I
n
t
e
r
n
at
ion
al
Jou
r
n
al
of
E
lec
t
r
ical
an
d
Com
p
u
t
e
r
E
n
gin
e
e
r
in
g
(
I
JE
CE
)
Vol.
16
,
No.
5
,
Oc
tober
20
26
,
pp
.
2473
~
2482
I
S
S
N:
2088
-
8708
,
DO
I
:
10
.
11591/i
jec
e
.
v
16
i
5
.
pp
2
473
-
2482
2473
Jou
r
n
al
h
omepage
:
ht
tp:
//
ij
e
c
e
.
iaes
c
or
e
.
c
om
In
f
or
m
at
io
n
a
r
c
h
ite
c
t
u
r
e
d
e
b
t
:
Wh
y
le
ga
c
y p
la
t
f
or
m
sc
h
e
m
a
d
e
c
is
io
n
s
c
on
s
t
r
ai
n
e
n
t
e
r
p
r
is
e
A
I
c
ap
a
b
il
ity
M
ih
ir
S
h
ah
I
nde
pe
nde
nt
R
e
s
e
a
r
c
he
r
, U
S
A
Ar
t
icle
I
n
f
o
AB
S
T
RA
CT
A
r
ti
c
le
h
is
tor
y
:
R
e
c
e
ived
J
un
28,
2026
R
e
vis
e
d
Aug
11,
2026
Ac
c
e
pted
Aug
19,
2026
E
n
t
erp
r
i
s
e
p
l
a
t
fo
rm
s
accu
m
u
l
a
t
e
a
s
p
ec
i
fi
c
ca
t
eg
o
ry
o
f
t
ech
n
i
ca
l
d
e
b
t
t
h
i
s
p
ap
er
t
erms
i
n
fo
rma
t
i
o
n
arc
h
i
t
ect
u
re
d
eb
t
.
Sch
ema
d
ec
i
s
i
o
n
s
o
p
t
i
m
i
zed
f
o
r
t
ran
s
act
i
o
n
al
eff
i
ci
e
n
cy
i
n
earl
i
er
c
o
mp
u
t
i
n
g
eras
p
ro
d
u
ce
s
t
r
u
c
t
u
ra
l
co
n
s
t
ra
i
n
t
s
t
h
a
t
l
i
m
i
t
w
h
at
art
i
fi
c
i
al
i
n
t
e
l
l
i
g
e
n
ce
can
acco
mp
l
i
s
h
o
n
t
h
o
s
e
p
l
a
t
fo
rm
s
,
l
arg
el
y
i
n
d
e
p
en
d
en
t
o
f
w
h
i
ch
mo
d
el
s
are
s
el
e
ct
ed
o
r
h
o
w
t
h
ey
are
o
rch
e
s
t
ra
t
ed
.
T
h
i
s
p
a
p
er
o
ffers
a
q
u
al
i
t
at
i
v
e,
co
n
cep
t
u
al
framew
o
rk
,
g
ro
u
n
d
ed
i
n
p
ract
i
t
i
o
n
er
e
x
p
er
i
en
ce
rat
h
er
t
h
an
an
e
mp
i
r
i
cal
s
t
u
d
y
,
an
d
p
o
s
i
t
i
o
n
s
i
n
f
o
rmat
i
o
n
arch
i
t
ect
u
re
d
e
b
t
a
s
a
cat
e
g
o
r
y
d
i
s
t
i
n
ct
fr
o
m
co
d
e
d
eb
t
an
d
i
n
fra
s
t
r
u
ct
u
re
d
eb
t
,
arg
u
i
n
g
t
h
at
t
h
e
d
i
s
t
i
n
c
t
i
o
n
mat
t
er
s
b
ecau
s
e
remed
i
a
t
i
o
n
l
o
cal
i
t
y
an
d
co
o
r
d
i
n
at
i
o
n
co
s
t
d
i
ffer
s
h
ar
p
l
y
.
A
f
i
v
e
-
i
n
d
i
ca
t
o
r
d
i
a
g
n
o
s
t
i
c
framew
o
rk
i
s
p
r
o
p
o
s
e
d
,
co
v
eri
n
g
d
u
p
l
i
ca
t
e
can
o
n
i
c
al
en
t
i
t
i
es
,
b
ro
k
en
s
ema
n
t
i
c
ch
a
i
n
s
,
p
ro
v
en
a
n
ce
g
a
p
s
,
en
f
o
rcem
en
t
as
y
mmet
r
y
,
an
d
co
n
s
u
mer
a
s
s
u
mp
t
i
o
n
d
i
v
erg
e
n
ce,
s
y
n
t
h
e
s
i
ze
d
t
h
r
o
u
g
h
a
s
t
r
u
ct
u
red
q
u
a
l
i
t
at
i
v
e
p
r
o
ces
s
d
es
cr
i
b
e
d
i
n
a
d
ed
i
ca
t
ed
met
h
o
d
o
l
o
g
y
s
ec
t
i
o
n
.
T
h
e
framew
o
r
k
s
u
p
p
o
r
t
s
a
cap
a
b
i
l
i
t
y
ce
i
l
i
n
g
h
y
p
o
t
h
e
s
i
s
,
p
re
s
en
t
ed
as
a
t
h
e
o
ret
i
cal
p
ro
p
o
s
i
t
i
o
n
ra
t
h
er
t
h
an
a
n
e
x
p
er
i
men
t
al
l
y
v
al
i
d
a
t
ed
c
l
ai
m
:
s
u
b
s
t
r
at
e
d
efec
t
s
i
mp
o
s
e
an
u
p
p
er
b
o
u
n
d
o
n
art
i
f
i
ci
al
i
n
t
e
l
l
i
g
e
n
ce
(A
I)
o
u
t
co
me
s
,
mo
s
t
d
i
rec
t
l
y
o
b
s
er
v
ab
l
e
i
n
ret
r
i
ev
a
l
-
au
g
me
n
t
e
d
an
d
rea
s
o
n
i
n
g
-
b
as
e
d
arch
i
t
ec
t
u
re
s
,
t
h
at
m
o
d
e
l
s
e
l
ect
i
o
n
al
o
n
e
d
o
es
n
o
t
ap
p
ear
ab
l
e
t
o
ex
ceed
.
A
s
eq
u
en
ce
d
reme
d
i
a
t
i
o
n
a
p
p
r
o
ach
f
o
l
l
o
w
s
,
p
ri
o
r
i
t
i
zi
n
g
e
n
t
i
t
y
res
o
l
u
t
i
o
n
,
s
eman
t
i
c
al
i
g
n
men
t
,
an
d
p
r
o
v
e
n
an
ce
i
n
s
t
ru
men
t
at
i
o
n
b
y
A
I
-
cap
ab
i
l
i
t
y
i
mp
ac
t
rat
h
er
t
h
an
ea
s
e
o
f
fi
x
,
w
h
i
l
e
ack
n
o
w
l
e
d
g
i
n
g
t
h
a
t
s
ev
ere
cas
es
can
req
u
i
re
t
h
e
s
e
q
u
e
n
ce
t
o
b
e
ad
a
p
t
e
d
.
G
ro
u
n
d
e
d
i
n
p
rac
t
i
t
i
o
n
er
ex
p
er
i
en
ce
s
t
ru
c
t
u
r
i
n
g
en
t
erp
r
i
s
e
i
n
f
o
rmat
i
o
n
arch
i
t
e
ct
u
re
acro
s
s
m
u
l
t
i
-
l
an
g
u
a
g
e,
mu
l
t
i
-
s
y
s
t
em
p
l
at
f
o
rms
,
t
h
e
framew
o
rk
o
ffe
rs
l
ead
er
s
d
i
ag
n
o
s
t
i
c
an
d
s
eq
u
en
c
i
n
g
t
o
o
l
s
a
n
d
es
t
ab
l
i
s
h
e
s
a
co
n
cep
t
u
al
fo
u
n
d
a
t
i
o
n
f
o
r
l
a
t
er
emp
i
r
i
cal
an
d
q
u
an
t
i
t
at
i
v
e
w
o
rk
.
K
e
y
w
o
r
d
s
:
Da
ta
pr
ove
na
nc
e
E
nter
pr
is
e
a
r
ti
f
icia
l
int
e
ll
igenc
e
E
nti
ty
r
e
s
olut
ion
I
nf
or
mation
a
r
c
hit
e
c
tur
e
de
bt
S
e
mantic
a
li
gnment
T
e
c
hnica
l
de
bt
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
ce
n
s
e.
C
or
r
e
s
pon
din
g
A
u
th
or
:
M
ihi
r
S
ha
h
I
nde
pe
nde
nt
R
e
s
e
a
r
c
he
r
United
S
tate
s
of
Ame
r
ica
E
mail:
mi
h
ir
s
ha
hc
onne
c
t@gm
a
il
.
c
om
1.
I
NT
RODU
C
T
I
ON
L
e
a
de
r
s
hip
c
onve
r
s
a
ti
ons
a
bout
e
nter
pr
is
e
a
r
ti
f
icia
l
int
e
ll
igenc
e
tend
to
c
onc
e
ntr
a
te
on
model
f
a
mi
li
e
s
,
ve
ndor
s
e
lec
ti
on,
a
nd
s
pe
c
ialis
t
tale
nt.
T
he
platf
or
m
s
ubs
tr
a
te
on
whic
h
thos
e
models
ope
r
a
te
r
e
c
e
ives
f
a
r
les
s
s
c
r
uti
ny.
T
his
a
s
ymm
e
tr
y
is
unde
r
s
tanda
ble,
be
c
a
us
e
model
c
a
pa
bil
it
y
is
vis
ibl
e
a
nd
im
pr
ove
s
on
a
publi
c
c
a
de
nc
e
,
while
the
s
c
he
ma
of
a
de
c
a
de
-
old
tr
a
ns
a
c
ti
ona
l
s
y
s
tem
is
ne
it
he
r
vis
ibl
e
nor
e
xc
it
ing.
A
ga
p
ope
ns
be
twe
e
n
wha
t
a
model
c
a
n
do
in
pr
inciple
a
nd
wha
t
the
unde
r
lyi
ng
da
ta
s
tr
uc
tur
e
will
pe
r
mi
t
in
p
r
a
c
ti
c
e
,
a
nd
that
ga
p
f
r
e
que
ntl
y
be
c
omes
the
bindi
ng
c
ons
tr
a
int
on
de
li
ve
r
e
d
va
lue.
S
tr
uc
tur
e
d
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
16
,
No.
5
,
Oc
tober
20
26
:
2473
-
2482
2474
a
na
lys
e
s
of
f
a
il
e
d
a
r
ti
f
icia
l
int
e
ll
igenc
e
pr
ogr
a
ms
r
e
pe
a
tedly
loca
te
r
oot
c
a
us
e
s
in
da
ta
r
e
a
di
ne
s
s
a
nd
or
ga
niza
ti
ona
l
a
li
gnment
r
a
ther
than
in
a
lgo
r
it
hm
c
hoice
[
1
]
.
T
he
c
ons
is
tenc
y
of
f
indi
ng
a
c
r
os
s
p
r
ogr
a
ms
s
ugge
s
ts
the
c
ons
tr
a
int
is
s
tr
uc
tur
a
l,
not
a
matter
of
indi
vidual
p
r
ojec
t
e
xe
c
uti
on.
T
his
pa
pe
r
int
r
oduc
e
s
in
f
or
mation
a
r
c
hit
e
c
tur
e
de
b
t
a
s
a
dis
ti
nc
t
c
a
tegor
y
of
tec
hnica
l
de
bt
a
nd
s
tate
s
plainly,
to
a
li
gn
r
e
a
de
r
e
xpe
c
tations
f
r
om
the
outs
e
t,
that
the
c
ontr
ibut
ion
is
a
qua
li
tative
c
o
nc
e
ptual
f
r
a
mew
or
k
r
a
ther
than
a
n
e
mpi
r
ica
l
s
tudy.
T
he
te
r
m
de
notes
the
a
c
c
umul
a
ted
c
os
t
of
s
c
he
ma
a
nd
s
tr
uc
tur
a
l
de
c
is
ions
that
we
r
e
r
e
a
s
on
a
ble
whe
n
made
f
or
tr
a
ns
a
c
ti
ona
l
e
f
f
icie
nc
y
ye
t
now
c
ons
tr
a
in
a
na
lyt
ica
l
a
nd
ge
ne
r
a
ti
ve
us
e
.
