec
u
tio
n
o
f
a
wh
o
le
jo
b
.
T
h
ese
s
tr
ag
g
ler
s
h
av
e
ad
v
e
r
s
e
ef
f
ec
ts
o
n
th
e
p
er
f
o
r
m
an
ce
an
d
r
e
s
o
u
r
c
e
ef
f
icien
cy
o
f
th
e
s
y
s
tem
as
a
wh
o
le.
C
o
n
s
eq
u
en
tly
,
p
r
o
p
er
i
d
en
tifi
ca
tio
n
an
d
co
n
tr
o
l
o
f
s
u
ch
ac
tiv
ities
ar
e
cr
itical,
an
d
s
u
p
er
io
r
f
r
a
m
ewo
r
k
s
an
d
s
tr
ateg
ies
ar
e
n
ec
ess
ar
y
t
o
m
ax
im
ize
r
eso
u
r
ce
allo
ca
tio
n
an
d
en
h
an
ce
t
h
e
ef
f
icien
cy
o
f
th
e
s
y
s
tem
in
g
en
er
al
[2
9
].
7.
CO
NCLU
SI
O
N
T
h
e
s
u
r
v
ey
h
as
ex
am
in
ed
Go
o
g
le
clu
s
ter
wo
r
k
lo
a
d
tr
ac
e
d
atasets
o
f
2
0
1
1
an
d
2
0
1
9
,
an
d
th
e
o
th
er
d
ata
s
ets
in
clu
d
in
g
t
h
e
Alib
ab
a,
MS
Azu
r
e
a
n
d
T
e
n
ce
n
t
clu
s
ter
tr
ac
e
s
.
I
t
co
m
p
ar
es
th
e
2
0
1
9
Go
o
g
le
tr
ac
e
t
o
b
e
lar
g
er
,
m
o
r
e
v
a
r
ied
,
an
d
cl
o
s
er
to
th
e
m
o
d
e
r
n
p
r
o
d
u
ctio
n
wo
r
k
lo
ad
s
co
m
p
ar
ed
t
o
th
e
2
0
1
1
tr
ac
e,
with
th
e
alter
n
ativ
e
d
atasets
o
f
f
er
in
g
s
u
p
p
lem
en
tar
y
in
s
ig
h
ts
o
f
co
-
l
o
ca
ted
wo
r
k
lo
a
d
s
,
clo
u
d
o
p
er
atio
n
s
,
an
d
lar
g
e
-
s
ca
le
s
ch
ed
u
lin
g
b
eh
av
i
o
r
.
N
ev
er
th
eless
,
th
er
e
ar
e
a
n
u
m
b
er
o
f
g
ap
s
in
r
esear
c
h
in
th
e
liter
atu
r
e.
T
o
b
eg
i
n
with
,
th
e
m
ajo
r
ity
o
f
th
e
cu
r
r
en
t
s
tu
d
ies
ar
e
d
ed
icate
d
to
o
n
e
d
ataset,
an
d
th
er
e
is
n
o
co
m
m
o
n
l
y
ad
o
p
ted
cr
o
s
s
-
tr
ac
e
b
en
c
h
m
ar
k
i
n
g
f
r
a
m
ewo
r
k
t
o
ass
ess
th
e
wo
r
k
lo
ad
an
aly
s
is
tech
n
i
q
u
es
in
v
ar
io
u
s
clo
u
d
s
ettin
g
s
.
Seco
n
d
,
p
r
ep
r
o
c
ess
in
g
c
o
n
v
e
n
tio
n
s
,
s
ch
em
a
m
ap
p
in
g
,
an
d
tr
ac
e
f
ilter
in
g
v
ar
y
s
ig
n
i
f
ican
tly
b
etwe
en
d
atasets
r
estrictin
g
r
ep
r
o
d
u
cib
ilit
y
an
d
d
ir
ec
t
co
m
p
ar
is
o
n
.
T
h
ir
d
,
th
e
tem
p
o
r
al
v
ar
iatio
n
s
in
th
e
n
a
tu
r
e
o
f
wo
r
k
lo
ad
s
,
s
ch
ed
u
le
p
atter
n
s
a
n
d
r
eso
u
r
ce
co
n
s
u
m
p
ti
o
n
ar
e
n
o
t
y
et
ad
eq
u
ately
i
n
v
esti
g
ated
,
p
ar
ti
cu
lar
ly
b
etwe
en
th
e
2
0
1
1
an
d
2
0
1
9
Go
o
g
le
tr
a
ce
s
.
L
astl
y
,
f
u
tu
r
e
s
tu
d
ies
o
u
g
h
t
t
o
f
o
c
u
s
o
n
tr
a
n
s
f
er
a
b
ilit
y
,
s
tan
d
ar
d
ized
m
ea
s
u
r
em
en
ts
an
d
s
tan
d
ar
d
i
ze
d
ev
alu
atio
n
p
r
o
ce
d
u
r
es
to
en
s
u
r
e
th
at
tr
ac
e
-
b
ased
m
o
d
els
ca
n
b
e
test
ed
ef
f
ic
ien
tly
u
s
in
g
h
eter
o
g
e
n
eo
u
s
p
r
o
d
u
ctio
n
d
atasets
.
Ov
er
all,
clu
s
ter
tr
ac
e
d
atasets
co
n
tin
u
e
t
o
b
e
cr
u
cial
in
co
m
p
r
eh
en
d
in
g
clo
u
d
w
o
r
k
lo
a
d
b
e
h
av
io
r
,
y
et
th
e
d
is
cip
lin
e
s
till
lack
s
s
tan
d
ar
d
ized
p
r
ac
tices
in
a
n
aly
s
is
an
d
v
ali
d
atio
n
ac
r
o
s
s
d
atasets
.
