I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
10
,
No
.
6
,
Decem
b
er
2
0
2
0
,
p
p
.
5
5
9
2
~
5
5
9
8
I
SS
N:
2088
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
10
i
6
.
pp
5
5
9
2
-
5
5
9
8
5592
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m/in
d
ex
.
p
h
p
/I
JE
C
E
Surv
ey
on
deep
l
e
a
rning
a
pplied to
predictive
m
a
i
nte
na
nce
Yo
us
s
ef
M
a
her
,
B
o
uje
m
a
a
Da
no
uj
De
p
a
rt
m
e
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
M
e
c
h
a
n
ica
l
En
g
i
n
e
e
rin
g
,
Un
iv
e
rsity
Ha
ss
a
n
1
,
M
o
r
o
c
c
o
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
No
v
2
5
,
2
0
1
9
R
ev
i
s
ed
Ma
y
4
,
2020
A
cc
ep
ted
Ma
y
1
8
,
2
0
2
0
P
r
o
g
n
o
sis
h
e
a
lt
h
m
o
n
it
o
ri
n
g
(P
H
M
)
p
lay
s
a
n
in
c
re
a
sin
g
l
y
i
m
p
o
rtan
t
ro
le
in
th
e
m
a
n
a
g
e
m
e
n
t
o
f
m
a
c
h
in
e
s
a
n
d
m
a
n
u
f
a
c
tu
re
d
p
ro
d
u
c
ts
in
t
o
d
a
y
’s
in
d
u
stry
,
a
n
d
d
e
e
p
lea
rn
in
g
p
lay
s
a
n
i
m
p
o
rtan
t
p
a
rt
b
y
e
sta
b
li
sh
in
g
th
e
o
p
ti
m
a
l
p
re
d
ictiv
e
m
a
in
ten
a
n
c
e
p
o
li
c
y
.
Ho
w
e
v
e
r,
trad
it
io
n
a
l
lea
rn
i
n
g
m
e
th
o
d
s
su
c
h
a
s
u
n
su
p
e
rv
ise
d
a
n
d
su
p
e
rv
ise
d
lea
rn
in
g
w
it
h
sta
n
d
a
rd
a
rc
h
it
e
c
tu
re
s
f
a
c
e
n
u
m
e
ro
u
s
p
ro
b
lem
s
w
h
e
n
e
x
p
lo
it
in
g
e
x
isti
n
g
d
a
ta.
T
h
e
re
f
o
re
,
in
th
is
e
ss
a
y
,
w
e
re
v
ie
w
th
e
sig
n
if
ic
a
n
t
imp
ro
v
e
m
e
n
ts
in
d
e
e
p
lea
rn
i
n
g
m
a
d
e
b
y
re
se
a
rc
h
e
rs
o
v
e
r
th
e
las
t
3
y
e
a
rs
in
so
lv
in
g
t
h
e
se
d
if
f
icu
lt
ies
.
We
n
o
te
th
a
t
re
se
a
rc
h
e
rs
a
re
striv
in
g
to
a
c
h
iev
e
o
p
ti
m
a
l
p
e
rf
o
r
m
a
n
c
e
in
e
stim
a
ti
n
g
th
e
re
m
a
in
in
g
u
s
e
f
u
l
li
f
e
(RU
L
)
o
f
m
a
c
h
in
e
h
e
a
lt
h
b
y
o
p
ti
m
izin
g
e
a
c
h
ste
p
f
ro
m
d
a
ta
to
p
re
d
ictiv
e
d
iag
n
o
stics
.
S
p
e
c
if
ica
ll
y
,
we
o
u
tl
in
e
th
e
c
h
a
ll
e
n
g
e
s
a
t
e
a
c
h
lev
e
l
w
it
h
th
e
ty
p
e
o
f
im
p
ro
v
e
m
e
n
t
th
a
t
h
a
s
b
e
e
n
m
a
d
e
,
a
n
d
w
e
f
e
e
l
th
a
t
th
is
is
a
n
o
p
p
o
rtu
n
it
y
to
try
to
se
lec
t
a
sta
te
-
of
-
th
e
-
a
rt
a
rc
h
it
e
c
tu
re
th
a
t
in
c
o
rp
o
ra
tes
th
e
se
c
h
a
n
g
e
s
so
e
a
c
h
re
se
a
rc
h
e
r
c
a
n
c
o
m
p
a
re
w
it
h
h
is
o
r
h
e
r
m
o
d
e
l.
In
a
d
d
it
io
n
,
p
o
st
-
RUL
re
a
so
n
in
g
a
n
d
t
h
e
u
se
o
f
d
istri
b
u
ted
c
o
m
p
u
ti
n
g
w
it
h
c
lo
u
d
tec
h
n
o
l
o
g
y
is
p
re
se
n
ted
,
w
h
ich
w
il
l
p
o
ten
ti
a
ll
y
im
p
ro
v
e
th
e
c
las
si
f
ica
ti
o
n
a
c
c
u
ra
c
y
in
m
a
in
ten
a
n
c
e
a
c
ti
v
it
ies
.
D
e
e
p
le
a
rn
in
g
w
il
l
u
n
d
o
u
b
ted
ly
p
ro
v
e
to
h
a
v
e
a
m
a
jo
r
im
p
a
c
t
in
u
p
g
ra
d
in
g
c
o
m
p
a
n
ies
a
t
th
e
lo
w
e
st co
st i
n
th
e
n
e
w
in
d
u
str
ial
re
v
o
lu
ti
o
n
,
I
n
d
u
stry
4
.
0
.
K
ey
w
o
r
d
s
:
Dee
p
lear
n
in
g
I
n
d
u
s
tr
y
4
.
0
P
r
ed
ictiv
e
m
ain
ten
a
n
ce
P
r
o
g
n
o
s
is
h
ea
l
th
m
o
n
i
to
r
in
g
R
e
m
a
in
i
n
g
u
s
ef
u
l li
f
e
es
ti
m
a
te
Co
p
y
rig
h
t
©
2
0
2
0
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Yo
u
s
s
e
f
Ma
h
er
,
Dep
ar
t
m
en
t o
f
E
lectr
ical
an
d
Me
ch
an
ical
E
n
g
i
n
ee
r
in
g
,
Un
i
v
er
s
it
y
Has
s
an
1
,
5
7
7
,
R
o
u
te
d
e
C
asa
-
,
Settat,
Mo
r
o
cc
o
.
E
m
ail:
m
a
h
er
-
y
o
u
s
s
e
f
@
h
o
t
m
a
il.c
o
m
1.
I
NT
RO
D
UCT
I
O
N
A
c
c
o
r
d
i
n
g
t
o
A
c
c
e
n
t
u
r
e
,
a
r
e
n
o
w
n
e
d
g
l
o
b
a
l
m
a
n
a
g
e
m
e
n
t
c
o
n
s
u
l
t
i
n
g
a
n
d
p
r
o
f
e
s
s
i
o
n
a
l
s
e
r
v
i
c
e
s
f
i
r
m
[
1
]
,
p
r
ed
ictiv
e
m
ai
n
te
n
an
ce
co
u
ld
s
av
e
u
p
to
1
2
%
o
n
s
ch
ed
u
led
r
ep
air
s
,
r
ed
u
ce
m
a
in
te
n
a
n
ce
c
o
s
ts
b
y
u
p
to
3
0
%
an
d
eli
m
in
ate
u
p
to
7
0
%
o
f
f
a
ilu
r
es.
Fo
r
ex
a
m
p
le,
a
s
t
u
d
y
b
y
th
e
n
atio
n
al
s
cie
n
ce
f
o
u
n
d
atio
n
(
NSF)
i
n
d
icate
s
th
at
t
h
e
ce
n
ter
f
o
r
in
te
llig
e
n
t
m
ai
n
ten
a
n
c
e
s
y
s
te
m
s
(
I
MS)
,
w
h
ic
h
is
a
lead
i
n
g
r
e
s
ea
r
ch
c
en
ter
in
t
h
e
f
ield
o
f
p
r
o
g
n
o
s
is
h
ea
lt
h
m
o
n
ito
r
in
g
(
P
HM
)
,
h
as
g
e
n
er
ated
an
ec
o
n
o
m
ic
i
m
p
ac
t
o
f
m
o
r
e
t
h
an
$
8
5
5
m
il
lio
n
t
h
r
o
u
g
h
th
e
d
ep
lo
y
m
en
t
o
f
P
HM
te
ch
n
o
lo
g
ies
to
ac
h
ie
v
e
n
ea
r
-
ze
r
o
u
n
p
la
n
n
ed
d
o
w
n
ti
m
e
a
n
d
m
o
r
e
o
p
ti
m
ized
m
ai
n
ten
a
n
ce
[
2
]
.
A
cc
o
r
d
in
g
to
t
h
e
I
n
ter
n
atio
n
al
Or
g
a
n
izatio
n
f
o
r
S
tan
d
ar
d
izatio
n
,
"
p
r
o
g
n
o
s
is
i
s
t
h
e
est
i
m
atio
n
o
f
th
e
f
ail
u
r
e
ti
m
e
a
n
d
t
h
e
r
is
k
f
o
r
o
n
e
o
r
m
o
r
e
e
x
is
ti
n
g
a
n
d
f
u
t
u
r
e
f
a
ilu
r
e
m
o
d
es"
[
3
]
.
T
h
er
e
ar
e
m
a
n
y
p
r
ed
icti
v
e
m
ai
n
ten
a
n
ce
s
tan
d
ar
d
s
,
s
u
c
h
as
th
e
m
ac
h
i
n
e
i
n
f
o
r
m
atio
n
m
an
a
g
e
m
e
n
t
allia
n
ce
(
MI
MO
S
A
)
,
w
h
ic
h
h
as
7
m
o
d
u
les
t
h
at
ar
e
u
s
ed
b
y
th
e
US
A
r
m
y
.
Fi
g
u
r
e
1
b
elo
w
d
escr
ib
e
s
th
e
d
if
f
er
en
t
p
r
o
g
n
o
s
tic
s
tep
s
in
o
r
d
er
m
o
n
ito
r
co
n
ti
n
u
o
u
s
l
y
t
h
e
s
ta
te
o
f
th
e
s
y
s
te
m
w
i
th
t
h
e
h
e
lp
o
f
th
e
d
is
tr
ib
u
ted
co
m
p
u
ti
n
g
tec
h
n
o
lo
g
y
.