P
r
a
c
ti
ti
one
r
e
xpe
r
ienc
e
s
tr
uc
tur
in
g
e
nter
pr
is
e
inf
o
r
mation
a
r
c
hit
e
c
tur
e
a
c
r
os
s
the
d
a
ta
laye
r
,
ba
c
k
e
nd,
a
nd
de
li
ve
r
y
pipeline
of
lar
ge
mul
t
i
-
s
ys
t
e
m
platf
o
r
ms
indi
c
a
tes
that
thes
e
c
ons
tr
a
int
s
a
r
e
s
y
s
tema
ti
c
r
a
ther
than
incide
ntal
.
T
he
s
a
me
s
tr
uc
tur
a
l
pa
tt
e
r
ns
r
e
c
ur
a
c
r
os
s
s
e
c
tor
s
,
a
nd
they
s
ur
f
a
c
e
mos
t
s
ha
r
pl
y
whe
n
a
c
r
os
s
-
f
unc
ti
ona
l
model
is
a
s
ke
d
to
r
e
a
s
on
ove
r
da
ta
that
s
e
ve
r
a
l
s
ys
tem
s
e
a
c
h
de
f
ine
dif
f
e
r
e
ntl
y.
R
e
c
ognizing
the
pa
tt
e
r
n
e
a
r
ly
in
tec
hnica
l
dis
c
ove
r
y,
be
f
or
e
mo
de
l
s
e
lec
ti
on,
a
ppe
a
r
s
to
c
ha
nge
whic
h
p
r
oblems
a
pr
ogr
a
m
c
a
n
r
e
a
li
s
ti
c
a
ll
y
s
olve.
T
he
pur
pos
e
of
thi
s
r
e
s
e
a
r
c
h
is
to
a
ddr
e
s
s
a
s
pe
c
if
ic
ga
p:
e
nter
pr
is
e
a
r
ti
f
icia
l
int
e
ll
igenc
e
(
AI
)
planning
r
outi
ne
ly
t
r
e
a
ts
the
da
ta
s
ubs
tr
a
te
a
s
a
ba
c
kgr
ound
c
ondit
ion
r
a
ther
than
a
s
the
pr
im
a
r
y
c
on
s
tr
a
int
on
a
c
hieva
ble
c
a
pa
bil
it
y,
a
nd
no
e
xis
ti
ng
tec
hnica
l
d
e
bt
c
a
tegor
y
c
a
ptur
e
s
the
c
r
os
s
-
s
y
s
tem,
s
tr
uc
tur
a
l
c
ha
r
a
c
ter
of
that
c
ons
tr
a
int
.
T
he
pa
pe
r
's
main
f
indi
ngs
a
r
e
a
f
ive
-
indi
c
a
tor
diagnos
ti
c
f
r
a
mew
or
k
c
a
pa
ble
of
s
ur
f
a
c
ing
the
c
ons
tr
a
int
be
f
or
e
it
de
r
a
il
s
a
model
-
s
e
lec
ti
on
-
f
ir
s
t
pr
ogr
a
m
,
a
nd
a
c
a
pa
bil
it
y
c
e
il
ing
hypoth
e
s
is
that
e
xplains
why
s
tr
onge
r
models
a
lone
do
not
r
e
s
olv
e
outcome
s
blocke
d
a
t
the
s
ubs
tr
a
te
leve
l
.
T
he
im
pli
c
a
ti
on
f
or
pr
a
c
ti
c
e
is
that
tec
hnica
l
dis
c
ove
r
y
s
hould
a
s
s
e
s
s
the
s
ubs
tr
a
te
be
f
o
r
e
e
va
luating
models
,
inve
r
ti
ng
the
s
e
que
nc
e
mos
t
pr
ogr
a
ms
c
ur
r
e
ntl
y
f
oll
ow.
T
he
a
r
gument
p
r
oc
e
e
ds
in
a
f
ixed
o
r
de
r
.
S
e
c
ti
o
n
2
r
e
view
s
r
e
late
d
li
te
r
a
tur
e
a
nd
pos
it
ions
the
c
onc
e
pt
withi
n
e
xis
ti
ng
tec
hnica
l
de
bt
s
c
holar
s
hip.
S
e
c
ti
on
3
de
s
c
r
ibes
the
methodology
us
e
d
to
d
e
r
ive
the
f
ive
-
indi
c
a
tor
f
r
a
mew
or
k
.
S
e
c
ti
on
4
pr
e
s
e
nts
the
diagnos
ti
c
f
r
a
mew
or
k
it
s
e
lf
.
S
e
c
ti
on
5
s
tate
s
the
c
a
pa
bil
it
y
c
e
il
ing
hypothes
is
,
a
nd
s
e
c
ti
on
6
de
ve
lops
a
s
e
que
nc
e
d
r
e
media
ti
on
f
r
a
mew
or
k
with
a
wo
r
ke
d
il
l
us
tr
a
ti
ve
s
c
e
na
r
io.
Or
ga
niza
ti
ona
l
dyna
mi
c
s
a
nd
s
e
c
tor
a
ppli
c
a
ti
ons
oc
c
upy
s
e
c
ti
on
7,
li
mi
tations
a
nd
f
utu
r
e
r
e
s
e
a
r
c
h
oc
c
upy
s
e
c
ti
on
8,
a
nd
s
e
c
ti
on
9
c
onc
ludes
.
2.
RE
L
AT
E
D
WORK
AN
D
CONC
E
P
T
UA
L
P
OS
I
T
I
ON
I
NG
2.
1.
T
e
c
h
n
ical
d
e
b
t
s
c
h
olars
h
ip
T
he
tec
hnica
l
de
bt
meta
phor
or
igi
na
tes
in
a
de
s
c
r
ipt
ion
of
de
f
e
r
r
e
d
de
s
ign
c
lea
nup
a
s
a
de
bt
that
a
c
c
r
ue
s
int
e
r
e
s
t
unti
l
it
is
r
e
pa
id
[
2]
.
L
a
ter
wor
k
e
x
tende
d
the
meta
phor
int
o
c
ode
de
bt,
de
s
ign
de
bt
,
t
e
s
t
de
bt,
doc
umenta
ti
on
de
bt,
a
nd
in
f
r
a
s
tr
uc
tur
e
de
bt,
a
nd
a
s
ys
tema
ti
c
r
e
view
in
s
ys
tems
e
nginee
r
ing
c
a
talogs
thi
s
pr
oli
f
e
r
a
ti
on
of
s
ubtypes
[
3]
.
M
a
c
hine
lea
r
ning
s
ys
tems
a
tt
r
a
c
ted
their
own
tr
e
a
tm
e
nt,
in
whi
c
h
da
ta
de
pe
nde
nc
ies
a
nd
c
onf
igur
a
ti
on
e
ntangle
ment
we
r
e
s
hown
to
ge
ne
r
a
te
de
bt
that
is
inv
is
ibl
e
a
t
the
c
o
de
leve
l
ye
t
c
os
tl
y
a
t
the
s
ys
tem
leve
l
[
4]
.
I
ntelli
ge
nt
tec
hniques
f
or
de
tec
ti
ng
a
nd
mana
ging
tec
hnica
l
de
bt
ha
ve
s
ince
be
e
n
mappe
d
a
c
r
os
s
the
s
of
twa
r
e
e
nginee
r
ing
li
ter
a
tur
e
[
5]
.
S
c
he
ma
e
volut
ion
r
e
s
e
a
r
c
h
doc
uments
how
pr
oduc
ti
on
da
taba
s
e
s
tr
uc
tur
e
s
os
s
if
y
ove
r
ti
me,
with
mi
gr
a
ti
ons
lagging
be
hind
c
ha
nging
r
e
quir
e
ments
[
6]
.
M
a
s
ter
da
ta
mana
ge
ment
s
c
holar
s
hip
a
ddr
e
s
s
e
s
e
n
ti
ty
a
nd
s
e
mantic
f
r
a
gmenta
ti
on
f
r
o
m
the
gove
r
na
nc
e
s
ide
[
7]
,
a
nd
da
ta
gove
r
na
nc
e
f
r
a
mew
or
ks
tar
ge
t
the
c
r
os
s
-
s
ys
t
e
m
c
oor
dination
that
li
ne
a
ge
pr
e
s
e
r
va
ti
on
r
e
quir
e
s
[
8]
.
E
nti
ty
matc
hing
ha
s
de
ve
loped
int
o
a
n
a
c
ti
ve
r
e
s
e
a
r
c
h
a
r
e
a
in
it
s
own
r
ight
,
with
ne
ur
a
l
a
nd
pr
e
-
tr
a
ined
langua
ge
model
a
ppr
oa
c
he
s
s
ur
ve
ye
d
e
xtens
ively
[
9
]
a
nd
e
xplaina
ble
va
r
iants
de
ve
loped
s
o
that
matc
hing
de
c
is
ions
c
a
n
be
a
udit
e
d
[
10]
.
I
nte
r
ope
r
a
bil
it
y
s
c
holar
s
hip
s
e
pa
r
a
tely
c
las
s
if
ie
s
the
mi
s
matc
he
s
t
ha
t
a
r
is
e
whe
n
bus
ines
s
c
onc
e
pts
f
a
il
to
tr
a
ns
late
a
c
r
os
s
s
y
s
tems
,
e
s
tablis
hing
that
s
e
mantic
int
e
r
ope
r
a
bil
it
y
is
dis
ti
nc
t
f
r
om,
a
nd
ha
r
de
r
to
a
c
hieve
than,
tec
hnica
l
c
onn
e
c
ti
vit
y
[
11
]
.
P
r
ove
na
nc
e
r
e
s
e
a
r
c
h
in
s
e
c
ur
it
y
a
n
d
pr
ivac
y
doc
uments
both
the
va
lue
of
li
ne
a
ge
a
nd
the
dif
f
i
c
ult
y
of
r
e
c
ons
tr
uc
ti
ng
it
onc
e
dis
c
a
r
de
d
[
12]
.
Da
t
a
qua
li
ty
mana
ge
ment
r
e
s
e
a
r
c
h
a
im
s
a
t
unif
or
m
e
nf
or
c
e
m
e
nt
of
va
li
da
ti
on
r
ules
a
c
r
os
s
s
our
c
e
s
[
13]
,
a
nd
t
he
da
ta
-
c
e
ntr
ic
view
of
a
r
ti
f
icia
l
int
e
ll
igenc
e
holds
mor
e
br
oa
dly
that
s
ys
tema
ti
c
e
nginee
r
ing
of
da
ta,
r
a
the
r
than
of
models
,
incr
e
a
s
ingl
y
gove
r
ns
outcome
s
[
14]
.
2.
2.
P
os
it
ion
in
g
in
f
or
m
a
t
ion
ar
c
h
it
e
c
t
u
r
e
d
e
b
t
r
e
lat
ive
t
o
d
at
a
d
e
b
t
I
nf
or
mation
a
r
c
hit
e
c
tu
r
e
de
bt
s
it
s
a
longs
ide
thes
e
c
a
tegor
ies
.
I
t
is
c
los
e
s
t
in
s
pir
it
to
da
ta
de
bt,
but
the
two
a
r
e
s
e
pa
r
a
ble
on
a
s
pe
c
if
ic
c
r
it
e
r
ion
:
da
t
a
de
bt
is
a
r
e
c
or
d
-
leve
l
or
s
ingl
e
-
s
ys
tem
qua
li
ty
pr
ope
r
ty,
while
inf
or
mation
a
r
c
hit
e
c
tur
e
de
bt
is
a
c
r
os
s
-
s
ys
t
e
m
s
tr
uc
tur
a
l
a
nd
s
e
mantic
pr
ope
r
ty
that
pe
r
s
is
ts
e
ve
n
whe
n
e
ve
r
y
indi
vidual
r
e
c
or
d
is
c
lea
n.
A
c
us
tom
e
r
r
e
c
or
d
with
no
m
is
s
ing
f
ields
,
c
or
r
e
c
t
f
or
matti
ng
,
a
nd
no
dupli
c
a
ti
on
withi
n
it
s
own
s
ys
tem
c
a
n
s
ti
ll
be
in
f
or
mation
-
a
r
c
hit
e
c
tur
e
-
indebte
d
if
a
s
e
c
ond
s
ys
tem
de
f
ines
"
c
us
tom
e
r
"
dif
f
e
r
e
ntl
y,
or
if
a
th
ir
d
s
ys
tem
holds
a
n
unr
e
c
onc
il
e
d
dupli
c
a
te
of
the
s
a
me
r
e
a
l
-
wor
ld
e
n
ti
ty.