T
h
e
f
ill
-
in
o
f
th
ese
g
ap
s
will
en
s
u
r
e
th
at
f
u
tu
r
e
t
r
ac
e
-
b
ased
s
ch
ed
u
lin
g
,
p
r
ed
ic
tio
n
,
an
d
r
eso
u
r
ce
-
m
an
a
g
em
e
n
t
s
tu
d
ies
b
ec
o
m
e
m
o
r
e
g
e
n
er
aliza
b
le
an
d
u
s
ef
u
l
in
p
r
ac
tice.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
Au
th
o
r
s
s
tate
n
o
f
u
n
d
in
g
in
v
o
lv
ed
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
tr
ib
u
to
r
R
o
les
T
ax
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
id
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Ak
ash
Patel
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Am
it Na
y
ak
✓
✓
✓
✓
✓
✓
Kh
u
s
h
i Pate
l
✓
✓
✓
✓
✓
An
an
d
Patel
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
ma
l
a
n
a
l
y
si
s
I
:
I
n
v
e
st
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
s
u
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
si
o
n
P
:
P
r
o
j
e
c
t
a
d
m
i
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
s
i
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
d
ec
lar
e
n
o
co
n
f
lict o
f
in
ter
est.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
I
SS
N:
2252
-
8
7
7
6
A
p
ilo
t revie
w
o
f th
e
G
o
o
g
le
cl
u
s
ter w
o
r
klo
a
d
tr
a
ce
2
0
1
9
,
me
th
o
d
o
lo
g
y
a
n
d
its
a
lter
n
a
tives
…
(
A
ka
s
h
P
a
tel
)
1175
DATA AV
AI
L
AB
I
L
I
T
Y
Data
g
en
er
ated
an
d
an
aly
ze
d
d
u
r
in
g
th
is
s
tu
d
y
ar
e
a
v
ailab
le
f
r
o
m
t
h
e
co
r
r
esp
o
n
d
i
n
g
a
u
th
o
r
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
T
h
e
d
ata
,
wh
ich
co
n
tain
in
f
o
r
m
atio
n
t
h
at
co
u
ld
c
o
m
p
r
o
m
is
e
th
e
p
r
iv
ac
y
o
f
r
esear
ch
p
ar
ticip
an
ts
,
ar
e
n
o
t
p
u
b
licl
y
av
ailab
le
d
u
e
to
ce
r
tain
r
estrictio
n
s
.
R
eq
u
ests
s
h
o
u
ld
b
e
d
ir
ec
ted
to
th
e
co
r
r
esp
o
n
d
in
g
a
u
th
o
r
’
s
co
n
tac
t d
etails.
RE
F
E
R
E
NC
E
S
[
1
]
F
.
H
.
B
a
p
p
y
,
T.
I
sl
a
m
,
T
.
S
.
Za
m
a
n
,
R
.
H
a
s
a
n
,
a
n
d
C
.
C
a
i
c
e
d
o
,
“
A
d
e
e
p
d
i
v
e
i
n
t
o
t
h
e
g
o
o
g
l
e
c
l
u
st
e
r
w
o
r
k
l
o
a
d
t
r
a
c
e
s
:
a
n
a
l
y
z
i
n
g
t
h
e
a
p
p
l
i
c
a
t
i
o
n
f
a
i
l
u
r
e
c
h
a
r
a
c
t
e
r
i
s
t
i
c
s
a
n
d
u
ser
b
e
h
a
v
i
o
r
s,
”
i
n
Pro
c
e
e
d
i
n
g
s
-
2
0
2
3
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
F
u
t
u
re
I
n
t
e
r
n
e
t
o
f
T
h
i
n
g
s
a
n
d
C
l
o
u
d
,
F
i
C
l
o
u
d
2
0
2
3
,
I
EE
E,
A
u
g
.
2
0
2
3
,
p
p
.
1
0
3
–
1
0
8
.
d
o
i
:
1
0
.
1
1
0
9
/
F
i
C
l
o
u
d
5
8
6
4
8
.
2
0
2
3
.
0
0
0
2
3
.
[
2
]
G
o
o
g
l
e
,
“
G
o
o
g
l
e
c
l
u
s
t
e
r
w
o
r
k
l
o
a
d
t
r
a
c
e
s
2
0
1
9
.
”
A
c
c
e
ss
e
d
:
M
a
y
2
0
,
2
0
2
6
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
r
e
se
a
r
c
h
.
g
o
o
g
l
e
/
r
e
s
o
u
r
c
e
s
/
d
a
t
a
set
s/
g
o
o
g
l
e
-
c
l
u
s
t
e
r
-
w
o
r
k
l
o
a
d
-
t
r
a
c
e
s
-
2
0
1
9
/
[
3
]
“
G
o
o
g
l
e
c
l
u
st
e
r
d
a
t
a
2
0
1
1
,
”
G
o
o
g
l
e
.
A
c
c
e
ss
e
d
:
M
a
y
2
0
,
2
0
2
6
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
r
e
sea
r
c
h
.
g
o
o
g
l
e
/
t
o
o
l
s/
d
a
t
a
se
t
s/
c
l
u
s
t
e
r
-
w
o
r
k
l
o
a
d
-
t
r
a
c
e
s/
[
4
]
N
.
D
e
z
h
a
b
a
d
,
S
.
G
a
n
t
i
,
a
n
d
G
.
S
h
o
j
a
,
“
C
l
o
u
d
w
o
r
k
l
o
a
d
c
h
a
r
a
c
t
e
r
i
z
a
t
i
o
n
a
n
d
p
r
o
f
i
l
i
n
g
f
o
r
r
e
s
o
u
r
c
e
a
l
l
o
c
a
t
i
o
n
,
”
i
n
Pr
o
c
e
e
d
i
n
g
o
f
t
h
e
2
0
1
9
I
EEE
8
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
l
o
u
d
N
e
t
w
o
rk
i
n
g
,
C
l
o
u
d
N
e
t
2
0
1
9
,
I
EEE,
N
o
v
.
2
0
1
9
,
p
p
.
1
–
4
,
d
o
i
:
1
0
.
1
1
0
9
/
C
l
o
u
d
N
e
t
4
7
6
0
4
.
2
0
1
9
.
9
0
6
4
1
3
8
.
[
5
]
A
l
i
b
a
b
a
,
“
C
l
u
s
t
e
r
d
a
t
a
c
o
l
l
e
c
t
e
d
f
r
o
m
p
r
o
d
u
c
t
i
o
n
c
l
u
s
t
e
r
s
i
n
A
l
i
b
a
b
a
f
o
r
c
l
u
st
e
r
m
a
n
a
g
e
me
n
t
r
e
sea
r
c
h
.
”
A
c
c
e
sse
d
:
M
a
y
2
0
,
2
0
2
6
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s:
/
/
g
i
t
h
u
b
.
c
o
m/
a
l
i
b
a
b
a
/
c
l
u
s
t
e
r
d
a
t
a
[
6
]
H
.
