I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
780
~
787
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
7
8
0
-
7
8
7
780
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
H
y
brid conv
o
luti
o
na
l net
wo
rks,
h
i
dden
M
a
rko
v
m
o
dels
,
a
nd
a
utoenco
ders for
enha
nced r
ecog
nition
Driss
Na
j
i
1
,
K
a
m
a
l El
ha
t
t
a
b
2
,
Abdela
li J
o
um
a
d
3
,
Abdelo
ua
hed Ait
I
der
4
,
A
bd
elk
bir
O
uis
a
a
da
ne
5
,
Azz
eddin
e
I
dh
m
a
d
6
1
TI
A
D
La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
me
n
t
C
o
mp
u
t
e
r
o
f
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
i
q
u
e
,
S
u
l
t
a
n
M
o
u
l
a
y
S
l
i
m
a
n
e
U
n
i
v
e
r
si
t
y
,
B
e
n
i
m
e
l
l
a
l
,
M
o
r
o
c
c
o
2
ELI
TES La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
me
n
t
C
o
mp
u
t
e
r
o
f
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
,
C
h
o
u
a
i
b
D
o
u
k
k
a
l
i
U
n
i
v
e
r
s
i
t
y
,
El
Ja
d
i
d
a
,
M
o
r
o
c
c
o
3
LA
R
O
S
ER
I
l
a
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
me
n
t
o
f
i
n
f
o
r
m
a
t
i
c
s,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
s
,
C
h
o
u
a
i
b
D
o
k
k
a
l
i
U
n
i
v
e
r
s
i
t
y
,
El
Ja
d
i
d
a
,
M
o
r
o
c
c
o
4
I
S
I
M
A
La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
me
n
t
C
o
m
p
u
t
e
r
o
f
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
P
o
l
y
d
i
sci
p
l
i
n
a
r
y
,
I
b
n
o
u
Z
o
h
r
U
n
i
v
e
r
s
i
t
y
,
Ta
r
o
u
d
a
n
t
,
M
o
r
o
c
c
o
5
LI
M
A
TI
La
b
o
r
a
t
o
r
y
,
D
e
p
a
r
t
m
e
n
t
o
f
M
a
t
h
e
ma
t
i
c
s a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
F
a
c
u
l
t
y
o
f
P
o
l
y
d
i
sc
i
p
l
i
n
a
r
y
,
S
u
l
t
a
n
M
o
u
l
a
y
S
l
i
ma
n
e
U
n
i
v
e
r
si
t
y
,
B
e
n
i
me
l
l
a
l
,
M
o
r
o
c
c
o
6
La
b
o
r
a
t
o
r
y
o
f
A
n
a
l
y
s
i
s,
G
e
o
m
e
t
r
y
,
a
n
d
A
p
p
l
i
c
a
t
i
o
n
s,
F
a
c
u
l
t
y
o
f
S
c
i
e
n
c
e
s
,
I
b
n
To
f
a
i
l
U
n
i
v
e
r
si
t
y
,
K
e
n
i
t
r
a
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Ma
r
9
,
2
0
2
5
R
ev
is
ed
No
v
8
,
2
0
2
5
Acc
ep
ted
J
an
1
0
,
2
0
2
6
Re
c
o
g
n
it
i
o
n
p
r
o
b
lem
s,
i
n
c
lu
d
in
g
o
b
jec
t
d
e
tec
ti
o
n
,
sc
e
n
e
u
n
d
e
rsta
n
d
in
g
,
a
n
d
fin
e
-
g
ra
in
e
d
c
a
teg
o
risa
ti
o
n
,
a
re
p
o
p
u
lar
s
u
b
jec
ts
i
n
c
o
m
p
u
te
r
v
isio
n
.
Ho
we
v
e
r,
it
is
c
h
a
ll
e
n
g
in
g
to
m
o
d
e
l
sp
a
ti
a
l
c
o
h
e
re
n
c
e
a
n
d
c
o
n
tex
t
u
a
l
d
e
p
e
n
d
e
n
c
ies
in
re
sp
o
n
se
to
c
h
a
n
g
e
s
in
c
o
n
fig
u
ra
ti
o
n
s.
H
u
m
a
n
Vs
c
o
m
p
u
ters
'
a
b
il
it
y
i
n
p
e
rc
e
p
ti
o
n
-
a
lt
h
o
u
g
h
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
two
rk
s
(CNN
s)
d
o
we
ll
i
n
t
h
e
e
x
trac
ti
o
n
o
f
fe
a
tu
re
s,
t
h
e
y
h
a
v
e
h
ig
h
d
e
p
e
n
d
e
n
c
e
o
n
lo
c
a
l
re
c
e
p
ti
v
e
field
s
a
n
d
a
re
n
o
t
a
b
le
to
c
a
p
tu
re
l
o
n
g
-
ra
n
g
e
sp
a
ti
a
l
re
latio
n
sh
i
p
s
a
n
d
h
i
g
h
-
o
rd
e
r
in
te
ra
c
ti
o
n
s.
To
a
ll
e
v
iate
t
h
e
sh
o
rtc
o
m
in
g
s
o
f
th
e
c
u
rre
n
t
a
p
p
ro
a
c
h
e
s,
we
p
r
e
se
n
t
a
n
e
n
h
a
n
c
e
d
h
y
b
r
id
CN
Ns
two
-
d
ime
n
sio
n
a
l
h
i
d
d
e
n
M
a
rk
o
v
m
o
d
e
l
(
2D
-
HMM
)
fra
m
e
wo
rk
th
a
t
c
o
m
b
in
e
s
2D
-
HMM
,
M
a
rk
o
v
ra
n
d
o
m
f
ield
s
(M
RF
)
a
n
d
v
a
riati
o
n
a
l
a
u
t
o
e
n
c
o
d
e
rs
(VA
Es)
in
t
o
a
si
n
g
le
m
o
d
e
l.
T
h
e
m
o
d
e
l
e
m
p
l
o
y
s
2
D
-
HM
M
s
f
o
r
p
a
irwise
sp
a
ti
a
l
m
o
d
e
ll
in
g
,
M
RF
s
fo
r
h
ig
h
e
r
o
rd
e
r
c
o
n
tex
t,
a
n
d
VA
Es
fo
r
st
a
b
le
late
n
t
re
p
re
se
n
tatio
n
lea
rn
i
n
g
.
Tes
ted
o
n
t
h
e
M
NIST
a
n
d
CIF
AR
-
1
0
b
e
n
c
h
m
a
rk
d
a
tas
e
ts,
o
u
r
a
p
p
ro
a
c
h
c
o
n
si
ste
n
tl
y
o
u
tp
e
rfo
rm
s
th
e
sta
te
-
of
-
th
e
-
a
rt
p
e
rfo
rm
a
n
c
e
b
y
9
8
.
2
%
a
n
d
8
9
.
5
%
,
re
sp
e
c
ti
v
e
l
y
,
wit
h
h
i
g
h
r
o
b
u
stn
e
ss
to
n
o
ise
a
n
d
o
c
c
lu
si
o
n
.
Re
su
lt
s
fro
m
a
b
latio
n
st
u
d
ies
fu
rt
h
e
r
sh
o
w
th
a
t
M
R
F
s
imp
ro
v
e
re
c
a
ll
b
y
1
.
6
%
a
n
d
VA
Es
imp
ro
v
e
p
re
c
isio
n
b
y
1
.
3
%
,
su
g
g
e
sti
n
g
th
a
t
th
e
y
c
o
m
p
lem
e
n
t
e
a
c
h
o
th
e
r
su
fficie
n
tl
y
with
re
sp
e
c
t
to
o
v
e
r
a
ll
tes
ti
n
g
p
e
rfo
rm
a
n
c
e
.
Th
is
wo
r
k
u
n
ifi
e
s
d
e
e
p
lea
rn
i
n
g
a
n
d
p
r
o
b
a
b
il
isti
c
g
ra
p
h
ica
l
m
o
d
e
ls,
lea
d
in
g
to
m
o
re
in
ter
p
r
e
tab
le,
sc
a
lab
le,
a
n
d
a
c
c
u
ra
te
re
c
o
g
n
i
ti
o
n
sy
ste
m
s.
K
ey
w
o
r
d
s
:
2
D
h
id
d
en
Ma
r
k
o
v
m
o
d
els
C
o
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
Hy
b
r
id
d
ee
p
lear
n
i
n
g
f
r
am
ewo
r
k
Ma
r
k
o
v
r
an
d
o
m
f
ield
s
Var
iatio
n
al
au
to
en
co
d
er
s
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
:
Dr
is
s
Naji
T
I
AD
L
ab
o
r
ato
r
y
,
Dep
a
r
tm
en
t Co
m
p
u
ter
o
f
Scien
ce
,
Facu
lty
o
f
Scien
ce
a
n
d
T
ec
h
n
iq
u
e
Su
ltan
Mo
u
lay
Sli
m
an
e
Un
iv
e
r
s
ity
B
en
i
-
Me
llal,
Mo
r
o
cc
o
E
m
ail: n
aji.
d
r
is
s
s
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
R
ec
o
g
n
itio
n
task
s
,
s
u
ch
as
o
b
ject
d
etec
tio
n
,
s
ce
n
e
u
n
d
er
s
t
an
d
in
g
,
a
n
d
f
in
e
-
g
r
ain
ed
clas
s
if
icatio
n
,
r
ep
r
esen
t
f
u
n
d
a
m
en
tal
c
h
allen
g
es
in
co
m
p
u
ter
v
is
io
n
a
n
d
p
atter
n
r
ec
o
g
n
itio
n
,
f
o
r
m
in
g
th
e
c
o
r
n
e
r
s
to
n
e
o
f
n
u
m
er
o
u
s
ap
p
licatio
n
s
r
an
g
in
g
f
r
o
m
au
to
n
o
m
o
u
s
v
eh
icl
es
to
m
ed
ical
d
iag
n
o
s
tics
.
Desp
ite
r
em
ar
k
ab
le
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
etw
o
r
ks,
h
id
d
en
Ma
r
ko
v
mo
d
els,
a
n
d
a
u
to
en
c
o
n
d
ers
fo
r
en
h
a
n
ce
d
…
(
Dri
s
s
N
a
ji
)
781
ad
v
an
ce
s
in
d
e
ep
lear
n
i
n
g
ar
ch
itectu
r
es,
p
er
s
is
ten
t
ch
al
len
g
es
r
em
ain
in
e
f
f
ec
tiv
ely
m
o
d
elin
g
s
p
atial
co
h
er
en
ce
an
d
co
n
tex
tu
al
d
ep
en
d
en
cies,
p
ar
ticu
lar
l
y
wh
en
d
ea
lin
g
with
co
m
p
lex
s
ce
n
ar
io
s
in
v
o
lv
in
g
o
cc
lu
s
io
n
,
n
o
is
e,
an
d
s
ig
n
if
i
ca
n
t
v
ar
iab
ilit
y
in
d
ata
d
is
tr
ib
u
tio
n
s
[
1
]
,
[
2
]
.
T
h
ese
ch
al
len
g
es
ar
e
f
u
r
th
er
co
m
p
o
u
n
d
e
d
b
y
t
h
e
in
h
er
en
t
lim
itatio
n
s
o
f
c
u
r
r
en
t
ap
p
r
o
ac
h
es
in
ca
p
tu
r
in
g
lo
n
g
-
r
an
g
e
s
p
atial
r
elatio
n
s
h
ip
s
an
d
h
ig
h
er
-
o
r
d
er
in
ter
ac
tio
n
s
t
h
at
ar
e
cr
u
cial
f
o
r
r
o
b
u
s
t r
ec
o
g
n
itio
n
p
e
r
f
o
r
m
an
ce
.
W
h
ile
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
etwo
r
k
s
(
C
NNs)
h
av
e
estab
lis
h
ed
th
em
s
elv
es
as
th
e
d
o
m
in
a
n
t
p
ar
ad
ig
m
f
o
r
f
ea
tu
r
e
ex
tr
ac
t
io
n
in
v
is
io
n
task
s
,
d
em
o
n
s
tr
atin
g
ex
ce
p
tio
n
al
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
d
iv
er
s
e
ap
p
licatio
n
s
,
th
eir
ar
ch
itectu
r
a
l
d
esig
n
in
tr
o
d
u
ce
s
f
u
n
d
am
en
t
al
lim
itatio
n
s
th
at
co
n
s
tr
ain
th
eir
ef
f
ec
tiv
en
ess
in
co
m
p
lex
r
ec
o
g
n
itio
n
s
ce
n
ar
io
s
.
Sp
ec
if
ically
,
C
NNs
r
ely
h
e
av
ily
o
n
lo
ca
l
r
ec
e
p
tiv
e
f
ield
s
,
wh
ich
i
n
h
er
en
tly
lim
its
th
eir
ab
ilit
y
to
ca
p
tu
r
e
lo
n
g
-
r
a
n
g
e
s
p
atial
r
elatio
n
s
h
ip
s
an
d
co
n
te
x
tu
al
d
ep
e
n
d
en
cies
th
at
ex
ten
d
b
ey
o
n
d
im
m
ed
iate
s
p
atial
n
eig
h
b
o
r
h
o
o
d
s
[
3
]
,
[
4
]
.
