I
n
d
on
e
s
ian
Jou
r
n
al
o
f
E
lec
t
r
ica
l
E
n
gin
e
e
r
in
g
a
n
d
Com
p
u
t
e
r
S
c
ience
Vo
l
.
38
,
N
o
.
2
,
M
a
y
20
25
,
pp.
1
324
~1
330
I
S
S
N:
2
502
-
4
7
52
,
DO
I
:
10
.
11591/i
j
e
e
cs
.v
38.
i
2
.
pp1
324
-
1
330
1324
Jou
r
n
al
h
o
m
e
page
:
ht
tp:
//
ij
e
e
cs
.
iaes
c
or
e
.
c
om
A
c
om
p
ar
at
iv
e
an
al
y
si
s of
GP
U
s, T
P
U
s, D
P
U
s, an
d
Q
P
U
s f
or
d
e
e
p
l
e
ar
n
in
g w
ith
p
y
t
h
on
Ayoub
Al
l
a
l
i,
Z
in
e
b
E
l
F
a
l
ah
,
Ayoub
S
gh
ir
,
Jaa
f
a
r
Abou
c
h
ab
ak
a,
Naj
at
Raf
al
ia
D
e
pa
r
tm
e
nt
of
C
o
mput
e
r
S
c
i
e
n
c
e
,
F
a
c
ul
t
y
of
S
c
i
e
nc
e
s
, I
bn
T
of
a
il
U
ni
ve
r
s
it
y
,
K
e
n
it
r
a
, M
or
o
c
c
o
Ar
t
ic
l
e
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
i
ve
d
Ju
n
19
,
202
4
R
e
vi
s
e
d
No
v
9
,
202
4
A
c
c
e
pt
e
d
N
o
v
24
,
202
4
I
n
t
h
e
rap
i
d
l
y
e
v
o
l
v
i
n
g
fi
el
d
o
f
d
ee
p
l
e
arn
i
n
g
,
t
h
e
c
o
m
p
u
t
at
i
o
n
al
d
em
an
d
s
fo
r
t
rai
n
i
n
g
s
o
p
h
i
s
t
i
c
at
e
d
mo
d
el
s
h
a
v
e
e
s
c
a
l
at
ed
,
p
ro
m
p
t
i
n
g
a
s
h
i
ft
t
o
w
ard
s
s
p
e
c
i
a
l
i
z
ed
h
ard
w
ar
e
a
cce
l
e
rat
o
rs
s
u
ch
as
g
rap
h
i
c
s
p
ro
ce
s
s
i
n
g
u
n
i
t
s
(G
PU
s
),
t
e
n
s
o
r
p
ro
ce
s
s
i
n
g
u
n
i
t
s
(T
PU
s
)
,
d
at
a
p
ro
ce
s
s
i
n
g
u
n
i
t
s
(
D
PU
s
),
a
n
d
q
u
an
t
u
m
p
ro
ce
s
s
i
n
g
u
n
i
t
s
(Q
PU
s
).
T
h
i
s
art
i
cl
e
p
ro
v
i
d
e
s
a
co
m
p
re
h
en
s
i
v
e
an
a
l
y
s
i
s
o
f
t
h
e
s
e
h
e
t
e
r
o
g
en
e
o
u
s
co
m
p
u
t
i
n
g
arch
i
t
ec
t
u
r
e
s
,
h
i
g
h
l
i
g
h
t
i
n
g
t
h
e
i
r
u
n
i
q
u
e
c
h
ara
c
t
e
ri
s
t
i
c
s
,
p
e
rfo
r
m
a
n
ce
me
t
ri
c
s
,
an
d
s
u
i
t
ab
i
l
i
t
y
fo
r
v
ar
i
o
u
s
d
ee
p
l
e
ar
n
i
n
g
t
as
k
s
.
B
y
l
ev
e
rag
i
n
g
p
y
t
h
o
n
,
a
p
r
e
d
o
m
i
n
an
t
p
r
o
g
ra
mm
i
n
g
l
a
n
g
u
ag
e
i
n
t
h
e
d
at
a
s
c
i
en
ce
d
o
m
a
i
n
,
t
h
e
i
n
t
eg
rat
i
o
n
a
n
d
o
p
t
i
mi
zat
i
o
n
t
ech
n
i
q
u
e
s
ap
p
l
i
c
ab
l
e
t
o
e
a
c
h
h
ard
w
are
p
l
at
fo
r
m
i
s
e
x
p
l
o
r
ed
,
o
ff
e
r
i
n
g
i
n
s
i
g
h
t
s
i
n
t
o
t
h
e
i
r
p
rac
t
i
c
a
l
i
m
p
l
i
c
at
i
o
n
s
fo
r
d
ee
p
l
e
ar
n
i
n
g
r
e
s
e
arch
a
n
d
ap
p
l
i
c
at
i
o
n
.
T
h
e
arc
h
i
t
ec
t
u
ra
l
d
i
ff
e
r
e
n
ce
s
t
h
at
i
n
fl
u
e
n
ce
c
o
m
p
u
t
at
i
o
n
al
e
ff
i
c
i
en
cy
i
s
e
x
ami
n
ed
,
p
aral
l
e
l
i
s
m
,
a
n
d
en
e
rg
y
co
n
s
u
m
p
t
i
o
n
,
a
l
o
n
g
s
i
d
e
d
i
s
cu
s
s
i
n
g
t
h
e
e
v
o
l
v
i
n
g
eco
s
y
s
t
em
o
f
s
o
ft
w
ar
e
t
o
o
l
s
an
d
l
i
b
rari
e
s
t
h
at
s
u
p
p
o
rt
d
ee
p
l
e
ar
n
i
n
g
o
n
t
h
e
s
e
p
l
at
fo
rm
s
.
T
h
ro
u
g
h
a
s
e
r
i
e
s
o
f
b
en
c
h
m
ar
k
s
a
n
d
c
as
e
s
t
u
d
i
e
s
,
t
h
i
s
s
t
u
d
y
ai
m
s
t
o
e
q
u
i
p
r
e
s
e
ar
ch
e
rs
a
n
d
p
rac
t
i
t
i
o
n
e
rs
w
i
t
h
t
h
e
k
n
o
w
l
e
d
g
e
t
o
m
ak
e
i
n
f
o
r
me
d
d
eci
s
i
o
n
s
w
h
e
n
s
e
l
ec
t
i
n
g
h
ar
d
w
ar
e
fo
r
t
h
e
i
r
d
ee
p
l
e
arn
i
n
g
p
ro
j
ec
t
s
,
u
l
t
i
m
at
el
y
co
n
t
ri
b
u
t
i
n
g
t
o
t
h
e
a
ccel
e
rat
i
o
n
o
f
mo
d
e
l
d
e
v
el
o
p
me
n
t
an
d
i
n
n
o
v
at
i
o
n
i
n
t
h
e
f
i
e
l
d
.
K
e
y
w
o
r
d
s
:
Da
t
a
pr
o
c
e
s
s
i
n
g
uni
t
s
De
e
p
l
e
a
r
ni
ng
Gr
a
phi
c
s
pr
o
c
e
s
s
i
ng
uni
t
s
P
y
t
h
o
n
Qua
n
t
u
m
pr
o
c
e
s
s
i
n
g
u
ni
t
s
T
e
n
s
o
r
pr
o
c
e
s
s
i
n
g
u
ni
t
s
Th
i
s
i
s
a
n
o
p
en
a
c
ces
s
a
r
t
i
c
l
e
u
n
d
e
r
t
h
e
CC
B
Y
-
SA
l
i
cen
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
Ay
o
u
b
All
a
l
i
De
pa
r
t
m
e
n
t
o
f
C
o
m
put
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
nc
e
s
,
I
b
n
T
o
f
a
il
U
ni
ve
r
s
i
t
y
K
e
ni
t
r
a
,
M
o
r
o
c
c
o
E
m
a
i
l
:
a
y
o
u
b
.
a
l
l
a
l
i
@u
i
t
.
a
c
.
m
a
1.
I
NT
RODU
C
T
I
ON
T
h
e
l
a
n
d
s
c
a
pe
o
f
de
e
p
l
e
a
r
ni
ng
h
a
s
w
i
t
n
e
s
s
e
d
a
r
e
m
a
r
ka
bl
e
t
r
a
n
s
f
o
r
m
a
t
i
o
n
o
v
e
r
t
h
e
l
a
s
t
de
c
a
de
,
l
a
r
ge
ly
f
ue
l
e
d
by
a
d
v
a
n
c
e
m
e
n
t
s
i
n
c
o
m
put
a
t
i
o
n
a
l
h
a
r
dwa
r
e
.
A
s
m
o
de
l
s
b
e
c
o
m
e
i
n
c
r
e
a
s
i
ng
ly
c
o
m
p
l
e
x
,
t
h
e
n
e
c
e
s
s
i
t
y
f
o
r
e
f
f
i
c
i
e
n
t
a
n
d
p
o
we
r
f
u
l
c
o
m
put
i
n
g
r
e
s
o
ur
c
e
s
h
a
s
n
e
v
e
r
b
e
e
n
m
o
r
e
a
ppa
r
e
n
t.
T
h
e
e
m
e
r
ge
n
c
e
o
f
s
pe
c
i
a
li
z
e
d
h
a
r
dwa
r
e
a
c
c
e
l
e
r
a
t
or
s
,
i
n
c
l
ud
i
ng
gr
a
phi
c
s
pr
o
c
e
s
s
i
n
g
u
ni
t
s
(
GPUs
)
,
t
e
n
s
o
r
pr
o
c
e
s
s
i
n
g
u
ni
t
s
(
T
P
Us
)
,
da
t
a
pr
o
c
e
s
s
i
n
g
uni
t
s
(
DPUs
)
,
a
n
d
qua
n
t
um
pr
o
c
e
s
s
i
ng
u
ni
t
s
(
QPUs
)
,
h
a
s
h
e
r
a
l
de
d
a
n
e
w
e
r
a
i
n
t
h
e
a
c
c
e
l
e
r
a
t
i
o
n
o
f
de
e
p
l
e
a
r
ni
ng
t
a
s
ks
.
T
h
e
s
e
t
e
c
h
n
o
l
o
g
i
e
s
o
f
f
e
r
s
u
b
s
t
a
n
t
i
a
l
im
pr
o
v
e
m
e
n
t
s
i
n
pr
o
c
e
s
s
in
g
s
pe
e
d
a
n
d
e
ne
r
g
y
e
f
f
i
c
i
e
n
c
y
,
e
n
a
bli
ng
r
e
s
e
a
r
c
h
e
r
s
a
n
d
pr
a
c
t
i
t
i
o
n
e
r
s
to
t
a
c
kl
e
m
o
r
e
s
o
phi
s
t
i
c
a
t
e
d
pr
o
bl
e
m
s
a
n
d
a
c
hi
e
v
e
b
r
e
a
kt
h
r
o
ugh
s
a
t
a
n
un
pr
e
c
e
de
n
t
e
d
pa
c
e
[
1]
.
C
o
r
r
a
l
e
t
al.
[
2]
e
v
a
l
ua
t
e
t
h
e
e
n
e
r
g
y
c
o
n
s
u
m
pt
i
o
n
a
n
d
pe
r
f
o
r
m
a
n
c
e
o
f
m
e
d
i
c
a
l
d
i
a
g
n
o
s
i
s
a
i
ds
im
p
l
e
m
e
n
t
e
d
o
n
e
m
b
e
dde
d
pr
o
c
e
s
s
o
r
s
,
s
pe
c
i
f
i
c
a
ll
y
Go
o
gl
e
'
s
E
dge
T
P
U
a
n
d
NV
I
DI
A'
s
M
a
x
we
ll
GPU.
Us
i
n
g
g
l
a
uc
o
m
a
de
t
e
c
t
i
o
n
f
r
o
m
c
o
l
o
r
f
u
n
dus
i
m
a
g
e
s
a
s
a
c
a
s
e
s
t
udy
,
t
h
e
y
de
m
o
n
s
t
r
a
t
e
t
h
e
f
e
a
s
i
b
il
i
t
y
o
f
r
e
a
l
-
t
i
m
e
im
a
ge
s
e
g
m
e
n
t
a
t
i
o
n
a
n
d
c
l
a
s
s
if
i
c
a
t
i
o
n
o
n
e
m
b
e
d
de
d
b
o
a
r
ds
.
T
h
e
s
t
udy
f
o
c
us
e
s
o
n
t
h
e
e
n
e
r
g
y
a
nd
t
i
mi
ng
pe
r
f
o
r
m
a
n
c
e
o
f
t
h
e
s
e
s
y
s
t
e
m
s
dur
i
n
g
o
pt
i
c
d
i
s
c
(
OD
)
a
n
d
c
up
(
OC
)
s
e
g
m
e
n
t
a
t
i
o
n
a
n
d
c
l
a
s
s
i
f
i
c
a
t
i
o
n
t
a
s
ks
,
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
c
ompar
ati
v
e
analys
is
of
GP
Us
,
T
P
Us
,
DP
Us
,
and
QP
Us
f
or
de
e
p
lear
ning
w
it
h
p
y
thon
(
A
y
oub
A
ll
a
li
)
1325
s
h
o
w
i
n
g
t
h
a
t
t
h
e
E
dge
T
P
U
i
s
m
o
r
e
e
n
e
r
g
y
-
e
f
f
ic
i
e
n
t
a
n
d
f
a
s
t
e
r
c
o
m
pa
r
e
d
t
o
t
h
e
M
a
x
w
e
l
l
GPU
f
o
r
b
o
t
h
s
e
g
m
e
n
t
a
t
i
o
n
a
n
d
c
l
a
s
s
if
i
c
a
t
i
o
n
.
Hua
n
g
e
t
al.