B
o
mm
a
l
a
,
V
.
U
ma
M
a
h
e
sw
a
r
i
,
R
.
A
l
u
v
a
l
u
,
a
n
d
S
.
M
u
d
r
a
k
o
l
a
,
“
M
a
c
h
i
n
e
l
e
a
r
n
i
n
g
j
o
b
f
a
i
l
u
r
e
a
n
a
l
y
s
i
s
a
n
d
p
r
e
d
i
c
t
i
o
n
mo
d
e
l
f
o
r
t
h
e
c
l
o
u
d
e
n
v
i
r
o
n
men
t
,
”
H
i
g
h
-
C
o
n
f
i
d
e
n
c
e
C
o
m
p
u
t
i
n
g
,
v
o
l
.
3
,
n
o
.
4
,
p
.
1
0
0
1
6
5
,
D
e
c
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
h
c
c
.
2
0
2
3
.
1
0
0
1
6
5
.
[
7
]
V
.
E
.
J
y
o
t
h
i
,
S
.
A
l
u
g
o
j
u
,
K
.
I
n
d
r
a
j
a
,
N
.
S
.
C
h
o
w
d
a
r
y
,
A
.
M
a
d
h
u
r
i
,
a
n
d
S
.
S
i
n
d
h
u
r
a
,
“
A
d
a
p
t
i
v
e
c
l
o
u
d
l
o
a
d
b
a
l
a
n
c
i
n
g
w
i
t
h
r
e
i
n
f
o
r
c
e
m
e
n
t
l
e
a
r
n
i
n
g
:
l
e
v
e
r
a
g
i
n
g
g
o
o
g
l
e
c
l
u
s
t
e
r
d
a
t
a
,
”
i
n
5
t
h
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
El
e
c
t
r
o
n
i
c
s
a
n
d
S
u
s
t
a
i
n
a
b
l
e
C
o
m
m
u
n
i
c
a
t
i
o
n
S
y
s
t
e
m
s,
I
C
E
S
C
2
0
2
4
-
Pro
c
e
e
d
i
n
g
s
,
I
EEE,
A
u
g
.
2
0
2
4
,
p
p
.
5
7
2
–
579
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
ESC
6
0
8
5
2
.
2
0
2
4
.
1
0
6
8
9
9
7
3
.
[
8
]
M
.
Y
i
l
d
i
z
a
n
d
A
.
B
a
i
o
c
c
h
i
,
“
D
a
t
a
-
d
r
i
v
e
n
w
o
r
k
l
o
a
d
g
e
n
e
r
a
t
i
o
n
b
a
s
e
d
o
n
g
o
o
g
l
e
d
a
t
a
c
e
n
t
e
r
me
a
su
r
e
me
n
t
s
,
”
i
n
I
EEE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
H
i
g
h
Pe
r
f
o
rm
a
n
c
e
S
w
i
t
c
h
i
n
g
a
n
d
R
o
u
t
i
n
g
,
H
PS
R
,
I
EEE,
J
u
l
.
2
0
2
4
,
p
p
.
1
4
3
–
148
,
d
o
i
:
1
0
.
1
1
0
9
/
H
P
S
R
6
2
4
4
0
.
2
0
2
4
.
1
0
6
3
5
9
2
5
.
[
9
]
D
.
S
h
a
h
mi
r
z
a
d
i
,
N
.
K
h
a
l
e
d
i
a
n
,
a
n
d
A
.
M
.
R
a
h
ma
n
i
,
“
A
n
a
l
y
z
i
n
g
t
h
e
i
m
p
a
c
t
o
f
v
a
r
i
o
u
s p
a
r
a
me
t
e
r
s o
n
j
o
b
sc
h
e
d
u
l
i
n
g
i
n
t
h
e
G
o
o
g
l
e
c
l
u
st
e
r
d
a
t
a
se
t
,
”
C
l
u
s
t
e
r
C
o
m
p
u
t
i
n
g
,
v
o
l
.
2
7
,
n
o
.
6
,
p
p
.
7
6
7
3
–
7
6
8
7
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s1
0
5
8
6
-
0
2
4
-
0
4
3
7
7
-
8.
[
1
0
]
S
.
Jo
ma
h
a
n
d
S
.
A
j
i
,
“
G
o
o
g
l
e
c
l
o
u
d
t
r
a
c
e
:
c
h
a
r
a
c
t
e
r
i
z
a
t
i
o
n
o
f
t
e
r
mi
n
a
t
e
d
j
o
b
s,
”
i
n
Pro
c
e
e
d
i
n
g
s
-
2
n
d
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ad
v
a
n
c
e
m
e
n
t
i
n
C
o
m
p
u
t
a
t
i
o
n
a
n
d
C
o
m
p
u
t
e
r
T
e
c
h
n
o
l
o
g
i
e
s
,
I
n
C
AC
C
T
2
0
2
4
,
I
EEE,
M
a
y
2
0
2
4
,
p
p
.
5
1
7
–
5
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
I
n
C
A
C
C
T6
1
5
9
8
.
2
0
2
4
.
1
0
5
5
1
1
4
6
.
[
1
1
]
S
.
G
u
p
t
a
a
n
d
A
.
D
.
D
i
l
e
e
p
,
“
Lo
n
g
r
a
n
g
e
d
e
p
e
n
d
e
n
c
e
i
n
c
l
o
u
d
ser
v
e
r
s
:
a
st
a
t
i
s
t
i
c
a
l
a
n
a
l
y
si
s
b
a
se
d
o
n
G
o
o
g
l
e
w
o
r
k
l
o
a
d
t
r
a
c
e
,
”
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
0
2
,
n
o
.
4
,
p
p
.
1
0
3
1
–
1
0
4
9
,
A
p
r
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
6
0
7
-
0
1
9
-
0
0
7
7
9
-
4.
[
1
2
]
Y
.
Li
n
,
A
.
B
a
r
k
e
r
,
a
n
d
S
.