T
h
is
lim
itatio
n
b
ec
o
m
es
p
ar
ticu
lar
ly
p
r
o
n
o
u
n
ce
d
in
clu
tter
e
d
s
ce
n
es
wh
er
e
C
NNs
o
f
ten
s
tr
u
g
g
le
to
d
is
tin
g
u
is
h
o
v
er
lap
p
in
g
o
b
jects
o
r
f
ail
to
in
f
er
c
o
n
tex
tu
al
c
u
es
th
at
r
eq
u
ir
e
u
n
d
er
s
tan
d
in
g
o
f
b
r
o
a
d
er
s
p
atial
co
n
tex
ts
[
5
]
.
Fo
r
in
s
tan
ce
,
in
s
ce
n
e
r
ec
o
g
n
itio
n
task
s
,
C
NNs
m
ay
m
is
class
if
y
o
b
jects
d
u
e
to
c
o
n
tex
tu
al
am
b
ig
u
ities
,
s
u
ch
as
d
is
tin
g
u
is
h
in
g
b
etwe
en
a
b
o
at
o
n
a
r
o
ad
v
e
r
s
u
s
a
b
o
at
o
n
wate
r
,
wh
er
e
g
lo
b
al
c
o
n
tex
t is ess
en
tial f
o
r
ac
cu
r
ate
class
if
icatio
n
.
T
h
e
r
ec
o
g
n
itio
n
o
f
t
h
ese
lim
itatio
n
s
h
as
m
o
tiv
ated
r
esear
ch
er
s
to
ex
p
lo
r
e
h
y
b
r
id
ap
p
r
o
ac
h
es
th
at
co
m
b
in
e
th
e
f
e
atu
r
e
e
x
tr
ac
tio
n
ca
p
ab
ilit
ies
o
f
C
NNs
with
p
r
o
b
ab
ilis
tic
m
o
d
els
ca
p
a
b
le
o
f
ex
p
licitly
m
o
d
elin
g
s
p
atial
an
d
co
n
tex
tu
al
r
elatio
n
s
h
ip
s
.
Pro
b
a
b
ilis
tic
m
o
d
els
s
u
ch
as
h
id
d
en
Ma
r
k
o
v
m
o
d
els
(
HM
Ms)
an
d
Ma
r
k
o
v
r
a
n
d
o
m
f
ield
s
(
MRF
s
)
h
av
e
s
h
o
wn
p
r
o
m
is
e
wh
en
in
teg
r
ated
with
C
NNs,
o
f
f
er
in
g
co
m
p
lem
e
n
tar
y
s
tr
en
g
th
s
in
s
tr
u
ctu
r
ed
p
r
ed
ict
io
n
an
d
s
p
atial
m
o
d
elin
g
.
R
ec
en
t
s
tu
d
ies
h
av
e
d
em
o
n
s
tr
ated
th
at
h
y
b
r
id
C
NN
-
HM
M
ar
ch
itectu
r
es
ca
n
s
ig
n
if
ican
tly
im
p
r
o
v
e
s
tr
u
ctu
r
ed
p
r
e
d
ictio
n
task
s
b
y
ef
f
ec
tiv
ely
m
o
d
elin
g
s
eq
u
en
tial
o
r
g
r
id
-
b
ased
d
ep
en
d
en
cies
[
6
]
.
Ho
wev
er
,
tr
ad
itio
n
al
f
ir
s
t
-
o
r
d
er
HM
Ms
an
d
MRF
s
,
wh
i
le
u
s
ef
u
l,
ar
e
o
f
ten
in
s
u
f
f
icien
t
f
o
r
c
o
m
p
lex
r
e
co
g
n
itio
n
task
s
th
at
r
eq
u
ir
e
m
o
d
elin
g
o
f
h
ig
h
er
-
o
r
d
e
r
in
ter
ac
tio
n
s
an
d
s
o
p
h
is
ticated
s
p
atial
r
elatio
n
s
h
ip
s
[
7
]
.
T
h
is
lim
itatio
n
n
ec
ess
i
tates
th
e
d
ev
elo
p
m
e
n
t
o
f
m
o
r
e
a
d
v
an
ce
d
h
y
b
r
i
d
f
r
am
ewo
r
k
s
th
at
ca
n
ef
f
ec
tiv
ely
lev
er
a
g
e
th
e
s
tr
en
g
th
s
o
f
m
u
ltip
le
m
o
d
elin
g
p
ar
a
d
ig
m
s
.
B
u
ild
in
g
u
p
o
n
th
ese
in
s
ig
h
ts
,
th
is
r
esear
ch
ad
v
an
ce
s
th
e
cu
r
r
en
t
s
tate
-
of
-
th
e
-
a
r
t
b
y
p
r
o
p
o
s
in
g
a
c
o
m
p
r
eh
e
n
s
iv
e
f
r
am
ewo
r
k
th
at
in
teg
r
ates
two
-
d
im
en
s
io
n
al
h
i
d
d
en
Ma
r
k
o
v
m
o
d
els
(
2
D
-
HM
Ms)
f
o
r
s
o
p
h
is
ticated
p
air
wis
e
s
p
atial
m
o
d
elin
g
,
MRF
s
f
o
r
ca
p
tu
r
in
g
h
ig
h
er
-
o
r
d
e
r
co
n
te
x
tu
al
r
elatio
n
s
h
ip
s
,
an
d
v
a
r
i
atio
n
al
au
to
en
co
d
er
s
(
VAE
s
)
f
o
r
r
o
b
u
s
t
laten
t
r
ep
r
esen
tatio
n
lear
n
i
n
g
with
i
n
a
u
n
if
ied
ar
ch
itectu
r
al
f
r
a
m
ewo
r
k
.
T
h
is
in
teg
r
atio
n
is
m
o
tiv
ated
b
y
r
ec
en
t
b
r
ea
k
th
r
o
u
g
h
s
in
s
ev
er
al
k
ey
ar
ea
s
o
f
r
esear
ch
.
First,
in
t
h
e
d
o
m
ain
o
f
s
p
atial
co
h
er
e
n
ce
m
o
d
elin
g
,
2
D
-
HM
Ms
h
av
e
d
em
o
n
s
tr
ated
s
ig
n
if
ican
t
s
u
cc
ess
in
ap
p
licatio
n
s
s
u
ch
as
im
ag
e
s
eg
m
en
tatio
n
[
8
]
an
d
d
o
c
u
m
en
t
an
aly
s
is
th
o
u
g
h
th
eir
ap
p
licatio
n
to
g
e
n
er
al
r
ec
o
g
n
itio
n
task
s
r
em
ain
s
r
elativ
ely
u
n
d
er
ex
p
lo
r
ed
a
n
d
p
r
esen
ts
o
p
p
o
r
tu
n
ities
f
o
r
n
o
v
el
c
o
n
tr
i
b
u
tio
n
s
[
9
]
,
[
1
0
]
.
Seco
n
d
,
r
eg
ar
d
in
g
h
ig
h
e
r
-
o
r
d
er
c
o
n
tex
t
m
o
d
elin
g
,
MRF
s
with
s
o
p
h
is
ticated
cliq
u
e
p
o
ten
tials
h
av
e
s
h
o
wn
co
n
s
id
er
a
b
le
p
r
o
m
is
e
in
s
em
an
tic
s
eg
m
en
tatio
n
[
1
1
]
a
n
d
m
ed
ical
im
ag
in
g
ap
p
licatio
n
s
[
1
2
]
,
[
1
3
]
th
o
u
g
h
th
eir
in
teg
r
atio
n
with
m
o
d
er
n
d
ee
p
lear
n
i
n
g
ar
c
h
itectu
r
es
co
n
tin
u
es
to
ev
o
lv
e
a
n
d
p
r
esen
ts
tech
n
ical
ch
allen
g
es.
T
h
ir
d
,
i
n
th
e
ar
e
a
o
f
r
o
b
u
s
t
laten
t
r
ep
r
esen
tati
o
n
lear
n
i
n
g
,
VAE
s
h
av
e
p
r
o
v
en
th
eir
ef
f
ec
tiv
e
n
e
s
s
in
en
h
an
cin
g
r
o
b
u
s
tn
ess
to
n
o
is
e
an
d
o
cc
lu
s
io
n
,
as
d
em
o
n
s
tr
ated
in
r
ec
en
t
s
tu
d
ies f
o
cu
s
in
g
o
n
s
em
i
-
s
u
p
e
r
v
is
ed
lear
n
in
g
[
1
4
]
,
[
1
5
]
a
n
d
an
o
m
aly
d
etec
tio
n
[
1
6
]
.
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
is
s
p
ec
if
ically
d
esig
n
ed
to
ad
d
r
ess
s
ev
er
al
cr
itical
g
ap
s
i
n
cu
r
r
en
t
r
ec
o
g
n
itio
n
s
y
s
tem
s
.
I
n
ter
m
s
o
f
s
p
atial
r
ea
s
o
n
in
g
ca
p
ab
ilit
ies,
th
e
f
r
am
ewo
r
k
c
ap
tu
r
es
b
o
th
l
o
ca
l
d
ep
en
d
e
n
cies
th
r
o
u
g
h
C
NN
f
ea
tu
r
e
ex
tr
ac
tio
n
a
n
d
g
lo
b
al
d
ep
en
d
e
n
cies
th
r
o
u
g
h
in
teg
r
a
ted
2
D
-
HM
M
an
d
MRF
m
o
d
elin
g
,
p
r
o
v
id
in
g
a
co
m
p
r
e
h
en
s
iv
e
ap
p
r
o
ac
h
to
s
p
atial
u
n
d
er
s
tan
d
in
g
.
Fro
m
an
in
ter
p
r
eta
b
ilit
y
p
er
s
p
ec
tiv
e,
t
h
e
in
c
o
r
p
o
r
atio
n
o
f
VAE
s
p
r
o
v
id
es
v
alu
ab
le
i
n
s
ig
h
ts
in
to
laten
t
f
ea
tu
r
e
d
is
tr
ib
u
tio
n
s
,
ali
g
n
in
g
with
cu
r
r
e
n
t
tr
e
n
d
s
to
war
d
e
x
p
lain
ab
le
ar
tific
ial
in
tellig
en
c
e
[
1
7
]
.
R
eg
ar
d
in
g
s
ca
lab
ilit
y
co
n
s
id
er
atio
n
s
,
t
h
e
f
r
am
ewo
r
k
i
n
co
r
p
o
r
ates
GPU
-
o
p
tim
ized
tr
ain
in
g
p
r
o
ce
d
u
r
es
th
at
en
ab
le
ef
f
icie
n
t
d
e
p
lo
y
m
en
t
o
n
h
i
g
h
-
r
eso
lu
tio
n
d
atasets
,
ad
d
r
ess
in
g
p
r
ac
tical
co
n
ce
r
n
s
ab
o
u
t
co
m
p
u
tatio
n
al
f
ea
s
ib
ilit
y
[
1
8
]
.
T
h
e
p
r
im
ar
y
o
b
jectiv
e
o
f
th
is
r
esear
ch
is
to
d
em
o
n
s
tr
ate
th
at
th
e
s
y
n
er
g
is
tic
co
m
b
in
atio
n
o
f
th
ese
co
m
p
l
em
en
tar
y
m
o
d
elin
g
ap
p
r
o
ac
h
es
ca
n
ac
h
iev
e
s
u
p
e
r
io
r
r
ec
o
g
n
itio
n
p
e
r
f
o
r
m
an
ce
wh
ile
m
ain
tain
in
g
co
m
p
u
tatio
n
al
ef
f
icien
cy
a
n
d
p
r
o
v
id
i
n
g
en
h
an
ce
d
in
ter
p
r
eta
b
ilit
y
co
m
p
ar
e
d
to
ex
is
tin
g
s
ta
te
-
of
-
th
e
-
a
r
t m
eth
o
d
s
.
2.
T
H
E
O
R
E
T
I
CA
L
F
O
UNDA
T
I
O
NS
2
.
1
.
Co
nv
o
lutio
na
l neura
l net
wo
rk
s
C
NNs
r
em
ain
th
e
b
ac
k
b
o
n
e
o
f
m
o
d
er
n
r
ec
o
g
n
itio
n
s
y
s
tem
s
d
u
e
to
th
eir
ab
ilit
y
to
lear
n
h
ier
ar
ch
ical
f
ea
tu
r
es.
R
ec
en
t
ar
ch
itectu
r
es,
s
u
ch
as
E
f
f
icien
tNet
[
1
8
]
an
d
v
is
io
n
tr
an
s
f
o
r
m
er
s
[
1
9
]
,
ac
h
i
ev
e
s
tate
-
of
-
th
e
-
a
r
t
r
esu
lts
b
y
b
alan
cin
g
d
ep
t
h
,
wid
th
,
an
d
r
eso
lu
tio
n
.