[
3]
f
o
c
us
o
n
i
m
pr
o
vi
n
g
t
h
e
e
f
f
i
c
i
e
n
c
y
o
f
hi
e
r
a
r
c
hi
c
a
l
t
uc
ke
r
(
HT
)
t
e
n
s
o
r
l
e
a
r
ni
ng
f
o
r
l
a
r
ge
-
s
c
a
l
e
,
hi
g
h
-
d
im
e
n
s
i
o
n
a
l
da
t
a
us
i
n
g
GPU
t
e
n
s
o
r
c
o
r
e
s
.
T
h
e
y
a
ddr
e
s
s
ke
y
c
h
a
ll
e
n
ge
s
i
n
o
p
t
i
mi
z
i
n
g
t
e
n
s
o
r
l
e
a
r
ni
ng
pr
i
m
i
t
i
ve
s
,
i
m
p
l
e
m
e
n
t
i
n
g
HT
t
e
n
s
o
r
a
l
g
o
r
i
t
hm
s
o
n
m
u
l
t
i
p
l
e
GPUs
,
a
n
d
h
a
n
d
li
ng
da
t
a
e
x
c
e
e
d
i
ng
GPU
m
e
m
o
r
y
.
T
h
e
y
pr
e
s
e
n
t
o
p
t
i
m
i
z
e
d
GPU
-
b
a
s
e
d
t
e
n
s
o
r
o
pe
r
a
t
i
o
ns
s
uc
h
a
s
c
o
n
t
r
a
c
t
i
o
n
s
,
m
a
t
r
i
c
i
z
a
t
i
o
n
s
,
a
n
d
s
i
n
gu
l
a
r
v
a
l
ue
de
c
o
m
po
s
i
t
i
o
n
(
S
VD
)
f
o
r
bi
g
da
t
a
a
n
a
ly
s
i
s
.
A
dd
i
t
i
o
na
l
ly
,
t
h
e
y
i
n
t
r
o
duc
e
a
n
HT
t
e
n
s
o
r
l
a
y
e
r
f
o
r
de
e
p
n
e
ur
a
l
ne
t
wor
ks
a
n
d
de
v
e
l
o
p
a
qua
n
t
u
m
m
a
c
hi
ne
l
e
a
r
ni
ng
a
l
go
r
i
t
hm
ba
s
e
d
o
n
a
t
e
ns
o
r
-
tr
e
e
s
t
r
uc
t
ur
e
,
a
c
hi
e
vi
ng
s
i
g
nif
i
c
a
n
t
s
pe
e
dups
o
v
e
r
e
xi
s
t
i
n
g
m
e
t
h
o
ds
o
n
NV
I
DI
A
GPUs
.
I
n
t
h
e
r
e
vi
e
w
e
d
l
i
t
e
r
a
t
ur
e
,
o
nl
y
GPU
s
a
n
d
T
P
Us
we
r
e
f
o
c
us
e
d
o
n
,
a
n
d
t
hi
s
m
a
y
n
o
t
b
e
e
n
o
ugh
i
n
t
h
e
f
ut
ur
e
,
b
e
c
a
us
e
a
s
t
h
e
da
t
a
v
o
l
u
m
e
i
nc
r
e
a
s
e
s
,
m
o
r
e
p
o
we
r
f
u
l
pr
o
c
e
s
s
i
ng
u
ni
t
s
a
r
e
r
e
qu
i
r
e
d
to
pr
o
c
e
s
s
t
hi
s
da
t
a
,
a
n
d
t
h
e
a
dv
a
n
t
a
ge
s
a
n
d
d
i
s
a
d
v
a
n
t
a
ge
s
o
f
e
a
c
h
uni
t
we
r
e
n
o
t
e
x
p
l
a
i
ne
d,
t
h
e
r
e
f
o
r
e
,
t
hi
s
a
r
t
i
c
l
e
e
x
p
l
a
i
n
s
t
h
e
f
o
l
l
o
w
i
ng:
i
)
e
n
e
r
g
y
a
n
d
t
i
m
e
c
o
n
s
u
m
pt
i
o
n
of
GPUs
,
T
P
Us
,
DPUs
,
a
n
d
QPUs
r
un
ni
ng
t
h
e
s
a
m
e
de
e
p
l
e
a
r
ni
ng
m
o
de
l
;
ii
)
h
o
w
to
i
n
t
e
gr
a
t
e
p
y
t
h
o
n
,
a
s
a
m
a
j
o
r
pr
o
g
r
a
m
m
i
ng
l
a
n
gua
ge
i
n
de
e
p
l
e
a
r
ni
ng,
wi
t
h
t
h
e
s
e
h
a
r
dwa
r
e
a
c
c
e
l
e
r
a
t
or
s
to
i
m
pr
o
v
e
pe
r
f
o
r
m
a
nc
e
a
nd
s
c
a
l
a
bil
i
t
y
;
ii
i
)
a
d
v
a
n
t
a
ge
s
a
n
d
d
i
s
a
d
v
a
n
t
a
ge
s
of
GPUs
,
T
P
Us
,
DPUs
,
a
n
d
QPUs
;
a
n
d
i
v
)
c
h
a
ll
e
n
ge
s
f
a
c
e
d
by
t
h
e
s
e
u
ni
t
s
.
2.
M
E
T
HO
D
T
h
e
pr
o
c
e
s
s
be
ga
n
w
i
t
h
t
h
e
c
r
e
a
t
i
o
n
o
f
a
de
e
p
l
e
a
r
ni
ng
m
o
de
l
i
n
p
y
t
h
o
n
,
w
hi
c
h
s
e
r
v
e
d
a
s
t
h
e
f
o
u
n
da
t
i
o
n
f
o
r
c
o
m
pa
r
i
s
o
n
a
c
r
o
s
s
di
f
f
e
r
e
n
t
pr
o
c
e
s
s
i
n
g
uni
t
s
.
T
hi
s
m
o
de
l
wa
s
we
l
l
-
de
f
i
ne
d
a
n
d
a
da
pt
a
bl
e
to
m
u
l
t
i
p
l
e
f
r
a
m
e
wo
r
ks
to
e
n
s
ur
e
c
o
n
s
i
s
t
e
n
c
y
.
T
h
e
c
h
o
i
c
e
o
f
t
h
e
m
o
de
l
c
o
u
l
d
h
a
v
e
b
e
e
n
s
o
m
e
t
hi
ng
c
o
m
put
a
t
i
o
n
a
ll
y
de
m
a
n
d
i
ng,
l
i
k
e
a
c
o
n
v
o
l
ut
i
o
n
a
l
n
e
ur
a
l
n
e
t
wo
r
k
(
C
NN
)
or
a
tr
a
n
s
f
o
r
m
e
r
,
de
pe
n
d
i
n
g
o
n
t
h
e
s
pe
c
i
f
i
c
t
a
s
ks
un
de
r
e
v
a
l
ua
t
i
o
n
.
T
h
e
p
y
t
h
o
n
c
o
de
wa
s
t
h
e
c
o
r
e
l
o
g
i
c
o
f
t
h
e
e
x
pe
r
i
m
e
n
t
a
n
d
wa
s
t
h
e
c
o
m
m
o
n
f
a
c
t
o
r
a
c
r
o
s
s
a
l
l
ha
r
dwa
r
e
p
l
a
t
f
o
r
m
s
,
e
ns
ur
i
n
g
th
a
t
t
h
e
c
o
m
pa
r
i
s
o
n
s
f
o
c
us
e
d
s
o
l
e
ly
o
n
t
h
e
pe
r
f
o
r
m
a
n
c
e
d
i
f
f
e
r
e
n
c
e
s
o
f
t
h
e
pr
o
c
e
s
s
i
n
g
u
ni
t
s
r
a
t
h
e
r
t
h
a
n
v
a
r
iat
i
o
n
s
i
n
m
o
de
l
im
p
l
e
m
e
n
t
a
t
i
o
n
[
4]
.
On
c
e
t
h
e
p
y
t
h
o
n
c
o
de
wa
s
r
e
a
dy
a
s
s
h
o
wn
i
n
F
i
g
ur
e
1
,
i
t
wa
s
de
p
l
o
y
e
d
o
n
a
GPU.
T
hi
s
s
t
e
p
us
e
d
e
i
t
h
e
r
T
e
ns
o
r
F
l
o
w/P
y
T
o
r
c
h
a
s
t
h
e
de
e
p
l
e
a
r
ni
n
g
f
r
a
m
e
wo
r
k,
i
n
c
o
nj
u
n
c
t
i
o
n
w
i
t
h
C
UD
A
,
t
h
e
pa
r
a
l
l
e
l
c
o
m
put
i
n
g
p
l
a
t
f
o
r
m
a
n
d
a
pp
l
i
c
a
t
i
o
n
pr
o
gr
a
m
mi
n
g
i
n
t
e
r
f
a
c
e
(
A
P
I
)
m
o
de
l
t
h
a
t
a
l
l
o
w
e
d
t
h
e
GPU
to
h
a
n
d
l
e
t
a
s
ks
a
t
a
m
uc
h
f
a
s
t
e
r
r
a
t
e
.
C
UD
A
e
n
a
bl
e
d
t
h
e
p
y
t
h
o
n
c
o
de
to
r
un
o
p
t
i
mi
z
e
d
o
pe
r
a
t
i
o
n
s
,
t
a
ki
n
g
f
u
l
l
a
dv
a
n
t
a
ge
o
f
t
h
e
GPU'
s
c
a
p
a
bi
li
t
y
t
o
pe
r
f
o
r
m
m
a
s
s
i
ve
ly
pa
r
a
l
l
e
l
c
o
m
put
a
t
i
o
ns
[
5]
.
T
h
e
GPU
wa
s
kno
wn
f
o
r
a
c
c
e
l
e
r
a
t
i
n
g
t
r
a
i
ni
ng
a
n
d
i
nf
e
r
e
n
c
e
t
a
s
ks
,
m
a
k
i
ng
i
t
i
de
a
l
f
o
r
m
a
ny
de
e
p
l
e
a
r
ni
ng
a
pp
l
i
c
a
t
i
o
ns
.
T
h
i
s
s
t
a
ge
m
e
a
s
ur
e
d
h
o
w
qu
i
c
k
ly
t
h
e
m
o
de
l
c
o
u
l
d
be
pr
o
c
e
s
s
e
d
o
n
t
h
e
GP
U
a
n
d
a
n
a
ly
z
e
d
t
h
e
e
n
e
r
g
y
c
o
ns
u
m
pt
i
o
n
i
nv
o
l
ve
d
[
6]
–
[
8]
.
T
h
e
m
o
de
l
w
a
s
a
da
pt
e
d
to
r
un
o
n
a
T
P
U
,
a
s
pe
c
i
a
li
z
e
d
h
a
r
dwa
r
e
a
c
c
e
l
e
r
a
t
or
de
s
i
g
n
e
d
to
h
a
n
d
le
m
a
t
r
i
x
o
pe
r
a
t
i
o
n
s
c
o
m
m
o
nl
y
us
e
d
i
n
de
e
p
l
e
a
r
ni
n
g.
T
e
n
s
o
r
F
l
o
w
wa
s
t
h
e
pr
e
f
e
r
r
e
d
f
r
a
m
e
wo
r
k
f
o
r
T
P
Us
due
to
i
t
s
t
i
g
h
t
i
n
t
e
gr
a
t
i
o
n
a
n
d
o
pt
i
m
i
z
a
t
i
o
n
f
o
r
T
P
U
a
r
c
hi
t
e
c
t
ur
e
.
T
h
e
T
P
Us
we
r
e
kn
o
wn
to
b
e
hi
g
hly
e
f
f
i
c
i
e
n
t
i
n
t
r
a
i
ni
ng
l
a
r
ge
m
o
de
l
s
,
e
s
pe
c
i
a
ll
y
f
o
r
o
pe
r
a
t
i
o
ns
l
ike
c
o
n
v
o
l
ut
i
o
n
a
n
d
m
a
t
r
i
x
m
u
l
t
i
p
li
c
a
t
i
o
n
.
B
y
r
u
nni
ng
t
h
e
m
o
de
l
o
n
a
T
P
U,
t
hi
s
p
h
a
s
e
pr
o
vi
d
e
d
i
n
s
i
g
h
t
s
i
n
t
o
h
o
w
T
P
Us
c
o
m
pa
r
e
d
w
i
t
h
GPUs
i
n
t
e
r
m
s
o
f
b
o
t
h
s
pe
e
d
a
n
d
e
n
e
r
g
y
e
f
f
i
c
i
e
n
c
y
,
e
s
pe
c
i
a
ll
y
f
o
r
l
a
r
ge
-
s
c
a
l
e
de
e
p
l
e
a
r
ni
ng
t
a
s
ks
[
9]
–
[
13]
.
F
o
r
t
h
e
DPU
,
o
u
r
pr
o
j
e
c
t
i
nv
o
l
v
e
d
s
e
t
t
i
n
g
up
a
r
o
b
us
t
da
t
a
p
r
o
c
e
s
s
i
ng
e
nvi
r
o
nm
e
n
t
us
i
n
g
VM
wa
r
e
to
c
r
e
a
t
e
a
n
d
m
a
n
a
g
e
10
vi
r
t
ua
l
m
a
c
hi
ne
s
(
VM
s
)
.
E
a
c
h
VM
wa
s
c
o
nf
i
gur
e
d
to
f
a
c
i
li
t
a
t
e
t
h
e
t
e
s
t
i
n
g
o
f
a
de
e
p
l
e
a
r
ni
ng
m
o
de
l
,
a
l
l
o
w
i
ng
us
to
e
v
a
l
ua
t
e
i
t
s
pe
r
f
o
r
m
a
n
c
e
a
n
d
s
c
a
l
a
bi
li
t
y
i
n
a
c
o
n
t
r
o
l
l
e
d
e
nvi
r
o
nm
e
n
t
.