C
e
e
s
a
y
,
“
E
x
p
l
o
r
i
n
g
c
h
a
r
a
c
t
e
r
i
st
i
c
s
o
f
i
n
t
e
r
-
c
l
u
s
t
e
r
mac
h
i
n
e
s
a
n
d
c
l
o
u
d
a
p
p
l
i
c
a
t
i
o
n
s
o
n
G
o
o
g
l
e
C
l
u
st
e
r
s,
”
i
n
Pr
o
c
e
e
d
i
n
g
s
-
2
0
2
0
I
EEE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Bi
g
D
a
t
a
,
Bi
g
D
a
t
a
2
0
2
0
,
I
EEE,
D
e
c
.
2
0
2
0
,
p
p
.
2
7
8
5
–
2
7
9
4
,
d
o
i
:
1
0
.
1
1
0
9
/
B
i
g
D
a
t
a
5
0
0
2
2
.
2
0
2
0
.
9
3
7
7
8
0
2
.
[
1
3
]
Q
.
W
e
n
g
e
t
a
l
.
,
“
M
La
a
S
i
n
t
h
e
w
i
l
d
:
w
o
r
k
l
o
a
d
a
n
a
l
y
s
i
s
a
n
d
sc
h
e
d
u
l
i
n
g
i
n
l
a
r
g
e
-
s
c
a
l
e
h
e
t
e
r
o
g
e
n
e
o
u
s
G
P
U
c
l
u
st
e
r
s,
”
Pr
o
c
e
e
d
i
n
g
s
o
f
t
h
e
1
9
t
h
U
S
EN
I
X
S
y
m
p
o
s
i
u
m
o
n
N
e
t
w
o
r
k
e
d
S
y
s
t
e
m
s De
si
g
n
a
n
d
I
m
p
l
e
m
e
n
t
a
t
i
o
n
,
N
S
D
I
2
0
2
2
,
p
p
.
9
4
5
–
9
6
0
,
2
0
2
2
.
[
1
4
]
T.
N
.
T
.
A
sm
a
w
i
,
A
.
I
smai
l
,
a
n
d
J.
S
h
e
n
,
“
C
l
o
u
d
f
a
i
l
u
r
e
p
r
e
d
i
c
t
i
o
n
b
a
se
d
o
n
t
r
a
d
i
t
i
o
n
a
l
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
,
”
J
o
u
rn
a
l
o
f
C
l
o
u
d
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
1
,
n
o
.
1
,
p
.
4
7
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
1
1
8
6
/
s
1
3
6
7
7
-
0
2
2
-
0
0
3
2
7
-
0.
[
1
5
]
S
.
M
.
A
l
i
a
n
d
G
.
K
e
c
sk
e
me
t
i
,
“
EF
e
c
t
i
o
n
:
e
f
f
e
c
t
i
v
e
n
e
ss
d
e
t
e
c
t
i
o
n
t
e
c
h
n
i
q
u
e
f
o
r
c
l
u
st
e
r
i
n
g
c
l
o
u
d
w
o
r
k
l
o
a
d
t
r
a
c
e
s,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
m
p
u
t
a
t
i
o
n
a
l
I
n
t
e
l
l
i
g
e
n
c
e
S
y
st
e
m
s
,
v
o
l
.
1
7
,
n
o
.
1
,
p
.
2
0
2
,
A
u
g
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
4
1
9
6
-
0
2
4
-
0
0
6
1
8
-
1.
[
1
6
]
A
.
U
mer,
A
.
N
.
M
i
a
n
,
a
n
d
O
.
R
a
n
a
,
“
P
r
e
d
i
c
t
i
n
g
m
a
c
h
i
n
e
b
e
h
a
v
i
o
r
f
r
o
m
G
o
o
g
l
e
c
l
u
st
e
r
w
o
r
k
l
o
a
d
t
r
a
c
e
s
,
”
C
o
n
c
u
r
re
n
c
y
a
n
d
C
o
m
p
u
t
a
t
i
o
n
:
Pr
a
c
t
i
c
e
a
n
d
Ex
p
e
ri
e
n
c
e
,
v
o
l
.
3
5
,
n
o
.
5
,
F
e
b
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
2
/
c
p
e
.
7
5
5
9
.
[
1
7
]
R
.
S
h
i
a
n
d
C
.
Ji
a
n
g
,
“
T
h
r
e
e
-
w
a
y
e
n
s
e
mb
l
e
p
r
e
d
i
c
t
i
o
n
f
o
r
w
o
r
k
l
o
a
d
i
n
t
h
e
d
a
t
a
c
e
n
t
e
r
,
”
I
EEE
A
c
c
e
ss
,
v
o
l
.
1
0
,
p
p
.
1
0
0
2
1
–
1
0
0
3
0
,
2
0
2
2
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
2
.
3
1
4
5
4
2
6
.
[
1
8
]
Y
.
Li
a
n
g
,
N
.
R
u
a
n
,
L.
Y
i
,
a
n
d
X
.
S
u
,
“
A
n
a
p
p
r
o
a
c
h
t
o
w
o
r
k
l
o
a
d
g
e
n
e
r
a
t
i
o
n
f
o
r
mo
d
e
r
n
d
a
t
a
c
e
n
t
e
r
s
:
a
v
i
e
w
f
r
o
m
A
l
i
b
a
b
a
t
r
a
c
e
,
”
Be
n
c
h
C
o
u
n
c
i
l
T
r
a
n
sa
c
t
i
o
n
s
o
n
B
e
n
c
h
m
a
r
k
s,
S
t
a
n
d
a
r
d
s
a
n
d
Ev
a
l
u
a
t
i
o
n
s
,
v
o
l
.
4
,
n
o
.
1
,
p
.
1
0
0
1
6
4
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
t
b
e
n
c
h
.
2
0
2
4
.
1
0
0
1
6
4
.
[
1
9
]
C
.
K
a
t
sa
k
i
o
r
i
s
,
C
.
A
l
v
e
r
t
i
,
K
.
N
i
k
a
s,
D
.
S
i
a
k
a
v
a
r
a
s
,
S
.
P
s
o
ma
d
a
k
i
s
,
a
n
d
N
.