Ho
w
ev
er
,
C
NNs
p
r
io
r
itize
lo
ca
l
f
ea
tu
r
es
an
d
lack
m
ec
h
an
is
m
s
to
ex
p
licitly
m
o
d
el
s
p
atial
r
elatio
n
s
h
ip
s
b
etwe
en
d
is
tan
t
r
eg
io
n
s
[
2
0
]
.
Fo
r
e
x
am
p
le,
in
s
ce
n
e
r
ec
o
g
n
itio
n
,
C
NNs
m
ay
m
is
class
if
y
o
b
jects
d
u
e
t
o
co
n
te
x
tu
al
am
b
ig
u
ities
(
e.
g
.
,
a
"b
o
at"
o
n
a
r
o
ad
v
s
.
wate
r
)
[
2
]
,
[
2
1
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
780
-
7
8
7
782
2
.
2
.
T
wo
-
dim
ens
io
na
l hidd
e
n
M
a
rk
o
v
m
o
dels
2D
-
HM
Ms
ex
ten
d
tr
ad
itio
n
a
l
HM
Ms
to
g
r
id
-
b
ased
d
at
a,
m
o
d
elin
g
s
tate
tr
an
s
itio
n
s
in
two
d
im
en
s
io
n
s
.
Un
lik
e
1
D
-
HM
M
,
wh
ich
p
r
o
ce
s
s
s
eq
u
e
n
ce
s
,
2
D
-
HM
Ms
ca
p
tu
r
e
s
p
atial
d
e
p
en
d
en
cies
in
im
ag
es
b
y
d
ef
i
n
in
g
s
tates o
v
er
p
ix
el
n
eig
h
b
o
r
h
o
o
d
s
[
6
]
.
R
ec
en
t w
o
r
k
ap
p
lies
2
D
-
HM
Ms to
:
−
Do
cu
m
en
t a
n
al
y
s
is
: r
ec
o
g
n
izin
g
h
an
d
wr
itten
tex
t b
y
m
o
d
elin
g
ch
ar
ac
te
r
co
-
o
cc
u
r
r
e
n
ce
s
[
2
2
]
.
−
Me
d
ical
im
ag
in
g
:
s
eg
m
en
tin
g
tu
m
o
r
s
b
y
e
n
co
d
i
n
g
s
p
at
ial
p
r
io
r
s
[
1
3
]
.
Ho
wev
er
2
D
-
HM
Ms
ar
e
co
m
p
u
tatio
n
ally
in
ten
s
iv
e
an
d
o
f
ten
r
e
q
u
ir
e
a
p
p
r
o
x
im
atio
n
s
f
o
r
lar
g
e
-
s
ca
le
task
s
[
2
3
]
.
2
.
3
.
M
a
r
k
o
v
r
a
nd
o
m
f
ields
MRF
s
m
o
d
el
h
ig
h
er
-
o
r
d
er
d
ep
en
d
en
cies
b
y
d
ef
i
n
in
g
p
o
t
en
tials
o
v
er
cliq
u
es
(
g
r
o
u
p
s
o
f
n
o
d
es)
.
R
ec
en
t a
d
v
an
ce
s
f
o
cu
s
o
n
:
−
Par
s
im
o
n
io
u
s
h
ig
h
er
-
o
r
d
er
M
R
Fs
: r
ed
u
cin
g
c
o
m
p
u
tatio
n
al
co
m
p
lex
ity
wh
ile
r
etai
n
in
g
ac
cu
r
ac
y
[
2
4
]
.
−
Dee
p
lear
n
in
g
in
teg
r
atio
n
:
co
m
b
in
in
g
MRF
s
with
C
NN
s
f
o
r
task
s
lik
e
im
ag
e
d
en
o
is
in
g
[
1
1
]
an
d
p
o
s
e
esti
m
atio
n
.
MRF
s
ex
ce
l
in
s
t
r
u
ctu
r
ed
p
r
ed
ictio
n
b
u
t
r
eq
u
ir
e
ca
r
ef
u
l
tu
n
in
g
o
f
p
o
ten
tial
f
u
n
ctio
n
s
t
o
av
o
id
o
v
er
f
itti
n
g
[
2
5
]
,
[
2
6
]
.
2
.
4
.
Va
ri
a
t
io
na
l a
ut
o
enco
ders
VAE
s
lear
n
p
r
o
b
ab
ilis
tic
lat
en
t
r
ep
r
esen
tatio
n
s
b
y
m
ax
i
m
izin
g
a
v
ar
iatio
n
al
lo
wer
b
o
u
n
d
[
1
5
]
.
R
ec
en
t
ex
ten
s
io
n
s
,
s
u
ch
as
co
n
d
itio
n
al
VAE
s
[
1
4
]
an
d
β
-
VAE
s
[
2
7
]
,
im
p
r
o
v
e
d
is
en
tan
g
lem
en
t
an
d
r
o
b
u
s
tn
ess
.
I
n
r
ec
o
g
n
itio
n
task
s
,
VAE
s
:
−
R
ed
u
ce
o
v
er
f
itti
n
g
:
b
y
r
eg
u
lar
izin
g
laten
t sp
ac
es [
1
7
]
.
−
Han
d
le
m
is
s
in
g
d
ata:
th
r
o
u
g
h
p
r
o
b
a
b
ilis
tic
in
f
er
en
ce
[
2
8
]
.
Fo
r
ex
am
p
le,
VAE
s
co
m
b
in
e
d
with
C
NNs
ac
h
iev
e
9
8
.
7
% a
cc
u
r
ac
y
o
n
MN
I
ST
with
3
0
% c
o
r
r
u
p
ted
p
ix
els [
2
]
.
3.
M
E
T
H
O
D
3
.
1
.
Arc
hite
ct
ure
des
ig
n
T
h
e
p
r
o
p
o
s
ed
en
h
an
ce
d
h
y
b
r
id
C
NN
-
2D
-
HM
M
f
r
am
ewo
r
k
in
teg
r
ates
f
o
u
r
co
m
p
lem
en
tar
y
co
m
p
o
n
en
ts
to
a
d
d
r
ess
th
e
li
m
itatio
n
s
o
f
in
d
iv
id
u
al
ap
p
r
o
ac
h
es
wh
ile
lev
er
ag
i
n
g
t
h
eir
r
esp
ec
tiv
e
s
tr
en
g
th
s
.
T
h
e
f
r
a
m
ewo
r
k
co
m
b
in
es
a
C
NNs
b
ac
k
b
o
n
e
f
o
r
h
ier
ar
ch
ical
f
ea
tu
r
e
ex
tr
ac
tio
n
,
2
D
-
HM
Ms
f
o
r
s
p
atial
d
ep
en
d
e
n
cy
m
o
d
elin
g
,
MRF
s
f
o
r
h
ig
h
er
-
o
r
d
er
c
o
n
tex
t
ca
p
tu
r
e,
an
d
VAE
s
f
o
r
r
o
b
u
s
t
laten
t
r
ep
r
esen
tatio
n
lear
n
in
g
.
T
h
e
ar
ch
itectu
r
e
d
esi
g
n
in
co
r
p
o
r
ates f
o
u
r
k
ey
c
o
m
p
o
n
en
ts
wo
r
k
in
g
in
s
y
n
er
g
y
:
−
C
NN:
a
R
e
s
Net
-
5
0
b
ac
k
b
o
n
e
ex
tr
ac
ts
f
ea
tu
r
es
[
2
9
]
.
−
2D
-
HM
M:
m
o
d
els s
p
atial
d
ep
en
d
en
cies u
s
in
g
f
o
r
war
d
-
b
ac
k
war
d
alg
o
r
ith
m
s
[
2
2
]
.
−
MRF
: c
ap
tu
r
es h
ig
h
er
-
o
r
d
er
in
ter
ac
tio
n
s
v
ia
cliq
u
e
p
o
ten
tials
[
3
0
]
.
−
VAE
: le
ar
n
s
laten
t r
ep
r
esen
tatio
n
s
to
r
ed
u
ce
o
v
e
r
f
itti
n
g
[
1
5
]
.
3
.
2
.
T
ra
ini
ng
s
t
ra
t
eg
y
T
h
e
tr
ain
in
g
s
tr
ateg
y
em
p
l
o
y
s
a
h
y
b
r
id
lo
s
s
f
u
n
ctio
n
th
at
c
o
m
b
in
es
m
u
ltip
le
o
b
jectiv
es
to
o
p
tim
ize
th
e
en
tire
f
r
am
ew
o
r
k
j
o
in
tly
.
T
h
e
lo
s
s
f
u
n
ctio
n
is
f
o
r
m
u
late
d
as
(
1
)
.
ℒ
=
ℒ
r
e
co
g
+
1
ℒ
H
MM
+
2
ℒ
MRF
+
3
ℒ
V
AE
(
1
)
W
h
er
e
ℒ
r
e
c
og
r
ep
r
esen
ts
th
e
p
r
im
a
r
y
r
ec
o
g
n
itio
n
lo
s
s
,
ℒ
HM
M
ca
p
tu
r
es
s
p
atial
co
h
er
en
ce
th
r
o
u
g
h
2
D
-
HM
M
lik
elih
o
o
d
,
ℒ
M
R
F
en
f
o
r
ce
s
h
ig
h
er
-
o
r
d
er
co
n
te
x
t
co
n
s
is
ten
cy
,
an
d
ℒ
VAE
p
r
o
v
id
es
laten
t
s
p
ac
e
r
eg
u
lar
izatio
n
.
T
h
e
h
y
p
e
r
p
a
r
a
m
e
t
e
r
s
₁
,
₂
,
a
n
d
₃
a
r
e
c
a
r
e
f
u
l
l
y
t
u
n
e
d
t
o
b
a
l
a
n
c
e
t
h
e
co
n
t
r
i
b
u
t
i
o
n
s
o
f
e
a
c
h
c
o
m
p
o
n
e
n
t
,
e
n
s
u
r
i
n
g
t
h
a
t
t
h
e
f
r
a
m
ew
o
r
k
b
e
n
e
f
i
ts
f
r
o
m
a
l
l
i
n
te
g
r
a
t
e
d
m
o
d
u
l
es
w
ith
o
u
t
a
n
y
s
i
n
g
l
e
c
o
m
p
o
n
e
n
t
d
o
m
i
n
a
t
i
n
g
t
h
e
l
e
a
r
n
i
n
g
p
r
o
c
e
s
s
.
T
h
e
o
p
t
i
m
i
z
a
ti
o
n
p
r
o
c
e
d
u
r
e
f
o
l
l
o
w
s
es
t
a
b
li
s
h
e
d
b
es
t
p
r
a
c
t
i
c
es
f
o
r
G
PU
p
r
o
g
r
a
m
m
i
n
g
[
3
1
]
.
4.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
4
.
1
.
Q
ua
ntit
a
t
iv
e
a
na
ly
s
is
T
h
e
p
r
o
p
o
s
ed
h
y
b
r
id
f
r
am
ewo
r
k
was
r
ig
o
r
o
u
s
ly
ev
alu
ated
o
n
th
e
MN
I
ST
[
3
2
]
an
d
C
I
FAR
-
10
[
3
3
]
d
atasets
,
ac
h
iev
in
g
s
tate
-
of
-
t
h
e
-
ar
t
p
er
f
o
r
m
a
n
ce
ac
r
o
s
s
m
u
ltip
le
ev
alu
atio
n
m
etr
ics.
T
h
e
co
m
p
r
eh
en
s
iv
e
ev
alu
atio
n
in
clu
d
ed
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
in
f
er
e
n
ce
tim
e
m
ea
s
u
r
e
m
en
ts
to
p
r
o
v
id
e
a
th
o
r
o
u
g
h
ass
ess
m
en
t
o
f
s
y
s
tem
p
er
f
o
r
m
a
n
ce
.
T
h
e
e
x
p
er
i
m
en
tal
s
etu
p
u
tili
ze
d
s
tan
d
ar
d
ized
tr
ain
in
g
a
n
d
test
in
g
p
r
o
to
c
o
ls
to
en
s
u
r
e
f
ai
r
co
m
p
a
r
is
o
n
with
e
x
is
tin
g
m
eth
o
d
s
,
an
d
all
ex
p
er
im
en
ts
w
er
e
co
n
d
u
cted
u
s
in
g
id
en
tical
h
ar
d
war
e
c
o
n
f
ig
u
r
ati
o
n
s
to
m
ain
tain
c
o
n
s
is
ten
cy
in
co
m
p
u
tatio
n
al
p
er
f
o
r
m
an
ce
m
ea
s
u
r
em
en
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
etw
o
r
ks,
h
id
d
en
Ma
r
ko
v
mo
d
els,
a
n
d
a
u
to
en
c
o
n
d
ers
fo
r
en
h
a
n
ce
d
…
(
Dri
s
s
N
a
ji
)
783
4
.
1
.
1
.
M
NIS
T
da
t
a
s
et
On
th
e
MN
I
ST
d
ataset,
th
e
f
r
am
ewo
r
k
ac
h
iev
ed
a
n
ac
cu
r
ac
y
o
f
9
8
.
2
%,
s
u
r
p
ass
in
g
al
l
b
aselin
e
m
o
d
els.