T
he
de
mar
c
a
ti
on,
c
onc
r
e
tely,
is
thi
s
:
a
da
ta
-
qua
li
ty
a
udit
r
un
a
ga
ins
t
a
s
ingl
e
s
ys
tem's
own
r
ules
will
no
t
s
ur
f
a
c
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
I
nfor
mation
ar
c
hit
e
c
tur
e
de
bt:
W
hy
legac
y
plat
for
m
s
c
he
ma
de
c
is
ions
c
ons
t
r
ain
…
(
M
ihi
r
Shah
)
2475
inf
or
mation
a
r
c
hit
e
c
tur
e
de
bt
,
be
c
a
us
e
the
de
bt
is
de
f
ined
by
the
r
e
lations
hip
be
twe
e
n
s
ys
tems
,
not
by
a
ny
one
s
ys
tem's
int
e
r
na
l
s
tate
.
S
e
mantic
e
r
os
ion
il
lus
tr
a
tes
the
dis
ti
nc
ti
on
s
ha
r
ply
.
A
bil
li
ng
s
ys
tem's
de
f
ini
ti
on
of
"
a
c
ti
ve
c
us
tom
e
r
"
may
be
pe
r
f
e
c
tl
y
e
nf
o
r
c
e
d
a
nd
pe
r
f
e
c
tl
y
doc
umente
d
with
in
bil
li
ng
,
a
nd
a
s
uppor
t
s
ys
tem's
de
f
ini
ti
on
may
be
e
qua
ll
y
we
ll
-
f
or
med
wi
thi
n
s
uppor
t
,
ye
t
the
two
s
ys
tems
c
a
n
diver
ge
in
a
wa
y
that
no
s
ingl
e
-
s
ys
tem
qua
li
ty
c
he
c
k
would
e
ve
r
f
lag.
I
t
is
the
mi
s
a
li
gnment
a
c
r
os
s
s
ys
tems
,
not
the
r
e
c
or
d
-
leve
l
qua
li
ty
withi
n
a
ny
one
of
them,
that
c
ons
tr
a
ins
mo
de
l
r
e
a
s
oning
a
c
r
os
s
the
bounda
r
y.
2.
3.
Ac
c
u
m
u
la
t
ion
m
e
c
h
an
is
m
s
T
hr
e
e
mec
ha
nis
ms
a
ppe
a
r
to
dr
ive
a
c
c
umul
a
ti
on.
T
he
f
ir
s
t
is
de
c
is
ion
lock
-
in:
a
s
c
he
me
c
hos
e
n
to
make
a
tr
a
ns
a
c
ti
on
f
a
s
t,
f
or
e
xa
mpl
e
a
c
us
tom
e
r
r
e
c
or
d
ke
ye
d
to
a
n
or
de
r
ing
s
ys
tem,
be
c
omes
a
de
pe
nde
nc
y
f
or
e
ve
r
y
downs
tr
e
a
m
c
ons
umer
,
a
nd
late
r
r
e
c
ons
i
de
r
a
ti
on
is
r
e
s
is
ted
be
c
a
us
e
the
c
os
t
of
c
ha
nge
f
a
ll
s
on
the
many
r
a
ther
than
the
f
e
w.
T
he
s
e
c
ond
is
s
e
mantic
e
r
os
ion:
a
bus
ines
s
c
onc
e
pt
s
uc
h
a
s
c
us
tom
e
r
me
a
ns
one
thi
ng
to
bil
li
ng,
a
nother
to
s
uppor
t,
a
nd
a
nother
to
mar
ke
ti
ng,
a
nd
the
diver
ge
nc
e
wide
ns
quietly
a
s
e
a
c
h
s
ys
tem
e
volves
on
it
s
own
s
c
he
dule.
T
he
th
ir
d
is
pr
ove
na
nc
e
los
s
:
da
ta
moves
thr
ough
t
r
a
ns
f
or
ma
ti
on
a
nd
int
e
gr
a
ti
on
s
teps
,
a
nd
li
ne
a
ge
is
dis
c
a
r
de
d
a
t
e
a
c
h
hop
be
c
a
us
e
the
tr
a
ns
a
c
ti
ona
l
pur
pos
e
did
not
r
e
quir
e
it
.
W
ha
t
unit
e
s
the
thr
e
e
is
that
e
a
c
h
is
loca
ll
y
r
a
ti
ona
l
a
nd
globally
c
os
tl
y
.
2.
4.
Dis
t
in
gu
is
h
i
n
g
T
he
pr
a
c
ti
c
a
l
dis
ti
nc
ti
on
f
r
om
other
tec
hnica
l
de
bt
c
a
tegor
ies
li
e
s
in
r
e
media
ti
on
loca
li
ty.
C
ode
de
bt
c
a
n
us
ua
ll
y
be
pa
id
down
withi
n
a
modul
e
by
the
tea
m
that
owns
it
,
a
nd
inf
r
a
s
tr
uc
tur
e
de
bt
c
a
n
be
r
e
duc
e
d
by
mi
gr
a
ti
ng
a
s
e
r
vice
,
both
be
ing
lar
ge
ly
loca
l
ope
r
a
t
ions
with
pr
e
dicta
ble
blas
t
r
a
dius
.
I
nf
or
mation
a
r
c
hit
e
c
tur
e
de
bt
r
e
s
is
ts
loca
l
r
e
pa
ir
.
A
s
ha
r
e
d
e
nti
ty
or
c
onc
e
pt
ha
s
many
de
pe
nde
nts
,
a
nd
c
ha
nging
it
r
e
quir
e
s
c
oor
dinate
d
c
ha
nge
a
c
r
os
s
a
ll
of
them
a
t
onc
e
.
C
oor
dination
c
os
t
ther
e
f
or
e
s
c
a
les
with
the
nu
mber
of
de
pe
nde
nts
r
a
ther
than
with
the
s
ize
of
a
ny
s
in
gle
c
ha
nge
;
a
s
a
r
ough
planning
he
ur
is
ti
c
,
the
pr
a
c
ti
c
a
l
c
oor
dination
bur
de
n
of
r
e
media
ti
ng
a
s
ha
r
e
d
e
nti
ty
tr
a
c
ks
the
pr
oduc
t
o
f
the
number
of
do
wns
tr
e
a
m
c
ons
umi
ng
s
ys
tems
a
nd
the
numbe
r
o
f
dis
ti
nc
t
or
ga
niza
ti
ona
l
tea
ms
that
own
thos
e
c
ons
umer
s
,
s
in
c
e
e
a
c
h
a
ddit
ional
owning
tea
m
a
dds
a
ne
goti
a
ti
on
the
t
e
c
hnica
l
c
ha
nge
it
s
e
lf
doe
s
not
r
e
quir
e
.
T
his
he
ur
is
ti
c
is
of
f
e
r
e
d
a
s
a
n
il
lus
tr
a
ti
ve
planning
a
id
r
a
ther
than
a
va
li
da
ted
model.
T
a
ble
1
s
umm
a
r
ize
s
the
c
ontr
a
s
t
a
c
r
os
s
r
e
media
ti
on
loca
li
ty,
c
oor
dination
c
os
t,
a
nd
typi
c
a
l
a
c
c
umul
a
ti
on
mec
ha
nis
m.
T
his
p
r
ope
r
ty
e
xplains
why
the
de
bt
pe
r
s
is
ts
e
ve
n
whe
n
indi
v
idual
tea
ms
r
e
c
ognize
it
c
lea
r
ly
,
s
ince
no
s
ingl
e
tea
m
c
a
n
r
e
pa
y
i
t
a
l
one
a
nd
none
is
ince
nti
vize
d
to
t
r
y.
T
ab
l
e
1
.
In
fo
rma
t
i
o
n
arc
h
i
t
ect
u
re
d
e
b
t
v
er
s
u
s
co
d
e
d
eb
t
v
ers
u
s
i
n
fras
t
ru
c
t
u
re
d
eb
t
P
r
ope
r
ty
C
ode
de
bt
I
nf
r
a
s
tr
uc
tu
r
e
de
bt
I
nf
or
ma
ti
on a
r
c
hi
te
c
tu
r
e
de
bt
R
e
me
di
a
ti
on l
oc
a
li
ty
L
oc
a
l
to
a
modul
e
or
t
e
a
m
L
oc
a
l
to
a
s
e
r
vi
c
e
or
mi
gr
a
ti
on
D
is
tr
ib
ut
e
d a
c
r
os
s
a
ll
de
p
e
nde
nt
s
of
a
s
ha
r
e
d e
nt
it
y
C
oor
di
na
ti
on c
os
t
L
ow
;
ow
ni
ng t
e
a
m a
c
ts
a
lo
ne
M
ode
r
a
te
;
bounde
d bl
a
s
t
r
a
di
u
s
H
ig
h;
s
c
a
le
s
w
it
h d
e
pe
nde
nt
s
×
ow
ni
ng t
e
a
ms
T
ypi
c
a
l
a
c
c
umul
a
ti
on
me
c
ha
ni
s
m
D
e
f
e
r
r
e
d c
le
a
nup a
nd s
hor
tc
ut
s
A
gi
ng pla
tf
or
ms
a
nd de
f
e
r
r
e
d
mi
gr
a
ti
on
D
e
c
is
io
n l
oc
k
-
in
, s
e
ma
nt
ic
e
r
os
io
n, pr
ove
na
nc
e
l
os
s
3.
M
E
T
HO
DOL
OG
Y
T
his
pa
pe
r
's
f
r
a
mew
or
k
is
de
r
ived
th
r
ough
a
qu
a
li
tative
s
ynthes
is
pr
oc
e
s
s
r
a
ther
than
a
f
or
mal
e
mpi
r
ica
l
s
tudy,
a
nd
the
pa
pe
r
make
s
no
c
laim
to
s
tatis
ti
c
a
l
ge
ne
r
a
li
z
a
bil
it
y;
it
of
f
e
r
s
a
c
onc
e
ptual
s
tr
uc
tur
e
int
e
nde
d
f
or
s
ubs
e
que
nt
e
mpi
r
ica
l
tes
ti
ng.
T
he
s
ynthes
is
dr
e
w
on
pr
a
c
ti
ti
one
r
e
xpe
r
ienc
e
s
tr
uc
tur
ing
e
nter
pr
is
e
inf
or
mation
a
r
c
hit
e
c
tur
e
a
c
r
os
s
the
da
ta
laye
r
,
ba
c
k
e
nd,
a
nd
de
li
ve
r
y
pipeline
of
lar
ge
,
mul
ti
-
s
ys
tem,
mul
ti
-
langua
ge
platf
or
ms
,
including
w
or
k
a
li
gning
s
c
he
mas
a
c
r
os
s
nine
-
langua
ge
loc
a
li
z
a
ti
on
s
ys
tems
a
nd
s
tanda
r
dizing
tens
of
thous
a
nds
of
e
nt
e
r
pr
is
e
pr
oc
e
s
s
maps
.
T
he
pr
oc
e
s
s
f
oll
owe
d
thr
e
e
s
tage
s
.
F
i
r
s
t,
r
e
c
ur
r
i
ng
tec
hnica
l
dis
c
ove
r
y
f
indi
ngs
we
r
e
c
a
talogue
d
a
c
r
os
s
e
nga
ge
ments
:
the
s
pe
c
if
ic
s
tr
uc
tur
a
l
pr
ob
l
e
ms
that
r
e
pe
a
tedly
obs
tr
uc
ted
AI
ini
ti
a
ti
ve
s
r
e
ga
r
dles
s
of
the
s
e
c
tor
or
platf
or
m
invol
ve
d.
S
e
c
ond,
thes
e
f
ind
ings
we
r
e
a
bs
tr
a
c
ted
int
o
c
a
ndidate
indi
c
a
tor
s
by
gr
ouping
obs
e
r
va
ti
ons
that
s
ha
r
e
d
a
c
omm
on
unde
r
lyi
ng
m
e
c
ha
nis
m
a
nd
a
c
omm
on
downs
tr
e
a
m
AI
f
a
il
ur
e
s
ignatur
e
,
dis
c
a
r
ding
c
a
ndidate
s
that
we
r
e
s
e
c
tor
-
s
pe
c
if
ic
r
a
ther
than
s
tr
uc
tur
a
l
.