K
o
z
i
r
i
s,
“
F
a
a
S
R
a
i
l
:
e
m
p
l
o
y
i
n
g
r
e
a
l
w
o
r
k
l
o
a
d
s
t
o
g
e
n
e
r
a
t
e
r
e
p
r
e
se
n
t
a
t
i
v
e
l
o
a
d
f
o
r
serv
e
r
l
e
ss
r
e
se
a
r
c
h,
”
i
n
H
PD
C
2
0
2
4
-
P
ro
c
e
e
d
i
n
g
s
o
f
t
h
e
3
3
rd
I
n
t
e
r
n
a
t
i
o
n
a
l
S
y
m
p
o
si
u
m
o
n
H
i
g
h
-
Pe
r
f
o
rm
a
n
c
e
Pa
r
a
l
l
e
l
a
n
d
D
i
st
ri
b
u
t
e
d
C
o
m
p
u
t
i
n
g
,
N
e
w
Y
o
r
k
,
N
Y
,
U
S
A
:
A
C
M
,
J
u
n
.
2
0
2
4
,
p
p
.
2
1
4
–
226
,
d
o
i
:
1
0
.
1
1
4
5
/
3
6
2
5
5
4
9
.
3
6
5
8
6
8
4
.
[
2
0
]
J.
Li
,
Q
.
W
a
n
g
,
P
.
P
.
C
.
Le
e
,
a
n
d
C
.
S
h
i
,
“
A
n
in
-
d
e
p
t
h
c
o
m
p
a
r
a
t
i
v
e
a
n
a
l
y
s
i
s
o
f
c
l
o
u
d
b
l
o
c
k
s
t
o
r
a
g
e
w
o
r
k
l
o
a
d
s:
f
i
n
d
i
n
g
s
a
n
d
i
mp
l
i
c
a
t
i
o
n
s
,
”
AC
M
T
r
a
n
s
a
c
t
i
o
n
s
o
n
S
t
o
r
a
g
e
,
v
o
l
.
1
9
,
n
o
.
2
,
p
p
.
1
–
3
2
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
7
2
7
7
9
.
[
2
1
]
C
.
Ji
a
n
g
,
G
.
H
a
n
,
J.
L
i
n
,
G
.
Ji
a
,
W
.
S
h
i
,
a
n
d
J.
W
a
n
,
“
C
h
a
r
a
c
t
e
r
i
st
i
c
s
o
f
C
o
-
a
l
l
o
c
a
t
e
d
o
n
l
i
n
e
s
e
r
v
i
c
e
s
a
n
d
b
a
t
c
h
j
o
b
s
i
n
i
n
t
e
r
n
e
t
d
a
t
a
c
e
n
t
e
r
s
:
a
c
a
se
st
u
d
y
f
r
o
m
A
l
i
b
a
b
a
c
l
o
u
d
,
”
I
E
E
E
Ac
c
e
ss
,
v
o
l
.
7
,
p
p
.
2
2
4
9
5
–
2
2
5
0
8
,
2
0
1
9
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
1
9
.
2
8
9
7
8
9
8
.
[
2
2
]
Q
.
W
a
n
g
,
Y
.
Ti
a
n
,
X
.
Y
u
,
L.
D
i
n
g
,
a
n
d
X
.
Z
h
a
n
g
,
“
W
h
e
r
e
i
s
t
h
e
Tr
a
f
f
i
c
G
o
i
n
g
?
A
c
o
mp
a
r
a
t
i
v
e
s
t
u
d
y
o
f
c
l
o
u
d
s
f
o
l
l
o
w
i
n
g
d
i
f
f
e
r
e
n
t
d
e
s
i
g
n
s,
”
I
EE
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
S
e
rv
i
c
e
s
C
o
m
p
u
t
i
n
g
,
v
o
l
.
1
6
,
n
o
.
2
,
p
p
.
1
4
7
3
–
1
4
8
4
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
TSC.
2
0
2
2
.
3
1
8
2
0
4
7
.
[
2
3
]
Q
.
Li
a
n
g
,
W
.
A
.
H
a
n
a
f
y
,
A
.
A
l
i
-
E
l
d
i
n
,
a
n
d
P
.
S
h
e
n
o
y
,
“
M
o
d
e
l
-
d
r
i
v
e
n
c
l
u
st
e
r
r
e
s
o
u
r
c
e
ma
n
a
g
e
m
e
n
t
f
o
r
A
I
w
o
r
k
l
o
a
d
s
i
n
e
d
g
e
c
l
o
u
d
s,
”
A
C
M
T
ra
n
s
a
c
t
i
o
n
s
o
n
Au
t
o
n
o
m
o
u
s
a
n
d
Ad
a
p
t
i
v
e
S
y
s
t
e
m
s
,
v
o
l
.
1
8
,
n
o
.
1
,
p
p
.
1
–
2
6
,
M
a
r
.
2
0
2
3
,
d
o
i
:
1
0
.
1
1
4
5
/
3
5
8
2
0
8
0
.
[
2
4
]
L.
R
u
a
n
,
X
.
X
u
,
L.
X
i
a
o
,
L
.
R
e
n
,
N
.
M
i
n
-
A
l
l
a
h
,
a
n
d
Y
.
X
u
e
,
“
E
v
a
l
u
a
t
i
n
g
p
e
r
f
o
r
m
a
n
c
e
v
a
r
i
a
t
i
o
n
s
c
r
o
ss
c
l
o
u
d
d
a
t
a
c
e
n
t
r
e
s
u
si
n
g
mu
l
t
i
v
i
e
w
c
o
m
p
a
r
a
t
i
v
e
w
o
r
k
l
o
a
d
t
r
a
c
e
s
a
n
a
l
y
s
i
s,
”
C
o
n
n
e
c
t
i
o
n
S
c
i
e
n
c
e
,
v
o
l
.
3
4
,
n
o
.
1
,
p
p
.
1
5
8
2
–
1
6
0
8
,
D
e
c
.
2
0
2
2
,
d
o
i
:
1
0
.
1
0
8
0
/
0
9
5
4
0
0
9
1
.
2
0
2
1
.
2
0
1
5
2
8
9
.
[
2
5
]
L.
S
l
i
w
k
o
,
“
C
l
u
s
t
e
r
w
o
r
k
l
o
a
d
a
l
l
o
c
a
t
i
o
n
:
a
p
r
e
d
i
c
t
i
v
e
a
p
p
r
o
a
c
h
l
e
v
e
r
a
g
i
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
e
f
f
i
c
i
e
n
c
y
,
”
I
EE
E
Ac
c
e
ss
,
v
o
l
.
1
2
,
p
p
.
1
9
4
0
9
1
–
1
9
4
1
0
7
,
2
0
2
4
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
4
.