T
h
e
b
r
ea
k
d
o
wn
o
f
p
e
r
f
o
r
m
a
n
ce
m
etr
ics
is
as
f
o
llo
w
s
in
T
ab
le
1
.
T
h
is
tab
le
s
u
m
m
ar
izes
th
e
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
an
d
in
f
er
e
n
ce
tim
e
o
f
v
ar
io
u
s
m
o
d
els
o
n
th
e
MN
I
ST
d
ataset.
Key
o
b
s
er
v
atio
n
s
:
i)
th
e
in
teg
r
atio
n
o
f
MRF
s
im
p
r
o
v
e
d
r
ec
all
b
y
~1
%,
h
ig
h
lig
h
tin
g
th
eir
ab
ilit
y
to
ca
p
tu
r
e
h
ig
h
e
r
-
o
r
d
er
d
ep
en
d
e
n
cies a
n
d
ii)
VAE
s
en
h
an
ce
d
p
r
ec
is
io
n
b
y
~0
.
8
%,
d
em
o
n
s
tr
atin
g
th
eir
ef
f
ec
tiv
e
n
e
s
s
in
r
ed
u
cin
g
n
o
is
e
an
d
im
p
r
o
v
in
g
r
o
b
u
s
tn
ess
.
Tab
le 1
.
P
e
rfo
rm
a
n
c
e
m
e
tri
c
s o
n
M
NIST
d
a
tas
e
t
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
I
n
f
e
r
e
n
c
e
t
i
me
(
ms)
S
t
a
n
d
a
l
o
n
e
C
N
N
9
6
.
5
9
6
.
3
9
6
.
4
9
6
.
3
2
.
1
C
N
N
-
2D
-
H
M
M
9
7
.
1
9
7
.
0
9
7
.
2
9
7
.
1
2
.
8
C
N
N
-
VAE
9
7
.
4
9
7
.
3
9
7
.
5
9
7
.
4
2
.
5
C
N
N
-
M
R
F
9
7
.
8
9
7
.
7
9
7
.
9
9
7
.
8
3
.
2
P
r
o
p
o
se
d
f
r
a
m
e
w
o
r
k
9
8
.
2
9
8
.
1
9
8
.
3
9
8
.
2
3
.
5
4
.
1
.
2
.
CIFAR
-
1
0
da
t
a
s
et
On
th
e
m
o
r
e
ch
allen
g
i
n
g
C
I
FAR
-
1
0
d
ataset,
th
e
f
r
am
e
wo
r
k
ac
h
iev
ed
an
ac
cu
r
ac
y
o
f
8
9
.
5
%,
o
u
tp
er
f
o
r
m
in
g
ex
is
tin
g
m
eth
o
d
s
.
T
h
e
r
esu
lts
ar
e
s
u
m
m
ar
iz
ed
in
T
ab
le
2
.
T
h
is
tab
le
p
r
esen
ts
th
e
ac
cu
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all,
F1
-
s
co
r
e,
a
n
d
in
f
er
en
ce
tim
e
o
f
d
if
f
er
e
n
t
m
o
d
els
o
n
t
h
e
m
o
r
e
c
h
allen
g
in
g
C
I
FAR
-
1
0
d
ataset.
Key
in
s
ig
h
ts
:
i)
th
e
co
m
b
in
atio
n
o
f
2
D
-
HM
M
an
d
MRF
s
s
ig
n
if
ican
tly
b
o
o
s
te
d
r
ec
all
b
y
~1
.
4
%,
ad
d
r
ess
in
g
ch
allen
g
es
p
o
s
ed
b
y
clu
tter
ed
s
ce
n
es
an
d
ii)
VAE
s
r
ed
u
ce
d
o
v
e
r
f
itti
n
g
,
im
p
r
o
v
in
g
g
en
e
r
aliza
tio
n
o
n
u
n
s
ee
n
test
d
ata.
Tab
le 2
.
P
e
rfo
rm
a
n
c
e
m
e
tri
c
s o
n
CIF
AR
-
1
0
d
a
tas
e
t
M
o
d
e
l
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
I
n
f
e
r
e
n
c
e
t
i
me
(
ms)
S
t
a
n
d
a
l
o
n
e
C
N
N
8
5
.
2
8
4
.
9
8
5
.
1
8
5
.
0
4
.
2
C
N
N
-
2D
-
H
M
M
8
6
.
7
8
6
.
5
8
6
.
8
8
6
.
6
4
.
9
C
N
N
-
VAE
8
7
.
3
8
7
.
1
8
7
.
4
8
7
.
2
4
.
6
C
N
N
-
M
R
F
8
8
.
1
8
8
.
0
8
8
.
2
8
8
.
1
5
.
3
P
r
o
p
o
se
d
f
r
a
m
e
w
o
r
k
8
9
.
5
8
9
.
4
8
9
.
6
8
9
.
5
5
.
7
4
.
2
.
Q
ua
lit
a
t
iv
e
a
na
ly
s
is
Fo
r
a
d
ee
p
e
r
u
n
d
er
s
tan
d
in
g
o
f
th
e
p
er
f
o
r
m
a
n
ce
an
d
th
e
b
eh
av
io
u
r
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el,
ex
ten
s
iv
e
q
u
alitativ
e
ev
alu
atio
n
s
wer
e
ca
r
r
ied
o
u
t
b
ased
o
n
a
d
v
an
c
ed
v
is
u
alis
atio
n
to
o
ls
an
d
d
e
tailed
ca
s
e
s
tu
d
ies.
T
h
ese
q
u
alitativ
e
an
aly
s
es
o
f
f
er
i
n
s
ig
h
ts
in
to
th
e
co
n
tr
i
b
u
tio
n
o
f
th
e
c
o
m
b
in
e
d
c
o
m
p
o
n
en
ts
f
o
r
s
y
s
tem
p
er
f
o
r
m
an
ce
a
n
d
c
o
n
s
tan
tly
p
r
o
v
e
s
ig
n
if
ican
t
i
n
u
n
d
er
s
tan
d
in
g
th
e
m
o
d
el
b
eh
av
i
o
u
r
u
n
d
er
d
if
f
er
e
n
t
s
itu
atio
n
s
.
Fo
r
ex
am
p
le,
th
e
atten
tio
n
h
ea
tm
ap
s
r
ev
ea
led
wh
ich
in
p
u
t
f
ea
tu
r
es
wer
e
d
e
em
ed
im
p
o
r
tan
t
f
o
r
d
ec
is
io
n
-
m
ak
in
g
b
y
th
e
m
o
d
e
l,
an
d
ca
s
e
s
tu
d
ies
s
h
o
wed
ed
g
e
ca
s
es
wh
er
e
th
e
m
o
d
el
p
er
f
o
r
m
ed
p
ar
ticu
lar
l
y
well
o
r
p
o
o
r
ly
at
a
g
iv
en
task
,
in
f
o
r
m
in
g
p
o
ten
tial f
o
cu
s
ed
i
m
p
r
o
v
e
m
en
ts
in
f
u
t
u
r
e
iter
atio
n
s
.
4
.
2
.
1
.
Vis
ua
liza
t
io
n o
f
la
t
ent
re
presenta
t
io
ns
W
e
a
n
a
l
y
ze
d
th
e
l
a
t
en
t
s
p
ac
e
l
e
a
r
n
e
d
b
y
t
h
e
V
A
E
co
m
p
o
n
en
t
u
s
i
n
g
t
-
d
i
s
t
r
ib
u
t
ed
s
t
o
c
h
a
s
t
i
c
n
e
i
g
h
b
o
r
e
m
b
e
d
d
i
n
g
(
t
-
S
N
E
)
[
3
4
]
.
T
h
e
v
i
s
u
a
l
iz
a
t
i
o
n
r
ev
e
a
l
ed
w
e
l
l
-
s
e
p
ar
a
t
ed
c
lu
s
t
er
s
f
o
r
e
a
ch
c
l
a
s
s
,
i
n
d
i
c
a
t
in
g
t
h
a
t
t
h
e
V
A
E
e
f
f
ec
t
i
v
e
l
y
d
i
s
en
t
a
n
g
l
ed
t
a
s
k
-
r
e
lev
a
n
t
f
e
a
tu
r
e
s
.
F
o
r
ex
a
m
p
l
e
:
i)
i
n
M
N
I
S
T
,
d
ig
i
t
s
w
i
t
h
s
i
m
i
l
ar
s
h
ap
e
s
(
e
.
g
.
,
"
3
"
a
n
d
"8
"
)
w
e
r
e
g
r
o
u
p
ed
c
lo
s
er
t
o
g
e
t
h
e
r
,
r
ef
l
e
c
t
in
g
t
h
e
ir
s
t
r
u
c
t
u
r
a
l
s
i
m
i
l
ar
i
t
i
e
s
a
n
d
i
i
)
i
n
C
I
F
A
R
-
1
0
,
o
b
j
e
c
t
s
w
i
t
h
s
h
ar
ed
a
t
tr
i
b
u
t
e
s
(
e
.
g
.
,
v
e
h
i
c
l
e
s
l
ik
e
c
a
r
s
a
n
d
t
r
u
c
k
s
)
f
o
r
m
ed
d
i
s
t
i
n
c
t
b
u
t
p
r
o
x
i
m
a
te
c
lu
s
t
e
r
s
.
4
.
2
.
2
.
At
t
ent
io
n ma
ps
Usi
n
g
th
e
SE
b
lo
ck
s
in
th
e
C
NN,
we
g
en
er
ated
atten
tio
n
m
ap
s
to
h
ig
h
lig
h
t
r
eg
io
n
s
o
f
in
ter
est.
T
h
ese
m
ap
s
d
em
o
n
s
tr
ated
th
at
th
e
m
o
d
el
f
o
cu
s
ed
o
n
d
is
cr
im
in
ativ
e
r
eg
io
n
s
.
Sp
ec
if
ically
,
it
h
ig
h
lig
h
ted
i)
th
e
ce
n
tr
al
s
tr
o
k
es
o
f
h
a
n
d
wr
itten
d
ig
its
in
MN
I
ST
an
d
ii)
k
e
y
o
b
ject
b
o
u
n
d
a
r
ies
in
C
I
FA
R
-
1
0
(
e.
g
.
,
ai
r
p
lan
e
win
g
s
,
b
ir
d
f
ea
t
h
er
s
)
.
4
.
2
.
3
.
F
a
ilu
re
ca
s
es
Desp
ite
its
h
ig
h
ac
c
u
r
ac
y
,
th
e
m
o
d
el
o
cc
asio
n
ally
m
is
class
i
f
ied
am
b
ig
u
o
u
s
s
am
p
les.
Fo
r
in
s
tan
ce
:
i
)
in
C
I
FAR
-
1
0
,
im
ag
es
with
h
ea
v
y
o
cc
lu
s
io
n
s
o
r
lo
w
c
o
n
tr
ast
(
e.
g
.
,
a
ca
t
p
a
r
tially
h
id
d
e
n
b
e
h
in
d
f
u
r
n
itu
r
e)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
780
-
7
8
7
784
wer
e
ch
allen
g
in
g
a
n
d
ii)
in
MN
I
ST,
h
ea
v
ily
s
ty
l
i
z
ed
d
ig
its
(
e.
g
.
,
a
"7
"
wr
itten
with
ad
d
itio
n
al
s
tr
o
k
es
)
ca
u
s
ed
co
n
f
u
s
io
n
.
T
h
ese
f
ailu
r
e
ca
s
es
u
n
d
er
s
co
r
e
th
e
im
p
o
r
tan
ce
o
f
i
n
co
r
p
o
r
atin
g
d
o
m
ai
n
-
s
p
ec
if
ic
p
r
i
o
r
s
o
r
au
g
m
en
tatio
n
s
to
h
an
d
le
e
d
g
e
ca
s
es.
4
.
3
.
Abla
t
io
n study
T
o
ev
alu
ate
th
e
c
o
n
tr
ib
u
tio
n
o
f
ea
ch
co
m
p
o
n
en
t,
we
p
er
f
o
r
m
ed
an
ab
latio
n
s
tu
d
y
b
y
s
y
s
tem
atica
lly
r
em
o
v
in
g
co
m
p
o
n
e
n
ts
f
r
o
m
th
e
f
r
am
ewo
r
k
.
T
h
e
r
esu
lts
o
n
C
I
FAR
-
1
0
ar
e
s
h
o
wn
i
n
T
ab
le
3
.
T
h
is
tab
le
s
h
o
ws
th
e
im
p
ac
t
o
f
r
em
o
v
i
n
g
i
n
d
iv
i
d
u
al
co
m
p
o
n
e
n
ts
(
VAE
,
MRF
,
an
d
2
D
-
HM
M)
f
r
o
m
th
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
,
h
ig
h
lig
h
tin
g
th
eir
co
n
tr
ib
u
tio
n
s
to
o
v
er
all
p
er
f
o
r
m
an
ce
.
K
ey
f
in
d
in
g
s
:
i)
r
em
o
v
in
g
th
e
VAE
led
to
a
1
.
3
%
d
r
o
p
in
ac
c
u
r
ac
y
,
em
p
h
asizin
g
its
r
o
le
in
r
o
b
u
s
tn
ess
;
ii)
r
em
o
v
in
g
t
h
e
MRF
r
esu
lte
d
in
a
1
.