T
hir
d
,
e
a
c
h
r
e
taine
d
c
a
ndi
da
te
wa
s
c
r
os
s
-
c
he
c
ke
d
a
ga
ins
t
the
tec
hnica
l
de
bt
,
e
nti
ty
r
e
s
olut
ion,
int
e
r
ope
r
a
bil
it
y,
a
nd
da
ta
gove
r
na
nc
e
li
ter
a
tur
e
s
r
e
view
e
d
in
S
e
c
ti
on
2,
to
c
onf
ir
m
that
the
indi
c
a
tor
c
or
r
e
s
ponde
d
to
a
r
e
c
ognize
d
s
tr
uc
tur
a
l
phe
nomenon
r
a
ther
than
a
n
idi
os
ync
r
a
ti
c
obs
e
r
va
ti
on
f
r
om
a
s
i
ngle
e
nga
ge
ment.
I
ndica
tor
s
that
c
ould
not
be
c
or
r
obor
a
ted
a
ga
ins
t
thi
s
li
ter
a
tur
e
,
or
that
a
ppli
e
d
to
f
e
we
r
t
ha
n
thr
e
e
indepe
nde
nt
e
nga
ge
ments
,
we
r
e
e
xc
luded.
F
ive
indi
c
a
tor
s
s
ur
vived
thi
s
pr
oc
e
s
s
;
they
a
r
e
p
r
e
s
e
nted
in
S
e
c
ti
on
4.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
16
,
No.
5
,
Oc
tober
20
26
:
2473
-
2482
2476
T
his
method
ha
s
a
c
lea
r
li
mi
tation,
a
ddr
e
s
s
e
d
f
ur
ther
in
S
e
c
ti
on
8:
it
doe
s
not
yield
a
s
tatis
ti
c
a
ll
y
va
li
da
ted
indi
c
a
tor
s
e
t,
a
nd
the
e
xc
lus
ion
th
r
e
s
holds
de
s
c
r
ibed
a
bove
we
r
e
a
ppli
e
d
qua
li
tatively
r
a
t
he
r
than
thr
ough
a
pr
e
-
r
e
gis
ter
e
d
pr
otocol
.
T
he
methodol
ogy
is
of
f
e
r
e
d
tr
a
ns
pa
r
e
ntl
y
s
o
that
the
indi
c
a
tor
s
c
a
n
be
tes
ted,
r
e
f
ined,
or
f
a
ls
if
ied
by
s
ubs
e
que
nt
e
mpi
r
ica
l
wor
k
r
a
ther
than
tr
e
a
ted
a
s
a
c
los
e
d
or
f
inal
li
s
t.
4.
T
HE
F
I
VE
-
I
ND
I
C
AT
OR
DI
AGNOS
T
I
C
F
RA
M
E
WORK
R
e
c
ognizing
the
de
bt
r
e
quir
e
s
obs
e
r
va
ble
s
ignals
.
T
he
f
ive
indi
c
a
tor
s
de
r
ived
in
S
e
c
ti
on
3
togethe
r
give
a
diagnos
ti
c
pictur
e
.
E
a
c
h
indi
c
a
tor
maps
to
a
c
ha
r
a
c
ter
is
ti
c
wa
y
that
a
r
ti
f
icia
l
int
e
ll
igenc
e
f
a
il
s
on
the
a
f
f
e
c
ted
platf
o
r
m,
a
nd
the
f
a
il
ur
e
typi
c
a
ll
y
pr
e
s
e
nts
a
s
a
model
de
f
e
c
t
while
or
igi
na
ti
ng
in
the
s
ubs
tr
a
te.
T
a
ble
2
c
ons
oli
da
tes
the
indi
c
a
tor
s
,
wha
t
e
a
c
h
mea
s
ur
e
s
,
a
nd
the
AI
f
a
il
u
r
e
mode
e
a
c
h
p
r
oduc
e
s
,
dis
ti
nguis
hing
whe
r
e
the
f
a
il
ur
e
s
ignatur
e
dif
f
e
r
s
be
twe
e
n
pr
e
dictive
mac
hine
lea
r
ning
s
ys
tems
a
nd
ge
ne
r
a
ti
ve
AI
s
ys
tems
,
a
dis
ti
nc
ti
on
that
matter
s
b
e
c
a
us
e
the
ha
ll
uc
ination
li
ter
a
tur
e
t
r
e
a
ts
ge
ne
r
a
ti
ve
f
a
il
ur
e
a
s
a
phe
nomenon
with
it
s
own
taxonomy
a
nd
c
a
us
e
s
r
a
ther
than
a
s
a
ge
ne
r
ic
a
c
c
ur
a
c
y
pr
oblem
[
15]
.
S
ur
ve
ys
of
da
tas
e
t
qua
li
ty
a
nd
o
f
da
ta
qua
li
ty
r
e
quir
e
me
nts
in
mac
hine
lea
r
ning
c
o
r
r
obor
a
te
that
thes
e
s
ubs
tr
a
te
pr
ope
r
ti
e
s
,
r
a
ther
than
model
a
r
c
hit
e
c
tur
e
,
tend
to
domi
na
te
outcome
qua
li
ty
in
pr
a
c
ti
c
e
[
16]
,
[
17]
.
T
he
indi
c
a
tor
s
a
r
e
int
e
nde
d
to
be
a
s
s
e
s
s
e
d
dur
ing
dis
c
ove
r
y,
be
f
o
r
e
a
model
is
c
hos
e
n.
T
ab
l
e
2
.
T
h
e
f
i
v
e
-
i
n
d
i
ca
t
o
r
d
i
ag
n
o
s
t
i
c
framew
o
rk
I
ndi
c
a
to
r
W
ha
t
it
me
a
s
ur
e
s
F
a
il
ur
e
mode
—
pr
e
di
c
ti
ve
M
L
F
a
il
ur
e
mode
—
ge
ne
r
a
ti
ve
A
I
D
upl
ic
a
te
c
a
noni
c
a
l
e
nt
it
ie
s
S
a
me
r
e
a
l
-
w
or
ld
e
nt
it
y he
ld
a
s
mul
ti
pl
e
unr
e
c
onc
il
e
d
r
e
c
or
ds
I
nc
ons
is
te
nt
f
e
a
tu
r
e
va
lu
e
s
de
gr
a
de
pr
e
di
c
ti
on a
c
c
ur
a
c
y
s
il
e
nt
ly
R
e
tr
ie
va
l
r
e
tu
r
ns
w
hi
c
he
ve
r
dupl
ic
a
te
i
t
e
nc
ount
e
r
s
, pr
oduc
in
g
a
ns
w
e
r
s
t
ha
t
r
e
a
d
a
s
ha
ll
uc
in
a
ti
on
B
r
oke
n s
e
ma
nt
ic
c
ha
in
s
B
us
in
e
s
s
c
onc
e
pt
s
t
ha
t
do not
ma
p c
le
a
nl
y a
c
r
os
s
s
y
s
te
ms
M
is
la
be
le
d or
mi
s
a
li
gne
d
tr
a
in
in
g t
a
r
ge
ts
a
c
r
os
s
s
our
c
e
s
ys
te
ms
I
nc
or
r
e
c
t
c
ombi
ne
d output
s
, or
t
he
mode
l
de
c
li
ne
s
t
o a
c
t
on t
he
a
mbi
gui
ty
P
r
ove
na
nc
e
ga
ps
D
a
ta
a
r
r
iv
in
g w
it
hout
or
ig
in
or
t
r
a
ns
f
or
ma
ti
on l
in
e
a
ge
M
ode
l
out
put
s
c
a
nnot
be
t
r
a
c
e
d
ba
c
k f
or
e
r
r
or
a
na
ly
s
is
O
ut
put
s
c
a
nnot
be
a
udi
te
d or
ma
de
c
ompl
ia
nt
, unde
r
mi
ni
ng t
r
us
t
in
ge
ne
r
a
te
d c
ont
e
nt
E
nf
or
c
e
me
nt
a
s
ymm
e
tr
y
D
a
ta
qua
li
ty
r
ul
e
s
a
ppl
ie
d
in
c
ons
is
te
nt
ly
a
c
r
os
s
s
y
s
te
ms
T
r
a
in
in
g da
ta
qua
li
ty
va
r
ie
s
unpr
e
di
c
ta
bl
y by s
our
c
e
s
y
s
te
m
B
r
it
tl
e
a
da
pt
iv
e
l
ogi
c
i
n a
ge
nt
ic
w
or
kf
lo
w
s
f
a
il
s
a
s
s
y
s
te
ms
dr
if
t
C
ons
ume
r
a
s
s
umpt
io
n
di
ve
r
ge
nc
e
P
r
iv
a
te
, unc
oor
di
na
te
d
dow
ns
tr
e
a
m wor
ka
r
ounds
F
e
a
tu
r
e
pi
pe
li
ne
s
s
il
e
nt
ly
di
ve
r
ge
f
r
om wha
t
th
e
mode
l
w
a
s
va
li
da
te
d on
H
id
de
n c
ont
r
a
di
c
ti
ons
s
ur
f
a
c
e
w
h
e
n
in
it
ia
ti
ve
s
c
r
os
s
f
unc
ti
ons
4.
1.
Dup
li
c
a
t
e
c
an
on
ical
e
n
t
it
ies
T
he
f
i
r
s
t
indi
c
a
tor
is
the
p
r
e
s
e
nc
e
of
the
s
a
me
r
e
a
l
-
wor
ld
e
nti
ty
a
s
mul
ti
ple
unr
e
c
onc
il
e
d
r
e
c
or
ds
.
W
he
n
c
us
tom
e
r
o
r
pr
oduc
t
e
xis
ts
thr
e
e
ti
mes
with
no
a
uthor
it
a
t
ive
r
e
s
olut
ion,
a
r
e
tr
ieva
l
or
r
e
a
s
oning
s
ys
tem
r
e
tur
ns
whic
he
ve
r
ve
r
s
ion
it
ha
ppe
ns
to
e
nc
ounter
,
a
nd
r
e
s
ult
s
a
ppe
a
r
incons
is
tent
to
the
us
e
r
.
T
ha
t
incons
is
tenc
y
r
e
a
ds
a
s
a
model
e
r
r
or
,
ye
t
it
s
or
igi
n
is
the
unr
e
s
olved
dupli
c
a
ti
on.
E
n
ti
ty
matc
hing
is
a
n
a
c
ti
ve
r
e
s
e
a
r
c
h
a
r
e
a
pr
e
c
is
e
ly
be
c
a
u
s
e
r
e
c
onc
il
iation
is
ha
r
d
a
t
e
nter
pr
is
e
s
c
a
le,
with
ne
ur
a
l
a
nd
pr
e
-
tr
a
ined
l
a
ngua
ge
model
a
ppr
oa
c
he
s
s
ur
ve
ye
d
e
xtens
ively
[
9]
,
a
nd
e
xplaina
ble
va
r
iants
de
ve
loped
s
o
that
the
matc
hi
ng
it
s
e
lf
c
a
n
be
a
udit
e
d
[
10]
.
M
a
s
ter
da
ta
mana
ge
ment
a
ddr
e
s
s
e
s
the
s
a
me
ne
e
d
f
r
om
the
gove
r
na
nc
e
s
ide
[
7]
.
P
r
a
c
ti
ti
one
r
e
xpe
r
ienc
e
a
s
s
e
mbl
ing
c
a
nonica
l
r
e
c
o
r
ds
a
c
r
os
s
s
e
pa
r
a
tely
buil
t
e
nter
pr
is
e
s
ys
tems
indi
c
a
tes
that
dupli
c
a
ti
on
is
r
a
r
e
ly
vis
ibl
e
unti
l
a
c
ons
umer
s
pa
nning
thos
e
s
ys
tem
s
e
xpos
e
s
it
.
T
he
diagnos
ti
c
tell
is
ther
e
f
or
e
not
the
dupli
c
a
ti
on
it
s
e
lf
but
the
pa
tt
e
r
n
of
a
ns
we
r
s
that
s
hif
t
de
pe
nding
on
whic
h
r
e
c
or
d
wa
s
r
e
tr
ieve
d.
4.
2.
B
r
ok
e
n
s
e
m
an
t
ic
c
h
ain
s
B
r
oke
n
s
e
mantic
c
ha
ins
f
or
m
the
s
e
c
ond
indi
c
a
tor
a
r
is
ing
whe
n
bus
ines
s
c
onc
e
pts
f
a
il
to
map
c
lea
nly
a
c
r
os
s
s
ys
tem
s
.
W
he
r
e
a
c
onc
e
pt
doe
s
not
tr
a
ns
late
be
twe
e
n
platf
or
ms
,
a
model
a
s
ke
d
to
c
ombi
ne
them
pr
oduc
e
s
incor
r
e
c
t
output
o
r
de
c
li
ne
s
to
a
c
t
on
the
a
mbi
guit
y.