B
y
l
e
v
e
r
a
g
i
ng
VM
wa
r
e
’
s
vi
r
t
ua
l
i
z
a
t
i
o
n
c
a
pa
bil
i
t
ies
,
we
e
n
s
ur
e
d
e
f
f
i
c
i
e
n
t
r
e
s
o
ur
c
e
a
l
l
o
c
a
t
i
o
n
a
n
d
i
s
o
l
a
t
i
o
n
a
m
o
n
g
t
h
e
VM
s
,
e
na
bl
i
ng
pa
r
a
ll
e
l
pr
o
c
e
s
s
i
ng
o
f
d
a
t
a
a
n
d
m
o
de
l
t
r
a
i
ni
n
g.
I
n
t
hi
s
s
t
e
p,
t
h
e
m
o
de
l
wa
s
a
da
pt
e
d
f
o
r
t
h
e
DPU
a
r
c
hi
t
e
c
t
ur
e
to
e
v
a
l
u
a
t
e
h
o
w
i
t
pe
r
f
o
r
m
e
d
i
n
r
e
a
l
-
t
i
m
e
i
nf
e
r
e
n
c
e
t
a
s
ks
,
pr
o
vi
d
i
ng
c
r
i
t
i
c
a
l
da
t
a
o
n
e
n
e
r
g
y
c
o
n
s
u
m
pt
i
o
n
,
s
pe
e
d,
a
n
d
pr
o
c
e
s
s
i
n
g
e
f
f
i
c
i
e
n
c
y
i
n
l
o
w
-
po
we
r
e
n
vi
r
o
nm
e
n
t
s
[
14]
–
[
17
]
.
QPUs
r
e
pr
e
s
e
n
t
e
d
a
c
u
tt
i
n
g
-
e
dge
a
ppr
o
a
c
h
to
c
om
put
a
t
i
o
n
,
l
e
v
e
r
a
g
i
ng
t
h
e
pr
i
n
c
i
p
l
e
s
o
f
qua
n
t
u
m
m
e
c
h
a
ni
c
s
to
p
ot
e
n
t
i
a
ll
y
s
o
l
ve
pr
o
bl
e
m
s
e
x
po
n
e
nt
i
a
ll
y
f
a
s
t
e
r
t
h
a
n
c
l
a
s
s
i
c
a
l
s
y
s
t
e
m
s
.
I
n
t
hi
s
s
t
e
p,
t
h
e
m
o
de
l
wa
s
r
e
i
m
a
g
i
ne
d
f
o
r
qua
n
t
u
m
pr
o
c
e
s
s
i
n
g
u
s
i
ng
qua
n
t
u
m
m
a
c
hi
ne
l
e
a
r
ni
ng
f
r
a
m
e
wo
r
ks
s
uc
h
a
s
Q
i
s
k
i
t
,
C
i
r
q,
o
r
P
e
nny
L
a
n
e
[
18]
–
[
20]
.
T
h
e
s
e
p
l
a
t
f
o
r
m
s
a
ll
o
we
d
t
he
m
o
de
l
t
o
b
e
e
x
e
c
ut
e
d
o
n
qu
a
n
t
um
ha
r
dwa
r
e
,
wh
e
r
e
qubi
t
s
we
r
e
us
e
d
to
pe
r
f
o
r
m
c
o
m
put
a
t
i
o
n
s
.
Qua
n
t
u
m
m
a
c
hi
ne
l
e
a
r
ni
ng
wa
s
s
t
i
ll
i
n
i
t
s
e
x
pe
r
i
m
e
n
t
a
l
s
t
a
ge
s
,
s
o
t
hi
s
s
t
e
p
a
s
s
e
s
s
e
d
h
o
w
we
l
l
qu
a
n
t
um
s
y
s
t
e
m
s
h
a
n
d
l
e
d
de
e
p
l
e
a
r
ni
ng
m
o
de
l
s
a
n
d
c
o
m
p
a
r
e
d
t
h
e
i
r
pe
r
f
o
r
m
a
n
c
e
t
o
c
l
a
s
s
i
c
a
l
s
y
s
t
e
m
s
.
S
pe
c
i
a
l
a
t
t
e
n
t
i
o
n
wa
s
pa
i
d
to
t
h
e
e
ne
r
g
y
us
a
ge
a
n
d
t
i
m
e
t
a
ke
n
f
o
r
t
a
s
ks
t
h
a
t
m
igh
t
h
a
ve
t
a
ke
n
s
i
g
nif
i
c
a
n
t
l
y
l
o
n
ge
r
o
n
c
l
a
s
s
i
c
a
l
h
a
r
dwa
r
e
[
14]
,
[
21]
–
[
23
]
.
Af
t
e
r
r
unni
ng
t
h
e
de
e
p
l
e
a
r
ni
ng
m
o
de
l
o
n
a
l
l
t
h
e
d
i
f
f
e
r
e
n
t
pr
o
c
e
s
s
i
ng
uni
t
s
,
GPU,
T
P
U,
DPU
,
a
n
d
QPU,
t
h
e
f
i
na
l
s
t
e
p
wa
s
to
a
n
a
l
y
z
e
t
h
e
r
e
s
u
l
t
s
.
T
hi
s
i
nv
o
l
v
e
d
c
o
l
l
e
c
t
i
n
g
da
t
a
o
n
t
h
e
t
i
m
e
t
a
ke
n
to
e
x
e
c
ut
e
t
h
e
m
o
de
l
,
t
h
e
e
n
e
r
g
y
c
o
n
s
u
m
e
d
by
e
a
c
h
ha
r
dwa
r
e
p
l
a
t
f
o
r
m
,
a
n
d
t
h
e
o
v
e
r
a
l
l
m
o
de
l
pe
r
f
o
r
m
a
n
c
e
i
n
t
e
r
m
s
o
f
a
c
c
ur
a
c
y
a
n
d
s
pe
e
d.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
38
,
N
o
.
2
,
M
a
y
20
25
:
1
324
-
1
330
1326
F
i
gur
e
1.
M
e
t
h
o
d
3.
RE
S
UL
T
S
AN
D
DI
S
CU
S
S
I
ON
T
h
e
l
a
n
ds
c
a
pe
o
f
h
a
r
dwa
r
e
a
c
c
e
l
e
r
a
t
o
r
s
f
o
r
de
e
p
l
e
a
r
ni
ng
i
s
r
i
c
h
a
n
d
v
a
r
i
e
d,
w
i
t
h
e
a
c
h
t
y
pe
o
f
a
c
c
e
l
e
r
a
t
or
o
f
f
e
r
i
ng
d
i
s
t
i
n
c
t
a
dv
a
n
t
a
ge
s
a
n
d
t
r
a
d
e
-
o
f
f
s
i
n
t
e
r
m
s
o
f
t
r
a
i
ni
ng
t
i
m
e
,
pa
r
a
l
l
e
li
s
m
,
a
nd
e
n
e
r
g
y
c
o
n
s
u
m
pt
i
o
n
.
T
hi
s
c
o
m
pa
r
a
t
i
v
e
a
n
a
ly
s
i
s
de
l
v
e
s
i
nto
GPUs
,
T
P
Us
,
DPUs
,
a
n
d
QPUs
,
hi
g
hl
i
g
h
t
i
n
g
ho
w
t
h
e
s
e
t
e
c
h
n
o
l
o
g
i
e
s
s
t
a
c
k
up
a
ga
i
ns
t
e
a
c
h
ot
h
e
r
i
n
t
h
e
s
e
c
r
i
t
i
c
a
l
a
r
e
a
s
.
3.
1.
T
r
ain
in
g
t
im
e
T
a
bl
e
1
s
h
o
ws
t
h
e
t
i
m
i
ng
p
e
r
f
o
r
m
a
n
c
e
o
f
t
r
a
i
ni
n
g
t
h
e
s
a
m
e
m
o
de
l
us
i
ng
d
if
f
e
r
e
n
t
uni
t
s
.
T
h
e
GPU
too
k
a
b
o
u
t
300
s
e
c
o
n
ds
,
whi
l
e
t
h
e
T
P
U
c
u
t
t
h
i
s
ti
m
e
i
n
h
a
lf
t
o
150
s
e
c
o
n
ds
,
r
e
f
l
e
c
t
i
n
g
i
t
s
s
pe
c
i
a
li
z
a
t
i
o
n
i
n
de
e
p
l
e
a
r
ni
ng
t
a
s
ks
.
T
h
e
DPU
wa
s
t
h
e
f
a
s
t
e
s
t
,
c
o
m
p
l
e
t
i
n
g
t
r
a
i
ni
ng
i
n
75
s
e
c
o
n
ds
,
t
h
a
n
ks
to
i
t
s
hi
g
hly
o
p
t
i
mi
z
e
d
a
r
c
hi
t
e
c
t
ur
e
f
o
r
n
e
ur
a
l
ne
t
wor
k
c
o
m
put
a
t
i
o
n
s
.
M
e
a
n
w
hil
e
,
t
h
e
Qua
n
t
u
m
QPU
i
s
s
t
i
ll
i
n
t
h
e
e
x
pe
r
im
e
n
t
a
l
s
t
a
ge
f
o
r
de
e
p
l
e
a
r
ni
ng,
a
n
d
s
pe
c
if
i
c
t
im
i
ng
da
t
a
i
s
n
o
t
y
e
t
a
v
a
il
a
bl
e
.
T
hi
s
c
o
m
pa
r
i
s
o
n
highl
i
g
h
t
s
t
h
e
s
upe
r
i
o
r
e
f
f
i
c
i
e
n
c
y
o
f
m
o
de
r
n
s
pe
c
i
a
li
z
e
d
DPUs
c
o
m
pa
r
e
d
to
tr
a
di
t
i
o
n
a
l
GPU
s
,
whi
l
e
QPU
r
e
m
a
i
n
s
a
t
t
h
e
f
o
r
e
f
r
o
n
t
o
f
c
utt
i
n
g
-
e
dge
i
nn
o
v
a
t
i
o
n
[
24]
–
[
30]
.
T
a
bl
e
1.
T
i
mi
ng
pe
r
f
o
r
m
a
n
c
e
o
f
GPU/T
P
U/DPU/
QPU
w
i
t
h
t
h
e
s
a
m
e
m
o
de
l
G
P
U
T
P
U
D
P
U
Q
P
U
T
im
e
(
s
e
c
o
nds
)
≈ 30
0s
≈ 15
0s
≈ 75
s
S
ti
ll
i
n
th
e
e
x
p
e
r
im
e
n
ta
l
pha
s
e
f
o
r
d
e
e
p l
e
a
r
ni
ng
3.
2
.
E
n
e
r
gy
c
on
s
u
m
p
t
ion
T
h
e
e
n
e
r
g
y
c
o
ns
u
m
e
d
by
e
a
c
h
pr
o
c
e
s
s
i
ng
uni
t
(
GPU,
T
P
U
,
DPU,
QPU)
dur
i
n
g
de
e
p
l
e
a
r
ni
ng
m
o
de
l
t
r
a
i
ni
ng
de
pe
n
d
s
o
n
s
e
v
e
r
a
l
f
a
c
t
o
r
s
,
s
uc
h
a
s
t
h
e
s
pe
c
i
f
i
c
ha
r
dwa
r
e
,
w
o
r
kl
o
a
d,
a
n
d
t
h
e
dur
a
t
i
o
n
o
f
t
r
a
i
ni
n
g.
T
h
e
tot
a
l
e
n
e
r
g
y
c
o
ns
u
m
e
d
E
(
i
n
J
o
u
l
e
s
o
r
ki
l
o
wa
t
t
-
h
o
u
r
s
)
c
a
n
b
e
e
s
t
i
m
a
t
e
d
us
i
n
g
t
h
e
f
o
l
l
o
w
i
ng
f
o
r
m
u
l
a
:
=
P
∗
T
(
1)
W
h
e
r
e
E
i
s
t
h
e
tot
a
l
e
n
e
r
gy
c
o
n
s
um
e
d
,
P
i
s
t
h
e
a
v
e
r
a
ge
p
owe
r
c
o
n
s
u
m
p
t
i
o
n
(
i
n
w
a
tt
s
)
,
T
i
s
t
h
e
t
i
m
e
o
f
op
e
r
a
t
i
o
n
(
i
n
s
e
c
o
n
ds
or
h
ou
r
s
)
[
1
8]
,
[
31
]
.
T
a
b
l
e
2
de
s
c
r
i
b
e
s
t
h
e
e
n
e
r
gy
c
o
n
s
u
m
p
t
i
o
n
i
n
j
ou
l
e
s
pe
r
h
ou
r
f
or
e
a
c
h
uni
t.
T
a
bl
e
2
.
C
o
m
pa
r
i
s
o
n
t
a
bl
e
U
ni
t
P
o
w
e
r
(
W
)
E
ne
r
g
y
f
o
r
1 h
o
ur
(
J
o
ul
e
s
/
MJ
)
G
P
U
400
1,440,000 J
(
1.44
)
T
P
U
300
1,080,000 J
(
1.08
)
D
P
U
75
270,000 J
(
0.27
)
Q
P
U
50
180,000 J
(
0.18
)
Hi
gh
-
p
e
r
f
or
m
a
n
c
e
G
P
Us
c
a
n
h
a
v
e
a
p
o
w
e
r
c
o
n
s
um
p
t
i
o
n
o
f
2
50
W
to
4
00
W
,
a
n
d
tr
a
i
ni
n
g
t
i
m
e
v
a
r
i
e
s
d
e
p
e
n
d
i
n
g
o
n
m
ode
l
c
o
m
p
l
e
x
i
t
y
,
da
ta
,
a
n
d
h
a
r
d
wa
r
e
b
u
t
T
=
1
h
o
u
r
(
3
,
60
0
s
e
c
o
n
ds
)
f
or
s
i
m
p
l
i
c
i
t
y
i
s
a
s
s
um
e
d
[
32
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
c
ompar
ati
v
e
analys
is
of
GP
Us
,
T
P
Us
,
DP
Us
,
and
QP
Us
f
or
de
e
p
lear
ning
w
it
h
p
y
thon
(
A
y
oub
A
ll
a
li
)
1327
=
P
∗
T
=
400
∗
3600
=
1
,
440
,
000
(
1
.