3
5
2
0
4
2
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
7
7
6
I
n
t J I
n
f
&
C
o
m
m
u
n
T
ec
h
n
o
l
,
Vo
l.
1
5
,
No
.
3
,
Sep
tem
b
er
20
2
6
:
1
1
6
7
-
1
1
7
8
1176
[2
6
]
S
.
O
f
e
k
a
n
d
A
.
S
o
me
c
h
,
“
E
x
p
l
a
i
n
i
n
g
b
l
a
c
k
-
b
o
x
c
l
u
st
e
r
i
n
g
p
i
p
e
l
i
n
e
s
w
i
t
h
c
l
u
st
e
r
-
e
x
p
l
o
r
e
r
,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
V
L
D
B
E
n
d
o
w
m
e
n
t
,
v
o
l
.
1
8
,
n
o
.
5
,
p
p
.
1
4
9
5
–
1
5
0
8
,
J
a
n
.
2
0
2
5
,
d
o
i
:
1
0
.
1
4
7
7
8
/
3
7
1
8
0
5
7
.
3
7
1
8
0
7
5
.
[
2
7
]
A
.
G
i
a
n
n
o
u
l
i
d
i
s,
A
.
G
o
u
n
a
r
i
s,
a
n
d
J.
P
a
p
a
r
r
i
z
o
s,
“
B
U
R
S
T:
r
e
n
d
e
r
i
n
g
c
l
u
s
t
e
r
i
n
g
t
e
c
h
n
i
q
u
e
s
su
i
t
a
b
l
e
f
o
r
e
v
o
l
v
i
n
g
st
r
e
a
ms,
”
Pro
c
e
e
d
i
n
g
s
o
f
t
h
e
VLD
B
E
n
d
o
w
m
e
n
t
,
v
o
l
.
1
8
,
n
o
.
1
1
,
p
p
.
4
0
5
4
–
4
0
6
3
,
Ju
l
.
2
0
2
5
,
d
o
i
:
1
0
.
1
4
7
7
8
/
3
7
4
9
6
4
6
.
3
7
4
9
6
7
5
.
[
2
8
]
J.
A
.
B
e
r
n
a
b
é
-
D
í
a
z
,
M
.
F
r
a
n
c
o
,
J.
M
.
V
i
v
o
,
a
n
d
J.
T
.
F
e
r
n
á
n
d
e
z
-
B
r
e
i
s,
“
O
p
t
i
mi
z
i
n
g
c
l
u
s
t
e
r
i
n
g
-
b
a
se
d
a
n
a
l
y
t
i
c
a
l
me
t
h
o
d
s
w
i
t
h
t
r
i
mm
e
d
a
n
d
s
p
a
r
se
c
l
u
st
e
r
i
n
g
,
”
C
o
m
p
u
t
e
rs
i
n
Bi
o
l
o
g
y
a
n
d
M
e
d
i
c
i
n
e
,
v
o
l
.
1
9
4
,
p
.
1
1
0
4
3
6
,
A
u
g
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
mp
b
i
o
m
e
d
.
2
0
2
5
.
1
1
0
4
3
6
.
[
2
9
]
T.
L
e
D
u
c
,
C.
N
g
u
y
e
n
,
a
n
d
P
.
-
O
.
Ö
st
b
e
r
g
,
“
E
n
e
r
g
y
-
a
w
a
r
e
st
r
a
g
g
l
e
r
t
a
s
k
m
a
n
a
g
e
me
n
t
f
r
a
m
e
w
o
r
k
f
o
r
s
u
st
a
i
n
a
b
l
e
c
l
o
u
d
e
n
v
i
r
o
n
m
e
n
t
,
”
A
u
t
h
o
re
a
Pre
p
ri
n
t
s
,
2
0
2
4
,
d
o
i
:
1
0
.
3
6
2
2
7
/
t
e
c
h
r
x
i
v
.
1
7
3
3
9
8
0
6
9
.
9
8
8
5
2
3
3
2
/
v
.
AP
P
E
NDI
X
T
ab
le
2
.
C
o
m
p
a
r
ativ
e
an
aly
s
is
R
e
f
e
r
e
n
c
e
a
n
d
p
u
b
l
i
c
a
t
i
o
n
w
i
t
h
y
e
a
r
O
b
j
e
c
t
i
v
e
s
o
f
t
h
e
r
e
s
e
a
r
c
h
C
l
u
st
e
r
t
r
a
c
e
d
a
t
a
se
t
u
se
d
F
i
n
d
i
n
g
s
o
f
t
h
e
r
e
se
a
r
c
h
Li
mi
t
a
t
i
o
n
s/
f
u
t
u
r
e
w
o
r
k
[
1
8
]
El
se
v
i
e
r
,
2
0
2
4
To
p
r
o
v
i
d
e
S
TWG
EN
,
a
u
n
i
q
u
e
a
p
p
r
o
a
c
h
t
h
a
t
u
s
e
s
st
a
t
i
st
i
c
a
l
t
r
a
c
e
d
a
t
a
t
o
g
e
n
e
r
a
t
e
w
o
r
k
l
o
a
d
,
t
o
so
l
v
e
t
h
i
s
c
h
a
l
l
e
n
g
e
.
A
l
i
b
a
b
a
Tr
a
c
e
2
0
1
8
S
TWG
EN
a
c
c
u
r
a
t
e
l
y
s
i
mu
l
a
t
e
s
b
a
t
c
h
t
a
s
k
r
e
s
o
u
r
c
e
u
t
i
l
i
z
a
t
i
o
n
b
y
i
n
t
e
g
r
a
t
i
n
g
t
a
s
k
-
l
e
v
e
l
d
a
t
a
w
i
t
h
ma
c
h
i
n
e
-
l
e
v
e
l
c
h
a
l
l
e
n
g
e
s.
Ex
t
e
n
si
v
e
e
x
p
e
r
i
m
e
n
t
s
c
o
n
f
i
r
m
t
h
e
su
p
e
r
i
o
r
i
t
y
o
f
t
h
e
S
TWG
EN
me
t
h
o
d
o
v
e
r
e
x
i
st
i
n
g
b
a
s
e
l
i
n
e
s.