T
he
c
onc
e
pt,
not
the
model
,
is
a
t
f
a
ult
.
I
nter
ope
r
a
bil
it
y
s
c
holar
s
hip
c
las
s
if
ies
thes
e
mi
s
matc
he
s
a
nd
c
onf
ir
ms
that
s
e
mantic
int
e
r
ope
r
a
bil
i
ty
i
s
dis
ti
nc
t
f
r
om,
a
nd
ha
r
de
r
than,
tec
hnica
l
c
onne
c
ti
vit
y
[
11
]
.
P
r
a
c
ti
ti
one
r
e
xpe
r
ienc
e
a
li
gning
s
c
he
mas
a
c
r
os
s
mul
ti
-
langua
ge
loca
li
z
a
ti
on
platf
o
r
ms
,
whe
r
e
a
s
ingl
e
c
ontent
c
onc
e
pt
ha
d
to
r
e
main
c
ohe
r
e
nt
a
c
r
os
s
nine
langua
ge
s
,
indi
c
a
tes
that
s
e
mantic
b
r
e
a
ks
a
r
e
f
r
e
qu
e
ntl
y
mi
s
take
n
f
o
r
tr
a
ns
lation
o
r
dis
play
de
f
e
c
ts
w
he
n
they
a
r
e
in
f
a
c
t
s
tr
uc
tu
r
a
l.
T
he
dis
ti
nc
ti
on
matter
s
,
be
c
a
us
e
a
tr
a
ns
lation
de
f
e
c
t
is
f
ixed
loca
ll
y
while
a
s
e
mantic
br
e
a
k
r
e
quir
e
s
a
li
gnment
a
c
r
os
s
e
ve
r
y
s
ys
tem
that
t
ouc
he
s
the
c
onc
e
pt.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
I
nfor
mation
ar
c
hit
e
c
tur
e
de
bt:
W
hy
legac
y
plat
for
m
s
c
he
ma
de
c
is
ions
c
ons
t
r
ain
…
(
M
ihi
r
Shah
)
2477
4.
3.
P
r
ove
n
an
c
e
gap
s
T
he
thi
r
d
indi
c
a
tor
is
the
a
r
r
ival
of
da
ta
without
li
n
e
a
ge
.
W
he
r
e
or
igi
n
a
nd
tr
a
ns
f
or
mation
his
tor
y
a
r
e
a
bs
e
nt,
a
udit
a
nd
c
ompl
ianc
e
us
e
c
a
s
e
s
c
a
nnot
be
s
uppor
ted
a
t
a
ll
,
a
nd
r
e
t
r
of
it
t
ing
l
inea
ge
a
f
ter
the
f
a
c
t
is
f
a
r
mor
e
e
xpe
ns
ive
than
ins
tr
umenting
it
in
plac
e
.
P
r
ove
na
nc
e
r
e
s
e
a
r
c
h
in
s
e
c
ur
it
y
a
nd
pr
ivac
y
doc
um
e
nts
both
the
va
lue
of
li
ne
a
ge
a
nd
the
dif
f
iculty
of
r
e
c
ons
tr
uc
ti
ng
it
onc
e
it
ha
s
be
e
n
los
t
[
12]
.
T
he
c
os
t
a
s
ymm
e
tr
y
is
qua
li
tative
but
c
ons
is
tent
a
c
r
os
s
e
nga
ge
ments
:
e
a
c
h
s
tage
of
de
lay
c
ompounds
the
r
e
tr
o
f
it
e
f
f
or
t,
be
c
a
us
e
mor
e
t
r
a
ns
f
or
mations
mus
t
be
r
e
ve
r
s
e
e
nginee
r
e
d
.
I
ns
tr
umenting
pr
ove
na
nc
e
while
a
f
low
is
be
ing
buil
t
is
c
ompar
a
ti
ve
ly
inexpe
ns
ive
a
nd
r
e
c
ons
tr
uc
ti
ng
it
ye
a
r
s
late
r
is
not.
P
r
a
c
ti
ti
one
r
e
xpe
r
ienc
e
a
c
r
o
s
s
mul
ti
-
s
ys
tem
e
nter
pr
is
e
platf
or
ms
indi
c
a
tes
that
the
f
low
s
mos
t
va
luable
to
AI
a
r
e
of
ten
the
one
s
whos
e
li
n
e
a
ge
wa
s
dis
c
a
r
de
d
e
a
r
li
e
s
t,
be
c
a
us
e
they
we
r
e
buil
t
f
or
s
pe
e
d
unde
r
t
r
a
ns
a
c
ti
ona
l
pr
e
s
s
ur
e
.
4.
4.
E
n
f
or
c
e
m
e
n
t
as
ym
m
e
t
r
y
E
nf
or
c
e
ment
a
s
ymm
e
tr
y,
the
f
our
th
indi
c
a
tor
,
a
ppe
a
r
s
whe
n
da
ta
qua
li
ty
r
ules
a
r
e
a
ppli
e
d
incons
is
tently
a
c
r
os
s
s
y
s
tems
.
W
he
n
one
s
ys
tem
e
nf
or
c
e
s
a
c
ons
tr
a
int
a
nd
a
nother
doe
s
not,
a
c
o
ns
umi
ng
model
mus
t
buil
d
br
it
t
le
a
da
pti
ve
logi
c
to
a
c
c
om
modate
both,
a
nd
that
log
ic
f
a
il
s
a
s
s
oon
a
s
e
it
he
r
s
ys
tem
dr
if
ts
.
T
he
a
s
ymm
e
tr
y
tends
to
c
ompound
ove
r
ti
me.
Da
ta
qua
li
ty
mana
ge
ment
methods
a
im
to
make
e
nf
or
c
e
ment
unif
o
r
m
a
c
r
os
s
s
our
c
e
s
[
13]
,
a
nd
t
he
ir
une
ve
n
a
dopti
on
in
p
r
a
c
ti
c
e
is
pr
e
c
is
e
ly
w
ha
t
thi
s
indi
c
a
tor
de
tec
ts
.
At
a
tec
hnica
l
leve
l,
r
e
s
olvi
ng
a
s
ymm
e
tr
ic
e
nf
or
c
e
ment
ge
ne
r
a
ll
y
invol
ve
s
one
of
a
s
mall
number
o
f
a
r
c
hit
e
c
tur
a
l
pa
tt
e
r
ns
:
a
c
e
ntr
a
li
z
e
d
va
li
da
ti
on
ga
tew
a
y
that
a
ll
wr
it
e
s
pa
s
s
thr
ough
r
e
ga
r
dles
s
of
or
igi
n
s
ys
tem,
s
c
he
ma
c
ontr
a
c
ts
that
f
or
mally
s
p
e
c
if
y
a
nd
ve
r
s
ion
the
c
ons
tr
a
int
s
e
a
c
h
pr
oduc
ing
s
ys
tem
a
gr
e
e
s
to
honor
,
or
a
s
ha
r
e
d
r
ule
e
ngine
that
mul
ti
ple
s
ys
tems
c
a
ll
r
a
ther
than
e
a
c
h
r
e
-
im
ple
menting
va
li
da
ti
on
indepe
nde
ntl
y.
None
of
thes
e
pa
tt
e
r
ns
is
univer
s
a
ll
y
a
ppli
c
a
ble,
a
nd
the
c
hoice
de
pe
nds
on
how
much
c
ontr
ol
the
r
e
media
ti
on
e
f
f
or
t
ha
s
ove
r
e
a
c
h
s
our
c
e
s
ys
tem;
lega
c
y
s
ys
tems
that
c
a
nnot
be
modi
f
ied
typi
c
a
ll
y
r
e
quir
e
a
ga
tew
a
y
-
ba
s
e
d
a
ppr
oa
c
h
a
ppli
e
d
a
t
the
int
e
gr
a
ti
on
laye
r
r
a
ther
than
a
t
the
s
ou
r
c
e
.
T
he
a
da
pti
ve
logi
c
on
the
c
ons
umi
ng
s
ide
is
it
s
e
lf
a
s
ympt
om
wor
th
wa
tching
f
o
r
,
be
c
a
us
e
it
s
pr
e
s
e
nc
e
s
ignals
that
ups
tr
e
a
m
e
nf
or
c
e
ment
c
a
nnot
be
tr
us
ted.
4.
5.
Cons
u
m
e
r
as
s
u
m
p
t
ion
d
iver
ge
n
c
e
T
he
f
i
f
th
indi
c
a
tor
is
the
qu
iet
pr
o
li
f
e
r
a
ti
on
of
pr
i
va
te,
unc
oor
dinate
d
wor
ka
r
ounds
in
downs
tr
e
a
m
s
ys
tems
.
E
a
c
h
c
ons
umer
c
ompens
a
tes
f
or
ups
tr
e
a
m
de
f
e
c
ts
in
it
s
own
wa
y,
a
nd
the
c
ompens
a
ti
ons
diver
ge
be
c
a
us
e
no
one
c
oor
dinate
s
them.
T
his
be
ha
vior
hides
the
de
bt
unti
l
a
c
r
os
s
-
f
unc
ti
ona
l
ini
ti
a
ti
ve
f
or
c
e
s
the
wor
ka
r
ounds
int
o
c
ontac
t,
a
t
whic
h
point
their
c
on
tr
a
dictions
s
ur
f
a
c
e
togethe
r
.
T
he
indi
c
a
tor
is
the
h
a
r
de
s
t
of
the
f
ive
to
obs
e
r
ve
dir
e
c
tl
y
,
be
c
a
us
e
e
a
c
h
wor
ka
r
ound
looks
loca
ll
y
r
e
a
s
ona
ble
a
nd
is
of
ten
undoc
umente
d.
I
ts
diagnos
ti
c
va
lue
is
high
pr
e
c
is
e
ly
be
c
a
us
e
it
s
dis
c
ove
r
y
us
ua
ll
y
c
oincide
s
with
the
mom
e
nt
a
n
a
mbi
ti
ous
AI
ini
ti
a
ti
ve
s
talls
.
5.
T
HE
CA
P
AB
I
L
I
T
Y
CE
I
L
I
NG
HYP
OT
HE
S
I
S
T
his
pa
pe
r
a
dva
nc
e
s
one
c
e
ntr
a
l
p
r
opos
it
ion,
of
f
e
r
e
d
a
s
a
theor
e
ti
c
a
l
c
laim
de
r
ived
f
r
om
c
onve
r
ging
pr
a
c
ti
ti
one
r
a
nd
li
ter
a
tu
r
e
e
videnc
e
r
a
ther
than
a
s
a
n
e
xpe
r
im
e
ntally
va
li
da
ted
r
e
s
ult
:
inf
o
r
mation
a
r
c
hit
e
c
tur
e
de
bt
im
pos
e
s
a
c
a
pa
bil
it
y
c
e
il
ing
that
model
s
e
lec
ti
on
a
lone
doe
s
not
a
ppe
a
r
a
ble
to
e
xc
e
e
d.
T
he
da
t
a
-
c
e
ntr
ic
view
of
a
r
ti
f
icia
l
int
e
ll
igenc
e
ho
lds
that
s
ys
tema
ti
c
e
nginee
r
ing
o
f
da
ta,
r
a
ther
than
of
models
,
incr
e
a
s
ingl
y
gove
r
ns
outcome
s
[
14]
,
a
nd
the
c
e
il
ing
hypothes
is
is
a
s
tr
uc
tur
a
l
c
or
oll
a
r
y
s
pe
c
if
ic
to
e
nter
pr
is
e
s
ubs
tr
a
te.
T
he
hypothes
is
is
s
c
ope
d
mos
t
dir
e
c
tl
y
to
r
e
tr
ieva
l
-
a
ugmente
d,
r
e
a
s
oning,
a
nd
a
ge
nti
c
a
r
c
hit
e
c
tur
e
s
,
whic
h
c
ons
ume
c
r
os
s
-
e
nti
ty
a
nd
c
r
os
s
-
s
ys
tem
da
ta
a
t
inf
e
r
e
nc
e
ti
me
a
nd
ther
e
f
or
e
inher
it
s
ubs
tr
a
te
de
f
e
c
ts
dir
e
c
tl
y
int
o
their
output
s
;
na
r
r
owe
r
p
r
e
dictive
models
tr
a
i
ne
d
on
a
s
ingl
e
c
u
r
a
ted
table
a
r
e
c
ompar
a
t
ively
i
ns
ulate
d,
s
ince
c
ur
a
ti
on
f
or
tr
a
ini
ng
c
a
n
mas
k,
without
r
e
s
ol
ving,
the
unde
r
lyi
ng
mi
s
a
li
gnment.