44
)
(
2
)
T
P
Us
a
r
e
m
o
r
e
p
o
we
r
-
e
f
f
i
c
i
e
n
t
f
o
r
l
a
r
ge
-
s
c
a
l
e
c
o
m
put
a
t
i
o
n
s
.
A
T
P
Uv
3
po
d
c
a
n
h
a
v
e
a
n
a
ve
r
a
ge
po
we
r
c
o
n
s
u
m
pt
i
o
n
o
f
a
r
o
un
d
200W
to
300W
pe
r
T
P
U
c
o
r
e
[
33]
.
=
P
∗
T
=
300
∗
3600
=
1
,
080
,
000
(
1
.
08
)
(
3
)
DPUs
l
i
ke
t
h
e
NV
I
DI
A
B
l
ue
F
i
e
l
d
-
2
ge
n
e
r
a
ll
y
c
o
n
s
u
m
e
a
r
o
un
d
50W
to
75W
de
pe
n
d
i
n
g
o
n
t
h
e
wo
r
kl
o
a
d.
H
o
we
v
e
r
,
s
i
n
c
e
DPUs
a
r
e
t
y
p
i
c
a
ll
y
n
o
t
us
e
d
f
o
r
m
o
de
l
t
r
a
i
ni
ng,
t
h
e
i
r
e
n
e
r
g
y
c
o
n
s
u
m
pt
i
o
n
i
n
t
hi
s
c
o
n
t
e
x
t
w
o
ul
d
r
e
l
a
t
e
to
da
t
a
h
a
n
d
li
ng
t
a
s
ks
[
34]
.
=
P
∗
T
=
75
∗
3600
=
270
,
0
0
0
(
0
.
27
)
(
4
)
QPUs
a
r
e
hi
g
hly
e
x
p
e
r
i
m
e
n
t
a
l
,
a
n
d
po
we
r
us
a
ge
v
a
r
i
e
s
s
i
g
nif
i
c
a
n
t
l
y
,
QPUs
i
n
r
e
s
e
a
r
c
h
e
nvi
r
o
nm
e
n
t
s
c
a
n
c
o
n
s
u
m
e
a
ny
w
h
e
r
e
f
r
o
m
10W
to
50W
o
r
m
o
r
e
,
b
u
t
t
h
e
y
a
r
e
ge
n
e
r
a
ll
y
n
o
t
us
e
d
f
o
r
p
r
o
l
o
n
ge
d
de
e
p
l
e
a
r
ni
ng
t
r
a
i
ni
ng
[
35]
–
[
37]
.
=
P
∗
T
=
50
∗
3600
=
180
,
0
0
0
(
0
.
18
)
(
5
)
3.
3.
P
ar
a
ll
e
l
is
m
P
a
r
a
l
l
e
li
s
m
i
s
a
m
e
a
s
ur
e
o
f
h
o
w
e
f
f
e
c
t
i
v
e
ly
a
c
om
put
i
n
g
s
y
s
t
e
m
c
a
n
pe
r
f
o
r
m
m
u
l
t
i
p
l
e
o
pe
r
a
t
i
o
ns
s
i
m
u
l
t
a
n
e
o
us
ly
.
GPU
s
e
x
c
e
l
i
n
t
hi
s
do
m
a
i
n
,
w
i
t
h
a
r
c
hi
t
e
c
t
ur
e
s
de
s
i
g
n
e
d
t
o
h
a
n
d
l
e
t
h
o
us
a
n
d
s
o
f
t
h
r
e
a
ds
i
n
pa
r
a
l
l
e
l
.
T
hi
s
c
a
pa
bil
i
t
y
m
a
ke
s
t
h
e
m
e
x
c
e
pt
i
o
n
a
ll
y
s
u
i
t
e
d
f
o
r
t
h
e
pa
r
a
l
l
e
l
n
a
t
ur
e
o
f
de
e
p
l
e
a
r
ni
ng
c
o
m
put
a
t
i
o
n
s
.
T
P
Us
a
l
s
o
o
f
f
e
r
s
i
g
ni
f
i
c
a
n
t
pa
r
a
l
le
l
pr
o
c
e
s
s
i
n
g
c
a
pa
bil
i
t
i
e
s
,
e
s
pe
c
i
a
ll
y
o
pt
i
mi
z
e
d
f
o
r
hi
g
h
-
v
o
l
u
m
e
,
l
o
w
-
pr
e
c
i
s
i
o
n
a
r
i
t
hm
e
t
i
c
o
pe
r
a
t
i
o
n
s
c
o
m
mo
n
i
n
de
e
p
l
e
a
r
ni
ng
a
l
go
r
i
t
hm
s
[
38]
–
[
43]
.
DPUs
c
o
n
tr
i
b
ut
e
to
pa
r
a
l
l
e
li
s
m
i
nd
i
r
e
c
t
l
y
by
o
f
f
lo
a
di
n
g
a
n
d
a
c
c
e
l
e
r
a
t
i
n
g
n
e
t
wor
ki
n
g,
s
e
c
ur
i
t
y
,
a
n
d
I
/O
o
pe
r
a
t
i
o
ns
,
whi
c
h
i
n
t
ur
n
e
n
a
bl
e
s
m
o
r
e
e
f
f
i
c
i
e
n
t
us
e
o
f
c
o
m
put
i
n
g
r
e
s
o
ur
c
e
s
f
o
r
pa
r
a
l
l
e
l
c
o
m
put
a
t
i
o
n
s
i
n
de
e
p
l
e
a
r
ni
ng
wo
r
kl
o
a
ds
.
QPUs
,
o
n
t
h
e
ot
h
e
r
h
a
nd,
o
f
f
e
r
a
f
o
r
m
o
f
pa
r
a
l
l
e
li
s
m
t
h
a
t
i
s
qua
li
t
a
t
i
v
e
ly
d
i
f
f
e
r
e
n
t
f
r
o
m
c
l
a
s
s
i
c
a
l
s
y
s
t
e
m
s
,
us
i
n
g
qua
n
t
u
m
e
n
t
a
n
g
le
m
e
n
t
a
n
d
s
upe
r
po
s
i
t
i
o
n
t
o
pe
r
f
o
r
m
a
v
a
s
t
n
um
be
r
o
f
c
a
l
c
u
l
a
t
i
o
ns
s
im
u
l
t
a
n
e
o
us
l
y
,
a
l
t
h
o
ugh
t
h
e
i
r
a
pp
li
c
a
t
i
o
n
i
n
de
e
p
l
e
a
r
ni
ng
i
s
s
t
i
ll
e
m
e
r
g
i
ng
[
44]
.
3.
4.
I
n
t
e
gr
a
t
ion
an
d
o
p
t
im
iz
at
ion
T
h
e
i
n
t
e
gr
a
t
i
o
n
o
f
t
h
e
s
e
a
c
c
e
l
e
r
a
t
o
r
s
i
n
t
o
de
e
p
l
e
a
r
ni
n
g
wo
r
k
f
l
o
ws
v
a
r
i
e
s
w
i
de
ly
.
GPUs
a
r
e
s
uppo
r
t
e
d
by
a
m
a
t
ur
e
e
c
o
s
y
s
t
e
m
o
f
t
oo
l
s
a
n
d
l
ib
r
a
r
i
e
s
,
s
uc
h
a
s
C
UD
A
f
o
r
NV
I
DI
A
GPUs
,
m
a
k
in
g
t
h
e
m
r
e
l
a
t
i
v
e
ly
s
t
r
a
i
g
h
t
f
o
r
wa
r
d
to
i
n
t
e
gr
a
t
e
i
n
t
o
e
xi
s
t
i
n
g
de
e
p
l
e
a
r
ni
ng
f
r
a
m
e
wo
r
ks
.
T
P
Us
a
r
e
a
l
s
o
we
l
l
-
s
u
ppo
r
t
e
d
,
pa
r
t
i
c
u
l
a
r
l
y
w
i
t
hi
n
Go
o
gl
e
’
s
e
c
o
s
y
s
t
e
m
,
w
i
t
h
T
e
ns
o
r
F
l
o
w
o
f
f
e
r
i
ng
s
e
a
ml
e
s
s
i
n
t
e
gr
a
t
i
o
n
.
DPUs
r
e
qu
i
r
e
m
o
r
e
s
pe
c
i
a
li
z
e
d
i
n
t
e
gr
a
t
i
o
n
e
f
f
o
r
t
s
,
f
o
c
us
i
ng
o
n
t
h
e
o
p
t
i
mi
z
a
t
i
o
n
o
f
da
t
a
c
e
n
t
e
r
o
pe
r
a
t
i
o
n
s
r
a
t
h
e
r
t
h
a
n
d
i
r
e
c
t
a
c
c
e
l
e
r
a
t
i
o
n
o
f
de
e
p
l
e
a
r
ni
ng
m
o
de
l
s
.
QPUs
,
b
e
i
ng
a
t
t
h
e
f
o
r
e
f
r
o
n
t
o
f
c
o
m
put
a
t
i
o
n
a
l
r
e
s
e
a
r
c
h
,
c
ur
r
e
n
t
l
y
r
e
qu
i
r
e
hi
g
hly
s
pe
c
i
a
li
z
e
d
k
n
o
w
l
e
dge
t
o
us
e
e
f
f
e
c
t
i
v
e
l
y
i
n
de
e
p
l
e
a
r
ni
ng
a
pp
l
i
c
a
t
i
o
ns
[
45]
,
[
46]
.
3.
5.
Chal
l
e
n
ge
s
I
n
de
e
p
l
e
a
r
ni
ng,
e
a
c
h
s
p
e
c
i
a
li
z
e
d
pr
o
c
e
s
s
i
ng
uni
t
GPU,
T
P
U,
D
P
U,
a
n
d
QPU
f
a
c
e
s
d
i
s
t
i
n
c
t
c
h
a
ll
e
n
g
e
de
s
c
r
i
be
s
i
n
T
a
bl
e
3
GPUs
a
r
e
po
we
r
f
u
l
f
o
r
pa
r
a
l
l
e
l
c
o
m
put
a
t
i
o
n
s
b
ut
s
tr
uggl
e
w
i
t
h
hi
g
h
e
n
e
r
g
y
c
o
n
s
u
m
pt
i
o
n
,
m
e
m
o
r
y
b
a
n
dw
i
dt
h
l
im
i
t
a
t
i
o
n
s
,
a
n
d
i
ne
f
f
i
c
i
e
n
c
y
i
n
s
m
a
ll
m
o
de
l
s
.
T
P
Us
,
de
s
i
g
n
e
d
f
o
r
t
e
n
s
o
r
o
pe
r
a
t
i
o
n
s
,
o
f
f
e
r
s
t
r
o
n
g
pe
r
f
o
r
m
a
nc
e
b
ut
a
r
e
l
im
it
e
d
by
t
h
e
i
r
l
a
c
k
o
f
f
l
e
xi
b
il
i
t
y
,
s
t
e
e
p
l
e
a
r
ni
ng
c
u
r
v
e
,
a
n
d
de
pe
n
d
e
n
c
e
o
n
l
a
r
ge
b
a
t
c
h
s
i
z
e
s
.
DPUs
,
whil
e
us
e
f
u
l
f
o
r
d
i
s
t
r
i
b
ut
e
d
s
y
s
t
e
m
s
by
h
a
n
d
li
ng
da
t
a
a
n
d
n
e
t
wo
r
k
t
a
s
ks
,
h
a
v
e
ni
c
h
e
us
e
c
a
s
e
s
,
c
o
m
p
l
e
x
s
o
f
t
wa
r
e
i
n
t
e
gr
a
t
i
o
n
,
a
n
d
l
i
mi
t
e
d
m
a
t
ur
i
t
y
.
QPUs
o
f
f
e
r
t
h
e
o
r
e
t
i
c
a
l
s
pe
e
d
-
ups
f
o
r
s
pe
c
i
f
i
c
t
a
s
ks
b
ut
f
a
c
e
m
a
j
o
r
h
ur
d
l
e
s
,
s
uc
h
a
s
h
a
r
dwa
r
e
i
mm
a
t
ur
i
t
y
,
i
n
t
e
gr
a
t
i
o
n
w
i
t
h
c
l
a
s
s
i
c
a
l
s
y
s
t
e
m
s
,
a
n
d
a
l
go
r
i
t
hm
i
c
de
v
e
l
o
p
m
e
n
t
,
m
a
k
i
ng
t
h
e
m
l
e
s
s
pr
a
c
t
i
c
a
l
f
o
r
c
ur
r
e
n
t
de
e
p
l
e
a
r
ni
ng
n
e
e
ds
[
4
7]
,
[
48]
.
T
a
bl
e
3
.
C
h
a
ll
e
n
ge
s
o
f
e
a
c
h
uni
t
U
ni
t
G
P
U
T
P
U
D
P
U
Q
P
U
M
a
jo
r
c
ha
ll
e
ng
e
s
H
ig
h
e
n
e
r
g
y
c
o
ns
umpt
i
o
n
M
e
mor
y
ba
ndw
id
th
bo
tt
l
e
n
e
c
ks
S
c
a
la
bi
li
t
y
i
s
s
ue
s
I
ne
f
f
i
c
ie
n
c
y
w
it
h
s
ma
ll
mo
d
e
ls
H
ig
h
c
o
s
t
L
im
it
e
d
f
l
e
x
ib
il
it
y
P
r
o
gr
a
mm
in
g
c
o
mpl
e
x
it
y
M
e
mor
y
b
o
tt
l
e
n
e
c
ks
C
lo
ud
-
o
nl
y
a
v
a
il
a
bi
li
t
y
B
a
tc
h s
i
z
e
s
e
ns
it
i
v
it
y
N
ic
h
e
us
e
c
a
s
e
s
S
of
twa
r
e
e
c
o
s
y
s
t
e
m
li
mi
ta
ti
o
ns
I
nt
e
g
r
a
ti
o
n
ove
r
he
a
d
C
o
s
t
M
a
tu
r
it
y
of
te
c
hn
o
l
o
g
y
H
a
r
dw
a
r
e
im
ma
tu
r
it
y
A
lg
o
r
i
th
mi
c
li
mi
ta
ti
o
ns
C
la
s
s
ic
a
l
qua
nt
um
in
te
gr
a
ti
o
n i
s
s
ue
s
L
im
it
e
d us
e
c
a
s
e
s
H
ig
h
c
o
s
t
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
38
,
N
o
.