D
ee
p
lear
n
in
g
is
r
ep
r
esen
ted
b
y
a
co
m
p
i
latio
n
o
f
m
ac
h
i
n
e
lear
n
in
g
al
g
o
r
ith
m
s
t
h
at
ca
n
m
o
d
el
h
i
g
h
lev
els
o
f
ab
s
tr
ac
tio
n
f
r
o
m
lar
g
e
a
m
o
u
n
ts
o
f
d
ata
t
h
r
o
u
g
h
t
h
e
u
s
e
o
f
m
u
ltil
a
y
er
ar
c
h
itect
u
r
es.
T
h
is
tech
n
iq
u
e
h
as
m
ad
e
co
n
s
id
er
ab
le
p
r
o
g
r
ess
i
n
s
ev
er
al
ar
ea
s
s
u
ch
as
i
m
ag
e
clas
s
i
f
icatio
n
,
s
p
ee
ch
r
ec
o
g
n
i
tio
n
,
i
n
s
tan
t
an
d
r
eliab
le
lan
g
u
a
g
e
tr
an
s
la
tio
n
[
4
]
an
d
ev
en
in
t
h
e
s
ea
r
c
h
f
o
r
n
e
w
ele
m
e
n
ts
i
n
p
ar
ticle
p
h
y
s
ic
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
S
u
r
ve
y
o
n
d
ee
p
lea
r
n
in
g
a
p
p
li
ed
to
p
r
ed
ictive
ma
in
ten
a
n
ce
(
Yo
u
s
s
ef
Ma
h
er
)
5593
Fig
u
r
e
1
.
P
r
o
g
n
o
s
tic
s
tep
s
[5
]
T
h
is
p
ap
er
is
s
tr
u
ctu
r
ed
as
f
o
llo
w
s
,
Sectio
n
1
d
escr
ib
es
th
e
li
m
it
s
o
f
R
U
L
esti
m
atio
n
an
d
th
e
n
e
w
ad
v
an
ce
s
i
n
d
ee
p
ar
ch
itectu
r
es
to
esti
m
ate
a
r
ea
lis
tic
R
U
L
.
Sectio
n
2
d
escr
ib
es
t
h
e
n
e
w
s
ta
te
o
f
th
e
ar
t
lear
n
in
g
ap
p
r
o
ac
h
es
a
n
d
s
h
o
w
s
n
e
w
o
p
p
o
r
tu
n
itie
s
f
o
r
u
s
i
n
g
clo
u
d
co
m
p
u
ti
n
g
i
n
d
ee
p
lear
n
in
g
.
Sec
tio
n
3
s
u
m
m
ar
izes t
h
e
co
n
cl
u
s
io
n
s
.
2.
AP
P
RO
XIM
AT
I
O
N
O
F
T
H
E
RUL B
Y
VAR
I
O
US A
R
C
H
I
T
E
CT
U
RE
S
T
h
e
r
esid
u
al
u
s
e
f
u
l
l
if
e
(
R
U
L
)
o
f
a
m
ac
h
i
n
e
o
r
co
m
p
o
n
en
t
is
t
h
e
p
r
e
d
ictio
n
o
f
th
e
ti
m
e
a
f
ter
w
h
ic
h
a
co
m
p
o
n
e
n
t
o
r
a
s
y
s
te
m
w
ill
n
o
lo
n
g
er
b
e
ab
le
to
m
ee
t
it
s
o
p
er
atio
n
al
n
ee
d
s
af
ter
t
h
e
o
b
s
er
v
atio
n
o
f
t
h
e
f
ir
s
t
f
ail
u
r
es
o
r
alar
m
s
tr
ig
g
er
ed
.
O
n
e
o
f
t
h
e
g
o
als
o
f
p
r
ed
ictiv
e
m
ai
n
ten
a
n
ce
i
s
to
o
b
tain
a
n
es
ti
m
ated
R
U
L
t
h
at
is
as
r
ea
lis
tic
as
p
o
s
s
ib
le.
Dee
p
lear
n
in
g
is
t
h
e
b
est
to
o
l
to
p
er
f
o
r
m
t
h
is
ta
s
k
.
T
h
is
s
ec
tio
n
p
r
o
v
id
es
an
o
v
er
v
ie
w
o
f
th
i
s
ar
ea
o
f
r
esear
ch
.
2
.
1
.
L
i
m
it
s
o
f
RUL est
i
m
a
t
i
o
n
by
curr
ent
m
et
ho
d
s
T
h
e
d
ec
is
io
n
m
ak
i
n
g
i
s
b
as
ed
o
n
th
e
R
U
L
co
n
f
id
en
ce
li
m
its
r
at
h
er
t
h
an
o
n
a
s
i
n
g
le
v
al
u
e.
R
U
L
p
r
ed
ictio
n
is
d
if
f
ic
u
lt d
u
e
to
s
ev
er
al
i
m
p
o
r
tan
t c
h
alle
n
g
es,
n
a
m
el
y
:
R
ea
l s
y
s
te
m
s
ar
e
co
m
p
le
x
,
an
d
th
eir
b
eh
av
io
r
s
ar
e
o
f
te
n
n
o
n
lin
ea
r
an
d
n
o
n
s
tat
io
n
ar
y
;
A
co
m
p
o
n
e
n
t
m
a
y
h
a
v
e
d
if
f
er
en
t
d
eg
r
ad
atio
n
cu
r
v
e
s
d
u
e
to
d
if
f
er
e
n
t
f
ail
u
r
e
m
o
d
es,
r
esu
lt
in
g
i
n
d
if
f
er
e
n
t
R
U
L
s
(
e.
g
.
,
b
ea
r
in
g
cr
ac
k
s
c
a
n
o
cc
u
r
in
th
e
i
n
n
er
r
i
n
g
,
i
n
t
h
e
o
u
ter
r
in
g
o
r
i
n
t
h
e
ca
g
e,
a
n
d
ea
ch
h
a
s
it
s
o
w
n
d
eg
r
ad
atio
n
cu
r
v
e)
;
T
h
e
ti
m
e
s
r
eq
u
ir
ed
to
ac
h
ie
v
e
th
e
s
a
m
e
le
v
el
o
f
d
e
g
r
ad
atio
n
b
y
m
ac
h
i
n
e
s
w
it
h
th
e
s
a
m
e
s
p
ec
if
icatio
n
s
ar
e
o
f
ten
d
i
f
f
er
e
n
t;
T
h
er
e
is
s
o
m
et
i
m
e
s
a
co
m
p
l
ex
te
m
p
o
r
al
d
ep
en
d
en
c
y
b
et
w
ee
n
s
e
n
s
o
r
s
;
f
o
r
ex
a
m
p
le,
a
ch
an
g
e
in
o
n
e
s
en
s
o
r
m
a
y
ca
u
s
e
a
ch
a
n
g
e
in
an
o
th
er
s
e
n
s
o
r
af
ter
a
d
ela
y
r
a
n
g
i
n
g
f
r
o
m
a
f
e
w
s
ec
o
n
d
s
to
h
o
u
r
s
;
T
h
e
cr
itical
ass
e
t
d
eg
r
ad
atio
n
h
i
s
to
r
y
i
s
s
o
m
eti
m
e
s
n
o
n
e
x
i
s
ten
t,
s
u
ch
as
a
co
o
lin
g
e
n
g
i
n
e
i
n
a
n
u
clea
r
p
o
w
er
p
lan
t;
in
t
h
ese
ca
s
e
s
,
m
ain
te
n
a
n
ce
co
n
s
i
s
ts
o
f
r
eg
u
lar
r
ep
lace
m
e
n
t
r
eg
ar
d
les
s
o
f
th
e
ac
tu
al
co
n
d
itio
n
s
o
f
t
h
e
as
s
ets
;
W
h
en
r
estar
ti
n
g
n
e
w
l
y
in
s
tall
ed
ass
ets,
it
w
ill
ta
k
e
a
lo
n
g
t
i
m
e
to
co
llect
p
ass
-
t
h
r
o
u
g
h
d
ata
to
ac
cu
r
atel
y
m
o
d
el
t
h
e
d
eg
r
ad
at
io
n
; a
n
d
T
h
e
ch
ar
ac
ter
is
tic
s
m
u
s
t
h
a
v
e
a
g
o
o
d
m
o
n
o
to
n
ic
ten
d
e
n
c
y
to
b
e
w
ell
co
r
r
elate
d
w
it
h
th
e
f
a
u
lt
p
r
o
p
ag
ati
o
n
p
r
o
ce
s
s
to
ac
cu
r
atel
y
ap
p
r
o
x
i
m
ate
t
h
e
R
U
L
.
I
n
co
n
tr
ast,
e
x
tr
ac
ted
en
titi
e
s
t
h
at
ten
d
to
h
a
v
e
o
n
l
y
d
r
a
m
atic
ch
an
g
es
n
ea
r
t
h
e
e
n
d
o
f
t
h
e
a
s
s
et
l
i
f
e
ar
e
n
o
t
ex
p
lo
itab
le.
F
ig
u
r
e
2
b
elo
w
s
h
o
w
s
t
h
e
d
i
f
f
er
en
ce
b
et
w
ee
n
g
o
o
d
an
d
b
ad
R
UL
p
r
ed
ictio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
6
,
Decem
b
er
2020
:
5
5
9
2
-
5
5
9
8
5594
Fig
u
r
e
2
.
C
h
ar
ac
ter
is
t
ics
o
f
R
UL
p
r
ed
ictio
n
: (
a)
id
ea
l,
(
b
)
n
o
t id
ea
l
[
6
]
2
.
2
.
So
lutio
ns
pro
v
ided by
t
he
d
ee
p
l
ea
rning
T
o
ad
d
r
ess
th
e
s
e
u
n
ce
r
tai
n
tie
s
,
p
r
ed
ictio
n
m
et
h
o
d
s
f
o
r
t
h
e
R
U
L
b
ased
o
n
d
ee
p
lear
n
i
n
g
h
av
e
b
ee
n
s
u
cc
e
s
s
f
u
ll
y
te
s
t
ed
an
d
ar
e
d
escr
ib
ed
as f
o
llo
w
s
:
2
.
2
.
1
.
Dee
p
belief
net
w
o
rk
(
DB
N)
a
nd
re
s
t
rict
ed
bo
lz
m
a
nn
m
a
chine
(
RB
M
)
I
n
[
6
]
u
s
e
an
i
m
p
r
o
v
ed
R
B
M
w
it
h
a
n
e
w
r
eg
u
lar
izatio
n
tec
h
n
iq
u
e
t
h
at
g
e
n
er
ates
c
h
ar
ac
te
r
is
tics
t
h
at
ar
e
th
en
e
m
p
lo
y
ed
a
s
R
B
M
i
n
p
u
t
d
ata.