Th
e
s
c
o
p
e
o
f
o
u
r
r
e
se
a
r
c
h
w
i
l
l
b
e
e
x
p
a
n
d
e
d
i
n
t
h
e
f
u
t
u
r
e
t
o
a
c
c
o
u
n
t
f
o
r
w
o
r
k
d
e
p
e
n
d
e
n
c
i
e
s
i
n
w
o
r
k
f
o
r
c
e
d
e
v
e
l
o
p
me
n
t
.
[
1
9
]
A
C
M
,
2
0
2
4
Th
i
s
r
e
se
a
r
c
h
a
i
ms
t
o
f
i
l
l
t
h
e
g
a
p
b
y
c
r
e
a
t
i
n
g
a
met
h
o
d
f
o
r
f
i
t
t
i
n
g
o
p
e
n
-
so
u
r
c
e
w
o
r
k
l
o
a
d
s
f
r
o
m
F
a
a
S
b
e
n
c
h
m
a
r
k
i
n
g
s
u
i
t
e
s
t
o
o
p
e
r
a
t
i
o
n
a
l
F
a
a
S
w
o
r
k
l
o
a
d
t
r
a
c
e
s
w
h
i
l
e
p
r
e
ser
v
i
n
g
t
h
e
i
r
e
sse
n
t
i
a
l
st
a
t
i
st
i
c
a
l
a
s
p
e
c
t
s.
M
S
A
z
u
r
e
Tr
a
c
e
I
t
a
n
a
l
y
s
e
s
t
h
e
e
x
e
c
u
t
i
o
n
mo
d
e
s
o
f
F
a
a
s
R
a
i
l
a
n
d
c
o
m
p
a
r
e
s
l
o
a
d
r
e
p
r
e
s
e
n
t
a
t
i
v
i
t
y
w
i
t
h
A
z
u
r
e
F
u
n
c
t
i
o
n
s,
H
u
a
w
e
i
’
s
p
r
o
d
u
c
t
i
o
n
-
sca
l
e
t
r
a
c
e
s,
a
n
d
e
x
i
st
i
n
g
s
e
r
v
e
r
l
e
ss
l
i
t
e
r
a
t
u
r
e
.
F
a
a
S
R
a
i
l
c
u
r
r
e
n
t
l
y
d
o
e
s
n
o
t
mo
d
i
f
y
f
u
n
c
t
i
o
n
i
n
p
u
t
s
,
r
e
su
l
t
i
n
g
i
n
c
o
n
s
i
st
e
n
t
e
x
p
e
c
t
e
d
e
x
e
c
u
t
i
o
n
t
i
mes.
E
x
p
e
r
i
m
e
n
t
i
n
g
w
i
t
h
d
i
f
f
e
r
e
n
t
i
n
p
u
t
s
i
s
a
p
o
t
e
n
t
i
a
l
n
e
x
t
st
e
p
.
N
o
t
a
l
l
e
x
p
e
r
i
m
e
n
t
s
b
e
n
e
f
i
t
f
r
o
m
sc
a
l
i
n
g
t
h
e
i
n
p
u
t
t
r
a
c
e
o
v
e
r
t
i
m
e
.
E
x
p
e
r
i
me
n
t
s
a
n
a
l
y
z
i
n
g
sa
n
d
b
o
x
c
a
c
h
i
n
g
st
r
a
t
e
g
i
e
s
o
r
p
r
e
d
i
c
t
i
v
e
p
r
e
a
l
l
o
c
a
t
i
o
n
o
p
t
i
mi
z
a
t
i
o
n
s ma
y
b
e
n
e
f
i
t
f
r
o
m
F
a
a
S
R
a
i
l
’
s
M
i
n
u
t
e
R
a
n
g
e
m
o
d
e
,
w
h
i
c
h
d
o
e
s
n
o
t
sca
l
e
r
e
q
u
e
s
t
o
v
e
r
t
i
me
.
[
2
0
]
A
C
M
,
2
0
2
3
Th
e
r
e
s
e
a
r
c
h
b
e
l
o
w
c
o
m
p
a
r
e
s
o
p
e
r
a
t
i
o
n
a
l
c
l
o
u
d
b
l
o
c
k
st
o
r
a
g
e
w
o
r
k
l
o
a
d
s
u
si
n
g
b
l
o
c
k
-
l
e
v
e
l
I
/
O
t
r
a
c
e
s
f
r
o
m
A
l
i
b
a
b
a
C
l
o
u
d
a
n
d
Te
n
c
e
n
t
C
l
o
u
d
B
l
o
c
k
S
t
o
r
a
g
e
,
a
n
a
l
y
z
i
n
g
b
i
l
l
i
o
n
s
o
f
r
e
q
u
e
st
s
.
A
l
i
b
a
b
a
C
l
o
u
d
&
Te
n
c
e
n
t
C
l
o
u
d
B
l
o
c
k
S
t
o
r
a
g
e
C
l
o
u
d
b
l
o
c
k
st
o
r
a
g
e
w
o
r
k
l
o
a
d
s
a
n
d
p
r
o
m
i
n
e
n
t
p
u
b
l
i
c
b
l
o
c
k
-
l
e
v
e
l
I
/
O
w
o
r
k
l
o
a
d
s
f
r
o
m
c
o
r
p
o
r
a
t
e
d
a
t
a
c
e
n
t
e
r
s
a
t
M
i
c
r
o
so
f
t
R
e
s
e
a
r
c
h
C
a
m
b
r
i
d
g
e
a
r
e
c
o
m
p
a
r
e
d
t
o
d
e
t
e
r
mi
n
e
t
h
e
i
r
si
mi
l
a
r
i
t
i
e
s
a
n
d
d
i
f
f
e
r
e
n
c
e
s
.
Th
e
s
e
o
b
j
e
c
t
i
v
e
s
a
r
e
a
c
h
i
e
v
e
d
t
h
r
o
u
g
h
t
h
e
o
v
e
r
a
l
l
p
r
o
v
i
si
o
n
o
f
s
i
x
t
e
e
n
d
e
t
a
i
l
e
d
d
a
t
a
p
o
i
n
t
s
a
n
d
si
x
h
i
g
h
-
l
e
v
e
l
d
a
t
a
p
o
i
n
t
s,
t
a
l
k
i
n
g
a
b
o
u
t
l
o
a
d
i
n
t
e
n
s
i
t
y
,
g
e
o
g
r
a
p
h
i
c
a
l
p
a
t
t
e
r
n
s,
a
n
d
t
e
m
p
o
r
a
l
p
a
t
t
e
r
n
s.