2
,
M
a
y
20
25
:
1
324
-
1
330
1328
L
a
s
t
l
y
,
s
e
l
e
c
t
i
n
g
t
h
e
r
i
g
h
t
h
a
r
dwa
r
e
a
c
c
e
l
e
r
a
to
r
f
o
r
a
de
e
p
l
e
a
r
ni
ng
pr
o
j
e
c
t
i
nv
o
l
ve
s
a
c
a
r
e
f
u
l
c
o
n
s
i
de
r
a
t
i
o
n
o
f
t
h
e
s
pe
c
i
f
i
c
r
e
qu
i
r
e
m
e
n
t
s
a
n
d
c
o
ns
t
r
a
i
n
t
s
o
f
t
h
e
pr
o
j
e
c
t
,
i
n
c
l
ud
i
ng
c
o
m
put
a
t
i
o
n
a
l
e
f
f
i
c
i
e
n
c
y
,
pa
r
a
l
l
e
li
s
m
,
e
n
e
r
g
y
c
o
n
s
u
m
pt
i
o
n
,
a
n
d
t
h
e
e
a
s
e
o
f
i
n
t
e
gr
a
t
i
o
n
.
A
s
t
hi
s
f
i
e
l
d
c
o
n
t
i
n
u
e
s
to
e
v
o
l
v
e
r
a
pi
d
ly
,
s
t
a
y
i
n
g
i
nf
o
r
m
e
d
a
b
o
ut
t
h
e
l
a
t
e
s
t
de
v
e
l
o
p
m
e
n
t
s
i
n
h
a
r
dwa
r
e
a
c
c
e
l
e
r
a
t
or
s
wi
ll
be
c
r
uc
i
a
l
f
o
r
m
a
xi
mi
z
i
n
g
t
h
e
pe
r
f
o
r
m
a
n
c
e
a
n
d
e
f
f
i
c
i
e
nc
y
o
f
de
e
p
l
e
a
r
ni
ng
a
pp
l
i
c
a
t
i
o
n
s
.
4.
CONC
L
USI
ON
W
hil
e
GPUs
,
T
P
Us
,
DPUs
,
a
n
d
QPUs
e
a
c
h
b
r
i
ng
s
i
g
nif
i
c
a
n
t
p
ot
e
n
t
i
a
l
t
o
a
c
c
e
l
e
r
a
t
e
de
e
p
l
e
a
r
ni
ng
t
a
s
ks
,
t
h
e
y
a
l
s
o
c
o
m
e
w
i
t
h
uni
qu
e
c
h
a
ll
e
n
ge
s
t
h
a
t
m
us
t
b
e
c
o
n
s
i
de
r
e
d
b
a
s
e
d
o
n
t
h
e
s
pe
c
i
f
i
c
us
e
c
a
s
e
.
GPUs
r
e
m
a
i
n
t
h
e
m
o
s
t
v
e
r
s
a
t
i
l
e
a
n
d
w
i
de
ly
us
e
d,
b
ut
t
h
e
i
r
e
ne
r
g
y
de
m
a
n
d
s
a
n
d
s
c
a
l
a
bi
li
t
y
i
s
s
ue
s
c
a
n
b
e
li
mi
t
i
n
g.
T
P
Us
o
f
f
e
r
s
pe
c
i
a
li
z
e
d
pe
r
f
o
r
m
a
n
c
e
a
dv
a
n
t
a
ge
s
,
pa
r
t
i
c
u
l
a
r
l
y
f
o
r
l
a
r
ge
-
s
c
a
l
e
m
o
de
l
s
,
b
ut
a
r
e
l
e
s
s
f
l
e
xi
b
l
e
a
n
d
m
o
r
e
d
i
f
f
i
c
u
l
t
to
pr
o
g
r
a
m
.
DPUs
s
e
r
ve
ni
c
he
r
o
l
e
s
i
n
d
i
s
t
r
i
b
ut
e
d
s
y
s
t
e
m
s
b
ut
r
e
qu
i
r
e
c
o
m
p
l
e
x
i
n
t
e
gr
a
ti
o
n
,
a
nd
QPUs
,
t
h
o
ugh
pr
o
m
i
s
i
ng,
a
r
e
s
t
i
l
l
i
n
t
h
e
e
x
pe
r
im
e
n
t
a
l
s
t
a
ge
a
n
d
n
ot
y
e
t
pr
a
c
t
i
c
a
l
f
o
r
ge
n
e
r
a
l
-
pur
p
o
s
e
de
e
p
l
e
a
r
ni
ng.
T
h
e
c
h
o
i
c
e
o
f
h
a
r
dwa
r
e
u
l
t
i
m
a
t
e
l
y
de
p
e
nds
o
n
b
a
l
a
n
c
i
ng
t
h
e
s
e
t
r
a
de
-
o
f
f
s
w
i
t
h
t
h
e
s
pe
c
i
f
i
c
n
e
e
d
s
o
f
t
h
e
de
e
p
l
e
a
r
ni
ng
a
pp
li
c
a
t
i
o
n
.
RE
F
E
R
E
NC
E
S
[
1]
N
. S
ha
h, L
.
I
.
G
. O
la
s
c
o
a
ga
, S
. Z
ha
o
, W
. M
e
e
r
t,
a
nd M
. V
e
r
he
l
s
t,
“
D
P
U
:
D
A
G
pr
o
c
e
s
s
in
g unit
f
o
r
ir
r
e
gul
a
r
g
r
a
phs
w
it
h pr
e
c
is
io
n
-
s
c
a
la
bl
e
p
o
s
it
a
r
it
hm
e
ti
c
in
28
nm,”
I
E
E
E
J
our
nal
of
Sol
id
-
St
at
e
C
ir
c
ui
ts
,
vo
l.
57,
n
o
.
8,
pp.
2586
–
2596,
20
22,
do
i:
10.1109/J
S
S
C
.2021.3134897.
[
2]
J
.
M
.
R
o
dr
íg
ue
z
C
o
r
r
a
l,
J
.
C
i
v
it
-
M
a
s
o
t,
F
.
L
una
-
P
e
r
e
jó
n,
I
.
D
ía
z
-
C
a
no
,
A
.
M
o
r
ga
d
o
-
E
s
té
ve
z
,
a
nd
M
.
D
o
m
ín
gue
z
-
M
o
r
a
le
s
,
“
E
ne
r
g
y
e
f
f
i
c
i
e
n
c
y
in
e
dge
T
P
U
v
s
.
e
mb
e
dd
e
d
G
P
U
f
or
c
omput
e
r
-
a
id
e
d
m
e
di
c
a
l
im
a
gi
ng
s
e
gme
nt
a
ti
o
n
a
nd
c
la
s
s
if
i
c
a
ti
o
n,”
En
gi
ne
e
r
in
g A
ppl
ic
at
io
ns
of
A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
, v
o
l.
127, J
a
n. 2024, do
i:
10.1016/j
.
e
nga
ppa
i.
2023.107298.
[
3]
H
.
H
ua
ng,
X
.
Y
.
L
iu
,
W
.
T
o
ng,
T
.
Z
ha
ng,
A
.
W
a
li
d,
a
nd
X
.
W
a
ng,
“
H
ig
h
pe
r
f
or
ma
n
c
e
p
e
r
f
or
ma
n
c
e
hi
e
r
a
r
c
h
ic
a
l
tu
c
k
e
r
t
e
ns
o
r
le
a
r
ni
ng
us
in
g
G
P
U
t
e
ns
or
c
o
r
e
s
,”
I
E
E
E
T
r
ans
ac
ti
ons
on
C
om
put
e
r
s
,
v
o
l.
72,
n
o
.
2,
pp.
452
–
465,
2023,
do
i:
10.1109/
T
C
.2022.3172895.
[
4]
S
.
C
.
M
a
ga
lh
ã
e
s
,
F
.
N
.
do
s
S
a
nt
o
s
,
P
.
M
a
c
ha
do
,
A
.
P
.
M
o
r
e
ir
a
,
a
nd
J
.
D
ia
s
,
“
B
e
nc
hma
r
ki
ng
e
dg
e
c
omput
in
g
de
v
i
c
e
s
f
or
gr
a
pe
bunc
h
e
s
a
nd
tr
unks
de
te
c
t
i
o
n
us
in
g
a
c
c
e
l
e
r
a
te
d
o
bj
e
c
t
de
te
c
ti
o
n
s
in
gl
e
s
h
o
t
mul
ti
b
ox
de
e
p
l
e
a
r
ni
ng
m
o
de
ls
,”
E
ngi
ne
e
r
in
g
A
ppl
ic
at
io
ns
of
A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
, v
o
l.
117, 2023, d
o
i:
10.1
016/
j.
e
nga
ppa
i.
2022.105604.
[
5]
S
.
L
a
ue
,
M
.
B
la
c
he
r
,
a
nd
J
.
G
i
e
s
e
n,
“
O
pt
im
iz
a
ti
o
n
f
o
r
c
la
s
s
ic
a
l
ma
c
hi
n
e
l
e
a
r
ni
ng
pr
o
bl
e
ms
o
n
th
e
G
P
U
,”
in
P
r
oc
e
e
di
ngs
of
th
e
36t
h A
A
A
I
C
onf
e
r
e
n
c
e
on A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
, A
A
A
I
2022
, 20
22, vo
l.
36, pp. 7300
–
7308, d
o
i:
10.1609/aa
a
i.
v
36i
7.20692.
[
6]
J
.
S
.
L
e
r
a
t,
S
.
A
.
M
a
hmo
udi
,
a
nd
S
.
M
a
hm
o
udi
,
“
S
in
gl
e
n
ode
d
e
e
p
l
e
a
r
ni
ng
f
r
a
m
e
w
o
r
ks
:
C
o
mpa
r
a
ti
ve
s
tu
d
y
a
nd
C
P
U
/
G
P
U
pe
r
f
or
ma
nc
e
a
na
l
y
s
is
,”
C
onc
ur
r
e
nc
y
and
C
om
put
at
io
n:
P
r
ac
ti
c
e
and E
x
pe
r
ie
nc
e
, v
o
l.
35, n
o
. 14, 2023, d
o
i:
10.1002/
c
p
e
.6730.
[
7]
S
.
T
h
o
ma
s
,
M
.
L
.
B
o
w
e
r
s
,
a
nd
W
.
L
ia
o
,
“
G
P
U
c
omput
in
g
a
nd
it
s
a
ppl
ic
a
ti
o
ns
in
A
I
a
nd
de
e
p
le
a
r
n
in
g,”
P
r
oc
e
e
di
ngs
of
th
e
W
e
s
t
V
ir
gi
ni
a A
c
ade
m
y
of
Sc
ie
n
c
e
, v
o
l.
90, n
o
. 1, Apr
. 2018, d
oi
:
10.
55632/pwv
a
s
.v
90i
1.408.
[
8]
D
.
Z
ha
ng,
Y
.
L
u
o
,
Y
.
W
a
ng,
X
.
K
ui
,
a
nd
J
.
R
e
n,
“
B
a
tOpt
:
O
p
ti
mi
z
in
g
G
P
U
-
ba
s
e
d
d
e
e
p
l
e
a
r
ni
ng
in
f
e
r
e
nc
e
us
in
g
d
y
na
mi
c
ba
tc
h
pr
oc
e
s
s
in
g,”
I
E
E
E
T
r
ans
ac
ti
ons
on C
lo
ud C
om
put
in
g
, v
o
l.
12,
no
. 1, pp. 174
–
185, 2024, d
o
i:
10.1109/
T
C
C
.2024.3350561.
[
9]
X
.
S
hi
,
X
.
P
e
ng,
L
.
H
e
,
Y
.
Z
ha
o
,
a
nd
H
.
J
in
,
“
W
a
te
r
w
a
ve
:
A
G
P
U
me
m
or
y
f
l
o
w
e
ngi
n
e
f
o
r
c
o
nc
u
r
r
e
nt
D
N
N
tr
a
in
in
g,”
I
E
E
E
T
r
ans
ac
ti
ons
on C
o
m
put
e
r
s
, v
ol
. 72, n
o
. 10, pp. 2938
–
2950, 20
23, do
i:
10.1109/
T
C
.2023.3278530.
[
10]
Y
.
W
a
ng
e
t
al
.
,
“
D
R
L
C
A
P
:
R
unt
im
e
G
P
U
f
r
e
que
n
c
y
c
a
p
pi
ng
w
it
h
d
e
e
p
r
e
in
f
or
c
e
m
e
nt
l
e
a
r
ni
ng,”
I
E
E
E
T
r
ans
ac
ti
ons
on
Sus
ta
in
abl
e
C
om
put
in
g
, v
ol
. 9, n
o
. 5, pp. 712
–
726, S
e
p. 2024,
do
i:
10.1109/
T
S
U
S
C
.2024.3362697.
[
11]
X
.
Z
ha
ng,
“
M
ix
tr
a
n:
A
n
e
f
f
ic
i
e
nt
a
nd
f
a
i
r
s
c
h
e
dul
e
r
f
or
mi
xe
d
de
e
p
l
e
a
r
ni
ng
w
o
r
kl
o
a
ds
in
h
e
te
r
o
g
e
ne
o
us
G
P
U
e
n
v
ir
o
nm
e
nt
s
,”
C
lu
s
te
r
C
om
put
in
g
, vo
l.
27, n
o
. 3, pp. 2775
–
2784, 2024, d
o
i:
1
0.1007/s
10586
-
023
-
04104
-
9.
[
12]
L.
D
e
ni
s
,
R
.
R
oy
e
n,
Q
.
B
o
ls
é
e
,
N
.
V
e
r
c
h
e
v
a
l,
A
.