Fin
a
ll
y
,
t
h
e
R
B
Ms
a
r
e
co
u
p
led
w
it
h
a
s
e
lf
-
o
r
g
a
n
iz
in
g
m
ap
(
SO
M)
to
i
m
p
r
o
v
e
th
e
p
r
ec
is
io
n
o
f
t
h
e
R
UL
p
r
ed
ictio
n
.
C
u
r
r
en
t
m
et
h
o
d
s
p
r
ev
en
t
f
ail
u
r
e
d
etec
tio
n
o
n
l
y
w
h
en
t
h
e
en
d
o
f
lif
e
o
f
th
e
eq
u
ip
m
e
n
t
is
v
er
y
c
lo
s
e.
I
n
[
7
]
p
r
o
p
o
s
e
a
n
ew
ap
p
r
o
ac
h
,
th
e
m
u
ltio
b
j
ec
tiv
e
DB
N
en
s
e
m
b
le,
w
h
ic
h
allo
w
s
f
o
r
a
co
m
p
r
o
m
i
s
e
b
et
w
ee
n
ac
c
u
r
ac
y
an
d
d
i
v
er
s
i
t
y
b
y
estab
li
s
h
in
g
a
n
o
v
er
al
l
m
o
d
el
f
o
r
R
U
L
esti
m
atio
n
w
i
th
o
u
ts
ta
n
d
in
g
p
er
f
o
r
m
an
ce.
2
.
2
.
2
.
Co
nv
o
lutio
na
l
neura
l
net
w
o
rk
(
CNN)
I
n
[
8
]
u
s
ed
a
n
o
v
el
d
ee
p
ar
ch
itect
u
r
e
C
N
N
-
b
ased
r
e
g
r
ess
o
r
to
esti
m
ate
th
e
R
U
L
b
y
e
m
p
lo
y
i
n
g
th
e
co
n
v
o
l
u
tio
n
an
d
p
o
o
lin
g
la
y
er
s
to
ca
p
tu
r
e
t
h
e
s
alie
n
t p
att
er
n
s
o
f
t
h
e
s
en
s
o
r
s
i
g
n
als
at
d
i
f
f
er
en
t
ti
m
e
s
ca
les,
un
i
f
y
i
n
g
th
e
m
a
n
d
f
in
a
ll
y
m
ap
p
in
g
t
h
e
m
in
to
th
e
m
o
d
el.
T
h
e
r
esu
ltin
g
R
U
L
es
ti
m
atio
n
is
e
f
f
icie
n
t
an
d
ac
cu
r
ate.
2
.
2
.
3
.
Va
ria
t
io
na
l
a
uto
-
enco
der
(
VAE)
I
n
[
9
]
d
escr
ib
ed
a
s
em
is
u
p
er
v
is
ed
lear
n
i
n
g
ap
p
r
o
ac
h
to
p
r
ed
ict
ass
et
f
ail
u
r
es
w
h
e
n
r
elati
v
el
y
litt
le
lab
el
in
f
o
r
m
atio
n
is
a
v
ailab
le.
T
h
e
ap
p
r
o
ac
h
u
s
e
s
th
e
n
o
n
lin
ea
r
e
m
b
ed
d
in
g
-
b
ased
V
A
E
as
a
d
ee
p
g
en
er
ativ
e
m
o
d
el,
w
h
ic
h
is
ea
s
y
to
tr
ain
.
T
h
e
VA
E
w
as
tr
ai
n
ed
f
o
l
lo
w
i
n
g
th
e
u
n
s
u
p
er
v
is
ed
lear
n
in
g
p
r
o
ce
s
s
w
h
ile
u
tili
zi
n
g
all
a
v
ailab
le
d
ata,
b
o
th
lab
eled
an
d
u
n
lab
eled
.
W
ith
th
is
ap
p
r
o
ac
h
,
th
e
p
r
ed
ictio
n
ac
cu
r
ac
y
w
as
v
er
y
h
ig
h
e
v
en
w
i
th
e
x
tr
e
m
el
y
li
m
i
ted
av
ailab
le
lab
el
in
f
o
r
m
atio
n
.
2
.
2
.
4
.
Rec
urre
nt
neura
l net
wo
rk
(
RNN)
I
n
[
1
0
]
p
r
esen
t
a
n
R
NN
HI
f
o
r
R
U
L
p
r
ed
ictio
n
in
a
s
et
o
f
e
x
p
er
i
m
e
n
tal
d
ata
f
o
r
ac
ce
ler
at
ed
b
ea
r
in
g
d
eg
r
ad
atio
n
an
d
SC
AD
A
d
at
a
o
f
w
i
n
d
t
u
r
b
in
e
d
eg
r
ad
atio
n
.
T
h
e
co
n
s
tr
u
ct
io
n
p
r
o
ce
s
s
w
a
s
co
m
p
o
s
ed
o
f
th
r
ee
s
tep
s
:
E
x
tr
ac
tio
n
o
f
1
4
ch
ar
ac
ter
is
tic
s
:
6
ch
ar
ac
ter
is
tic
s
o
f
r
elate
d
s
i
m
ilar
it
y
(
1
te
m
p
o
r
al
an
d
5
f
r
eq
u
en
c
y
)
a
n
d
8
ti
m
e
-
f
r
eq
u
en
c
y
f
u
n
ctio
n
s
ar
e
co
m
b
i
n
ed
to
g
et
h
er
;
Selectio
n
o
f
s
e
n
s
iti
v
e
f
ea
t
u
r
es;
m
o
n
o
to
n
ic
a
n
d
co
r
r
elatio
n
m
etr
ic
s
s
elec
t
t
h
e
m
o
s
t
s
en
s
iti
v
e
f
a
ilu
r
e
ch
ar
ac
ter
is
tic
s
; a
n
d
T
h
e
b
u
ild
in
g
o
f
t
h
e
R
NN
-
HI
:
th
e
s
elec
ted
f
ea
t
u
r
es
ar
e
m
er
g
ed
in
to
o
n
e
HI
(
R
NN
-
HI
)
v
ia
an
R
NN.
T
h
ey
u
s
ed
an
L
ST
M
n
et
w
o
r
k
to
s
o
l
v
e
t
h
e
p
r
o
b
le
m
o
f
r
ap
id
ly
in
cr
ea
s
in
g
o
r
v
a
n
i
s
h
in
g
g
r
ad
ien
t
s
a
n
d
a
d
o
u
b
le
ex
p
o
n
en
tial
m
o
d
el
to
v
alid
ate
th
e
e
f
f
ec
ti
v
en
e
s
s
o
f
t
h
e
p
r
o
p
o
s
ed
R
NN
-
HI
ap
p
r
o
ac
h
.
T
h
e
r
esu
lts
s
h
o
w
t
h
at
t
h
e
R
NN
-
HI
o
b
tai
n
s
r
elat
iv
el
y
h
i
g
h
m
o
n
o
to
n
ic
an
d
co
r
r
elatio
n
v
alu
e
s
a
n
d
be
tter
p
er
f
o
r
m
a
n
ce
in
R
U
L
p
r
ed
ictio
n
th
a
n
w
it
h
a
SOM
H
I
m
e
th
o
d
.
I
n
[
1
1
]
u
s
e
th
e
R
N
N
en
co
d
er
-
d
ec
o
d
er
(
R
NN
-
E
D)
w
it
h
u
n
s
u
p
er
v
is
ed
lear
n
i
n
g
o
n
a
s
et
o
f
air
cr
af
t
e
n
g
i
n
e
a
n
d
p
u
m
p
d
ata.
T
h
e
R
N
N
en
co
d
er
ex
tr
ac
t
s
th
e
i
m
p
o
r
ta
n
t p
atter
n
s
i
n
t
h
e
ti
m
e
s
er
ies
s
u
b
s
eq
u
e
n
ce
s
o
f
t
h
e
en
tire
o
p
er
atio
n
al
lif
e
o
f
th
e
m
ac
h
in
e
s
.
T
h
e
R
NN
d
ec
o
d
er
r
eb
u
ild
s
th
e
n
o
r
m
al
b
eh
av
io
r
,
b
u
t
it
d
o
es
n
o
t
w
o
r
k
w
ell
f
o
r
th
e
r
ec
o
n
s
tr
u
ctio
n
o
f
ab
n
o
r
m
al
b
eh
av
io
r
.
T
h
e
tr
a
j
ec
to
r
ies
o
f
2
0
en
g
in
es
ar
e
tr
u
n
ca
ted
at
f
i
v
e
lo
ca
tio
n
s
to
o
b
tain
f
iv
e
d
i
f
f
er
e
n
t
i
n
s
ta
n
c
es.
T
h
e
R
NN
th
e
n
m
ap
s
th
e
s
e
n
s
o
r
r
ea
d
in
g
s
o
n
a
h
ea
lth
i
n
d
ex
(
HI
)
tr
en
d
cu
r
v
e
to
ca
lcu
late
th
e
w
ei
g
h
ted
av
er
ag
e
o
f
t
h
e
R
U
L
ap
p
r
o
x
im
a
tio
n
s
f
r
o
m
t
h
e
f
aile
d
in
s
ta
n
ce
s
an
d
th
e
n
o
b
tai
n
s
t
h
e
late
s
t
R
U
L
e
s
ti
m
atio
n
.
T
h
e
R
NN
-
E
D
h
a
n
d
le
s
th
e
r
ea
d
in
g
o
f
n
o
i
s
y
s
en
s
o
r
s
,
m
is
s
i
n
g
d
ata
an
d
t
h
e
lack
o
f
p
r
ev
io
u
s
k
n
o
w
led
g
e
ab
o
u
t t
h
e
d
eg
r
ad
atio
n
tr
en
d
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
S
u
r
ve
y
o
n
d
ee
p
lea
r
n
in
g
a
p
p
li
ed
to
p
r
ed
ictive
ma
in
ten
a
n
ce
(
Yo
u
s
s
ef
Ma
h
er
)
5595
2
.
2
.
5
.