T
hr
e
e
pa
tt
e
r
ns
il
lus
tr
a
te
the
mec
ha
nis
m
withi
n
it
s
s
c
ope
d
r
a
nge
.
A
r
e
tr
ieva
l
-
a
ugmente
d
s
y
s
tem
ope
r
a
ti
ng
ove
r
dupli
c
a
te,
un
r
e
c
onc
il
e
d
e
nti
ti
e
s
r
e
tur
ns
incons
is
tent
a
ns
we
r
s
r
e
ga
r
dles
s
of
the
ge
ne
r
a
ti
ve
model,
be
c
a
us
e
the
r
e
tr
ieva
l
c
or
pus
it
s
e
lf
dis
a
gr
e
e
s
with
it
s
e
lf
,
a
pa
tt
e
r
n
c
ons
is
tent
with
s
ur
ve
y
e
vid
e
nc
e
that
ha
ll
uc
ination
in
lar
ge
langua
ge
models
is
f
r
e
que
n
tl
y
tr
a
c
e
a
ble
to
c
onf
l
icts
or
ga
ps
in
the
s
our
c
e
m
a
ter
ial
a
s
ys
tem
r
e
tr
ieve
s
f
r
o
m
r
a
ther
than
to
the
ge
ne
r
a
ti
on
s
tep
it
s
e
lf
[
15
]
.
An
a
utonom
ous
a
ge
nt
r
e
a
s
on
ing
ove
r
diver
ge
nt
s
e
mantics
r
e
a
c
he
s
c
ontr
a
dictions
it
c
a
nnot
r
e
s
olve,
be
c
a
us
e
the
c
ontr
a
diction
li
ve
s
in
the
da
ta
model
r
a
ther
than
in
the
r
e
a
s
oning.
A
model
op
e
r
a
ti
ng
on
p
r
ove
na
nc
e
-
ga
ppe
d
da
ta
pr
oduc
e
s
output
s
that
c
a
nnot
be
a
udit
e
d,
be
c
a
us
e
the
li
ne
a
ge
r
e
quir
e
d
f
o
r
a
udit
wa
s
not
c
a
ptur
e
d
[
12]
.
E
a
c
h
of
thes
e
is
a
s
ubs
tr
a
te
f
a
il
ur
e
c
omm
on
ly
mi
s
diagnos
e
d
a
s
a
model
f
a
il
ur
e
,
a
nd
s
tr
uc
tu
r
e
d
a
na
lys
e
s
of
AI
p
r
ojec
t
f
a
i
lur
e
a
r
e
c
ons
is
tent
with
that
mi
s
a
tt
r
ibut
ion
[
1]
.
T
he
c
e
il
ing
c
a
n
be
r
a
is
e
d
only
by
pa
ying
down
the
de
bt,
thr
ough
s
c
he
ma
r
a
ti
ona
li
z
a
ti
on,
s
e
mantic
a
li
gnment,
e
nti
ty
r
e
s
olut
ion,
a
nd
p
r
ove
na
nc
e
ins
tr
umenta
ti
on.
S
wa
pping
one
model
f
or
a
s
tr
onge
r
o
ne
lea
ve
s
the
c
e
il
ing
untouche
d
withi
n
it
s
s
c
ope
d
r
a
nge
,
whic
h
is
why
c
a
pa
bil
it
y
-
f
oc
us
e
d
pr
ogr
a
ms
that
ignor
e
the
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
16
,
No.
5
,
Oc
tober
20
26
:
2473
-
2482
2478
s
ubs
tr
a
te
tend
to
plate
a
u.
A
c
omm
on
f
ield
pa
tt
e
r
n
make
s
the
mi
s
a
tt
r
ibut
ion
c
onc
r
e
te.
A
pil
ot
pe
r
f
o
r
m
s
we
ll
on
a
c
ur
a
ted
s
li
c
e
of
da
ta,
the
pr
ogr
a
m
s
c
a
les
it
to
the
f
ull
e
s
tate
,
qua
li
ty
de
gr
a
de
s
,
a
nd
the
de
gr
a
da
ti
on
is
a
tt
r
ibut
e
d
to
the
model
r
a
ther
than
to
the
s
ubs
tr
a
te
that
the
c
u
r
a
ted
s
li
c
e
ha
d
quietly
hidden
.
T
he
c
o
r
r
e
c
ti
ve
is
to
tr
e
a
t
dis
c
ove
r
y
a
s
s
ub
s
tr
a
te
a
s
s
e
s
s
ment
f
ir
s
t
a
nd
model
e
va
luation
s
e
c
ond,
whic
h
inver
ts
the
us
u
a
l
or
de
r
of
e
nter
pr
is
e
AI
planning.
As
a
theor
e
ti
c
a
l
pr
opos
it
i
on,
the
hypothes
is
is
s
tate
d
he
r
e
to
be
tes
ted,
not
a
s
s
umed;
s
e
c
ti
on
8
r
e
tur
ns
to
wha
t
a
n
e
xpe
r
im
e
ntal
e
va
luatio
n
of
i
t
would
r
e
quir
e
.
6.
T
HE
RE
M
E
D
I
AT
I
ON
F
RA
M
E
WORK
6.
1.
S
e
q
u
e
n
c
e
d
r
e
m
e
d
iat
ion
logi
c
R
e
media
ti
on
s
hould
be
s
e
que
nc
e
d
by
AI
-
c
a
pa
bil
it
y
im
pa
c
t
r
a
ther
than
by
e
a
s
e
o
f
f
ix,
be
c
a
us
e
the
objec
ti
ve
is
to
r
a
is
e
the
c
e
il
ing
whe
r
e
models
a
r
e
a
c
tually
blocke
d.
T
hr
e
e
wa
ve
s
f
oll
ow
f
r
om
that
p
r
i
nc
ipl
e
a
s
a
de
f
a
ult
he
ur
is
ti
c
r
a
ther
than
a
n
invar
iant
r
u
le.
T
h
e
f
i
r
s
t
wa
ve
r
e
s
olves
c
a
nonica
l
e
nti
ti
e
s
f
o
r
the
e
nti
t
ies
mos
t
he
a
vil
y
us
e
d
by
A
I
c
ons
umer
s
,
typi
c
a
ll
y
c
us
tom
e
r
,
e
mpl
oye
e
,
p
r
oduc
t,
c
ontr
a
c
t,
a
nd
loca
ti
on.
T
h
e
s
e
c
ond
wa
ve
a
li
gns
s
e
mantics
f
or
the
de
pe
nde
nt
c
onc
e
p
ts
that
thos
e
e
nti
ti
e
s
r
e
ly
on
,
including
ti
me,
ge
ogr
a
phy,
hier
a
r
c
hy,
a
nd
li
f
e
c
yc
le
s
tate
.
T
he
thi
r
d
wa
ve
i
ns
tr
uments
pr
ove
na
nc
e
f
or
the
da
ta
f
lows
that
f
e
e
d
AI
c
ons
umer
s
,
s
o
that
output
s
be
c
ome
a
udit
a
ble.
T
his
de
f
a
ult
or
de
r
ing
a
s
s
umes
s
e
mantic
diver
ge
n
c
e
a
nd
pr
ove
na
nc
e
los
s
a
r
e
moder
a
te
e
nough
that
e
nti
ty
r
e
s
olut
ion
c
a
n
p
r
oc
e
e
d
without
them;
th
a
t
a
s
s
umpt
ion
doe
s
not
a
lwa
ys
hold.
W
he
r
e
s
e
mantic
diver
ge
nc
e
is
s
e
ve
r
e
e
nough
that
the
s
a
me
identi
f
i
e
r
r
e
s
olves
to
c
onc
e
ptually
di
f
f
e
r
e
nt
thi
ngs
a
c
r
os
s
s
ys
tems
,
pa
r
ti
a
l
s
e
mantic
tr
iage
s
hould
pr
e
c
e
de
f
ull
e
nti
ty
r
e
s
olut
ion,
s
ince
r
e
s
olvi
ng
e
nti
ti
e
s
unde
r
a
n
uns
table
c
onc
e
pt
de
f
ini
ti
on
pr
oduc
e
s
a
c
a
nonica
l
r
e
c
or
d
that
it
s
e
lf
ne
e
ds
to
be
r
e
done
onc
e
the
c
onc
e
pt
is
c
lar
if
ied
.
S
im
il
a
r
ly,
whe
r
e
pr
ove
na
nc
e
is
s
o
s
pa
r
s
e
that
the
or
igi
n
of
c
onf
l
icting
r
e
c
or
ds
c
a
nnot
be
e
s
tablis
he
d,
a
mi
nim
a
l
pr
ove
na
nc
e
ba
s
e
li
ne
—
e
ve
n
s
im
ple
s
our
c
e
-
s
ys
t
e
m
tagging
without
f
ull
li
ne
a
ge
—
is
of
ten
a
pr
a
c
ti
c
a
l
pr
e
r
e
quis
it
e
f
o
r
tr
us
twor
thy
e
nti
ty
r
e
s
olut
ion
,
s
ince
r
e
s
olvi
ng
dup
li
c
a
tes
r
e
quir
e
s
s
ome
ba
s
is
f
o
r
judgi
ng
whic
h
r
e
c
or
d
is
a
uthor
it
a
ti
ve
.
I
n
thes
e
s
e
ve
r
e
c
a
s
e
s
the
thr
e
e
wa
ve
s
a
r
e
be
tt
e
r
unde
r
s
tood
a
s
it
e
r
a
ti
ve
a
nd
pa
r
ti
a
ll
y
ove
r
lapping
r
a
ther
than
s
tr
ictly
s
e
que
nti
a
l.
T
a
ble
3
r
e
c
or
ds
the
tar
ge
t
a
nd
r
a
ti
ona
le
o
f
e
a
c
h
wa
ve
unde
r
the
de
f
a
ult
c
a
s
e
.
E
nt
it
y
r
e
s
olut
ion
r
e
s
e
a
r
c
h
of
f
e
r
s
matur
e
buil
ding
blocks
f
or
the
f
i
r
s
t
wa
ve
,
including
pr
e
-
tr
a
ined
langua
ge
model
matc
he
r
s
[
18]
,
dua
l
-
objec
ti
ve
f
ine
-
tuni
ng
of
tr
a
ns
f
or
me
r
e
nc
ode
r
s
[
19]
,
s
e
lf
-
s
upe
r
vis
e
d
mul
ti
-
f
e
a
tur
e
c
ol
labor
a
ti
on
f
r
a
mew
or
ks
[
20
]
,
a
nd
pa
r
a
ll
e
li
z
e
d
r
e
s
olut
ion
ove
r
dyna
mi
c
da
ta
[
21
]
.
S
e
que
nc
ing
thi
s
wa
y
a
c
c
e
pts
t
ha
t
s
ome
e
a
s
y
f
ixes
will
wa
it
on
the
gr
ounds
that
a
n
e
a
s
y
f
ix
with
no
c
a
pa
bil
it
y
im
pa
c
t
is
not
the
c
ons
tr
a
int
.
T
a
ble
3.
S
e
que
nc
e
d
r
e
media
ti
on
wa
ve
s
(
de
f
a
ult
c
a
s
e
)
W
a
ve
T
a
r
ge
t
R
a
ti
ona
le
W
a
ve
1
C
a
noni
c
a
l
e
nt
it
y r
e
s
ol
ut
io
n f
or
hi
gh
-
AI
-
us
e
e
nt
it
ie
s
(
c
us
to
me
r
, e
mpl
oye
e
, pr
oduc
t,
c
ont
r
a
c
t,
l
oc
a
ti
on)
D
upl
ic
a
te
e
nt
it
ie
s
c
a
p r
e
tr
ie
va
l
a
nd r
e
a
s
oni
ng qua
li
ty
f
ir
s
t;
r
e
s
ol
vi
ng t
he
m r
a
is
e
s
t
he
c
e
il
in
g mos
t
W
a
ve
2
S
e
ma
nt
ic
a
li
gnme
nt
f
or
de
pe
nde
nt
c
onc
e
pt
s
(
ti
me
,
ge
ogr
a
phy, hie
r
a
r
c
hy, l
if
e
c
yc
le
s
ta
te
)
A
li
gne
d e
nt
it
ie
s
s
ti
ll
f
a
il
w
he
n t
he
c
onc
e
pt
s
r
e
la
ti
ng t
he
m
di
ve
r
ge
a
c
r
os
s
s
ys
te
ms
W
a
ve
3
P
r
ove
na
nc
e
i
ns
tr
ume
nt
a
ti
on f
or
A
I
-
f
e
e
di
ng da
ta
f
lo
w
s
A
udi
ta
bi
li
ty
a
nd c
ompl
ia
nc
e
r
e
qui
r
e
l
in
e
a
ge
t
ha
t
tr
a
ns
a
c
ti
ona
l
pur
pos
e
s
ne
ve
r
c
a
pt
ur
e
d
6.