P
i
ž
ur
i
c
a
,
a
nd
A
.
M
unt
e
a
nu,
“
G
P
U
r
a
s
te
r
iz
a
ti
o
n
-
ba
s
e
d
3D
L
iDA
R
s
im
ul
a
ti
o
n
f
or
de
e
p l
e
a
r
n
in
g,”
Se
ns
or
s
, v
o
l.
23, n
o
. 19, 2023, d
o
i:
10.3390/s
23
198130.
[
13]
R
.
L
.
C
a
s
tr
o
,
D
.
A
ndr
a
de
,
a
nd
B
.
B
.
F
r
a
gu
e
la
,
“
S
T
uni
ng
-
DL
:
M
o
de
l
-
dr
i
ve
n
a
ut
o
tu
ni
ng
of
s
pa
r
s
e
G
P
U
k
e
r
n
e
ls
f
o
r
de
e
p
l
e
a
r
ni
ng,”
I
E
E
E
A
c
c
e
s
s
, v
ol
. 12, 2024, d
o
i:
10.1109/AC
C
E
S
S
.2024.34023
26.
[
14]
A
.
R
a
v
ik
uma
r
,
H
.
S
r
ir
a
ma
n,
P
.
M
.
S
a
i
S
a
ke
th
,
S
.
L
o
k
e
s
h,
a
nd
A
.
K
a
r
a
na
m,
“
E
f
f
e
c
t
of
ne
u
r
a
l
ne
tw
o
r
k
s
tr
u
c
tu
r
e
in
a
c
c
e
l
e
r
a
ti
ng
pe
r
f
or
ma
nc
e
a
nd
a
c
c
u
r
a
c
y
of
a
c
o
n
vo
lu
ti
o
na
l
ne
u
r
a
l
ne
tw
o
r
k
w
it
h
G
P
U
/
T
P
U
f
o
r
im
a
g
e
a
na
l
y
ti
c
s
,”
P
e
e
r
J
C
om
put
e
r
Sc
ie
nc
e
,
vo
l.
8, M
a
r
. 2022, d
o
i:
10.7717/p
e
e
r
j
-
c
s
.909.
[
15]
Z
.
Y
e
e
t
al
.
,
“
D
e
e
p
l
e
a
r
ni
ng
w
o
r
kl
o
a
d
s
c
h
e
dul
in
g
in
G
P
U
d
a
ta
c
e
nt
e
r
s
:
A
s
ur
ve
y
,”
A
C
M
C
om
put
in
g
Sur
v
e
y
s
,
v
o
l.
56,
no.
6,
pp. 1
–
38, J
un. 2024, do
i:
10.1145/3638757.
[
16]
Z
.
P
e
ng,
J
.
D
u,
a
nd
Y
.
Q
ia
o
,
“
D
e
s
ig
n
of
G
P
U
ne
tw
or
k
-
on
-
c
h
ip
f
o
r
r
e
a
l
-
ti
me
vi
de
o
s
up
e
r
-
r
e
s
o
lu
ti
o
n
r
e
c
o
ns
tr
u
c
ti
o
n,”
M
ic
r
om
ac
hi
ne
s
, v
o
l.
14, n
o
. 5, M
a
y
2023, d
o
i:
10.3390
/m
i1
405
1055.
[
17]
J
.
R
a
y
m
o
nd
e
t
al
.
,
“
H
y
b
r
id
qua
nt
um
a
nne
a
li
ng
f
or
la
r
g
e
r
-
th
a
n
-
Q
P
U
qua
nt
um
a
nne
a
li
ng
f
o
r
la
r
g
e
r
-
th
a
n,”
A
C
M
T
r
ans
ac
ti
ons
on
Q
uant
um
C
om
put
in
g
, vo
l.
4, n
o
. 3, pp. 1
–
30, S
e
p. 2023, d
o
i:
10
.1145/3579368.
[
18]
J
.
V
á
z
que
z
-
P
é
r
e
z
,
C
.
P
iñ
e
ir
o
,
J
.
C
.
P
ic
h
e
l,
T
.
F
.
P
e
na
,
a
n
d
A
.
G
ó
me
z
,
“
Q
P
U
in
t
e
gr
a
ti
o
n
in
O
pe
nC
L
f
o
r
he
te
r
o
ge
n
e
o
us
pr
o
g
r
a
mm
in
g,”
T
he
J
our
nal
of
Supe
r
c
om
put
in
g
, v
o
l.
80, n
o
. 8,
pp. 11682
–
11703, M
a
y
2024, d
o
i
:
10.1007/s
11227
-
023
-
05879
-
9.
[
19]
T
.
T
a
n
a
nd
G
.
C
a
o
,
“
T
h
e
r
ma
l
-
a
w
a
r
e
s
c
he
dul
in
g
f
o
r
d
e
e
p
l
e
a
r
n
in
g
o
n
m
o
b
il
e
d
e
v
i
c
e
s
w
it
h
N
P
U
,”
I
E
E
E
T
r
ans
ac
ti
ons
on
M
o
bi
le
C
om
put
in
g
, vo
l.
23, n
o
. 12, pp. 10706
–
10719, D
e
c
. 2024, d
o
i
:
10.1109/
T
M
C
.2024.3379501.
[
20]
A
.
M
a
r
tí
n
-
M
a
r
tí
n
e
t
al
.
,
“
H
a
r
dw
a
r
e
im
pl
e
m
e
nt
a
ti
o
ns
of
a
d
e
e
p
l
e
a
r
ni
ng
a
ppr
o
a
c
h
t
o
o
pt
im
a
l
c
o
n
f
ig
u
r
a
ti
o
n
of
r
e
c
o
n
f
ig
ur
a
bl
e
in
te
ll
ig
e
n
c
e
s
ur
f
a
c
e
s
,”
Se
ns
or
s
, v
o
l.
24, n
o
. 3, J
a
n. 2024, do
i:
1
0.3390/s
24030899.
[
21]
A
.
M
o
r
ni
ngs
ta
r
e
t
al
.
,
“
S
im
ul
a
ti
o
n
of
qua
nt
um
ma
n
y
-
b
o
d
y
d
y
na
mi
c
s
w
it
h
te
ns
o
r
p
r
o
c
e
s
s
in
g
uni
ts
:
F
l
o
qu
e
t
pr
e
th
e
r
ma
li
z
a
ti
o
n,”
P
R
X
Q
uant
um
,
vo
l.
3, n
o
. 2, M
a
y
2022, d
o
i:
10.1103/P
R
X
Q
ua
nt
um.3.020331.
[
22]
O
.
H
ua
ng
a
nd
M
.
L
.
P
a
lm
e
r
i,
“
T
P
U
ba
s
e
d
d
e
e
p
l
e
a
r
ni
n
g
im
a
ge
e
nha
nc
e
m
e
nt
f
or
r
e
a
l
-
ti
m
e
p
o
in
t
-
of
-
c
a
r
e
ul
t
r
a
s
o
u
nd,”
I
E
E
E
T
r
ans
a
c
ti
ons
on C
om
put
at
io
nal
I
m
agi
ng
, vo
l.
10, pp. 461
–
468, 2024, do
i:
10.1109/
T
C
I
.2024.3372445.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
I
S
S
N:
2
5
0
2
-
4
7
52
A
c
ompar
ati
v
e
analys
is
of
GP
Us
,
T
P
Us
,
DP
Us
,
and
QP
Us
f
or
de
e
p
lear
ning
w
it
h
p
y
thon
(
A
y
oub
A
ll
a
li
)
1329
[
23]
H
.
C
h
o
i,
B
.
H
.
L
e
e
,
S
.
Y
.
C
hun,
a
nd
J
.
L
e
e
,
“
T
o
w
a
r
ds
a
c
c
e
le
r
a
ti
ng
m
o
d
e
l
pa
r
a
ll
e
li
s
m
in
di
s
tr
ib
ut
e
d
d
e
e
p
l
e
a
r
ni
ng
s
y
s
t
e
ms
,”
P
L
O
S O
N
E
,
vo
l.
18, n
o
. 11, N
ov
. 20
23, d
o
i:
10.1371/j
o
ur
na
l.
p
one
.0293338.
[
24]
A
.
E
r
c
i
y
a
s
a
nd
N
.
B
a
r
iş
ç
i,
“
A
me
ta
-
a
na
l
y
s
is
o
n
di
a
b
e
ti
c
r
e
ti
no
pa
th
y
a
nd
d
e
e
p
l
e
a
r
ni
ng
a
ppl
ic
a
ti
o
ns
,”
M
ul
ti
m
e
di
a
T
ool
s
and
A
ppl
ic
at
io
ns
, vo
l.
83, n
o
. 19, pp. 57429
–
57448, D
e
c
. 2023, d
o
i:
10.1007/s
11042
-
023
-
17784
-
7
.
[
25]
F
.
G
.
T
a
n,
“
T
h
e
im
pa
c
t
of
de
e
p
l
e
a
r
ni
ng
a
nd
tr
a
ns
f
e
r
l
e
a
r
ni
n
g
a
lg
o
r
i
th
ms
o
n
dr
o
n
e
d
e
t
e
c
t
i
o
n
pe
r
f
o
r
ma
nc
e
,
(
I
n
T
u
r
ki
s
h:
D
e
r
in
öğr
e
nm
e
ve
öğr
e
nm
e
a
kt
a
r
ım
ı
a
lg
o
r
it
ma
la
r
ın
ın
d
r
o
n
e
a
lg
ıl
a
ma
pe
r
f
or
ma
ns
ı
üz
e
r
in
e
e
tk
is
i)
,”
G
az
i
J
our
nal
o
f
E
ngi
ne
e
r
in
g
Sc
ie
nc
e
s
, v
o
l.
9, n
o
. 4, pp. 1
–
13, De
c
. 2023, d
o
i:
10.30855
/g
mb
d.0705S
01.
[
26]
G
.
A
kka
d,
A
.
M
a
ns
o
ur
,
a
nd
E
.
I
na
t
y
,
“
E
mb
e
dd
e
d
d
e
e
p
l
e
a
r
ni
n
g
a
c
c
e
l
e
r
a
t
o
r
s
:
A
s
ur
ve
y
o
n
r
e
c
e
nt
a
dv
a
n
c
e
s
,”
I
E
E
E
T
r
ans
ac
ti
ons
on A
r
ti
f
ic
ia
l
I
nt
e
ll
ig
e
nc
e
, v
o
l.
5, n
o
. 5, pp. 1954
–
1972, M
a
y
20
24, do
i:
10.1109/
T
A
I
.2023.3311776.
[
27]
Y
.
W
u
e
t
al
.
,
“
E
la
s
ti
c
d
e
e
p
l
e
a
r
ni
ng
in
mul
ti
-
t
e
na
nt
G
P
U
c
l
us
te
r
s
,”
I
E
E
E
T
r
ans
ac
ti
ons
on
P
ar
al
le
l
and
D
is
tr
ib
ut
e
d
Sy
s
te
m
s
,
vo
l.
33, n
o
. 1, pp. 144
–
158, J
a
n. 2022, doi:
10.1109/
T
P
D
S
.202
1.3064966.
[
28]
H
.
C
hoi
a
nd
J
.
L
e
e
,
“
E
f
f
i
c
i
e
nt
us
e
of
G
P
U
m
e
m
o
r
y
f
or
la
r
g
e
-
s
c
a
l
e
d
e
e
p
l
e
a
r
ni
ng
m
o
d
e
l
tr
a
in
in
g,”
A
pp
li
e
d
Sc
ie
nc
e
s
,
v
o
l.
11,
no
. 21, N
ov
. 2021, d
o
i:
10.3390/app112110
377.
[
29]
M
.
P
a
nde
y
e
t
al
.
,
“
T
he
t
r
a
ns
f
or
ma
ti
o
na
l
r
o
le
of
G
P
U
c
o
m
put
in
g
a
nd
de
e
p
l
e
a
r
ni
ng
in
dr
ug
di
s
c
ov
e
r
y
,”
N
at
ur
e
M
ac
hi
ne
I
nt
e
ll
ig
e
nc
e
, vo
l.
4, n
o
. 3, pp. 211
–
221, M
a
r
. 2022, d
o
i:
10.103
8/
s
42256
-
022
-
00463
-
x.
[
30]
Z
.
C
he
n,
X
.
Z
ha
o
,
C
.
Z
hi
,
a
nd
J
.
Y
in
,
“
D
e
e
pB
oo
t:
D
y
na
mi
c
s
c
h
e
dul
in
g
s
y
s
t
e
m
f
o
r
t
r
a
in
in
g
a
nd
in
f
e
r
e
n
c
e
de
e
p
l
e
a
r
ni
ng
ta
s
ks
in
G
P
U
c
lu
s
t
e
r
,”
I
E
E
E
T
r
ans
ac
ti
ons
on
P
ar
al
le
l
and
D
is
tr
ib
ut
e
d
Sy
s
te
m
s
,
vo
l.
34,
n
o
.
9,
pp.
2553
–
2567,
S
e
p.
2
023,
do
i:
10.1109/
T
P
D
S
.2023.3293835.
[
31]
Y
.
L
ia
o
,
J
.
W
u,
W
.
L
u,
X
.
L
i,
a
nd
G
.
Y
a
n,
“
D
P
U
-
di
r
e
c
t:
U
nl
e
a
s
hi
ng
r
e
m
o
t
e
a
c
c
e
le
r
a
t
o
r
s
v
ia
e
nha
n
c
e
d
R
D
M
A
f
o
r
di
s
a
ggr
e
g
a
te
d
da
ta
c
e
nt
e
r
s
,”
I
E
E
E
T
r
ans
ac
ti
ons
on C
om
put
e
r
s
, v
o
l.
73, n
o
. 8,
pp. 2081
–
2095, Aug. 2024, do
i:
10.1109/
T
C
.2024.3404089.
[
32]
B
.