L
o
ng
s
ho
rt
-
t
er
m
m
e
mo
ry
(
L
ST
M
)
I
n
[
1
2
]
p
r
o
p
o
s
ed
u
s
i
n
g
th
e
L
ST
M
f
o
r
NAS
A
d
u
al
-
f
lo
w
c
o
n
d
itio
n
m
o
n
i
to
r
in
g
an
d
ac
h
i
ev
ed
h
i
g
h
p
er
f
o
r
m
a
n
ce
in
f
a
u
lt
d
iag
n
o
s
i
s
a
n
d
p
r
ed
ictio
n
u
n
d
er
co
m
p
l
ex
w
o
r
k
in
g
co
n
d
itio
n
s
,
m
ix
ed
d
ef
ec
ts
an
d
n
o
i
s
y
en
v
ir
o
n
m
e
n
t
s
.
T
h
e
s
ta
n
d
ar
d
L
ST
M
s
h
o
w
ed
i
m
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
o
v
er
t
h
e
R
NN
b
y
p
r
o
v
id
in
g
ac
c
u
r
ate
in
f
o
r
m
atio
n
ab
o
u
t
t
h
e
R
U
L
f
o
r
ea
ch
f
a
u
lt
s
i
m
u
lta
n
eo
u
s
l
y
an
d
also
ab
o
u
t
t
h
e
p
r
o
b
ab
ilit
y
o
f
o
cc
u
r
r
en
ce
o
f
d
ef
ec
ts
u
n
d
er
co
m
p
lex
o
p
er
ati
o
n
al
m
o
d
es a
n
d
m
u
ltip
le
d
eg
r
ad
atio
n
s
.
I
n
[
1
3
]
p
r
o
p
o
s
e
an
u
n
s
u
p
er
v
is
ed
tech
n
iq
u
e
f
o
r
r
ec
o
n
s
tr
u
cti
n
g
m
u
lti
v
ar
iate
ti
m
e
s
er
ies
co
r
r
esp
o
n
d
in
g
to
n
o
r
m
a
l
b
eh
av
io
r
to
o
b
tain
a
h
ea
lth
i
n
d
ex
(
HI
)
.
T
o
esti
m
ate
th
e
R
U
L
,
th
e
y
i
m
p
le
m
e
n
t
a
n
L
ST
M
en
co
d
er
-
d
ec
o
d
er
(
L
ST
M
-
E
D)
th
at
wo
r
k
s
as
f
o
llo
w
s
:
an
L
ST
M
en
co
d
er
is
u
s
ed
to
m
ap
a
m
u
ltiv
ar
ia
te
in
p
u
t
s
eq
u
en
ce
to
b
u
ild
a
f
ix
ed
-
d
i
m
en
s
io
n
a
l
v
ec
to
r
r
ep
r
esen
tati
o
n
,
an
d
t
h
e
L
ST
M
d
ec
o
d
er
th
en
u
s
es
th
i
s
v
ec
to
r
r
ep
r
esen
tatio
n
to
p
r
o
d
u
ce
th
e
tar
g
et
s
eq
u
en
ce
.
T
h
e
L
ST
M
-
ED
-
b
ased
HI
m
o
d
el
p
r
ed
icts
f
u
tu
r
e
ti
m
e
-
s
er
ie
s
,
u
s
e
s
th
e
p
r
ed
ictio
n
er
r
o
r
s
to
e
s
ti
m
ate
th
e
h
ea
lt
h
o
r
n
o
v
elt
y
o
f
a
p
o
in
t
an
d
f
i
n
all
y
u
tili
ze
s
cu
r
v
e
m
atc
h
in
g
to
esti
m
ate
th
e
R
U
L
.
3.
NE
W
S
T
A
T
E
O
F
T
H
E
ART
DE
E
P
L
E
AR
NIN
G
AP
P
RO
ACH
E
S
T
h
e
m
ai
n
ad
v
a
n
tag
e
o
f
d
ee
p
lear
n
in
g
i
s
its
f
lex
ib
ilit
y
,
w
h
ic
h
p
r
o
v
id
es
o
p
p
o
r
tu
n
itie
s
f
o
r
i
m
p
r
o
v
e
m
en
t.
E
n
h
a
n
ce
m
en
ts
b
y
u
s
i
n
g
n
e
w
lear
n
in
g
ap
p
r
o
ac
h
es,
n
e
w
t
y
p
es
o
f
ar
ch
itect
u
r
es,
an
d
n
e
w
co
m
p
u
ti
n
g
n
et
w
o
r
k
s
en
s
u
r
e
co
n
ti
n
u
o
u
s
i
m
p
r
o
v
e
m
e
n
t a
n
d
o
p
p
o
r
tu
n
ities
f
o
r
d
ee
p
lear
n
in
g
at
al
l le
v
el
s
.
3
.
1
.
T
ra
ns
f
er
lea
rning
T
h
e
tr
an
s
f
er
lear
n
in
g
ab
ilit
ie
s
th
at
ar
e
lear
n
ed
a
n
d
ac
cu
m
u
lated
d
u
r
in
g
p
r
ev
io
u
s
tas
k
s
i
m
p
r
o
v
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
o
t
h
er
n
e
u
r
al
n
et
w
o
r
k
s
w
h
en
t
h
e
y
ar
e
ap
p
lied
to
a
n
e
w
tas
k
w
it
h
a
r
ed
u
c
ed
tr
ain
in
g
d
ataset.
T
r
a
d
itio
n
al
tr
a
n
s
f
er
lear
n
i
n
g
m
et
h
o
d
s
ass
ig
n
t
h
e
f
ir
s
t
n
la
y
e
r
s
o
f
a
w
el
l
-
f
o
r
m
ed
n
et
w
o
r
k
t
o
th
e
tar
g
e
t
n
e
t
w
o
r
k
,
w
h
ile
t
h
e
last
la
y
er
s
i
n
th
e
tar
g
et
n
et
w
o
r
k
ar
e
lef
t
u
n
tr
ai
n
ed
.
T
h
ey
ar
e
th
e
n
tr
ain
ed
u
s
i
n
g
t
h
e
lear
n
i
n
g
d
ata
o
f
th
e
n
e
w
ta
s
k
.
I
n
[
1
4
]
u
s
e
in
f
r
ar
ed
th
er
m
al
i
m
a
g
e
tr
an
s
f
er
lear
n
i
n
g
in
t
h
e
d
etec
tio
n
o
f
m
ac
h
i
n
e
d
ef
ec
t
s
an
d
also
in
th
e
p
r
ed
ictio
n
o
f
th
e
o
il
lev
e
l.
T
h
e
y
u
s
e
a
m
o
d
i
f
ied
VGG
n
et
w
o
r
k
(
n
eu
r
al
n
et
w
o
r
k
cr
ea
ted
b
y
th
e
Vis
u
al
Geo
m
etr
y
Gr
o
u
p
at
Ox
f
o
r
d
Un
i
v
er
s
it
y
)
,
w
h
ic
h
is
th
e
b
es
t
n
et
w
o
r
k
f
o
r
i
m
a
g
e
d
ata.
T
h
e
s
tan
d
ar
d
VGG
is
a
v
er
y
d
ee
p
C
NN
w
i
th
1
6
la
y
er
s
w
it
h
li
n
ea
r
ac
ti
v
atio
n
f
u
n
c
tio
n
s
i
n
ea
c
h
la
y
er
e
x
ce
p
t
th
e
last
o
n
e,
w
h
ich
is
a
f
u
ll
y
co
n
n
ec
ted
la
y
er
w
it
h
a
s
o
f
t
m
ax
ac
tiv
a
tio
n
f
u
n
ct
io
n
.
T
h
e
last
la
y
er
i
s
r
ep
lace
d
b
y
a
n
e
w
f
u
ll
y
co
n
n
ec
ted
la
y
er
w
ith
a
lo
w
er
w
ei
g
h
t
an
d
f
e
w
er
class
e
s
to
ac
co
m
m
o
d
ate
lear
n
in
g
tr
an
s
f
er
.
Valu
ab
le
in
f
o
r
m
at
io
n
o
n
i
m
p
o
r
tan
t
r
eg
io
n
s
o
f
t
h
er
m
a
l
i
m
ag
e
s
ca
n
b
e
o
b
tain
ed
b
y
ap
p
l
y
in
g
Z
e
iler
'
s
m
e
th
o
d
to
th
is
n
e
w
VG
G.
T
h
is
lead
s
to
a
n
e
w
le
v
el
o
f
u
n
d
er
s
tan
d
in
g
o
f
t
h
e
p
h
y
s
ical
f
ie
ld
.
I
n
[
1
5
]
p
r
o
p
o
s
e
th
r
ee
d
ee
p
tr
a
n
s
f
er
s
tr
ateg
ie
s
(
DT
L
m
et
h
o
d
s
)
b
ased
o
n
an
SA
E
au
to
e
n
co
d
er
:
w
ei
g
h
t
tr
an
s
f
er
,
tr
an
s
f
er
lear
n
i
n
g
o
f
ch
ar
ac
ter
is
t
ics
a
n
d
w
ei
g
h
t
u
p
d
ate.
A
n
S
A
E
n
et
w
o
r
k
is
f
ir
s
t
tr
ai
n
ed
w
it
h
th
e
f
ail
u
r
e
d
ata
h
i
s
to
r
y
o
f
a
c
u
tti
n
g
to
o
l
in
a
n
o
f
f
lin
e
p
r
o
ce
s
s
,
a
n
d
th
i
s
n
et
w
o
r
k
i
s
t
h
en
e
m
p
lo
y
ed
w
it
h
a
n
e
w
to
o
l
f
o
r
an
o
n
li
n
e
R
U
L
p
r
ed
ictio
n
.
T
h
e
DT
L
o
f
f
er
s
t
h
e
p
o
s
s
ib
ilit
y
o
f
ex
tr
ac
tin
g
t
h
e
c
h
ar
ac
ter
is
tic
s
f
r
o
m
h
is
to
r
ical
f
a
u
lt
d
ata,
ad
ap
tin
g
an
d
tr
an
s
f
er
r
i
n
g
th
e
m
to
a
n
e
w
to
o
l
an
d
u
lti
m
atel
y
p
r
o
v
id
i
n
g
an
e
f
f
ec
ti
v
e
R
U
L
p
r
ed
ictio
n
w
it
h
li
m
ited
h
i
s
to
r
i
ca
l f
au
lt d
ata.
I
n
[
1
6
]
p
r
esen
t
a
tr
an
s
f
er
lear
n
in
g
ap
p
r
o
ac
h
b
ased
o
n
a
p
r
et
r
ain
ed
d
ee
p
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
w
o
r
k
th
at
is
u
s
ed
to
au
to
m
atica
l
l
y
ex
tr
ac
t
in
p
u
t
ch
ar
ac
ter
i
s
tics
.