2.
B
r
ownf
ield
-
gr
e
e
n
f
ield
t
r
ad
e
o
f
f
A
gr
e
e
nf
ield
,
A
I
-
na
ti
ve
platf
or
m
is
a
n
a
ppe
a
li
ng
r
e
s
pons
e
,
ye
t
it
tends
to
unde
r
e
s
ti
mate
the
c
os
t
o
f
maintaining
two
e
nvir
onments
while
the
lega
c
y
e
s
t
a
te
c
onti
nue
s
to
s
e
r
ve
li
ve
tr
a
ns
a
c
ti
ons
.
A
hybr
id
a
ppr
oa
c
h
a
ppe
a
r
s
mor
e
de
f
e
ns
ibl
e
f
or
mos
t
e
nter
pr
is
e
s
.
A
thi
n
c
a
nonica
l
laye
r
buil
t
ove
r
the
lega
c
y
platf
or
ms
c
a
n
e
xpos
e
r
e
s
olved
e
nti
ti
e
s
,
a
li
gne
d
s
e
mantics
,
a
nd
ins
tr
umente
d
pr
ove
na
nc
e
to
AI
c
ons
umer
s
,
while
th
e
lega
c
y
s
ys
tems
ke
e
p
s
e
r
ving
their
tr
a
ns
a
c
ti
ona
l
pur
pos
e
u
nc
ha
nge
d
be
ne
a
th
it
.
I
n
p
r
a
c
ti
c
e
,
thi
s
laye
r
is
typi
c
a
ll
y
ke
pt
c
ur
r
e
nt
th
r
o
ugh
e
ve
nt
-
dr
iven
c
ha
nge
-
da
ta
-
c
a
ptur
e
f
e
e
ds
f
r
om
the
s
our
c
e
s
ys
tems
r
a
ther
than
s
ync
hr
onous
qu
e
r
ies
,
whic
h
a
voids
a
dding
r
e
a
d
or
wr
i
te
late
nc
y
to
the
unde
r
lyi
ng
t
r
a
ns
a
c
ti
ona
l
s
ys
tems
;
r
e
s
e
a
r
c
h
on
c
ha
nge
-
da
ta
-
c
a
ptur
e
a
r
c
hit
e
c
tur
e
s
c
onf
ir
ms
that
t
r
a
dit
ional
poll
ing
-
ba
s
e
d
a
ppr
oa
c
he
s
s
tr
uggle
to
de
li
ve
r
c
ha
ng
e
s
with
low
late
nc
y,
whic
h
is
the
unde
r
lyi
ng
r
e
a
s
on
e
ve
nt
-
dr
iven
f
e
e
ds
a
r
e
pr
e
f
e
r
r
e
d
f
or
thi
s
laye
r
[
22]
.
T
h
e
tr
a
de
of
f
is
that
the
c
a
nonica
l
laye
r
ope
r
a
tes
on
e
ve
ntual
r
a
ther
than
im
media
te
c
ons
is
tenc
y,
whic
h
is
a
c
c
e
ptable
f
or
mos
t
AI
-
c
ons
umpt
ion
us
e
c
a
s
e
s
(
r
e
tr
ieva
l,
r
e
a
s
oning,
r
e
por
ti
ng
)
but
would
not
s
uit
us
e
c
a
s
e
s
r
e
quir
ing
s
tr
ict
r
e
a
l
-
ti
me
c
ons
is
tenc
y
with
the
tr
a
n
s
a
c
ti
ona
l
s
our
c
e
.
Know
ledge
gr
a
ph
int
e
gr
a
ti
on
laye
r
s
de
mons
tr
a
te
thi
s
pa
tt
e
r
n
f
or
manuf
a
c
tur
ing
da
ta,
whe
r
e
a
s
e
mantic
laye
r
pr
ovides
unif
ied
a
c
c
e
s
s
without
r
e
plac
ing
the
s
our
c
e
s
ys
tems
it
dr
a
ws
f
r
om
[
23
]
.
T
he
da
ta
mes
h
pa
r
a
digm
of
f
e
r
s
a
n
or
ga
niza
ti
ona
l
a
na
log
to
thi
s
tec
hnica
l
a
ppr
oa
c
h,
tr
e
a
ti
ng
da
ta
a
s
a
p
r
oduc
t
owne
d
by
domain
tea
ms
unde
r
f
e
de
r
a
ted
gove
r
na
nc
e
r
a
th
e
r
than
c
e
nt
r
a
li
z
ing
it
outr
ight
,
a
nd
a
s
ys
tema
ti
c
r
e
view
of
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
I
nfor
mation
ar
c
hit
e
c
tur
e
de
bt:
W
hy
legac
y
plat
for
m
s
c
he
ma
de
c
is
ions
c
ons
t
r
ain
…
(
M
ihi
r
Shah
)
2479
indus
tr
ial
da
ta
mes
h
p
r
a
c
ti
c
e
c
onf
ir
ms
do
main
owne
r
s
hip
a
nd
f
e
de
r
a
ted
gove
r
na
nc
e
a
s
the
pa
r
a
digm
's
de
f
ini
ng
pr
inciples
[
24
]
.
T
he
hybr
id
ke
e
ps
r
e
media
ti
on
incr
e
menta
l,
whic
h
matter
s
be
c
a
us
e
a
f
ull
r
e
pl
a
c
e
ment
c
onc
e
ntr
a
tes
r
is
k
int
o
a
s
ingl
e
c
utover
that
mos
t
or
ga
niza
ti
ons
c
a
nnot
a
bs
or
b.
6.
3.
Wor
k
e
d
il
lu
s
t
r
at
ive
s
c
e
n
ar
io
T
he
f
oll
owing
is
a
c
ompos
it
e
il
lus
tr
a
ti
ve
s
c
e
na
r
io
c
ons
tr
uc
ted
to
de
mons
tr
a
te
how
the
diagnos
ti
c
a
nd
s
e
que
nc
ing
logi
c
a
bove
would
be
a
ppli
e
d;
it
d
oe
s
not
r
e
por
t
f
indi
ngs
f
r
om
a
s
pe
c
if
ic
dis
c
los
e
d
c
li
e
nt
e
nga
ge
ment
or
a
r
e
lea
s
e
d
e
mpi
r
ica
l
da
tas
e
t,
a
li
mi
tation
dis
c
us
s
e
d
f
ur
ther
in
s
e
c
ti
on
8
.
C
ons
ider
a
ge
ne
r
ic
e
nter
pr
is
e
with
thr
e
e
c
us
tom
e
r
-
f
a
c
ing
s
ys
tems
a
nd
dupli
c
a
te,
unr
e
c
onc
il
e
d
c
us
tom
e
r
r
e
c
or
ds
s
pr
e
a
d
a
c
r
os
s
them.
A
dis
c
ove
r
y
a
s
s
e
s
s
ment
s
c
or
e
s
e
a
c
h
indi
c
a
tor
by
s
e
ve
r
it
y
a
nd
yields
a
s
e
que
nc
e
d
r
e
c
omm
e
nda
ti
on.
T
a
ble
4
pr
e
s
e
nts
the
c
onde
ns
e
d
a
s
s
e
s
s
ment,
c
r
os
s
-
r
e
f
e
r
e
nc
e
d
to
the
f
a
il
ur
e
modes
in
T
a
ble
2.
T
he
r
e
c
omm
e
nde
d
pa
th
r
e
s
olves
the
c
us
tom
e
r
e
nti
ty
f
ir
s
t,
a
li
gns
the
de
pe
nde
nt
s
e
mantics
ne
xt,
a
nd
ins
tr
uments
pr
ove
na
nc
e
f
or
the
AI
-
f
e
e
ding
f
lows
las
t,
with
a
n
incr
e
menta
l
a
s
s
is
tant
de
ploym
e
nt
r
e
lea
s
e
d
a
s
e
a
c
h
wa
ve
c
ompl
e
tes
s
o
that
va
lue
is
de
mons
tr
a
ted
c
onti
nuou
s
ly
r
a
ther
than
de
f
e
r
r
e
d
to
a
s
ingl
e
d
is
tant
c
utover
.
T
a
ble
4.
W
or
ke
d
il
lus
tr
a
ti
ve
s
c
e
na
r
io
—
s
e
ve
r
it
y
a
s
s
e
s
s
ment
I
ndi
c
a
to
r
S
e
ve
r
it
y
B
r
ie
f
f
in
di
ng
C
or
r
e
s
ponding T
a
bl
e
2 f
a
il
ur
e
mode
D
upl
ic
a
te
c
a
noni
c
a
l
e
nt
it
ie
s
H
ig
h
C
us
to
me
r
e
xi
s
ts
a
c
r
os
s
t
hr
e
e
s
ys
te
m
s
w
it
h no a
ut
hor
it
a
ti
ve
r
e
s
ol
ut
io
n
I
nc
ons
is
te
nt
r
e
tr
ie
va
l/
r
e
a
s
oni
ng
r
e
s
ul
ts
r
e
a
d a
s
mod
e
l
e
r
r
or
B
r
oke
n s
e
ma
nt
ic
c
h
a
in
s
M
e
di
um
A
c
c
ount
s
ta
tu
s
a
nd l
if
e
c
yc
le
s
t
a
te
de
f
in
e
d
di
f
f
e
r
e
nt
ly
pe
r
s
ys
te
m
I
nc
or
r
e
c
t
c
ombi
ne
d output
s
, or
mode
l
de
c
li
ne
s
t
o a
c
t
P
r
ove
na
nc
e
ga
ps
H
ig
h
I
nt
e
gr
a
ti
on f
lo
w
s
di
s
c
a
r
d l
in
e
a
ge
;
a
udi
t
us
e
c
a
s
e
s
un
s
uppor
te
d
O
ut
put
s
c
a
nnot
be
a
udi
te
d or
ma
de
c
ompl
ia
nt
E
nf
or
c
e
me
nt
a
s
ymm
e
tr
y
M
e
di
um
V
a
li
da
ti
on e
nf
or
c
e
d i
n one
s
ys
te
m,
a
bs
e
nt
in
t
he
ot
he
r
t
w
o
B
r
it
tl
e
a
da
pt
iv
e
l
ogi
c
f
a
il
s
a
s
s
ys
te
ms
dr
if
t
C
ons
ume
r
a
s
s
umpt
io
n di
ve
r
ge
nc
e
L
ow
I
s
ol
a
te
d w
or
ka
r
ounds
pr
e
s
e
nt
but
not
ye
t
in
c
onf
li
c
t
H
id
de
n c
ont
r
a
di
c
ti
ons
s
ur
f
a
c
e
w
h
e
n
in
it
ia
ti
ve
s
c
r
os
s
f
unc
ti
ons
7.
ORGANI
Z
AT
I
ONAL
DY
NA
M
I
CS
AN
D
S
E
CT
OR
AP
P
L
I
CA
T
I
ONS
T
he
de
b
t
is
c
r
e
a
te
d
b
y
de
c
e
nt
r
a
l
ize
d
de
c
is
ions
a
nd
r
e
pa
i
d
onl
y
thr
ough
c
e
n
tr
a
li
z
e
d
c
oor
dina
ti
o
n,
whic
h
p
r
o
duc
e
s
a
s
t
r
uc
t
ur
a
l
m
is
a
li
g
nmen
t
o
f
in
c
e
nti
ve
s
.
T
he
a
r
c
h
it
e
c
t
or
c
hie
f
da
ta
o
f
f
ice
r
wh
o
owns
r
e
med
iati
on
is
r
a
r
e
ly
the
pa
r
ty
t
ha
t
be
ne
f
it
s
mos
t
f
r
om
i
t,
a
nd
the
tea
ms
whos
e
s
c
he
ma
c
hoice
s
c
r
e
a
te
d
the
de
bt
a
r
e
r
a
r
e
ly
the
one
s
a
s
ke
d
to
f
u
nd
it
s
r
e
pa
ir
.