C
h
o
pr
a
,
“
E
nha
n
c
in
g
ma
c
h
in
e
le
a
r
n
in
g
pe
r
f
o
r
ma
nc
e
:
T
he
r
o
l
e
of
G
P
U
-
ba
s
e
d
A
I
c
o
mput
e
a
r
c
hi
t
e
c
tu
r
e
s
,”
J
our
nal
of
K
now
le
dge
L
e
ar
ni
ng and
Sc
ie
nc
e
T
e
c
hnol
og
y
, v
ol
. 3, n
o
. 3, pp. 29
–
42, M
a
r
. 2024, do
i:
10.60087/j
k
ls
t.
vo
l3
.n3.p40.
[
33]
Z
.
P
a
n,
F
.
Z
ha
ng,
H
.
L
i,
C
.
Z
ha
ng,
X
.
D
u,
a
nd
D
.
D
e
ng,
“
G
-
S
L
I
D
E
:
A
G
P
U
-
ba
s
e
d
s
ub
-
li
n
e
a
r
d
e
e
p
l
e
a
r
ni
ng
e
ngi
n
e
v
ia
L
S
H
s
pa
r
s
if
i
c
a
ti
o
n,”
I
E
E
E
T
r
ans
ac
ti
ons
on
P
ar
al
l
e
l
and
D
is
tr
ib
ut
e
d
Sy
s
te
m
s
,
v
o
l.
33,
n
o
.
11,
pp.
3015
--
3027,
20
21,
do
i:
10.1109/
T
P
D
S
.2021.3132493.
[
34]
P
.
M
a
il
la
r
d
e
t
al
.
,
“
R
a
di
a
ti
o
n
t
o
l
e
r
a
nt
d
e
e
p
l
e
a
r
ni
ng
p
r
oc
e
s
s
or
uni
t
(
D
P
U
)
ba
s
e
d
pl
a
t
f
or
m
us
in
g
X
il
in
x
20n
m
K
in
t
e
x
U
lt
r
a
S
c
a
le
TM
F
P
G
A
,”
I
E
E
E
T
r
ans
ac
ti
on
s
on N
uc
le
a
r
Sc
ie
nc
e
, v
o
l.
70, n
o
. 4,
pp. 714
–
721, Apr
. 2023, do
i:
10.1109
/
T
N
S
.2022.3216360.
[
35]
C
.
L
i,
R
.
X
u,
Y
.
L
v
,
Y
.
Z
ha
o
,
a
nd
W
.
J
in
g,
“
E
dge
r
e
a
l
-
ti
m
e
obj
e
c
t
de
te
c
ti
o
n
a
nd
dpu
-
ba
s
e
d
ha
r
dw
a
r
e
i
mpl
e
m
e
nt
a
ti
o
n
f
or
o
p
ti
c
a
l
r
e
m
o
t
e
s
e
ns
in
g i
ma
g
e
s
,”
R
e
m
ot
e
S
e
ns
in
g
, v
ol
. 15, n
o
. 16, Aug. 2023, do
i:
10.3390/
r
s
15163975.
[
36]
Z
.
W
a
ng,
C
.
W
a
ng,
a
nd
L
.
W
a
ng,
“
D
P
U
B
e
n
c
h:
A
n
a
ppl
i
c
a
ti
o
n
-
d
r
i
v
e
n
s
c
a
la
bl
e
b
e
n
c
hma
r
k
s
ui
t
e
f
or
c
o
mp
r
e
h
e
ns
i
ve
D
P
U
e
v
a
lu
a
ti
o
n,”
B
e
nc
hC
ounc
il
T
r
ans
a
c
ti
ons
on
B
e
nc
hm
ar
k
s
,
St
andar
ds
and
E
v
al
uat
io
ns
,
v
o
l.
3,
no
.
2,
J
un.
20
23,
do
i:
10.1016/j
.
tb
e
n
c
h.2023.100120.
[
37]
R
.
W
a
ng,
X
.
Y
u,
Q
.
W
u,
C
.
Y
i,
P
.
W
a
ng,
a
nd
D
.
N
i
y
a
t
o
,
“
E
f
f
ic
i
e
nt
d
e
pl
oy
m
e
nt
of
pa
r
ti
a
l
pa
r
a
ll
e
li
z
e
d
s
e
r
v
i
c
e
f
un
c
ti
o
n
c
ha
in
s
i
n
C
P
U
+
D
P
U
-
ba
s
e
d
he
te
r
o
g
e
n
e
o
us
N
F
V
pl
a
tf
o
r
ms
,”
I
E
E
E
T
r
ans
ac
ti
on
s
on
M
obi
le
C
om
put
in
g
,
v
o
l.
23,
no
.
10,
pp.
9090
–
9107, Oc
t.
2024, do
i:
10.1109/
T
M
C
.2024.3357796.
[
38]
X
.
G
e
ng,
H
.
Z
ha
ng,
Z
.
Z
ha
o
,
a
nd
H
.
M
a
,
“
I
nt
e
r
f
e
r
e
nc
e
-
a
w
a
r
e
pa
r
a
ll
e
li
z
a
ti
o
n
f
or
de
e
p
l
e
a
r
ni
ng
w
o
r
kl
o
a
d
in
G
P
U
c
lu
s
t
e
r
,”
C
lu
s
te
r
C
om
put
in
g
, vo
l.
23, n
o
. 4,
pp. 2689
–
2702, De
c
. 2020, d
o
i:
10.1007/s
10586
-
019
-
03037
-
6.
[
39]
A
.
K
a
la
nt
a
r
,
Z
.
Z
im
me
r
ma
n,
a
nd
P
.
B
r
is
k,
“
F
P
G
A
-
ba
s
e
d
a
c
c
e
le
r
a
ti
o
n
of
ti
m
e
s
e
r
i
e
s
s
im
il
a
r
it
y
p
r
e
di
c
ti
o
n
:
F
r
om
c
l
o
ud
t
o
e
d
ge
,”
A
C
M
T
r
ans
ac
ti
ons
on R
e
c
on
f
ig
ur
abl
e
T
e
c
hnol
ogy
and S
y
s
te
m
s
,
vo
l.
16
, n
o
. 1, pp. 1
–
27, M
a
r
. 2023, d
o
i:
10.1145/3555810.
[
40]
S.
-
O
.
P
a
r
k,
O
.
K
w
o
n,
Y
.
K
im
,
S
.
K
.
C
ha
,
a
nd
H
.
Y
oo
n,
“
M
i
nd
c
o
nt
r
o
l
a
tt
a
c
k
:
U
nde
r
mi
ni
ng
d
e
e
p
l
e
a
r
ni
ng
w
it
h
G
P
U
me
mo
r
y
e
x
pl
o
it
a
ti
o
n,”
C
om
put
e
r
s
&
Se
c
ur
it
y
, vo
l.
102, M
a
r
. 2021, do
i
:
10.1016/j
.c
o
s
e
.2020
.102115.
[
41]
Z
.
C
he
n,
“
R
I
F
L
I
N
G
:
A
r
e
in
f
o
r
c
e
m
e
nt
l
e
a
r
ni
ng‐
ba
s
e
d
G
P
U
s
c
h
e
dul
e
r
f
o
r
d
e
e
p
l
e
a
r
ni
ng
r
e
s
e
a
r
c
h
a
nd
de
ve
l
o
pm
e
nt
pl
a
t
f
or
ms
,”
Sof
tw
ar
e
:
P
r
ac
ti
c
e
and E
x
pe
r
ie
nc
e
, v
o
l.
52, n
o
. 6, pp. 1319
–
13
36, J
un. 2022, do
i:
10.1002/s
pe
.3066.
[
42]
T
.
T
.
J
.
K
ir
a
n,
“
D
e
e
p
r
a
ns
f
o
r
m
l
e
a
r
ni
ng
v
is
i
o
n
a
c
c
ur
a
c
y
a
na
ly
s
is
o
n
G
P
U
us
in
g
te
ns
o
r
f
l
o
w
,”
I
nt
e
r
nat
io
nal
J
our
nal
of
R
e
c
e
nt
T
e
c
hnol
ogy
and E
ngi
ne
e
r
in
g (
I
J
R
T
E
)
,
vo
l.
9, n
o
. 3, pp. 224
–
22
7, S
e
p. 2020, d
o
i:
10.35940/i
jr
t
e
.C
4402.099320.
[
43]
H
.
L
iu
,
S
.
L
iu
,
C
.
W
e
n,
a
nd
W
.
E
.
W
o
ng,
“
T
B
E
M
:
T
e
s
ti
ng
-
ba
s
e
d
G
P
U
-
me
m
or
y
c
o
ns
umpt
i
o
n
e
s
ti
ma
ti
o
n
f
or
d
e
e
p
le
a
r
ni
n
g,”
I
E
E
E
A
c
c
e
s
s
, v
ol
. 10, pp. 39674
–
39680, 2022, d
o
i:
10.1109/A
C
C
E
S
S
.2022.3164510.
[
44]
J
.
W
e
i,
X
.
Z
ha
ng,
L
.
W
a
ng,
a
nd
Z
.
W
e
i,
“
F
a
s
te
ns
or
:
O
pt
im
is
e
th
e
t
e
ns
or
I
/O
pa
th
f
r
o
m
S
S
D
t
o
G
P
U
f
o
r
d
e
e
p
l
e
a
r
ni
ng
tr
a
in
in
g,”
A
C
M
T
r
ans
ac
ti
ons
on A
r
c
hi
te
c
tu
r
e
and C
ode
O
pt
imi
z
at
io
n
, vo
l
. 20, no
. 4, pp. 1
–
25, D
e
c
. 2023, d
oi
:
10.1145/3630108.
[
45]
D.
-
K
.
K
a
ng,
K
.
-
B
.
L
e
e
,
a
nd
Y
.
-
C
.
K
i
m,
“
C
o
s
t
e
f
f
ic
i
e
nt
G
P
U
c
lu
s
te
r
ma
na
ge
m
e
nt
f
or
t
r
a
in
in
g
a
nd
in
f
e
r
e
n
c
e
of
d
e
e
p
l
e
a
r
ni
n
g,”
E
ne
r
gi
e
s
, v
o
l.
15, n
o
. 2, J
a
n. 2022, d
o
i:
10.3390/
e
n15020474.
[
46]
D
.
V
a
s
a
n,
M
.
H
a
mm
o
ud
e
h,
a
nd
M
.
A
la
z
a
b,
“
B
r
o
a
d
l
e
a
r
ni
ng
:
A
G
P
U
-
f
r
e
e
im
a
ge
-
ba
s
e
d
ma
lwa
r
e
c
la
s
s
if
i
c
a
ti
o
n,”
A
ppl
ie
d
Sof
t
C
om
put
in
g
, vo
l.
154, M
a
r
. 2024, d
o
i:
10.1016/j
.a
s
oc
.2024.1114
01.
[
47]
W
.
G
a
o
,
Z
.
Y
e
,
P
.
S
un,
T
.
Z
ha
ng,
a
nd
Y
.
W
e
n,
“
U
ni
S
c
h
e
d:
A
uni
f
ie
d
s
c
he
dul
e
r
f
or
d
e
e
p
l
e
a
r
ni
ng
t
r
a
in
in
g
j
o
bs
w
it
h
di
f
f
e
r
e
nt
us
e
r
de
ma
nds
,”
I
E
E
E
T
r
ans
ac
ti
ons
on C
om
put
e
r
s
, v
o
l.
73, n
o
. 6, pp.
1500
–
1515, J
un. 2024, do
i:
10.110
9/
T
C
.2024.3371794.
[
48]
P
.
V
.
L
a
ks
hmi
,
T
.
A
.
S
.
S
r
in
i
v
a
s
,
A
.
D
.
D
o
na
ld
,
S
.
A
bi
da
,
I
.
D
w
a
r
a
ka
,
a
nd
S
r
ih
it
h,
“
G
P
U
ma
gi
c
:
T
u
r
b
oc
ha
r
gi
ng
y
ou
r
ma
c
hi
ne
le
a
r
ni
ng m
o
d
e
ls
,”
C
om
put
e
r
G
r
aphi
c
s
and M
ul
ti
m
e
di
a T
e
c
hnol
ogy
, 5A
D
.
B
I
OG
RA
P
HI
E
S
OF
AU
T
HO
RS
A
y
o
u
b
A
l
l
a
l
i
b
o
rn
o
n
A
u
g
u
s
t
1
,
1
9
9
7
,
i
n
Si
d
i
K
a
cem
,
Mo
r
o
cc
o
,
h
as
p
u
rs
u
ed
a
n
ac
ad
emi
c
a
n
d
r
e
s
e
ar
ch
c
ar
ee
r
i
n
t
h
e
fi
el
d
s
o
f
co
m
p
u
t
er
s
c
i
en
ce
a
n
d
b
i
g
d
at
a.
H
e
c
o
m
p
l
e
t
e
d
h
i
s
Bac
h
el
o
r
's
d
eg
r
ee
i
n
Mat
h
em
at
i
c
al
a
n
d
Co
m
p
u
t
e
r
S
c
i
e
n
ce
s
fro
m
t
h
e
D
e
p
art
me
n
t
o
f
Co
m
p
u
t
e
r
Sci
en
ce
at
t
h
e
Fa
c
u
l
t
y
o
f
S
c
i
en
ce
s
,
I
b
n
T
o
fai
l
U
n
i
v
e
rs
i
t
y
i
n
K
en
i
t
ra,
Mo
r
o
cc
o
,
i
n
2
0
1
9
.
Co
n
t
i
n
u
i
n
g
h
i
s
e
d
u
c
at
i
o
n
at
t
h
e
s
ame
i
n
s
t
i
t
u
t
i
o
n
,
A
y
o
u
b
e
ar
n
e
d
a
Mas
t
e
r's
d
e
g
r
ee
i
n
B
i
g
D
at
a
an
d
Cl
o
u
d
Co
m
p
u
t
i
n
g
i
n
2
0
2
1
.