T
h
ey
t
h
e
n
g
o
t
h
r
o
u
g
h
a
f
u
ll
y
co
n
n
ec
ted
s
tep
to
class
i
f
y
th
e
c
h
ar
ac
ter
is
tic
s
o
b
tain
ed
u
s
i
n
g
ex
p
er
i
m
en
tal
d
ata
co
m
p
o
s
ed
o
f
g
ea
r
d
ef
ec
t
s
.
T
h
e
d
ee
p
C
NN
n
et
w
o
r
k
ad
o
p
ted
as
th
e
b
asic
ar
ch
itectu
r
e
f
o
r
A
le
x
n
et
(
f
i
v
e
co
n
v
o
lu
t
io
n
al
la
y
er
s
an
d
t
h
r
ee
f
u
ll
y
co
n
n
ec
ted
la
y
er
s
)
i
n
t
h
is
s
t
u
d
y
w
as
o
r
ig
i
n
all
y
p
r
o
p
o
s
ed
b
y
[
1
7
]
f
o
r
o
b
j
ec
t
r
ec
o
g
n
itio
n
(
s
o
u
r
ce
d
o
m
a
in
)
w
it
h
I
m
ag
e
Net
I
L
SV
R
C
-
2
0
1
2
(
1
0
0
0
o
b
j
ec
t
class
es
an
d
>
1
m
il
lio
n
i
m
a
g
es).
T
h
e
class
i
f
icatio
n
ac
c
u
r
ac
y
o
f
th
e
p
r
o
p
o
s
ed
ap
p
r
o
ac
h
[
1
8
]
ex
ce
ed
s
th
o
s
e
o
f
o
th
er
m
eth
o
d
s
,
s
u
ch
as
t
h
e
lo
ca
ll
y
f
o
r
m
ed
co
n
v
o
l
u
tio
n
al
n
eu
r
al
n
e
t
w
o
r
k
an
d
SVM.
T
h
e
p
r
ec
is
io
n
o
b
tain
ed
in
d
icate
s
t
h
at
t
h
i
s
ap
p
r
o
ac
h
i
s
n
o
t
o
n
l
y
r
o
b
u
s
t
b
u
t
ca
n
also
b
e
ap
p
lied
to
f
a
u
lt
d
iag
n
o
s
i
s
i
n
o
th
er
s
y
s
te
m
s
.
I
n
[
1
9
]
p
r
o
p
o
s
e
an
u
n
s
u
p
er
v
is
ed
d
o
m
ain
ad
ap
tatio
n
th
a
t
d
o
e
s
n
o
t
r
eq
u
ir
e
an
y
lab
el
in
f
o
r
m
atio
n
f
r
o
m
t
h
e
tar
g
et
d
o
m
ai
n
b
y
o
n
l
y
m
o
d
if
y
i
n
g
t
h
e
s
tati
s
tics
o
f
t
h
e
B
N
lay
er
.
T
h
is
A
d
aB
N
m
o
d
el
g
iv
e
s
h
i
g
h
-
le
v
el
g
en
er
aliza
tio
n
ab
il
it
y
to
t
h
e
D
NN
b
y
tr
an
s
f
er
r
i
n
g
lear
n
ed
f
e
atu
r
es
f
r
o
m
a
s
o
u
r
ce
d
o
m
ain
t
o
th
e
tar
g
et
d
o
m
ai
n
w
it
h
o
u
t
f
i
n
e
-
tu
n
i
n
g
o
r
ad
d
itio
n
al
co
m
p
o
n
e
n
t
s
.
Go
o
d
f
ello
w
I
,
B
en
g
io
Y
a
n
d
C
o
u
r
v
ill
e
A
[
2
0
]
d
escr
ib
e
a
co
m
p
r
o
m
i
s
e
b
et
w
ee
n
'
b
i
as
an
d
v
ar
ia
n
ce
'
w
it
h
b
en
ef
icial
g
en
er
a
liza
tio
n
p
r
o
p
er
ties
.
Ma
x
i
m
izi
n
g
th
e
p
r
ed
ictio
n
in
v
o
lv
e
s
o
p
ti
m
i
zin
g
t
h
e
m
o
d
el
b
y
i
n
co
r
p
o
r
atin
g
e
x
ter
n
a
l in
f
o
r
m
at
io
n
.
3
.
2
.
M
ultim
o
da
l
lea
rni
ng
I
n
co
n
n
ec
tio
n
w
it
h
tr
a
n
s
f
er
lear
n
i
n
g
,
m
u
lt
i
m
o
d
al
lear
n
i
n
g
ca
n
id
en
ti
f
y
f
ea
t
u
r
es
t
h
a
t
d
escr
ib
e
co
m
m
o
n
co
n
ce
p
ts
f
r
o
m
d
if
f
er
en
t in
p
u
t t
y
p
e
s
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
6
,
Decem
b
er
2020
:
5
5
9
2
-
5
5
9
8
5596
W
ith
lar
g
e
u
n
lab
eled
d
ata,
d
ee
p
g
r
ap
h
ical
m
o
d
els
s
u
c
h
as D
B
Ns ca
n
b
e
p
r
etr
ain
ed
in
an
u
n
s
u
p
er
v
is
ed
w
a
y
an
d
t
h
en
ad
j
u
s
ted
to
a
s
m
a
lle
r
n
u
m
b
er
o
f
lab
eled
d
ata
b
ec
a
u
s
e
th
e
y
lear
n
a
j
o
in
t
p
r
o
b
ab
i
lit
y
d
is
tr
ib
u
tio
n
f
r
o
m
t
h
e
in
p
u
ts
; a
n
d
C
NNs
ar
e
m
o
s
tl
y
u
s
ed
w
it
h
lab
eled
d
ata
b
ec
au
s
e
th
e
y
ca
n
b
e
tr
ain
ed
f
r
o
m
en
d
to
en
d
w
it
h
a
b
ac
k
p
r
o
p
ag
atio
n
f
u
n
ctio
n
an
d
d
em
o
n
s
tr
ate
p
ea
k
p
er
f
o
r
m
a
n
ce
in
m
a
n
y
d
i
v
er
s
e
ta
s
k
s
.
I
n
[
2
1
]
p
r
o
p
o
s
e
a
n
o
v
el
m
u
lti
m
o
d
e
f
a
u
lt
cla
s
s
i
f
icat
io
n
m
et
h
o
d
b
ased
o
n
DNN
t
h
at
r
eso
lv
es
th
e
p
r
o
b
lem
o
f
t
h
e
lo
ad
,
m
o
d
e
an
d
ch
an
g
in
g
en
v
ir
o
n
m
e
n
t
in
w
h
ic
h
t
h
e
m
ac
h
i
n
er
y
eq
u
ip
m
e
n
t
o
p
er
ates.
I
t
is
a
h
ier
ar
ch
ical
DNN
m
o
d
el
co
m
p
o
s
ed
o
f
th
r
ee
p
ar
ts
:
th
e
f
ir
s
t
h
ier
ar
c
h
y
u
s
ed
f
o
r
th
e
m
o
d
e
p
ar
titi
o
n
,
th
e
s
ec
o
n
d
co
m
p
r
is
i
n
g
a
s
et
o
f
DNNs,
w
h
ic
h
is
d
ev
i
s
ed
to
ex
tr
ac
t
f
ea
t
u
r
es
o
f
d
i
f
f
er
e
n
t
m
o
d
es
s
ep
ar
ately
a
n
d
d
iag
n
o
s
e
t
h
e
f
au
lt
s
o
u
r
ce
,
a
n
d
th
e
th
ir
d
,
w
h
ic
h
co
n
s
i
s
ts
o
f
a
n
o
th
er
s
et
of
DNN
s
d
esi
g
n
ed
to
d
is
tin
g
u
is
h
th
e
s
ev
er
it
y
o
f
th
e
f
au
lt
in
a
g
iv
e
n
m
o
d
e.
T
h
is
ap
p
r
o
ac
h
allo
w
s
f
o
r
m
o
d
e
p
ar
titi
o
n
in
g
,
w
h
ich
h
elp
s
in
th
e
p
r
ed
ictiv
e
m
a
in
te
n
a
n
ce
o
f
m
ac
h
in
er
y
an
d
eq
u
ip
m
en
t.
3
.
3
.
M
ultit
a
s
k
lea
rning
Mu
ltit
a
s
k
lear
n
i
n
g
is
co
m
p
le
m
en
tar
y
to
m
u
lt
i
m
o
d
al
a
n
d
tr
an
s
f
er
lear
n
i
n
g
.
I
n
t
h
e
m
o
v
ie
Kar
ate
Kid
(
1
9
8
4
)
,
Mr
.
Miy
a
g
i
teac
h
es
t
h
e
ch
ild
to
d
o
th
in
g
s
th
at
ar
e
n
o
t
r
elate
d
to
k
ar
ate,
s
u
ch
as
s
a
n
d
in
g
th
e
f
lo
o
r
an
d
w
a
x
i
n
g
a
ca
r
.
I
n
h
i
n
d
s
i
g
h
t,
th
e
s
e
k
in
d
s
o
f
ta
s
k
s
p
r
o
v
e
to
b
e
i
n
v
al
u
ab
le
f
o
r
ac
q
u
ir
i
n
g
k
ar
a
te
s
k
ills
.
T
h
e
p
u
r
p
o
s
e
o
f
an
au
x
iliar
y
ta
s
k
i
n
MT
L
is
to
allo
w
t
h
e
m
o
d
el
to
lear
n
r
ep
r
esen
tatio
n
s
th
at
ar
e
s
h
ar
ed
o
r
u
s
ef
u
l
f
o
r
th
e
m
ain
tas
k
[
2
2
]
.
T
h
e
y
u
s
e
w
h
at
th
e
y
ca
ll
cr
o
s
s
-
s
titc
h
u
n
it
s
to
allo
w
t
h
e
m
o
d
el
to
d
eter
m
i
n
e
h
o
w
task
-
s
p
ec
i
f
ic
n
et
w
o
r
k
s
ex
p
lo
it
k
n
o
w
led
g
e
o
f
an
o
t
h
er
tas
k
b
y
lear
n
i
n
g
a
li
n
ea
r
co
m
b
in
at
i
o
n
o
f
t
h
e
p
r
ev
io
u
s
la
y
er
o
u
tp
u
t
s
.
I
n
[
2
3
]
co
n
clu
d
e
th
at
in
m
u
lt
i
task
lear
n
i
n
g
,
s
u
cc
ess
d
ep
en
d
s
lar
g
el
y
o
n
"
th
e
s
i
m
ilar
it
y
o
f
th
e
s
o
u
r
ce
s
e
m
a
n
tic
s
a
n
d
tar
g
et
d
ata
s
ets.