6
%
d
r
o
p
,
h
ig
h
lig
h
tin
g
its
im
p
o
r
tan
ce
i
n
m
o
d
elin
g
h
ig
h
er
-
o
r
d
e
r
c
o
n
te
x
t;
an
d
iii)
r
em
o
v
in
g
th
e
2
D
-
HM
M
ca
u
s
ed
a
3
%
d
r
o
p
,
u
n
d
er
s
co
r
in
g
its
cr
itical
r
o
le
in
s
p
atial
m
o
d
elin
g
.
T
ab
le
3
.
Ab
latio
n
s
tu
d
y
r
esu
lt
s
o
n
C
I
FAR
-
10
C
o
n
f
i
g
u
r
a
t
i
o
n
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
F
u
l
l
f
r
a
mew
o
r
k
8
9
.
5
8
9
.
4
8
9
.
6
8
9
.
5
W
i
t
h
o
u
t
V
A
E
8
8
.
2
8
8
.
1
8
8
.
3
8
8
.
2
W
i
t
h
o
u
t
M
R
F
8
7
.
9
8
7
.
8
8
8
.
0
8
7
.
9
W
i
t
h
o
u
t
2
D
-
H
M
M
8
6
.
5
8
6
.
4
8
6
.
6
8
6
.
5
C
N
N
o
n
l
y
8
5
.
2
8
4
.
9
8
5
.
1
8
5
.
0
4
.
4
.
Co
m
pu
t
a
t
io
na
l e
f
f
iciency
W
h
ile
th
e
p
r
o
p
o
s
ed
f
r
am
ew
o
r
k
ac
h
ie
v
es
s
u
p
er
io
r
ac
c
u
r
a
cy
,
it
in
cu
r
s
ad
d
itio
n
al
co
m
p
u
tatio
n
al
o
v
er
h
ea
d
co
m
p
ar
ed
to
s
tan
d
a
lo
n
e
C
NNs.
T
h
e
in
f
er
en
ce
tim
es
p
er
im
a
g
e
ar
e
s
u
m
m
ar
iz
ed
in
T
ab
le
4
.
T
h
is
tab
le
co
m
p
a
r
es
th
e
in
f
er
en
ce
tim
es
(
in
m
illi
s
ec
o
n
d
s
)
o
f
th
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
an
d
b
ase
lin
e
m
o
d
els
o
n
b
o
th
MN
I
ST
an
d
C
I
FAR
-
1
0
d
atasets
.
Desp
ite
th
e
in
cr
ea
s
ed
co
m
p
u
tatio
n
,
t
h
e
f
r
a
m
ewo
r
k
r
e
m
ain
s
p
r
ac
tical
f
o
r
r
ea
l
-
tim
e
ap
p
licatio
n
s
,
esp
ec
ia
lly
wh
en
d
e
p
lo
y
ed
o
n
m
o
d
er
n
GPUs
.
T
ab
le
4
.
C
o
m
p
u
tatio
n
al
ef
f
icie
n
cy
co
m
p
ar
is
o
n
M
o
d
e
l
M
N
I
S
T
(
ms)
C
I
F
A
R
-
1
0
(
ms)
S
t
a
n
d
a
l
o
n
e
C
N
N
2
.
1
4
.
2
C
N
N
-
2D
-
H
M
M
2
.
8
4
.
9
C
N
N
-
VAE
2
.
5
4
.
6
C
N
N
-
M
R
F
3
.
2
5
.
3
P
r
o
p
o
se
d
f
r
a
m
e
w
o
r
k
3
.
5
5
.
7
4
.
5
.
Co
m
pa
riso
n wit
h sta
t
e
-
of
-
t
he
-
a
r
t
m
e
t
ho
ds
W
e
f
u
r
th
er
ev
alu
ated
o
u
r
f
r
a
m
ewo
r
k
u
s
in
g
th
e
C
I
FAR
-
1
0
b
en
ch
m
ar
k
s
an
d
co
m
p
ar
e
d
it
with
o
th
er
m
eth
o
d
s
to
d
e
m
o
n
s
tr
ate
t
h
at
it
s
p
er
f
o
r
m
an
ce
in
im
ag
e
class
i
f
icatio
n
is
co
m
p
etitiv
e.
W
e
co
m
p
ar
e
o
u
r
m
eth
o
d
with
v
is
io
n
tr
an
s
f
o
r
m
er
s
,
E
f
f
i
cien
tNet,
an
d
R
esNet
o
n
ac
c
u
r
ac
y
,
p
r
ec
is
io
n
,
r
ec
all
an
d
F
1
-
s
co
r
e
in
T
ab
le
5
s
h
o
win
g
th
at
o
u
r
m
o
d
el
h
as
b
etter
o
r
s
im
ilar
p
er
f
o
r
m
a
n
ce
o
n
all
m
etr
ics.
No
tab
ly
,
o
u
r
m
o
d
el
ac
h
iev
es
a
h
ig
h
er
p
r
ec
is
io
n
an
d
F1
-
s
co
r
e
co
m
p
ar
ed
to
v
is
io
n
tr
an
s
f
o
r
m
er
,
an
d
its
r
elativ
ely
lo
w
co
s
t
in
in
f
er
en
ce
m
ak
es
it
m
o
r
e
e
f
f
icien
t
f
o
r
p
r
ac
tical
d
ep
lo
y
m
e
n
t.
C
I
FAR
-
1
0
ac
h
ie
v
es
s
tate
-
of
-
th
e
-
ar
t
ac
cu
r
ac
y
o
f
9
8
.
2
%
an
d
8
9
.
5
%,
r
esp
ec
tiv
ely
,
in
a
n
o
is
e/o
cc
lu
s
io
n
r
o
b
u
s
t
s
ettin
g
,
co
m
p
a
r
ed
to
r
ec
en
t
m
eth
o
d
s
.
Ab
latio
n
s
tu
d
ies
s
h
o
w
th
at
MRF
s
in
cr
ea
s
e
r
ec
all
b
y
1
.
6
%,
an
d
VAE
s
im
p
r
o
v
e
p
r
ec
is
io
n
b
y
1
.
3
%,
s
h
o
win
g
t
h
eir
co
m
p
lem
en
tar
ity
.
T
h
is
wo
r
k
estab
lis
h
es
a
c
o
n
n
ec
ti
o
n
b
etwe
en
d
ee
p
lear
n
in
g
a
n
d
p
r
o
b
a
b
ilis
tic
g
r
ap
h
ical
m
o
d
els,
en
ab
lin
g
t
h
e
d
ev
elo
p
m
e
n
t
o
f
in
ter
p
r
etab
le,
s
ca
lab
le,
an
d
m
o
r
e
ac
c
u
r
ate
r
ec
o
g
n
itio
n
s
y
s
tem
s
.
Ou
r
f
r
a
m
ewo
r
k
o
u
tp
er
f
o
r
m
s
b
o
th
tr
ad
itio
n
al
C
NNs
an
d
tr
an
s
f
o
r
m
er
-
b
ased
ar
ch
itectu
r
e
s
,
d
em
o
n
s
tr
atin
g
th
e
s
y
n
er
g
y
b
etwe
en
C
NNs,
2D
-
HM
M,
MRF
s
,
an
d
VAE
s
.
Tab
le 5
.
Co
m
p
a
riso
n
wit
h
sta
te
-
of
-
th
e
-
a
rt
m
e
th
o
d
s o
n
CIF
AR
-
10
M
e
t
h
o
d
A
c
c
u
r
a
c
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
R
e
c
a
l
l
(
%)
F1
-
sc
o
r
e
(
%)
V
i
si
o
n
t
r
a
n
sf
o
r
m
e
r
s [2
]
8
8
.
4
8
8
.
3
8
8
.
5
8
8
.
4
Ef
f
i
c
i
e
n
t
N
e
t
-
B
0
[
3
]
8
7
.
9
8
7
.
8
8
8
.
0
8
7
.
9
R
e
sN
e
t
-
5
0
[
4
]
8
6
.
5
8
6
.
4
8
6
.
6
8
6
.
5
P
r
o
p
o
se
d
f
r
a
m
e
w
o
r
k
8
9
.
5
8
9
.
4
8
9
.
6
8
9
.
5
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
etw
o
r
ks,
h
id
d
en
Ma
r
ko
v
mo
d
els,
a
n
d
a
u
to
en
c
o
n
d
ers
fo
r
en
h
a
n
ce
d
…
(
Dri
s
s
N
a
ji
)
785
5.
CO
NCLU
SI
O
N
I
n
th
is
p
ap
er
,
we
h
a
v
e
a
way
to
an
en
h
an
ce
d
h
y
b
r
id
C
NN
-
2D
-
HM
M
f
r
am
ewo
r
k
ex
p
lo
it
in
g
MRF
s
an
d
VAE
s
in
o
r
d
e
r
to
e
n
h
a
n
ce
s
p
atial
m
o
d
elin
g
a
n
d
s
y
s
tem
r
o
b
u
s
tn
ess
f
o
r
r
ec
o
g
n
it
io
n
p
u
r
p
o
s
es.
T
h
e
p
r
o
p
o
s
ed
f
r
a
m
ewo
r
k
tack
le
s
s
ev
er
al
ch
allen
g
es
s
u
ch
as
lo
n
g
-
r
an
g
e
s
p
atial
d
e
p
en
d
en
cies;
h
i
g
h
er
o
r
d
er
-
c
o
n
tex
tu
al
r
elatio
n
s
h
ip
s
an
d
n
o
is
e
s
u
p
p
r
ess
if
icatio
n
b
y
e
x
p
lo
itin
g
th
e
s
y
n
er
g
y
a
m
o
n
g
C
NN
f
ea
tu
r
e
ex
tr
ac
tio
n
,
2
D
-
HM
M
b
ased
s
p
atial
m
o
d
elin
g
,
MRF
co
n
tex
tu
al
an
aly
s
is
an
d
VAE
la
ten
t
r
ep
r
esen
tatio
n
lear
n
in
g
.
I
t
r
ep
o
r
ted
s
tate
-
of
-
t
h
e
-
ar
t
p
e
r
f
o
r
m
an
ce
o
n
MN
I
S
T
with
9
8
.
2
%
ac
cu
r
ac
y
,
an
d
with
C
I
FAR
-
1
0
th
e
ac
cu
r
ac
y
was
8
9
.
5
%.
T
h
r
o
u
g
h
an
ex
ten
s
iv
e
ex
p
er
im
en
tal
an
aly
s
is
,
we
d
em
o
n
s
tr
ate
th
e
clea
r
ad
v
an
tag
e
o
f
ea
ch
co
m
p
o
n
en
t:
2
D
-
HM
M
s
p
atial
m
o
d
ellin
g
elev
ates
ac
cu
r
ac
y
b
y
3
%,
MRF
co
n
tex
t
ca
p
tu
r
e
b
y
a
n
o
th
er
1
.
6
%
an
d
VAE
r
o
b
u
s
tn
ess
b
y
f
u
r
th
er
1
.
3
%.
T
h
e
p
r
o
p
o
s
ed
f
r
am
ewo
r
k
p
o
s
s
ess
es
co
m
p
u
tatio
n
al
ef
f
icien
cy
an
d
ac
h
iev
es
s
ig
n
if
ican
t
p
er
f
o
r
m
a
n
ce
en
h
an
ce
m
e
n
ts
av
ailab
le
f
o
r
ac
tu
al
ap
p
licatio
n
s
.
Salien
t
f
u
tu
r
e
d
ir
ec
tio
n
s
m
ay
in
clu
d
e:
s
ca
lin
g
u
p
th
r
o
u
g
h
m
o
r
e
ef
f
icien
t
2
D
-
HM
M
ap
p
r
o
x
im
atio
n
alg
o
r
ith
m
;
h
ar
d
war
e
-
f
r
ien
d
ly
ar
ch
itectu
r
e,
c.
f
.
E
f
f
icien
tNe
t
o
r
v
is
io
n
tr
an
s
f
o
r
m
er
s
(
c
o
m
p
ar
e
lin
ea
r
-
s
am
p
ler
)
,
d
o
m
ain
ad
ap
tatio
n
i
n
m
ed
ical/satellit
e
im
ag
in
g
ap
p
licatio
n
s
,
th
eo
r
etica
l
u
n
d
e
r
s
tan
d
in
g
o
f
th
e
h
y
b
r
id
lo
s
s
f
u
n
ctio
n
tr
ain
(
)
.
M
o
r
e
im
p
o
r
tan
tly
,
th
is
f
r
a
m
ewo
r
k
ca
n
in
s
p
ir
e
o
th
er
r
ea
l
-
tim
e
c
o
n
d
itio
n
al
o
p
tim
izatio
n
task
s
an
o
th
e
r
p
r
o
m
is
in
g
d
ir
ec
tio
n
is
to
im
p
r
o
v
e
th
e
ex
p
lain
ab
ilit
y
with
in
tr
icate
p
r
o
b
ab
ilis
tic
co
m
p
o
n
en
ts
f
o
r
in
ter
p
r
etab
le
p
r
ed
ictio
n
s
an
d
u
n
ce
r
tain
ty
q
u
a
n
tific
atio
n
.