Si
n
ce
2
0
2
2
,
h
e
h
as
b
ee
n
a
d
o
c
t
o
ra
l
s
t
u
d
e
n
t
at
t
h
e
C
o
m
p
u
t
e
r
R
e
s
e
arch
L
ab
o
rat
o
r
y
(
L
aRI
)
w
i
t
h
i
n
t
h
e
D
e
p
art
men
t
o
f
Co
m
p
u
t
er
Sci
e
n
ce
at
t
h
e
Fa
c
u
l
t
y
o
f
S
c
i
en
ce
s
,
I
b
n
T
o
fai
l
U
n
i
v
e
rs
i
t
y
.
H
i
s
r
e
s
e
ar
c
h
i
n
t
e
r
e
s
t
s
are
f
o
c
u
s
e
d
o
n
b
i
g
d
at
a
a
n
d
d
ee
p
l
e
arn
i
n
g
,
i
n
d
i
c
at
i
n
g
a
s
t
ro
n
g
c
o
mmi
t
men
t
t
o
a
d
v
a
n
c
i
n
g
k
n
o
w
l
e
d
g
e
i
n
t
h
e
s
e
c
ri
t
i
c
a
l
ar
e
as
o
f
d
at
a
s
c
i
en
ce
.
H
e
c
a
n
b
e
c
o
n
t
ac
t
e
d
at
em
a
i
l
:
a
y
o
u
b
.
al
l
a
l
i
@
u
i
t
.
ac.
m
a.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2
5
0
2
-
4
7
52
I
n
do
n
e
s
i
a
n
J
E
l
e
c
E
n
g
&
C
o
m
p
S
c
i
,
Vo
l
.
38
,
N
o
.
2
,
M
a
y
20
25
:
1
324
-
1
330
1330
Zi
n
eb
El
F
a
l
a
h
b
o
rn
o
n
F
e
b
ru
ar
y
2
4
,
1
9
9
7
,
i
n
M
ek
n
e
s
,
Mo
r
o
cc
o
,
h
as
d
e
d
i
c
at
e
d
h
e
r
ac
ad
emi
c
a
n
d
r
e
s
e
ar
c
h
c
ar
ee
r
t
o
t
h
e
fi
el
d
s
o
f
c
o
m
p
u
t
er
s
c
i
en
ce
a
n
d
b
i
g
d
at
a.
Sh
e
o
b
t
ai
n
ed
h
e
r
Mas
t
e
r's
d
eg
r
ee
i
n
B
i
g
D
at
a
an
d
Cl
o
u
d
Co
m
p
u
t
i
n
g
i
n
2
0
2
0
.
Si
n
ce
2
0
2
1
,
s
h
e
h
as
b
ee
n
a
Ph
.
D
.
s
t
u
d
e
n
t
at
t
h
e
C
o
m
p
u
t
e
r
S
c
i
en
ce
R
e
s
e
ar
ch
L
ab
o
rat
o
r
y
(L
aRi
)
i
n
t
h
e
Fa
cu
l
t
y
o
f
S
ci
e
n
ce
at
I
b
n
T
o
fai
l
U
n
i
v
e
rs
i
t
y
.
H
e
r
r
e
s
e
ar
c
h
f
o
c
u
s
e
s
o
n
d
at
a
an
a
l
y
s
i
s
an
d
d
eci
s
i
o
n
-
m
ak
i
n
g
i
n
b
i
g
d
at
a,
u
t
i
l
i
z
i
n
g
art
i
fi
ci
al
i
n
t
el
l
i
g
en
ce
,
m
a
ch
i
n
e
l
e
ar
n
i
n
g
,
an
d
d
ee
p
l
e
arn
i
n
g
.
T
h
i
s
d
emo
n
s
t
rat
e
s
h
e
r
s
t
ro
n
g
c
o
mmi
t
men
t
t
o
a
d
v
a
n
c
i
n
g
k
n
o
w
l
e
d
g
e
i
n
t
h
e
s
e
c
r
i
t
i
c
al
a
re
as
o
f
d
at
a
s
ci
e
n
ce
.
Sh
e
c
an
b
e
co
n
t
a
c
t
ed
at
e
m
ai
l
:
z
i
n
e
b
.
el
fal
a
h
@
u
i
t
.
a
c
.
m
a.
A
y
o
u
b
Sg
hi
r
w
as
b
o
rn
i
n
1
9
9
6
i
n
Sal
e.
H
e
r
ecei
v
ed
h
i
s
Mas
t
e
r’s
d
e
g
r
ee
i
n
c
o
m
p
u
t
e
r
s
c
i
e
n
ce,
b
i
g
d
at
a
c
l
o
u
d
c
o
m
p
u
t
i
n
g
fr
o
m
I
b
n
T
o
fai
l
U
n
i
v
e
rs
i
t
y
,
K
en
i
t
ra,
Mo
r
o
cc
o
.
H
e
i
s
a
Ph
.
D
.
s
t
u
d
e
n
t
i
n
Co
m
p
u
t
e
r
R
e
s
e
ar
c
h
L
ab
o
rat
o
r
y
(
L
aR
I
)
at
I
b
n
T
o
fai
l
.
H
i
s
r
e
s
e
arch
i
n
t
e
r
e
s
t
s
i
n
c
l
u
d
e
b
i
g
d
at
a,
d
at
a
s
t
o
rag
e
,
c
l
o
u
d
co
m
p
u
t
i
n
g
,
an
d
d
i
s
t
ri
b
u
t
e
d
co
m
p
u
t
i
n
g
.
H
e
c
an
b
e
co
n
t
a
c
t
ed
at
em
ai
l
:
ay
o
u
b
.
s
g
h
i
r
@
u
i
t
.
ac.
m
a.
J
a
a
f
a
r
A
bo
u
ch
a
ba
k
a
b
o
rn
o
n
F
e
b
ru
ar
y
2
7
,
1
9
6
8
,
i
n
G
u
e
rs
i
f,
Mo
r
o
cc
o
,
i
s
a
d
i
s
t
i
n
g
u
i
s
h
e
d
ac
a
d
em
i
c
w
i
t
h
a
p
ro
l
i
fi
c
c
ar
ee
r
i
n
m
at
h
emat
i
c
s
an
d
co
m
p
u
t
e
r
s
ci
e
n
ce.
H
e
em
b
ark
e
d
o
n
h
i
s
ed
u
c
at
i
o
n
a
l
j
o
u
rn
ey
at
K
ad
i
A
y
a
d
U
n
i
v
e
rs
i
t
y
i
n
Marrak
ec
h
,
M
o
ro
cco
,
w
h
e
r
e
h
e
o
b
t
ai
n
ed
h
i
s
Ba
ch
e
l
o
r's
d
eg
r
ee
i
n
F
u
n
d
a
me
n
t
al
Mat
h
em
at
i
c
s
fro
m
t
h
e
D
e
p
art
me
n
t
o
f
Mat
h
em
at
i
c
s
,
Fa
c
u
l
t
y
o
f
S
c
i
en
ce
s
,
i
n
1
9
9
2
.
H
e
f
u
rt
h
e
r
e
d
h
i
s
s
t
u
d
i
e
s
at
Mo
h
amme
d
V
U
n
i
v
e
rs
i
t
y
i
n
Rab
at
,
Mo
ro
cco
,
e
arn
i
n
g
a
Po
s
t
g
rad
u
at
e
D
i
p
l
o
m
a
(D
E
A
)
i
n
Mat
h
em
a
t
i
c
s
i
n
1
9
9
4
,
an
d
l
at
e
r
a
D
o
c
t
o
ra
t
e
i
n
Co
m
p
u
t
e
r
S
c
i
en
c
e
s
A
p
p
l
i
ed
t
o
Mat
h
em
at
i
c
s
fro
m
t
h
e
D
e
p
art
men
t
o
f
C
o
m
p
u
t
e
r
S
c
i
en
ce
i
n
2
0
0
1
.
Si
n
ce
2
0
0
5
,
he
h
as
h
e
l
d
t
h
e
p
o
s
i
t
i
o
n
o
f
Pr
o
f
e
s
s
o
r
i
n
t
h
e
D
e
p
art
me
n
t
o
f
C
o
m
p
u
t
e
r
S
c
i
en
ce
s
at
I
b
n
T
o
fai
l
U
n
i
v
e
rs
i
t
y
i
n
K
en
i
t
ra,
M
o
ro
cco
.
H
i
s
r
e
s
e
ar
c
h
i
n
t
ere
s
t
s
ar
e
b
ro
ad
an
d
i
m
p
ac
t
fu
l
,
f
o
c
u
s
i
n
g
o
n
c
o
n
c
u
rr
en
t
an
d
p
aral
l
el
p
ro
g
ra
mm
i
n
g
,
d
i
s
t
ri
b
u
t
e
d
s
y
s
t
em
s
,
art
i
f
i
c
i
a
l
i
n
t
e
l
l
i
g
e
n
ce,
a
n
d
b
i
g
d
at
a
.
O
v
e
r
t
h
e
c
o
u
rs
e
o
f
h
i
s
c
ar
ee
r
,
h
e
h
as
co
n
t
ri
b
u
t
e
d
s
i
g
n
i
fi
c
an
t
l
y
t
o
t
h
e
a
c
a
d
em
i
c
co
mm
u
n
i
t
y
,
au
t
h
o
ri
n
g
mo
r
t
h
an
7
0
p
a
p
e
rs
an
d
co
m
p
l
e
t
i
n
g
2
t
h
e
s
e
s
.
H
e
c
an
b
e
co
n
t
a
c
t
ed
at
e
m
a
i
l
:
j
aafar.
ab
o
u
ch
ab
ak
a@
u
i
t
.
a
c
.
m
a.
N
a
ja
t
R
a
f
a
l
i
a
b
o
rn
o
n
N
o
v
em
b
e
r
4
,
1
9
6
8
,
i
n
K
en
i
t
ra,
M
o
ro
cco
,
h
as
e
s
t
ab
l
i
s
h
e
d
h
e
rs
e
l
f
as
a
p
ro
mi
n
en
t
fi
g
u
r
e
i
n
t
h
e
f
i
e
l
d
s
o
f
c
o
m
p
u
t
e
r
s
c
i
en
c
e
an
d
ap
p
l
i
e
d
m
at
h
em
at
i
c
s
.
H
e
r
ac
ad
emi
c
j
o
u
r
n
ey
b
e
g
a
n
at
Mo
h
a
mmed
V
U
n
i
v
e
rs
i
t
y
i
n
Rab
at
,
Mo
ro
cco
,
w
h
e
r
e
s
h
e
r
ec
ei
v
ed
h
e
r
Bac
h
el
o
r
's
d
e
g
r
ee
i
n
A
p
p
l
i
ed
Mat
h
em
at
i
c
s
fro
m
t
h
e
D
e
p
art
men
t
o
f
Mat
h
em
at
i
c
s
,
Fa
cu
l
t
y
o
f
Sci
en
ce
s
,
i
n
1
9
9
2
.
Sh
e
c
o
n
t
i
n
u
e
d
at
t
h
e
s
ame
i
n
s
t
i
t
u
t
i
o
n
t
o
e
ar
n
a
Po
s
t
g
rad
u
at
e
D
i
p
l
o
m
a
(D
E
A
)
i
n
Co
m
p
u
t
e
r
S
c
i
en
ce
i
n
1
9
9
4
an
d
l
at
e
r
a
D
o
c
t
o
rat
e
i
n
Co
m
p
u
t
e
r
S
ci
e
n
ce
s
fro
m
t
h
e
D
e
p
a
rt
me
n
t
o
f
Co
m
p
u
t
e
r
S
c
i
en
ce
i
n
1
9
9
7
.
Si
n
ce
1
9
9
7
,
s
h
e
h
as
s
e
rv
e
d
as
a
Pr
o
f
e
s
s
o
r
i
n
t
h
e
D
e
p
art
me
n
t
o
f
Co
m
p
u
t
e
r
S
ci
e
n
ce
s
at
I
b
n
T
o
fa
i
l
U
n
i
v
e
rs
i
t
y
.
Fu
rt
h
er
ad
v
a
n
c
i
n
g
h
e
r
e
x
p
e
rt
i
s
e
,
s
h
e
h
a
d
h
e
r
p
o
s
t
d
o
c
t
o
ral
t
h
e
s
i
s
i
n
2
0
1
3
an
d
c
o
m
p
l
e
t
e
d
h
e
r
t
h
i
r
d
D
o
ct
o
rat
e
i
n
Co
m
p
u
t
e
r
S
ci
e
n
ce
s
at
I
b
n
T
o
fai
l
U
n
i
v
e
rs
i
t
y
i
n
K
en
i
t
ra,
Mo
ro
cco
,
i
n
2
0
1
7
.
H
e
r
r
e
s
e
ar
ch
i
n
t
e
r
e
s
t
s
are
d
ee
p
l
y
ro
o
t
ed
i
n
b
i
g
d
at
a
an
al
y
t
i
c,
art
i
fi
ci
al
i
n
t
el
l
i
g
en
ce
an
d
i
t
s
ap
p
l
i
c
at
i
o
n
s
,
i
n
t
ern
e
t
o
f
t
h
i
n
g
s
,
d
i
s
t
ri
b
u
t
e
d
s
y
s
t
em
s
,
m
u
l
t
i
-
ag
en
t
s
y
s
t
em
s
,
co
n
cu
rr
e
n
t
an
d
p
aral
l
el
p
ro
g
ra
mmi
n
g
.
T
h
ro
u
g
h
o
u
t
h
e
r
c
ar
ee
r,
she
h
as
m
ad
e
s
i
g
n
i
fi
c
an
t
c
o
n
t
ri
b
u
t
i
o
n
s
t
o
h
e
r
fi
el
d
,
au
t
h
o
ri
n
g
mo
r
t
h
a
n
7
0
p
a
p
e
rs
an
d
c
o
m
p
l
e
t
i
n
g
3
t
h
e
s
e
s
.
S
h
e
c
an
b
e
co
n
t
a
c
t
ed
at
em
ai
l
:
n
aj
at
.
rafa
l
i
a
@
u
i
t
.
ac
.
m
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.