”
W
h
en
o
n
l
y
s
m
all
a
m
o
u
n
ts
o
f
tar
g
et
d
ata
ar
e
a
v
ailab
le,
t
h
e
s
h
ar
i
n
g
o
f
co
n
cr
ete
p
ar
am
eter
s
ca
n
b
e
co
n
s
id
er
ed
as lea
r
n
in
g
w
ith
a
m
ea
n
co
n
s
tr
ain
t,
i
n
w
h
ic
h
p
ar
ts
o
f
all
m
o
d
els (
u
s
u
all
y
h
id
d
en
la
y
er
s
)
ar
e
f
o
r
ce
d
to
b
e
th
e
s
a
m
e
as
t
h
e
av
er
a
g
e.
T
h
e
m
u
l
t
itas
k
lear
n
i
n
g
ar
ch
itectu
r
e
u
s
e
d
i
s
a
b
id
ir
ec
tio
n
al
L
ST
M
th
at
co
n
s
is
t
s
o
f
a
s
i
n
g
le
h
id
d
en
la
y
er
o
f
1
0
0
d
i
m
en
s
io
n
s
s
h
ar
ed
b
et
w
ee
n
1
0
w
o
r
d
-
p
r
o
ce
s
s
in
g
tas
k
s
.
3
.
4
.
Dee
p lea
rning
o
pti
m
iza
t
io
n by
clo
ud
co
m
pu
t
ing
C
lo
u
d
co
m
p
u
tin
g
co
u
ld
p
la
y
a
k
e
y
r
o
le
in
s
ca
lin
g
u
p
d
ee
p
le
ar
n
in
g
.
Ho
w
ev
er
,
th
e
m
a
i
n
w
e
ak
n
e
s
s
i
s
th
at
r
ec
o
r
d
in
g
t
h
e
d
ata
f
r
o
m
an
i
n
cr
ea
s
i
n
g
n
u
m
b
er
o
f
p
iec
es
o
f
eq
u
ip
m
en
t
a
n
d
s
e
n
d
in
g
th
e
d
ata
d
ir
ec
tl
y
to
th
e
clo
u
d
cr
ea
tes
a
p
r
o
b
le
m
o
f
s
y
s
te
m
o
v
er
lo
ad
,
in
d
u
ci
n
g
p
r
o
b
lem
s
w
it
h
s
p
ee
d
,
co
s
t
an
d
s
ec
u
r
it
y
.
O
n
e
w
a
y
to
ad
d
r
ess
th
is
w
ea
k
n
ess
i
s
f
o
g
c
o
m
p
u
ti
n
g
tec
h
n
o
lo
g
y
.
Fo
g
co
m
p
u
ti
n
g
is
a
d
ec
en
tr
ali
ze
d
co
m
p
u
ti
n
g
in
f
r
astr
u
ctu
r
e
i
n
w
h
ic
h
d
ata,
co
m
p
u
tatio
n
,
s
t
o
r
ag
e
an
d
ap
p
licatio
n
s
ar
e
d
is
tr
ib
u
ted
in
th
e
m
o
s
t
e
f
f
icie
n
t
an
d
lo
g
ic
al
p
lace
b
etw
ee
n
t
h
e
d
ata
s
o
u
r
ce
an
d
th
e
clo
u
d
.
T
h
is
tech
n
o
lo
g
y
r
ed
u
ce
s
t
h
e
r
eso
u
r
ce
s
al
lo
ca
ted
an
d
k
ee
p
s
th
e
b
an
d
w
id
th
n
et
w
o
r
k
at
n
o
r
m
al
o
p
er
atin
g
co
n
d
itio
n
s
,
e
s
p
ec
iall
y
f
o
r
d
ee
p
lear
n
in
g
.
I
n
f
ac
t,
t
h
e
i
n
f
er
e
n
ce
is
n
o
t
d
o
n
e
at
th
e
clo
u
d
l
ev
el
b
u
t
at
t
h
e
f
o
g
lev
el
w
it
h
s
u
f
f
icie
n
t
m
e
m
o
r
y
an
d
ca
lcu
latio
n
r
eso
u
r
ce
s
,
an
d
th
e
f
o
g
l
e
v
el
w
ill
o
n
l
y
tr
an
s
f
e
r
th
e
n
ec
ess
ar
y
d
ata
to
th
e
clo
u
d
,
s
u
c
h
as
t
h
e
w
eig
h
t
u
p
d
ates
a
n
d
th
e
ex
tr
ac
ted
f
ea
tu
r
es.
T
h
e
r
esear
ch
er
s
in
[
2
4
]
p
r
o
p
o
s
ed
a
n
e
w
an
d
ef
f
icien
t
w
a
y
to
ex
p
lo
it
d
ee
p
lear
n
in
g
b
y
u
s
i
n
g
f
o
g
-
clo
u
d
co
m
p
u
ti
n
g
a
n
d
th
e
ca
p
ab
ilit
ies
o
f
in
telli
g
e
n
t
d
ev
ices
(
ed
g
e
d
ev
ices)
.
Fi
g
u
r
e
3
b
elo
w
d
escr
ib
es
th
e
f
o
g
co
m
p
u
ti
n
g
ar
c
h
itect
u
r
e
co
n
n
e
ctin
g
th
e
d
if
f
er
e
n
t
n
et
w
o
r
k
le
v
els
a
n
d
d
is
p
la
y
s
th
e
co
m
p
u
tatio
n
,
s
to
r
ag
e
an
d
lo
ca
l
co
m
m
u
n
icatio
n
b
etw
ee
n
ed
g
e
d
ev
ice
s
.
Fig
u
r
e
4
b
elo
w
s
h
o
w
s
th
e
ad
v
an
tag
e
s
o
f
u
s
i
n
g
th
e
f
o
g
co
m
p
u
ti
n
g
i
n
ter
m
s
o
f
s
ec
u
r
it
y
,
en
er
g
y
a
n
d
s
p
ee
d
.
Fig
u
r
e
3
.
T
h
e
clo
u
d
-
f
o
g
tec
h
n
o
lo
g
y
an
d
t
h
e
in
ter
ac
tio
n
s
b
et
w
ee
n
lev
e
ls
[
2
5
]
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2088
-
8708
S
u
r
ve
y
o
n
d
ee
p
lea
r
n
in
g
a
p
p
li
ed
to
p
r
ed
ictive
ma
in
ten
a
n
ce
(
Yo
u
s
s
ef
Ma
h
er
)
5597
Fig
u
r
e
4
.
Op
ti
m
izi
n
g
t
h
e
s
e
n
s
o
r
n
et
w
o
r
k
w
i
th
f
o
g
co
m
p
u
ti
n
g
[
2
6
]
T
h
e
au
th
o
r
s
u
s
ed
a
h
y
b
r
id
ap
p
r
o
ac
h
th
at
co
m
b
i
n
es
t
h
e
co
m
p
r
e
s
s
io
n
p
r
o
p
er
ties
o
f
n
e
u
r
al
n
et
w
o
r
k
s
s
u
c
h
as
C
NN,
R
NN
a
n
d
L
ST
M
an
d
t
h
e
i
n
co
r
p
o
r
atio
n
o
f
f
o
g
co
m
p
u
ti
n
g
m
ad
e
p
o
s
s
ib
le
b
y
t
h
e
i
n
te
g
r
atio
n
o
f
th
e
i
n
cr
ea
s
i
n
g
n
u
m
b
er
s
o
f
C
P
Us
an
d
GP
Us
i
n
d
ev
ices.
T
h
e
u
s
e
o
f
th
e
f
o
g
i
n
s
u
p
p
o
r
t
o
f
d
ee
p
lear
n
in
g
i
m
p
r
o
v
es
t
h
e
u
s
er
ex
p
er
ie
n
ce
an
d
o
p
ti
m
izes
t
h
e
e
x
p
lo
itatio
n
o
f
t
h
e
clo
u
d
.
T
h
e
f
o
g
e
m
b
e
d
d
ed
in
th
e
s
y
s
te
m
w
il
l
ab
s
o
r
b
th
e
r
ea
l
-
ti
m
e
d
at
a
f
r
o
m
t
h
e
s
e
n
s
o
r
s
an
d
ac
t
q
u
ick
l
y
th
r
o
u
g
h
t
h
e
ac
tu
ato
r
s
.
T
h
is
tec
h
n
iq
u
e
i
s
ap
p
lica
b
le
to
th
e
f
ield
o
f
h
ea
lth
m
ac
h
in
e
p
r
o
g
n
o
s
i
s
u
s
i
n
g
th
e
C
NN
f
o
r
i
m
a
g
e
cla
s
s
i
f
ic
atio
n
o
f
th
e
h
ea
lt
h
co
n
d
itio
n
o
f
m
ac
h
in
e
s
b
y
i
n
c
r
ea
s
in
g
th
e
p
r
ec
is
io
n
b
y
5
%
c
o
m
p
ar
ed
to
tr
ad
itio
n
al
m
eth
o
d
s
an
d
s
av
in
g
d
ata,
en
er
g
y
an
d
tr
af
f
ic,
w
h
ic
h
m
a
k
es it r
o
b
u
s
t f
o
r
ap
p
lica
tio
n
s
i
n
th
e
n
e
w
i
n
ter
n
et
o
f
t
h
i
n
g
s
(
I
o
T
)
in
d
u
s
tr
y
.
4.
CO
NCLU
SI
O
N
T
o
q
u
ick
l
y
i
n
te
g
r
ate
th
e
I
n
d
u
s
tr
y
4
.
0
r
ev
o
lu
tio
n
,
t
h
e
P
HM
s
y
s
te
m
n
ee
d
s
to
ad
o
p
t
estab
lis
h
ed
a
n
d
h
ig
h
-
p
er
f
o
r
m
a
n
ce
p
r
ed
ictiv
e
m
ai
n
ten
a
n
ce
s
tan
d
ar
d
s
,
s
u
c
h
as
th
e
US
A
r
m
y
's
MI
MO
S
A
n
o
r
m
,
to
e
n
s
u
r
e
o
p
tim
a
l
m
ac
h
in
e
m
o
n
ito
r
in
g
an
d
h
ig
h
-
q
u
alit
y
m
a
n
u
f
ac
t
u
r
e
d
p
r
o
d
u
cts.