T
h
e
f
r
a
m
e
w
o
r
k
i
s
a
m
a
j
o
r
m
o
v
e
t
o
w
a
r
d
c
l
o
s
i
n
g
t
h
e
g
a
p
b
e
tw
e
e
n
d
e
e
p
l
e
a
r
n
i
n
g
a
n
d
p
r
o
b
a
b
i
l
i
s
t
i
c
g
r
a
p
h
i
c
al
m
o
d
e
l
s
,
w
h
i
c
h
s
e
r
v
es
a
s
a
s
o
l
i
d
f
o
u
n
d
a
t
i
o
n
f
o
r
m
o
r
e
a
d
v
a
n
c
e
d
,
i
n
t
e
r
p
r
e
t
a
b
l
e
a
n
d
r
o
b
u
s
t
r
e
c
o
g
n
i
t
i
o
n
s
y
s
t
e
m
s
i
n
r
e
a
l
-
w
o
r
l
d
v
is
u
a
l
t
as
k
s
.
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
Dr
is
s
Naji
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Kam
al
E
lh
attab
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ab
d
elali
J
o
u
m
ad
✓
✓
✓
✓
✓
✓
✓
✓
✓
Ab
d
elo
u
ah
e
d
Ait I
d
er
✓
✓
✓
✓
✓
✓
✓
✓
Ab
d
elk
b
ir
Ou
is
aa
d
an
e
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Azz
ed
in
e
I
d
h
m
ad
✓
✓
✓
✓
✓
✓
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
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
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
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
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
Au
th
o
r
s
s
tate
n
o
co
n
f
lict o
f
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
ata
th
at
s
u
p
p
o
r
t
th
e
f
in
d
in
g
s
o
f
th
is
s
tu
d
y
ar
e
p
u
b
li
cly
av
ailab
le
b
en
ch
m
ar
k
d
atasets
.
T
h
e
MN
I
ST
d
ataset
is
o
p
en
ly
ac
ce
s
s
ib
le
at
h
ttp
://y
an
n
.
lecu
n
.
co
m
/ex
d
b
/m
n
is
t/,
an
d
th
e
C
I
FAR
-
1
0
d
ataset
is
o
p
en
ly
ac
ce
s
s
ib
le
at
h
ttp
s
://www.
cs.to
r
o
n
to
.
ed
u
/~k
r
iz/cif
ar
.
h
tm
l.
T
h
e
tr
ain
e
d
m
o
d
el
wei
g
h
ts
,
ex
p
er
im
e
n
tal
co
n
f
ig
u
r
atio
n
s
u
s
ed
in
th
is
s
tu
d
y
ar
e
a
v
ailab
le
f
r
o
m
th
e
c
o
r
r
esp
o
n
d
in
g
au
th
o
r
,
[
DN]
,
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
M
.
A
l
y
,
A
.
G
h
a
l
l
a
b
,
a
n
d
I
.
S
.
F
a
t
h
i
,
“
En
h
a
n
c
i
n
g
f
a
c
i
a
l
e
x
p
r
e
ssi
o
n
r
e
c
o
g
n
i
t
i
o
n
s
y
st
e
m
i
n
o
n
l
i
n
e
l
e
a
r
n
i
n
g
c
o
n
t
e
x
t
u
si
n
g
e
f
f
i
c
i
e
n
t
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
,
”
I
EEE
Ac
c
e
ss
,
v
o
l
.
1
1
,
p
p
.
1
2
1
4
1
9
–
1
2
1
4
3
3
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
0
9
/
A
C
C
ESS
.
2
0
2
3
.
3
3
2
5
4
0
7
.
[
2
]
M
.
S
h
a
f
i
q
a
n
d
Z.
G
u
,
“
D
e
e
p
r
e
si
d
u
a
l
l
e
a
r
n
i
n
g
f
o
r
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
:
a
s
u
r
v
e
y
,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
2
,
n
o
.
1
8
,
S
e
p
.
2
0
2
2
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
2
1
8
8
9
7
2
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
780
-
7
8
7
786
[
3
]
N
.
D
e
n
g
,
Z.
L.
X
.
X
u
,
C
.
G
a
o
,
a
n
d
X
.
W
a
n
g
,
“
D
e
e
p
l
e
a
r
n
i
n
g
a
n
d
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
:
f
a
c
e
r
e
c
o
g
n
i
t
i
o
n
a
p
p
r
o
a
c
h
b
a
s
e
d
o
n
t
h
e
DS
-
C
D
C
N
a
l
g
o
r
i
t
h
m,
”
A
p
p
l
i
e
d
S
c
i
e
n
c
e
s
,
v
o
l
.
1
4
,
n
o
.
1
3
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
a
p
p
1
4
1
3
5
7
3
9
.
[
4
]
A
.
D
o
s
o
v
i
t
sk
i
y
e
t
a
l
.
,
“
A
n
i
ma
g
e
i
s
w
o
r
t
h
1
6
x
1
6
w
o
r
d
s
:
t
r
a
n
sf
o
r
mers
f
o
r
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
a
t
sc
a
l
e
,
”
Ju
n
.
2
0
2
1
,
a
rXi
v
:
2
0
1
0
.
1
1
9
2
9
.
[
5
]
J.
Z
h
a
n
g
,
Q
.
a
n
d
R
.
J.
Z
h
a
n
g
,
Y
.
Zh
a
o
,
a
n
d
J
.
L
i
u
,
“
S
p
a
t
i
a
l
-
c
o
n
t
e
x
t
-
a
w
a
r
e
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
m
u
l
t
i
-
c
l
a
ss
i
m
a
g
e
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
i
n
I
C
AS
S
P
2
0
2
2
-
2
0
2
2
I
E
EE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
Ac
o
u
st
i
c
s,
S
p
e
e
c
h
a
n
d
S
i
g
n
a
l
Pro
c
e
ssi
n
g
(
I
C
A
S
S
P)
,
2
0
2
2
,
p
p
.
1
9
6
0
–
1
9
6
4
.
d
o
i
:
1
0
.
1
1
0
9
/
I
C
A
S
S
P
4
3
9
2
2
.
2
0
2
2
.
9
7
4
6
9
2
1
.
[
6
]
N
.
M
a
n
o
u
c
h
e
h
r
i
a
n
d
N
.
B
o
u
g
u
i
l
a
,
“
H
u
ma
n
a
c
t
i
v
i
t
y
r
e
c
o
g
n
i
t
i
o
n
w
i
t
h
a
n
H
M
M
-
b
a
s
e
d
g
e
n
e
r
a
t
i
v
e
mo
d
e
l
,
”
S
e
n
s
o
rs
,
v
o
l
.
2
3
,
n
o
.
3
,
Jan
.
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
s2
3
0
3
1
3
9
0
.
[
7
]
G
.
M
a
n
o
h
a
r
a
n
a
n
d
K
.
S
i
v
a
k
u
m
a
r
,
“
A
mo
d
i
f
i
e
d
h
i
d
d
e
n
M
a
r
k
o
v
m
o
d
e
l
f
o
r
o
u
t
l
i
e
r
d
e
t
e
c
t
i
o
n
i
n
m
u
l
t
i
v
a
r
i
a
t
e
d
a
t
a
se
t
s,”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
En
g
i
n
e
e
ri
n
g
S
y
s
t
e
m
s
M
o
d
e
l
l
i
n
g
a
n
d
S
i
m
u
l
a
t
i
o
n
,
v
o
l
.
1
5
,
n
o
.
3
,
p
p
.
1
2
1
–
1
2
8
,
2
0
2
4
,
d
o
i
:
1
0
.
1
5
0
4
/
I
JES
M
S
.
2
0
2
4
.
1
3
8
2
8
7
.
[
8
]
I
.
M
u
r
a
t
a
j
e
t
a
l
.
,
“
H
y
p
e
r
b
o
l
i
c
met
a
mat
e
r
i
a
l
s
v
i
a
h
i
e
r
a
r
c
h
i
c
a
l
b
l
o
c
k
c
o
p
o
l
y
mer
n
a
n
o
s
t
r
u
c
t
u
r
e
s,
”
Ad
v
a
n
c
e
d
O
p
t
i
c
a
l
M
a
t
e
ri
a
l
s
,
v
o
l
.
9
,
n
o
.
7
,
A
p
r
.
2
0
2
1
,
d
o
i
:
1
0
.
1
0
0
2
/
a
d
o
m.
2
0
2
0
0
1
9
3
3
.
[
9
]
Y
.
Y
u
a
n
e
t
a
l
.
,
“
Tw
o
-
d
i
me
n
si
o
n
a
l
n
a
n
o
m
a
t
e
r
i
a
l
s
a
s
e
n
h
a
n
c
e
d
s
u
r
f
a
c
e
p
l
a
s
mo
n
r
e
s
o
n
a
n
c
e
s
e
n
si
n
g
p
l
a
t
f
o
r
ms:
d
e
si
g
n
p
e
r
sp
e
c
t
i
v
e
s
a
n
d
i
l
l
u
s
t
r
a
t
i
v
e
a
p
p
l
i
c
a
t
i
o
n
s,
”
B
i
o
se
n
s
o
rs
a
n
d
Bi
o
e
l
e
c
t
r
o
n
i
c
s
,
v
o
l
.
2
4
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
b
i
o
s.
2
0
2
3
.
1
1
5
6
7
2
.
[
1
0
]
F
.
Li
u
,
W
.
Y
a
n
g
,
a
n
d
J.
L
i
,
“
A
f
i
n
i
t
e
e
l
e
m
e
n
t
me
t
h
o
d
f
o
r
h
y
p
e
r
b
o
l
i
c
m
e
t
a
m
a
t
e
r
i
a
l
s
w
i
t
h
a
p
p
l
i
c
a
t
i
o
n
s
f
o
r
h
y
p
e
r
l
e
n
s,
”
S
I
A
M
J
o
u
rn
a
l
o
n
N
u
m
e
r
i
c
a
l
An
a
l
y
si
s
,
v
o
l
.
6
2
,
n
o
.
3
,
p
p
.
1
4
2
0
–
1
4
4
2
,
J
u
n
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
3
7
/
2
3
M
1
5
9
1
2
0
7
.
[
1
1
]
P
.
P
a
n
,
C
.
Z
h
a
n
g
,
J.
S
u
n
,
a
n
d
L.
G
u
o
,
“
M
u
l
t
i
-
s
c
a
l
e
c
o
n
v
-
a
t
t
e
n
t
i
o
n
U
-
N
e
t
f
o
r
me
d
i
c
a
l
i
ma
g
e
se
g
me
n
t
a
t
i
o
n
,
”
S
c
i
e
n
t
i
f
i
c
Re
p
o
r
t
s
,
v
o
l
.
1
5
,
n
o
.
1
,
A
p
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
5
-
9
6
1
0
1
-
8.
[
1
2
]
W
.
D
o
n
g
,
B
.
D
u
,
a
n
d
Y
.
X
u
,
“
S
h
a
p
e
-
i
n
t
e
n
si
t
y
-
g
u
i
d
e
d
U
-
n
e
t
f
o
r
m
e
d
i
c
a
l
i
m
a
g
e
s
e
g
me
n
t
a
t
i
o
n
,
”
N
e
u
r
o
c
o
m
p
u
t
i
n
g
,
v
o
l
.
6
1
0
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
n
e
u
c
o
m.
2
0
2
4
.
1
2
8
5
3
4
.
[
1
3
]
J.
S
h
e
d
b
a
l
k
a
r
a
n
d
K
.
P
r
a
b
h
u
s
h
e
t
t
y
,
“
D
e
e
p
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
m
o
d
e
l
f
o
r
b
r
a
i
n
t
u
mo
r
se
g
me
n
t
a
t
i
o
n
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
U
N
e
t
a
n
d
c
h
o
p
p
e
d
V
G
G
N
e
t
,
”
I
n
d
o
n
e
si
a
n
J
o
u
rn
a
l
o
f
E
l
e
c
t
ri
c
a
l
En
g
i
n
e
e
r
i
n
g
a
n
d
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
3
3
,
n
o
.
3
,
p
p
.
1
4
0
5
–
1
4
1
5
,
M
a
r
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
e
c
s
.
v
3
3
.
i
3
.
p
p
1
4
0
5
-
1
4
1
5
.
[
1
4
]
S
.
C
h
e
n
a
n
d
W
.
G
u
o
,
“
A
u
t
o
-
e
n
c
o
d
e
r
s
i
n
d
e
e
p
l
e
a
r
n
i
n
g
—
a
r
e
v
i
e
w
w
i
t
h
n
e
w
p
e
r
s
p
e
c
t
i
v
e
s,”
Ma
t
h
e
m
a
t
i
c
s
,
v
o
l
.
1
1
,
n
o
.
8
,
2
0
2
3
,
d
o
i
:
1
0
.
3
3
9
0
/
mat
h
1
1
0
8
1
7
7
7
.
[
1
5
]
Y
.