T
h
e
a
n
a
l
y
si
s
f
o
c
u
s
e
s
o
n
h
o
w
t
h
e
f
i
n
d
i
n
g
s
i
n
f
l
u
e
n
c
e
l
o
a
d
b
a
l
a
n
c
i
n
g
i
n
c
l
o
u
d
b
l
o
c
k
st
o
r
a
g
e
,
t
h
e
e
f
f
i
c
i
e
n
c
y
o
f
c
a
c
h
i
n
g
,
a
n
d
t
h
e
a
d
mi
n
i
s
t
r
a
t
i
o
n
o
f
c
l
u
st
e
r
s
.
Tr
a
c
e
a
n
a
l
y
s
i
s
h
a
s
v
a
r
i
o
u
s
l
i
m
i
t
a
t
i
o
n
s.
B
e
c
a
u
s
e
t
h
e
t
r
a
c
e
s
c
o
l
l
e
c
t
e
d
b
y
A
l
i
C
l
o
u
d
a
n
d
Te
n
c
e
n
t
C
l
o
u
d
c
a
p
t
u
r
e
t
h
e
r
e
sp
o
n
s
e
t
i
me
d
i
f
f
e
r
e
n
t
l
y
t
h
a
n
t
h
e
t
r
a
c
e
s
c
a
p
t
u
r
e
d
b
y
M
S
R
C
,
it
’
s
i
m
p
o
ssi
b
l
e
t
o
d
o
l
a
t
e
n
c
y
a
n
a
l
y
si
s
f
o
r
I
/
O
r
e
q
u
e
s
t
s
d
u
r
i
n
g
a
r
e
a
l
d
e
p
l
o
y
me
n
t
.
B
o
t
h
o
f
t
h
e
m
a
l
s
o
d
o
n
o
t
e
x
p
o
s
e
w
h
i
c
h
p
r
o
g
r
a
ms
a
r
e
r
u
n
n
i
n
g
o
n
a
p
a
r
t
i
c
u
l
a
r
d
i
s
k
,
t
h
u
s
l
e
a
v
i
n
g
o
u
t
t
h
e
a
ss
o
c
i
a
t
i
o
n
o
f
i
n
d
i
v
i
d
u
a
l
w
o
r
k
l
o
a
d
s
o
f
a
n
a
p
p
l
i
c
a
t
i
o
n
t
o
t
h
e
i
r
c
o
r
r
e
s
p
o
n
d
i
n
g
I
/
O
p
a
t
t
e
r
n
.
[
2
1
]
I
EEE,
2
0
2
3
H
a
r
mo
n
y
i
s
a
d
e
e
p
l
e
a
r
n
i
n
g
-
d
r
i
v
e
n
M
L
c
l
u
st
e
r
s
c
h
e
d
u
l
e
r
t
h
a
t
o
p
t
i
m
i
z
e
s
p
e
r
f
o
r
ma
n
c
e
b
y
sch
e
d
u
l
i
n
g
h
e
t
e
r
o
g
e
n
e
o
u
s
t
r
a
i
n
i
n
g
t
a
sk
s
u
s
i
n
g
p
a
r
a
m
e
t
e
r
serv
e
r
o
r
a
l
l
-
r
e
d
u
c
e
a
r
c
h
i
t
e
c
t
u
r
e
t
o
mi
n
i
mi
z
e
i
n
t
e
r
f
e
r
e
n
c
e
a
n
d
i
n
c
r
e
a
s
e
p
e
r
f
o
r
m
a
n
c
e
(
i
.
e
.
,
t
r
a
i
n
i
n
g
c
o
m
p
l
e
t
i
o
n
t
i
me
r
e
d
u
c
t
i
o
n
)
.
R
e
a
l
M
L
W
o
r
k
l
o
a
d
s
i
n
K
u
b
e
r
n
e
t
e
s
C
l
u
st
e
r
Th
e
m
o
d
e
l
i
s
b
u
i
l
t
f
o
r
o
n
l
i
n
e
j
o
b
a
r
r
i
v
a
l
u
s
i
n
g
a
p
a
r
a
m
e
t
e
r
ser
v
e
r
o
r
a
l
l
-
r
e
d
u
c
e
a
r
c
h
i
t
e
c
t
u
r
e
.
A
t
w
o
-
st
e
p
l
e
a
r
n
i
n
g
m
e
c
h
a
n
i
sm
i
s
a
p
p
l
i
e
d
t
o
l
e
a
r
n
t
h
e
p
l
a
c
e
me
n
t
p
o
l
i
c
y
w
i
t
h
o
u
t
a
n
i
n
t
e
r
f
e
r
e
n
c
e
m
o
d
e
l
o
f
w
o
r
k
l
o
a
d
.
I
n
i
t
i
a
l
l
y
,
t
h
e
s
c
a
r
c
e
h
i
st
o
r
i
c
a
l
d
a
t
a
i
s
u
se
d
f
o
r
t
r
a
i
n
i
n
g
a
r
e
w
a
r
d
n
e
u
r
a
l
n
e
t
w
o
r
k
,
a
n
d
t
h
e
n
D
R
L
i
s
t
r
a
i
n
e
d
o
n
p
l
a
c
e
me
n
t
d
e
c
i
si
o
n
w
i
t
h
r
e
w
a
r
d
samp
l
e
s
o
b
t
a
i
n
e
d
f
r
o
m
t
h
e
r
e
w
a
r
d
mo
d
e
l
.
A
c
c
o
r
d
i
n
g
t
o
t
h
e
e
v
a
l
u
a
t
i
o
n
o
f
a
K
u
b
e
r
n
e
t
e
s
c
l
u
st
e
r
,
H
a
r
m
o
n
y
r
e
d
u
c
e
s
t
h
e
c
o
mp
l
e
t
i
o
n
t
i
me
f
o
r
j
o
b
s
b
y
1
6
%
-
4
2
%
c
o
m
p
a
r
e
d
w
i
t
h
r
e
p
r
e
s
e
n
t
a
t
i
v
e
s
c
h
e
d
u
l
i
n
g
me
t
h
o
d
s.
-
Evaluation Warning : The document was created with Spire.PDF for Python.