T
h
is
w
ill
en
s
u
r
e
co
m
p
lete
m
ac
h
i
n
e
m
o
n
ito
r
i
n
g
t
h
r
o
u
g
h
w
e
ll
-
s
tr
u
ctu
r
ed
co
m
m
u
n
icatio
n
s
y
s
te
m
s
a
n
d
p
r
o
to
co
ls
,
g
iv
in
g
an
ed
g
e
in
th
e
h
i
g
h
l
y
co
m
p
eti
tiv
e
ec
o
n
o
m
ic
en
v
ir
o
n
m
en
t.
W
ith
th
e
d
e
v
elo
p
m
e
n
t
o
f
d
ee
p
lear
n
in
g
,
m
ai
n
te
n
an
ce
w
i
ll
b
ec
o
m
e
m
o
r
e
ef
f
icien
t
an
d
m
o
r
e
r
ea
ctiv
e
in
th
e
n
ea
r
f
u
t
u
r
e
an
d
f
it
i
n
to
t
h
e
v
is
io
n
o
f
I
n
d
u
s
tr
y
4
.
0
b
ec
au
s
e
d
ee
p
lear
n
i
n
g
,
w
it
h
its
m
u
ltip
le
ad
v
an
ce
d
ar
ch
itect
u
r
es,
eli
m
in
ate
s
th
e
co
m
p
lex
it
y
o
f
d
ata
an
al
y
s
i
s
an
d
th
e
n
ee
d
f
o
r
ex
p
er
tis
e
an
d
ex
ce
s
s
i
v
e
w
o
r
k
.
T
h
at
s
ai
d
,
d
ev
elo
p
in
g
a
s
o
-
ca
l
led
"
in
tellig
e
n
t"
s
y
s
te
m
i
n
r
ea
l
ti
m
e
i
s
ess
e
n
tia
l
to
m
ak
e
co
m
p
a
n
ies
in
ter
ested
b
y
c
h
o
o
s
in
g
to
f
r
a
m
e
t
h
e
s
tu
d
ied
s
y
s
te
m
ac
co
r
d
in
g
to
co
s
ts
an
d
b
en
e
f
its
b
ec
au
s
e
th
e
f
u
ll
p
o
ten
tial o
f
D
L
h
a
s
n
o
t
y
et
b
ee
n
ex
p
lo
r
ed
.
RE
F
E
R
E
NC
E
S
[1
]
P
.
Da
u
g
h
e
rty
,
e
t
a
l.
,
"
Driv
in
g
Un
c
o
n
v
e
n
ti
o
n
a
l
G
ro
w
th
th
r
o
u
g
h
th
e
I
n
d
u
strial
I
n
tern
e
t
o
f
T
h
in
g
s
,
"
Acc
e
n
tu
re
T
e
c
h
n
o
l
o
g
y
,
2
0
1
5
.
[2
]
G
ra
y
,
D.,
Riv
e
rs,
a
n
d
G
.
V
e
r
m
o
n
t,
"
M
e
a
su
rin
g
th
e
Eco
n
o
m
ic
I
m
p
a
c
ts
o
f
th
e
NSF
In
d
u
stry
/Un
iv
e
rs
it
y
Co
o
p
e
ra
ti
v
e
Re
se
a
rc
h
Ce
n
ters
P
ro
g
ra
m
:
A
F
e
a
sib
il
it
y
S
tu
d
y
,
"
F
in
a
l
Re
p
o
rt,
In
d
u
stry
-
Un
ive
rs
it
y
Co
o
p
e
ra
ti
v
e
Res
e
a
rc
h
Ce
n
ter
Pro
g
ra
m N
a
ti
o
n
a
l
S
c
ien
c
e
Fo
u
n
d
a
ti
o
n
Arli
n
g
to
n
,
2
0
1
9
.
[3
]
T
o
b
o
n
-
M
e
ji
a
,
D.A
.,
K.
M
e
d
jah
e
r,
a
n
d
N.
Zerh
o
u
n
i,
"
T
h
e
IS
O
1
3
3
8
1
-
1
sta
n
d
a
rd
'
s
f
a
il
u
re
p
ro
g
n
o
st
ics
p
r
o
c
e
ss
th
ro
u
g
h
a
n
e
x
a
m
p
le,"
in
2
0
1
0
Pro
g
n
o
st
ics
a
n
d
S
y
ste
m He
a
lt
h
M
a
n
a
g
e
me
n
t
Co
n
fer
e
n
c
e
,
2
0
1
0
.
[4
]
W
u
,
Y.,
e
t
a
l.
,
"
G
o
o
g
le'
s
N
e
u
ra
l
M
a
c
h
in
e
T
ra
n
sla
ti
o
n
S
y
ste
m
:
B
rid
g
in
g
th
e
G
a
p
b
e
t
we
e
n
Hu
m
a
n
a
n
d
M
a
c
h
i
n
e
T
ra
n
sla
ti
o
n
,"
a
rXiv p
re
p
ri
n
t
a
rXi
v
:1
6
0
9
.
0
8
1
4
4
,
2
0
1
6
.
[5
]
L
e
b
o
ld
,
M
.
a
n
d
K.
Re
ich
a
rd
,
"
OSA
-
CBM
A
rc
h
it
e
c
tu
re
De
v
e
lo
p
m
e
n
t
w
it
h
E
m
p
h
a
sis
o
n
XML
Im
p
le
m
e
n
tatio
n
s,"
M
a
in
ten
a
n
c
e
An
d
Reli
a
b
i
li
ty Co
n
fer
e
n
c
e
,
2
0
0
2
.
[6
]
L
iao
,
L
.
,
W
.
Jin
,
a
n
d
R.
P
a
v
e
l,
"
En
h
a
n
c
e
d
Re
stricte
d
B
o
lt
z
m
a
n
n
M
a
c
h
in
e
W
it
h
P
r
o
g
n
o
sa
b
i
li
ty
Re
g
u
lariz
a
ti
o
n
f
o
r
Pr
o
g
n
o
s
t
i
c
s
a
n
d
H
e
a
l
t
h
A
s
s
e
s
s
m
e
n
t
,
"
I
E
E
E
T
r
a
n
s
a
c
t
i
o
n
s
o
n
I
n
d
u
s
t
r
i
a
l
E
l
e
c
t
r
o
n
i
c
s
,
v
o
l
.
63
,
n
o
.
1
1
,
p
p
.
7076
-
7083
,
2
0
1
6
.
[7
]
Zh
a
n
g
,
C.
,
e
t
a
l.
,
M
u
l
ti
o
b
jec
ti
v
e
De
e
p
Be
li
e
f
Ne
t
w
o
rk
s
En
se
m
b
le
f
o
r
Re
m
a
in
in
g
Us
e
f
u
l
L
i
fe
Esti
m
a
ti
o
n
i
n
P
r
o
g
n
o
stics
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ne
u
r
a
l
Ne
two
rk
s
a
n
d
L
e
a
r
n
in
g
S
y
ste
ms
,
v
o
l.
28
,
n
o
.
1
0
,
p
p
.
2
3
0
6
-
2
3
1
8
,
2
0
1
7
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
10
,
No
.
6
,
Decem
b
er
2020
:
5
5
9
2
-
5
5
9
8
5598
[8
]
S
a
tee
sh
Ba
b
u
,
G
.
,
P
.
Zh
a
o
,
a
n
d
X
.
-
L
.
L
i
,
"
De
e
p
Co
n
v
o
lu
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
Ba
se
d
Re
g
re
ss
i
o
n
A
p
p
r
o
a
c
h
f
o
r
Esti
m
a
ti
o
n
o
f
R
e
m
a
in
in
g
Us
e
f
u
l
L
ife
,
"
Da
ta
b
a
se
S
y
ste
ms
fo
r
Ad
v
a
n
c
e
d
A
p
p
li
c
a
ti
o
n
s
,
C
h
a
m:
S
p
ri
n
g
e
r
In
ter
n
a
t
io
n
a
l
Pu
b
li
sh
i
n
g
,
p
p
.
2
1
4
-
2
2
8
,
2
0
1
6
.
[9
]
S
.
Yo
o
n
,
A
.
,
e
t
a
l.
,
"
S
e
m
i
-
su
p
e
rv
ise
d
L
e
a
rn
in
g
w
it
h
De
e
p
G
e
n
e
ra
ti
v
e
M
o
d
e
ls
f
o
r
A
s
se
t
F
a
il
u
re
P
re
d
icti
on
,"
a
rXiv p
re
p
ri
n
t
a
rX
iv:1
7
0
9
.
0
0
8
4
5
,
2
0
1
7
.
[1
0
]
G
u
o
,
L
.
,
e
t
a
l.
,
"
A
re
c
u
rre
n
t
n
e
u
r
a
l
n
e
tw
o
rk
b
a
se
d
h
e
a
lt
h
i
n
d
ica
to
r
f
o
r
re
m
a
in
in
g
u
se
f
u
l
li
f
e
p
re
d
ictio
n
o
f
b
e
a
rin
g
s,"
Ne
u
ro
c
o
mp
u
ti
n
g
,
v
o
l
.
2
4
0
,
p
p
.
9
8
-
1
0
9
,
2
0
1
7
.
[1
1
]
G
u
g
u
lo
th
u
,
N.,
"
P
re
d
ictin
g
Re
m
a
in
in
g
Us
e
f
u
l
L
if
e
u
sin
g
T
i
m
e
S
e
ries
Em
b
e
d
d
in
g
s
b
a
se
d
o
n
Re
c
u
rre
n
t
Ne
u
ra
l
Ne
tw
o
rk
s
,
"
a
rXiv p
re
p
rin
t
a
rXiv:
1
7
0
9
.
0
1
0
7
3
,
2
0
1
7
.
[1
2
]
Yu
a
n
,
M
.
,
Y.
W
u
,
a
n
d
L
.
L
in
,
"
F
a
u
lt
d
iag
n
o
sis
a
n
d
re
m
a
in
in
g
u
se
f
u
l
li
f
e
e
sti
m
a
ti
o
n
o
f
a
e
ro
e
n
g
in
e
u
sin
g
L
S
T
M
n
e
u
ra
l
n
e
tw
o
rk
,
"
IEE
E
In
ter
n
a
ti
o
n
a
l
C
o
n
fe
re
n
c
e
o
n
A
irc
ra
ft
Util
i
ty S
y
ste
ms
(
AUS
)
,
Ch
in
a
,
p
p
.
1
3
5
-
1
4
0
,
2
0
1
6
.
[1
3
]
M
a
lh
o
tra,
P
.