Z
h
a
o
a
n
d
S
.
L
i
n
d
e
r
ma
n
,
“
R
e
v
i
s
i
t
i
n
g
s
t
r
u
c
t
u
r
e
d
v
a
r
i
a
t
i
o
n
a
l
a
u
t
o
e
n
c
o
d
e
r
s,”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
,
2
0
2
3
,
p
p
.
4
2
0
4
6
–
4
2
0
5
7
.
[
1
6
]
J.
H
u
a
n
g
,
W
.
Y
a
n
,
G
.
L
i
,
T.
Li
,
a
n
d
S
.
Li
u
,
“
L
e
a
r
n
i
n
g
d
i
se
n
t
a
n
g
l
e
d
r
e
p
r
e
s
e
n
t
a
t
i
o
n
f
o
r
m
u
l
t
i
-
v
i
e
w
3
D
o
b
j
e
c
t
r
e
c
o
g
n
i
t
i
o
n
,
”
I
EEE
T
ra
n
s
a
c
t
i
o
n
s
o
n
C
i
r
c
u
i
t
s
a
n
d
S
y
st
e
m
s
f
o
r
V
i
d
e
o
T
e
c
h
n
o
l
o
g
y
,
v
o
l
.
3
2
,
n
o
.
2
,
p
p
.
6
4
6
–
6
5
9
,
2
0
2
1
,
d
o
i
:
1
0
.
1
1
0
9
/
TC
S
V
T
.
2
0
2
1
.
3
0
6
2
1
9
0
[
1
7
]
I
.
D
.
M
i
e
n
y
e
a
n
d
T.
G
.
S
w
a
r
t
,
“
A
c
o
mp
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
d
e
e
p
l
e
a
r
n
i
n
g
:
a
r
c
h
i
t
e
c
t
u
r
e
s
,
r
e
c
e
n
t
a
d
v
a
n
c
e
s
,
a
n
d
a
p
p
l
i
c
a
t
i
o
n
s,
”
I
n
f
o
rm
a
t
i
o
n
,
v
o
l
.
1
5
,
n
o
.
1
2
,
N
o
v
.
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
i
n
f
o
1
5
1
2
0
7
5
5
.
[
1
8
]
M
.
T
a
n
a
n
d
Q
.
V
.
Le
,
“
Ef
f
i
c
i
e
n
t
N
e
t
:
r
e
t
h
i
n
k
i
n
g
m
o
d
e
l
sca
l
i
n
g
f
o
r
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
c
o
n
f
e
re
n
c
e
o
n
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
,
S
e
p
.
2
0
2
0
,
p
p
.
6
1
0
5
–
6
1
1
4
.
[
1
9
]
Q
.
Z
h
a
n
g
,
Y
.
X
u
,
J.
Z
h
a
n
g
,
a
n
d
D
.
Ta
o
,
“
V
i
TA
E
v
2
:
v
i
s
i
o
n
t
r
a
n
sf
o
r
mer
a
d
v
a
n
c
e
d
b
y
e
x
p
l
o
r
i
n
g
i
n
d
u
c
t
i
v
e
b
i
a
s
f
o
r
i
ma
g
e
r
e
c
o
g
n
i
t
i
o
n
a
n
d
b
e
y
o
n
d
,
”
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
e
r
V
i
si
o
n
,
v
o
l
.
1
3
1
,
n
o
.
5
,
p
p
.
1
1
4
1
–
1
1
6
2
,
M
a
y
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
1
2
6
3
-
022
-
0
1
7
3
9
-
w.
[
2
0
]
M
.
A
.
H
a
s
a
n
a
n
d
K
.
D
e
y
,
"
D
e
p
t
h
w
i
se
sep
a
r
a
b
l
e
c
o
n
v
o
l
u
t
i
o
n
s
w
i
t
h
d
e
e
p
r
e
s
i
d
u
a
l
c
o
n
v
o
l
u
t
i
o
n
s,"
N
o
v
.
1
2
,
2
0
2
4
,
a
r
Xi
v
:
2
4
1
1
.
0
7
5
4
4
.
[
2
1
]
D
.
C
a
k
i
r
a
n
d
N
.
A
r
i
c
a
,
“
C
a
sc
a
d
i
n
g
C
N
N
s
f
o
r
f
a
c
i
a
l
a
c
t
i
o
n
u
n
i
t
d
e
t
e
c
t
i
o
n
,
”
En
g
i
n
e
e
ri
n
g
S
c
i
e
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
a
n
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
,
v
o
l
.
4
7
,
N
o
v
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
e
s
t
c
h
.
2
0
2
3
.
1
0
1
5
5
3
.
[
2
2
]
J.
M
a
,
Z
.
W
a
n
g
,
a
n
d
J
.
D
u
,
“
A
n
o
p
e
n
-
so
u
r
c
e
l
i
b
r
a
r
y
o
f
2
D
-
G
M
M
-
H
M
M
b
a
sed
o
n
K
a
l
d
i
T
o
o
l
k
i
t
a
n
d
i
t
s a
p
p
l
i
c
a
t
i
o
n
t
o
h
a
n
d
w
r
i
t
t
e
n
C
h
i
n
e
se
c
h
a
r
a
c
t
e
r
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
I
m
a
g
e
a
n
d
G
ra
p
h
i
c
s
,
2
0
2
1
,
p
p
.
2
3
5
–
2
4
4
.
d
o
i
:
1
0
.
1
0
0
7
/
9
7
8
-
3
-
0
3
0
-
8
7
3
5
5
-
4
_
2
0
.
[
2
3
]
A
.
Jo
u
ma
d
,
A
.
El
M
o
u
t
a
o
u
a
k
k
i
l
,
A
.
N
a
sr
o
a
l
l
a
h
,
O
.
B
o
u
t
k
h
o
u
m,
F
.
R
u
s
t
a
m,
a
n
d
I
.
A
sh
r
a
f
,
“
U
n
s
u
p
e
r
v
i
se
d
st
a
t
i
st
i
c
a
l
i
ma
g
e
seg
m
e
n
t
a
t
i
o
n
u
s
i
n
g
b
i
-
d
i
me
n
si
o
n
a
l
h
i
d
d
e
n
M
a
r
k
o
v
c
h
a
i
n
s
m
o
d
e
l
w
i
t
h
a
p
p
l
i
c
a
t
i
o
n
t
o
m
a
mm
o
g
r
a
p
h
y
i
ma
g
e
s,
”
J
o
u
r
n
a
l
o
f
K
i
n
g
S
a
u
d
U
n
i
v
e
rsi
t
y
-
C
o
m
p
u
t
e
r
a
n
d
I
n
f
o
r
m
a
t
i
o
n
S
c
i
e
n
c
e
s
,
v
o
l
.
3
5
,
n
o
.
9
,
O
c
t
.
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
j
k
s
u
c
i
.
2
0
2
3
.
1
0
1
7
1
5
.
[
2
4
]
Q
.
H
u
,
F
.
W
a
n
g
,
J
.
F
a
n
g
,
a
n
d
Y
.
L
i
,
“
S
e
ma
n
t
i
c
l
a
b
e
l
i
n
g
o
f
h
i
g
h
-
r
e
s
o
l
u
t
i
o
n
i
m
a
g
e
s
c
o
m
b
i
n
i
n
g
a
se
l
f
-
c
a
s
c
a
d
e
d
mu
l
t
i
m
o
d
a
l
f
u
l
l
y
c
o
n
v
o
l
u
t
i
o
n
n
e
u
r
a
l
n
e
t
w
o
r
k
w
i
t
h
f
u
l
l
y
c
o
n
d
i
t
i
o
n
a
l
r
a
n
d
o
m f
i
e
l
d
,
”
Re
m
o
t
e
S
e
n
si
n
g
,
v
o
l
.
1
6
,
n
o
.
1
7
,
2
0
2
4
,
d
o
i
:
1
0
.
3
3
9
0
/
r
s
1
6
1
7
3
3
0
0
.
[
2
5
]
Y
.
W
a
n
g
,
H
.
Z
h
a
n
g
,
S
.
W
a
n
g
,
Y
.
L
o
n
g
,
a
n
d
L
.
Y
a
n
g
,
“
S
e
ma
n
t
i
c
c
o
m
b
i
n
e
d
n
e
t
w
o
r
k
f
o
r
z
e
r
o
‐
s
h
o
t
sc
e
n
e
p
a
r
si
n
g
,
”
I
ET
I
m
a
g
e
Pro
c
e
ssi
n
g
,
v
o
l
.
1
4
,
n
o
.
4
,
p
p
.
7
5
7
–
7
6
5
,
M
a
r
.
2
0
2
0
,
d
o
i
:
1
0
.
1
0
4
9
/
i
e
t
-
i
p
r
.
2
0
1
9
.
0
8
7
0
.
[
2
6
]
W
.
S
h
a
f
i
k
,
A
.
T
u
f
a
i
l
,
C
.
L
.
D
.
S
i
l
v
a
,
a
n
d
R
.
A
.
A
.
H
.
M
.
A
p
o
n
g
,
“
A
n
o
v
e
l
h
y
b
r
i
d
i
n
c
e
p
t
i
o
n
-
x
c
e
p
t
i
o
n
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
f
o
r
e
f
f
i
c
i
e
n
t
p
l
a
n
t
d
i
se
a
se
c
l
a
ss
i
f
i
c
a
t
i
o
n
a
n
d
d
e
t
e
c
t
i
o
n
,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
r
t
s
,
v
o
l
.
1
5
,
n
o
.
1
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
3
8
/
s
4
1
5
9
8
-
0
2
4
-
8
2
8
5
7
-
y.
[
2
7
]
H
.
M
e
sh
k
i
n
e
t
a
l
.
,
“
H
a
r
n
e
ss
i
n
g
l
a
r
g
e
l
a
n
g
u
a
g
e
mo
d
e
l
s’
z
e
r
o
-
s
h
o
t
a
n
d
f
e
w
-
sh
o
t
l
e
a
r
n
i
n
g
c
a
p
a
b
i
l
i
t
i
e
s
f
o
r
r
e
g
u
l
a
t
o
r
y
r
e
se
a
r
c
h
,
”
Bri
e
f
i
n
g
s i
n
Bi
o
i
n
f
o
rm
a
t
i
c
s
,
v
o
l
.
2
5
,
n
o
.
5
,
Ju
l
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
9
3
/
b
i
b
/
b
b
a
e
3
5
4
.
[
2
8
]
T.
C
h
e
n
,
S
.
K
o
r
n
b
l
i
t
h
,
M
.
N
o
r
o
u
z
i
,
a
n
d
G
.
H
i
n
t
o
n
,
“
A
s
i
m
p
l
e
f
r
a
mew
o
r
k
f
o
r
c
o
n
t
r
a
st
i
v
e
l
e
a
r
n
i
n
g
o
f
v
i
s
u
a
l
r
e
p
r
e
sen
t
a
t
i
o
n
s,”
i
n
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
Ma
c
h
i
n
e
L
e
a
r
n
i
n
g
,
Ju
l
.
2
0
2
0
,
p
p
.
1
5
9
7
–
1
6
0
7
.
[
2
9
]
S
.
R
.
K
e
m
p
a
n
n
a
e
t
a
l
.
,
“
R
e
v
o
l
u
t
i
o
n
i
z
i
n
g
b
r
a
i
n
t
u
m
o
r
d
i
a
g
n
o
s
e
s:
a
R
e
sN
e
t
1
8
a
n
d
f
o
c
a
l
l
o
ss
a
p
p
r
o
a
c
h
t
o
m
a
g
n
e
t
i
c
r
e
s
o
n
a
n
c
e
i
ma
g
i
n
g
-
b
a
s
e
d
c
l
a
ss
i
f
i
c
a
t
i
o
n
i
n
n
e
u
r
o
-
o
n
c
o
l
o
g
y
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
r
n
a
l
o
f
El
e
c
t
r
i
c
a
l
a
n
d
C
o
m
p
u
t
e
r
E
n
g
i
n
e
e
ri
n
g
,
v
o
l
.
1
4
,
n
o
.
6
,
pp
.
6
5
5
1
-
6
5
5
9
,
D
e
c
.
2
0
2
4
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
c
e
.
v
1
4
i
6
.
p
p
6
5
5
1
-
6
5
5
9
.
[
3
0
]
Y
.
W
a
n
g
,
C
.
Y
i
n
g
,
X
.
L
u
o
,
a
n
d
T
.
Y
u
,
“
N
e
u
r
o
L
i
f
t
i
n
g
:
n
e
u
r
a
l
i
n
f
e
r
e
n
c
e
o
n
M
a
r
k
o
v
r
a
n
d
o
m
f
i
e
l
d
s
a
t
sc
a
l
e
,
”
M
a
y
2
0
2
5
,
a
rXi
v
:
2
4
1
1
.
1
8
9
5
4
.
[
3
1
]
G
.
X
u
,
X
.
W
a
n
g
,
X
.
W
u
,
X
.
Le
n
g
,
a
n
d
Y
.
X
u
,
“
D
e
v
e
l
o
p
me
n
t
o
f
r
e
si
d
u
a
l
l
e
a
r
n
i
n
g
i
n
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
s
f
o
r
c
o
mp
u
t
e
r
v
i
s
i
o
n
:
a
su
r
v
e
y
,
”
En
g
i
n
e
e
ri
n
g
A
p
p
l
i
c
a
t
i
o
n
s
o
f
Art
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
,
v
o
l
.
1
4
2
,
F
e
b
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
n
g
a
p
p
a
i
.
2
0
2
4
.
1
0
9
8
9
0
.
[
3
2
]
P
.
Tsi
r
t
sa
k
i
s
,
G
.
Za
c
h
a
r
i
s,
G
.
S
.
M
a
r
a
sl
i
d
i
s
,
a
n
d
G
.
F
.
F
r
a
g
u
l
i
s
,
“
D
e
e
p
l
e
a
r
n
i
n
g
f
o
r
o
b
j
e
c
t
r
e
c
o
g
n
i
t
i
o
n
:
a
c
o
m
p
r
e
h
e
n
s
i
v
e
r
e
v
i
e
w
o
f
mo
d
e
l
s
a
n
d
a
l
g
o
r
i
t
h
ms,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
C
o
g
n
i
t
i
v
e
C
o
m
p
u
t
i
n
g
i
n
E
n
g
i
n
e
e
r
i
n
g
,
v
o
l
.
6
,
p
p
.
2
9
8
–
3
1
2
,
D
e
c
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
i
j
c
c
e
.
2
0
2
5
.
0
1
.
0
0
4
.
[
3
3
]
N
.
B
h
a
t
t
,
N
.
B
h
a
t
t
,
P
.
P
r
a
j
a
p
a
t
i
,
V
.
S
o
r
a
t
h
i
y
a
,
S
.
A
l
s
h
a
t
h
r
i
,
a
n
d
W
.
E
l
-
S
h
a
f
a
i
,
“
A
d
a
t
a
-
c
e
n
t
r
i
c
a
p
p
r
o
a
c
h
t
o
i
m
p
r
o
v
e
p
e
r
f
o
r
ma
n
c
e
o
f
d
e
e
p
l
e
a
r
n
i
n
g
mo
d
e
l
s,
”
S
c
i
e
n
t
i
f
i
c
R
e
p
o
rt
s
,
v
o
l
.
1
4
,
n
o
.
1
,
S
e
p
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
5
9
8
-
024
-
7
3
6
4
3
-
x.
[
3
4
]
L.
X
i
a
,
C
.
Le
e
,
a
n
d
J.
J.
L
i
,
“
S
t
a
t
i
st
i
c
a
l
me
t
h
o
d
s
c
D
EED
f
o
r
d
e
t
e
c
t
i
n
g
d
u
b
i
o
u
s
2
D
si
n
g
l
e
-
c
e
l
l
e
m
b
e
d
d
i
n
g
s
a
n
d
o
p
t
i
m
i
z
i
n
g
t
-
S
N
E
a
n
d
U
M
A
P
h
y
p
e
r
p
a
r
a
m
e
t
e
r
s,”
N
a
t
u
re
C
o
m
m
u
n
i
c
a
t
i
o
n
s
,
v
o
l
.
1
5
,
n
o
.
1
,
F
e
b
.
2
0
2
4
,
d
o
i
:
1
0
.
1
0
3
8
/
s4
1
4
6
7
-
0
2
4
-
4
5
8
9
1
-
y.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
Hyb
r
id
co
n
vo
lu
tio
n
a
l n
etw
o
r
ks,
h
id
d
en
Ma
r
ko
v
mo
d
els,
a
n
d
a
u
to
en
c
o
n
d
ers
fo
r
en
h
a
n
ce
d
…
(
Dri
s
s
N
a
ji
)
787
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Dr
iss
Na
ji
re
c
e
iv
e
d
h
is
d
o
c
t
o
r
a
t
d
e
g
re
e
in
Co
m
p
u
ter
S
c
ie
n
c
e
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
iq
u
e
,
S
u
lt
a
n
M
o
u
la
y
S
l
ima
n
e
Un
iv
e
rsit
y
,
Be
n
i
M
e
ll
a
l,
M
o
ro
c
c
o
,
i
n
2
0
2
4
.
Cu
rre
n
tl
y
,
h
e
is
e
n
r
o
ll
e
d
i
n
th
e
T
IAD
lab
o
ra
t
o
ry
a
t
th
e
sa
m
e
u
n
i
v
e
rsity
,
w
h
e
re
t
h
e
ir
re
se
a
rc
h
fo
c
u
se
s
o
n
th
e
AI
a
n
d
Io
T
.
Th
e
ir
wo
rk
a
ims
to
a
d
v
a
n
c
e
th
e
in
teg
ra
ti
o
n
o
f
AI
a
n
d
I
o
T
tec
h
n
o
l
o
g
ies
,
c
o
n
tri
b
u
ti
n
g
t
o
th
e
d
e
v
e
lo
p
m
e
n
t
o
f
in
n
o
v
a
ti
v
e
so
lu
ti
o
n
s
i
n
t
h
e
se
field
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
n
a
ji
.
d
risss
@g
m
a
il
.
c
o
m
.
K
a
m
a
l
El
h
a
tt
a
b
re
c
e
iv
e
d
a
P
h
.
D.
d
e
g
re
e
in
Co
m
p
u
ter
S
c
ien
c
e
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
,
Ch
o
u
a
i
b
Do
u
k
k
a
li
Un
iv
e
rsity
,
El
Ja
d
id
a
,
M
o
ro
c
c
o
,
in
2
0
2
4
.
C
u
rre
n
tl
y
,
h
e
is
e
n
ro
ll
e
d
in
th
e
EL
I
TE
S
lab
o
ra
t
o
r
y
a
t
t
h
e
sa
m
e
u
n
iv
e
rsit
y
,
w
h
e
re
th
e
ir
re
se
a
rc
h
fo
c
u
se
s
o
n
th
e
in
tern
e
t
o
f
t
h
in
g
s
(Io
T)
a
n
d
a
rti
fi
c
ial
in
telli
g
e
n
c
e
.
Th
e
ir
w
o
rk
a
im
s
to
a
d
v
a
n
c
e
th
e
i
n
teg
ra
ti
o
n
o
f
I
o
T
a
n
d
AI
tec
h
n
o
lo
g
ies
,
c
o
n
t
rib
u
ti
n
g
t
o
th
e
d
e
v
e
l
o
p
m
e
n
t
o
f
i
n
n
o
v
a
t
iv
e
s
o
lu
ti
o
n
s
i
n
th
e
se
field
s.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
k
a
m
a
l.
e
lh
a
tt
a
b
@g
m
a
il
.
c
o
m
.
Abd
e
la
li
J
o
u
m
a
d
re
c
e
iv
e
d
a
P
h
.
D.
d
e
g
re
e
in
M
a
th
e
m
a
ti
c
s
Ap
p
li
e
d
t
o
C
o
m
p
u
ter
S
c
ien
c
e
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
i
e
n
c
e
,
Ch
o
u
a
i
b
Do
u
k
k
a
li
Un
i
v
e
rsity
,
El
Ja
d
id
a
,
M
o
r
o
c
c
o
,
in
2
0
2
4
.
Cu
rre
n
tl
y
,
h
e
is
e
n
r
o
ll
e
d
i
n
t
h
e
LAROSE
RI
lab
o
ra
t
o
ry
a
t
t
h
e
sa
m
e
u
n
i
v
e
rsity
,
wh
e
re
th
e
ir
re
se
a
rc
h
sp
e
c
ialize
s
in
a
d
v
a
n
c
e
d
c
o
m
p
u
tatio
n
a
l
m
e
th
o
d
s
fo
r
ima
g
e
a
n
a
ly
sis,
with
a
p
a
rti
c
u
lar
fo
c
u
s
o
n
p
ro
b
a
b
il
ist
i
c
m
o
d
e
li
n
g
a
n
d
m
a
c
h
in
e
lea
rn
in
g
tec
h
n
iq
u
e
s.
His
wo
rk
e
x
p
lo
re
s
th
e
i
n
teg
ra
ti
o
n
o
f
h
id
d
e
n
M
a
r
k
o
v
m
o
d
e
ls,
d
e
e
p
lea
rn
i
n
g
a
rc
h
it
e
c
tu
re
s,
a
n
d
h
y
b
ri
d
fra
m
e
wo
rk
s
to
a
d
d
re
ss
c
o
m
p
lex
c
h
a
ll
e
n
g
e
s
i
n
ima
g
e
se
g
m
e
n
tati
o
n
a
n
d
p
a
t
tern
re
c
o
g
n
it
i
o
n
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
jo
u
m
a
d
@y
a
h
o
o
.
fr
.
Abd
e
lo
u
a
h
e
d
Ait
Ider
re
c
e
iv
e
d
a
P
h
.
D.
d
e
g
re
e
in
Co
m
p
u
ter
S
c
ien
c
e
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
s
a
n
d
Tec
h
n
o
lo
g
ies
,
S
u
l
tan
M
o
u
lay
S
li
m
a
n
e
Un
iv
e
rsity
,
Be
n
i
-
m
e
ll
a
l,
M
o
ro
c
c
o
,
in
2
0
1
8
.
P
r
o
fe
ss
o
r
o
f
Co
m
p
u
ter
S
c
ien
c
e
a
t
P
o
l
y
d
isc
ip
li
n
a
ry
F
a
c
u
l
ty
o
f
Taro
u
d
a
n
t,
Ib
n
o
u
Z
o
h
r
Un
i
v
e
rsity
,
M
o
r
o
c
c
o
.
His
re
se
a
rc
h
sp
a
n
s
a
rti
ficia
l
in
telli
g
e
n
c
e
,
m
a
c
h
in
e
lea
rn
in
g
,
d
a
ta
sc
ien
c
e
a
n
d
c
o
m
p
u
ter
v
isio
n
,
wi
th
a
fo
c
u
s
o
n
d
e
e
p
lea
rn
i
n
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
.
a
it
id
e
r@u
iz.ac
.
m
a
.
Abd
e
lk
b
ir
O
u
isa
a
d
a
n
e
is
a
d
o
c
to
r
in
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
re
se
a
rc
h
e
r
in
Artifi
c
ial
In
telli
g
e
n
c
e
a
t
S
u
lt
a
n
M
o
u
la
y
S
l
ima
n
e
Un
i
v
e
rsity
(US
M
S
)
i
n
LI
M
ATI
lab
o
ra
to
r
y
,
re
c
e
iv
e
d
a
B.
S
c
.
d
e
g
re
e
in
M
a
th
e
m
a
ti
c
s
a
n
d
Co
m
p
u
ter
S
c
ie
n
c
e
a
n
d
a
n
M
.
S
c
.
d
e
g
re
e
in
a
p
p
li
e
d
m
a
th
e
m
a
ti
c
s
fro
m
th
e
F
a
c
u
lt
y
o
f
S
c
ien
c
e
a
n
d
Tec
h
n
ics
Be
n
i
M
e
l
lal,
M
o
r
o
c
c
o
,
i
n
2
0
1
0
a
n
d
2
0
1
4
,
re
sp
e
c
ti
v
e
l
y
.
He
is
c
u
rre
n
tl
y
p
r
o
fe
ss
o
r
o
f
M
a
th
e
m
a
ti
c
s
a
n
d
re
se
a
rc
h
e
r
in
C
o
m
p
u
ter
S
c
ien
c
e
a
n
d
S
ig
n
a
l
P
ro
c
e
ss
in
g
fro
m
S
u
lt
a
n
M
o
u
la
y
S
li
m
a
n
e
Un
iv
e
rsit
y
,
M
o
r
o
c
c
o
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
sp
e
e
c
h
a
n
d
s
p
e
a
k
e
r
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
b
d
e
lk
b
ir.
o
u
isa
a
d
a
n
e
@g
m
a
il
.
c
o
m
.
Az
z
e
d
d
in
e
Idh
m
a
d
re
c
e
iv
e
d
a
P
h
.
D.
d
e
g
re
e
in
Co
m
p
u
ter
S
c
ien
c
e
fro
m
th
e
Lab
o
ra
to
r
y
o
f
An
a
ly
sis,
G
e
o
m
e
try
,
a
n
d
Ap
p
li
c
a
ti
o
n
s
F
a
c
u
lt
y
o
f
S
c
ien
c
e
s
Ib
n
To
fa
il
Un
iv
e
rsity
Ke
n
it
ra
,
M
o
r
o
c
c
o
.
His
re
se
a
rc
h
sp
a
n
s
a
rti
ficia
l
i
n
telli
g
e
n
c
e
,
m
a
c
h
in
e
lea
rn
in
g
,
d
a
ta
sc
ien
c
e
a
n
d
c
o
m
p
u
ter
v
isio
n
,
wit
h
a
fo
c
u
s
o
n
d
e
e
p
lea
rn
i
n
g
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
a
z
z
e
d
d
in
e
.
id
h
m
a
d
@u
it
.
a
c
.
m
a
.
Evaluation Warning : The document was created with Spire.PDF for Python.