,
e
t
a
l.
,
"
M
u
lt
i
-
S
e
n
s
o
r
P
ro
g
n
o
stics
u
sin
g
a
n
Un
su
p
e
rv
ise
d
He
a
lt
h
In
d
e
x
b
a
se
d
o
n
LS
T
M
En
c
o
d
e
r
-
De
c
o
d
e
r
,"
a
rXiv p
re
p
rin
t
a
rX
iv:1
6
0
8
.
0
6
1
5
4
,
2
0
1
6
.
[1
4
]
Ja
n
ss
e
n
s,
O.,
e
t
a
l.
,
"
De
e
p
Lea
rn
in
g
f
o
r
In
f
ra
re
d
T
h
e
r
m
a
l
I
m
a
g
e
B
a
se
d
M
a
c
h
in
e
He
a
lt
h
M
o
n
it
o
ri
n
g
,
"
IEE
E/
AS
M
E
T
ra
n
sa
c
ti
o
n
s
o
n
M
e
c
h
a
tro
n
ics
,
v
o
l.
23
,
n
o
.
1
,
p
p
.
1
5
1
-
1
5
9
,
2
0
1
8
.
[1
5
]
S
u
n
,
C.
,
e
t
a
l.
,
''
De
e
p
T
ra
n
s
f
e
r
Lea
rn
in
g
Ba
se
d
o
n
S
p
a
rse
A
u
to
-
e
n
c
o
d
e
r
f
o
r
Re
m
a
in
in
g
Us
e
f
u
l
L
i
fe
P
re
d
ictio
n
o
f
T
o
o
l
in
M
a
n
u
f
a
c
tu
rin
g
,
"
IEE
E
T
r
a
n
sa
c
ti
o
n
s
o
n
In
d
u
stri
a
l
I
n
fo
rm
a
t
ics
,
p
p
.
1
-
1
,
2
0
1
8
.
[1
6
]
Ca
o
,
P
.
,
S
.
Z
h
a
n
g
,
a
n
d
J.
T
a
n
g
,
"
P
re
p
r
o
c
e
ss
in
g
-
F
re
e
G
e
a
r
F
a
u
lt
Dia
g
n
o
sis
Us
in
g
S
m
a
ll
Da
t
a
se
ts
W
it
h
De
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
-
Ba
se
d
T
ra
n
s
f
e
r
L
e
a
rn
in
g
,
"
IEE
E
Acc
e
ss
,
v
o
l.
6
,
p
p
.
2
6
2
4
1
-
2
6
2
5
3
,
2
0
1
8
.
[1
7
]
Kriz
h
e
v
sk
y
,
A
.
,
I.
S
u
tsk
e
v
e
r,
a
n
d
G
.
E.
Hin
t
o
n
,
"
Im
a
g
e
Ne
t
C
las
sif
ic
a
ti
o
n
w
it
h
De
e
p
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
s
,
"
Ad
v
a
n
c
e
s in
n
e
u
ra
l
in
fo
rm
a
ti
o
n
p
ro
c
e
ss
in
g
sy
ste
ms
,
v
o
l.
2
5
,
p
p
.
1
0
9
7
-
1
1
0
5
,
2
0
1
2
.
[1
8
]
L
iu
,
J.,
Y.
W
a
n
g
,
a
n
d
Y.
Qia
o
,
"
S
p
a
rse
De
e
p
T
r
a
n
sfe
r
L
e
a
rn
in
g
f
o
r
Co
n
v
o
l
u
ti
o
n
a
l
Ne
u
ra
l
Ne
tw
o
rk
,
"
T
h
irty
-
Fi
rs
t
AA
AI
Co
n
fer
e
n
c
e
o
n
Arti
fi
c
i
a
l
I
n
t
e
ll
ig
e
n
c
e
,
p
p
.
2
2
4
5
-
2
2
5
1
,
2
0
1
7
.
[1
9
]
L
i,
Y.,
e
t
a
l.
,
"
A
d
a
p
ti
v
e
Ba
tch
No
rm
a
li
z
a
ti
o
n
f
o
r
p
ra
c
ti
c
a
l
d
o
m
a
in
a
d
a
p
tati
o
n
,
"
P
a
tt
e
rn
Rec
o
g
n
it
io
n
,
v
o
l.
80
,
p
p
.
1
0
9
-
117
,
2
0
1
8
.
[2
0
]
G
o
o
d
f
e
ll
o
w
,
Ia
n
,
Yo
sh
u
a
Be
n
g
io
,
a
n
d
A
a
ro
n
Co
u
rv
il
le.
,
"
De
e
p
L
e
a
rn
in
g
,
"
M
IT
Pre
ss
,
2
0
1
6
.
[2
1
]
Zh
o
u
,
F
.
,
Y.
G
a
o
,
a
n
d
C.
W
e
n
,
"
A
No
v
e
l
M
u
lt
im
o
d
e
F
a
u
lt
Cl
a
ss
if
i
c
a
ti
o
n
M
e
t
h
o
d
Ba
se
d
o
n
D
e
e
p
L
e
a
rn
in
g
,
"
J
o
u
rn
a
l
o
f
Co
n
tro
l
S
c
ien
c
e
a
n
d
E
n
g
i
n
e
e
rin
g
,
v
o
l.
2
0
1
7
,
n
o
.
1
,
p
p
.
1
-
14,
2
0
1
7
.
[
2
2
]
R
u
d
e
r
,
S
.
,
"
A
n
O
v
e
r
v
i
e
w
o
f
M
u
l
t
i
-
T
a
s
k
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
,"
a
r
X
i
v
p
r
e
p
r
i
n
t
a
r
X
i
v
:
1
7
0
6
.
0
5
0
9
8
,
2017.
[2
3
]
Bin
g
e
l,
J.
a
n
d
A
.
S
ø
g
a
a
rd
,
"
Id
e
n
ti
fy
in
g
b
e
n
e
f
icia
l
tas
k
re
latio
n
s
f
o
r
m
u
lt
i
-
tas
k
lea
rn
in
g
in
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s
,
"
a
r
X
i
v
p
r
e
p
r
i
n
t
a
r
X
i
v
:
1
7
0
2
.
0
8
3
0
3
,
2
0
1
7
.
[2
4
]
Bin
Qa
isa
r,
S
.
a
n
d
M
.
Us
m
a
n
,
"
F
o
g
Ne
t
w
o
rk
in
g
f
o
r
M
a
c
h
in
e
H
e
a
lt
h
P
r
o
g
n
o
sis:
A
De
e
p
Lea
rn
in
g
P
e
rsp
e
c
ti
v
e
,
"
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
m
p
u
t
a
t
i
o
n
a
l
S
c
i
e
n
c
e
a
n
d
I
t
s
A
p
p
l
i
c
a
t
i
o
n
s
,
p
p
.
2
1
2
-
2
1
9
,
2
0
1
7
.
[2
5
]
ERP
INN
EW
S
,
"
F
o
g
c
o
m
p
u
ti
n
g
v
s
e
d
g
e
c
o
m
p
u
ti
n
g
,
"
2
0
1
8
.
[On
li
n
e
]
,
A
v
a
il
a
b
le:
h
tt
p
s://
e
rp
i
n
n
e
w
s.c
o
m
/
f
o
g
-
c
o
m
p
u
ti
n
g
-
vs
-
e
d
g
e
-
c
o
m
p
u
ti
n
g
.
[2
6
]
Ch
e
ll
iah
,
P
.
R.
a
n
d
J.
P
u
s
h
p
a
,
"
Ex
p
o
u
n
d
in
g
t
h
e
e
d
g
e
/f
o
g
c
o
m
p
u
ti
n
g
in
f
ra
stru
c
tu
re
s f
o
r
d
a
ta sc
ien
c
e
,
in
c
lo
u
d
a
n
d
f
o
g
c
o
m
p
u
ti
n
g
in
f
ra
stru
c
tu
re
s f
o
r
d
a
ta sc
ien
c
e
,
"
IGI Glo
b
a
l,
p
p
.
1
-
32
,
2
0
1
8
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Yo
u
ss
e
f
M
a
h
e
r
is
a
re
se
a
rc
h
sc
h
o
lar
in
th
e
De
p
a
rt
m
e
n
t
o
f
El
e
c
tri
c
a
l
a
n
d
M
e
c
h
a
n
ica
l
En
g
in
e
e
rin
g
,
f
ro
m
th
e
Un
iv
e
rsity
Ha
ss
a
n
1
,
S
e
tt
a
t
,
M
o
ro
c
c
o
.
He
h
a
s
a
d
e
g
re
e
in
i
n
d
u
strial
a
n
d
m
e
c
h
a
n
ica
l
e
n
g
in
e
e
rin
g
.
His
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
Da
ta
M
in
i
n
g
,
Big
d
a
ta
a
n
a
l
y
ti
c
s,
De
e
p
L
e
a
rn
in
g
,
Ne
u
ra
l
Ne
t
w
o
rk
s P
re
d
ictiv
e
a
n
a
ly
ti
c
s
.
B
o
u
je
m
a
a
D
a
n
o
u
j
Do
c
to
r
En
g
in
e
e
r
in
m
e
c
h
a
n
ica
l
e
n
g
in
e
e
rin
g
.
G
r
a
d
u
a
te o
f
th
e
Na
ti
o
n
a
l
S
c
h
o
o
l
o
f
El
e
c
tri
c
it
y
a
n
d
M
e
c
h
a
n
ics
Ca
sa
b
lan
c
a
,
M
o
ro
c
c
o
.
M
a
ste
r
o
f
P
h
il
o
so
p
h
y
(M
.
P
h
il
)
f
ro
m
M
id
d
les
e
x
Un
iv
e
rsit
y
,
UK
.
P
h
.
D.
f
ro
m
ÉT
S
(Éco
le
d
e
T
e
c
h
n
o
lo
g
ie
S
u
p
é
rieu
re
),
M
o
n
tréa
l,
Ca
n
a
d
a
.
E
x
p
e
rti
se
s:
F
re
e
f
o
r
m
c
u
rv
e
s
a
n
d
su
rf
a
c
e
s
m
o
d
e
li
n
g
,
p
a
rti
a
l
d
isc
h
a
rg
e
s
in
h
ig
h
v
o
lt
a
g
e
m
a
c
h
in
e
s,
d
e
e
p
lea
rn
i
n
g
a
n
d
p
re
d
ictiv
e
m
a
in
ten
a
n
c
e
,
M
e
tro
l
o
g
y
a
n
d
to
lera
n
c
in
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.