E
x
e
c
u
ti
ve
s
pons
o
r
s
hip
is
the
r
e
f
or
e
a
pr
e
c
ondi
ti
o
n,
a
nd
i
t
is
be
s
t
s
e
c
ur
e
d
wit
h
c
on
c
r
e
te
de
mons
t
r
a
t
ions
of
un
r
e
a
li
z
e
d
AI
va
lue
r
a
the
r
t
ha
n
with
a
bs
t
r
a
c
t
a
r
gume
nts
a
b
out
s
c
he
ma
hygiene
.
P
r
a
c
t
it
ione
r
e
xpe
r
ienc
e
lea
din
g
lar
ge
-
s
c
a
le
pr
oc
e
s
s
-
a
r
c
hit
e
c
tu
r
e
s
tanda
r
diz
a
ti
on
a
c
r
os
s
tens
o
f
thous
a
nds
of
e
nte
r
p
r
is
e
p
r
oc
e
s
s
maps
ind
ica
tes
t
ha
t
r
e
medi
a
ti
o
n
s
uc
c
e
e
ds
w
he
n
i
t
is
f
r
a
med
a
s
c
a
pa
bil
it
y
unlocke
d,
not
a
s
c
lea
nup
pe
r
f
or
med
.
T
he
ince
nti
ve
p
r
ob
lem
is
ge
nuine
,
be
c
a
us
e
de
bt
-
c
r
e
a
ti
ng
tea
ms
a
r
e
a
s
ke
d
to
be
a
r
a
r
e
med
iati
on
c
os
t
f
or
a
de
b
t
they
d
id
no
t
kn
owing
ly
c
r
e
a
te
.
T
he
m
is
matc
h
is
s
t
r
uc
tu
r
a
l
,
not
pe
r
s
ona
l.
S
e
c
tor
c
ontext
s
ha
pe
s
indi
c
a
tor
s
e
ve
r
it
y.
W
it
hi
n
f
inanc
ial
s
e
r
vice
s
,
c
us
tom
e
r
e
nti
ty
r
e
s
olut
ion,
tr
a
ns
a
c
ti
on
li
ne
a
ge
,
a
nd
r
e
gulator
y
r
e
por
ti
ng
p
r
ov
e
na
nc
e
tend
to
pr
e
s
e
nt
a
ll
f
ive
indi
c
a
tor
s
a
t
high
s
e
ve
r
it
y,
a
nd
r
e
por
ti
ng
obli
ga
ti
ons
make
p
r
ove
na
nc
e
ga
ps
e
s
pe
c
ially
c
os
tl
y
to
lea
ve
una
ddr
e
s
s
e
d
[
12]
.
I
n
he
a
lt
hc
a
r
e
,
pa
ti
e
nt
e
nti
ty
r
e
s
olut
ion
a
nd
c
li
nica
l
taxonomy
a
li
gnment
a
r
e
c
e
ntr
a
l
c
onc
e
r
ns
,
ye
t
r
e
c
onc
il
iation
is
c
ompl
ica
ted
by
r
e
gulator
y
c
ons
tr
a
int
s
that
li
mi
t
ho
w
f
r
e
e
ly
r
e
c
or
ds
may
be
mer
ge
d.
Gove
r
nment
s
e
tt
ings
a
dd
a
f
ur
ther
w
r
inkl
e
,
be
c
a
us
e
c
it
ize
n
e
nti
ty
r
e
s
olut
io
n
a
c
r
os
s
a
ge
nc
ies
a
nd
s
e
r
vice
-
c
a
tegor
y
a
li
gnment
mee
ts
a
tens
ion
be
twe
e
n
tr
a
ns
pa
r
e
nc
y
a
nd
pr
ivac
y,
a
nd
pr
oc
ur
e
ment
s
tr
uc
tur
e
s
r
a
is
e
c
oor
dination
c
os
t
s
ti
ll
higher
.
I
n
both
he
a
lt
hc
a
r
e
a
nd
gove
r
n
ment
c
ontexts
,
pr
ivac
y
-
pr
e
s
e
r
ving
a
r
c
hit
e
c
tur
e
int
e
r
s
e
c
ts
dir
e
c
tl
y
wi
th
s
e
mantic
a
li
gnment
wor
k:
da
ta
mi
nim
iza
ti
on
pr
inciples
c
o
ns
tr
a
in
how
much
of
a
r
e
c
or
d
c
a
n
be
s
ha
r
e
d
a
c
r
os
s
the
s
ys
tems
be
ing
a
li
gne
d,
a
nd
ne
e
d
-
to
-
know
a
c
c
e
s
s
c
ontr
ols
mea
n
s
e
mantic
a
li
gnment
of
ten
mus
t
be
a
c
hieve
d
thr
ough
f
e
de
r
a
ted
o
r
pr
ivac
y
-
pr
e
s
e
r
ving
matc
hing
tec
hniques
.
T
he
s
e
a
r
e
a
ppli
c
a
ti
ons
of
the
f
r
a
mew
o
r
k
r
a
ther
than
de
li
ve
r
e
d
e
nga
ge
ments
,
a
nd
the
va
lue
of
the
f
r
a
mew
or
k
is
that
it
s
f
ive
indi
c
a
tor
s
tr
a
ns
f
e
r
a
c
r
os
s
a
ll
thr
e
e
without
modi
f
ica
ti
on
.
8.
L
I
M
I
T
AT
I
ONS
AN
D
F
UT
UR
E
RE
S
E
AR
CH
T
he
f
r
a
mew
or
k
is
qua
li
tative
a
nd
diagnos
ti
c
r
a
th
e
r
than
qua
nti
tative,
a
nd
s
e
c
ti
on
3
de
s
c
r
ibes
the
qua
li
tative
s
ynthes
is
pr
oc
e
s
s
by
whic
h
it
wa
s
de
r
ived;
that
pr
oc
e
s
s
is
a
de
f
e
n
s
ibl
e
s
tar
ti
ng
point
but
not
a
s
ubs
ti
tut
e
f
or
a
pr
e
-
r
e
gis
ter
e
d
e
mpi
r
ica
l
s
tudy,
a
nd
the
indi
c
a
tor
s
e
t
s
hould
be
tr
e
a
ted
a
s
a
hypot
he
s
is
f
or
tes
ti
ng
r
a
ther
than
a
s
e
tt
led
taxonomy.
T
he
c
a
pa
bil
it
y
c
e
il
ing
hypothes
is
in
s
e
c
ti
on
5
is
s
im
il
a
r
ly
a
theor
e
ti
c
a
l
pr
opos
it
ion.
T
e
s
ti
ng
it
e
xpe
r
im
e
ntally
would
r
e
q
uir
e
,
a
t
mi
nim
um
,
a
c
ontr
oll
e
d
c
ompar
is
on
holdi
ng
model
c
hoice
c
ons
tant
while
va
r
y
ing
s
ubs
tr
a
te
qua
li
ty
a
c
r
os
s
the
f
ive
indi
c
a
tor
s
,
with
outcome
mea
s
ur
e
s
a
ppr
opr
iate
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
16
,
No.
5
,
Oc
tober
20
26
:
2473
-
2482
2480
to
the
A
I
s
ys
tem
type
(
tas
k
a
c
c
ur
a
c
y
or
c
ons
is
tenc
y
f
or
r
e
t
r
ieva
l
a
nd
r
e
a
s
oning
s
ys
tems
,
a
udit
a
bil
it
y
c
ove
r
a
ge
f
or
pr
ove
na
nc
e
-
de
pe
nde
nt
us
e
c
a
s
e
s
)
,
a
nd
idea
ll
y
a
longi
tudi
na
l
de
s
ign
that
tr
a
c
ks
the
s
a
me
e
nter
pr
is
e
platf
or
m
thr
ough
a
r
e
media
ti
on
pr
ogr
a
m
to
obs
e
r
ve
whe
ther
c
a
pa
bil
it
y
r
is
e
s
a
s
d
e
bt
is
pa
id
down.
No
s
uc
h
s
tudy
e
xis
ts
ye
t,
a
nd
thi
s
pa
pe
r
doe
s
not
c
laim
to
s
ubs
ti
tut
e
f
or
one
.
T
he
f
r
a
mew
or
k
identif
ies
indi
c
a
tor
s
a
nd
s
e
que
nc
e
s
r
e
media
ti
on,
but
it
doe
s
not
ye
t
mea
s
ur
e
them
qua
nti
tatively.
Us
e
f
ul
met
r
ics
would
include
e
nti
t
y
dupli
c
a
ti
on
r
a
te,
s
e
mantic
a
li
gnment
de
ns
it
y,
pr
ove
na
nc
e
c
ove
r
a
ge
,
e
nf
or
c
e
ment
c
ons
is
tenc
y,
a
nd
c
ons
umer
a
s
s
umpt
ion
diver
ge
nc
e
.
As
a
n
il
lus
tr
a
ti
ve
s
tar
ti
ng
point
f
or
f
utur
e
qua
nti
tative
wo
r
k,
a
s
e
ve
r
it
y
s
c
or
e
f
or
a
giv
e
n
e
nti
ty
or
f
low
mi
ght
c
ombi
ne
a
n
or
d
inal
s
e
ve
r
it
y
r
a
ti
ng
(
a
s
us
e
d
qua
li
tatively
in
T
a
ble
4
)
with
a
we
ight
ing
f
or
the
nu
mber
o
f
downs
tr
e
a
m
de
pe
nde
nts
;
thi
s
is
of
f
e
r
e
d
only
a
s
a
dir
e
c
ti
on
f
or
va
li
da
ti
on
,
no
t
a
s
a
tes
ted
ins
tr
ument.
R
e
quir
e
ments
tec
hnica
l
de
bt
r
e
s
e
a
r
c
h
of
f
e
r
s
a
template
f
or
mov
ing
f
r
om
qua
li
tative
identif
ica
ti
on
towa
r
d
mea
s
ur
e
ment
[
25
]
,
a
nd
da
tas
e
t
qua
li
ty
s
ur
ve
ys
s
ugge
s
t
c
a
ndidate
metr
ics
that
c
ould
be
a
da
pted
to
the
s
ubs
tr
a
te
s
e
tt
ing
[
16]
.
T
he
il
lus
t
r
a
t
ive
s
c
e
na
r
io
in
s
e
c
ti
on
6
.
3
is
a
c
o
mpos
it
e
c
o
ns
tr
uc
ted
f
o
r
e
xpos
i
ti
o
n
r
a
t
he
r
t
ha
n
a
dis
c
los
e
d
c
a
s
e
s
tud
y
o
r
r
e
lea
s
e
d
da
tas
e
t
,
whic
h
l
i
mi
ts
w
ha
t
c
a
n
be
s
a
id
a
bout
the
f
r
a
mew
or
k's
r
e
pr
o
duc
ib
il
i
ty
a
nd
pr
a
c
ti
c
a
l
va
li
dit
y
unt
il
r
e
a
l
-
wo
r
ld
c
a
s
e
e
v
idenc
e
be
c
omes
a
va
i
lable
.
T
he
f
r
a
mew
o
r
k
a
ls
o
t
r
e
a
ts
i
nf
or
ma
ti
on
a
r
c
hi
tec
tu
r
e
de
bt
in
is
ola
ti
o
n,
a
nd
it
doe
s
n
ot
a
ddr
e
s
s
how
ta
lent
,
c
ha
nge
c
a
pa
c
i
ty
,
a
n
d
go
ve
r
na
nc
e
c
o
ns
tr
a
i
nts
c
ombi
ne
wi
th
it
in
a
r
e
a
l
p
r
og
r
a
m
.
A
f
ur
the
r
l
i
mi
tat
ion
is
tempo
r
a
l
:
t
he
f
r
a
mew
o
r
k
is
t
ied
to
t
h
e
c
ur
r
e
n
t
bounda
r
y
o
f
AI
c
a
pa
bil
it
y
,
a
nd
it
s
s
e
ve
r
it
y
t
hr
e
s
ho
l
ds
wi
ll
ne
e
d
r
e
c
a
l
ibr
a
ti
o
n
a
s
m
ode
ls
i
mpr
ove
.
9.
CONC
L
USI
ON
I
nf
or
mation
a
r
c
hit
e
c
tur
e
de
bt
a
ppe
a
r
s
to
be
t
he
s
ubs
tr
a
te
c
ons
tr
a
int
on
e
nter
p
r
is
e
a
r
ti
f
icia
l
int
e
ll
igenc
e
,
a
nd
it
is
dis
ti
nc
t
f
r
om
the
c
ode
a
nd
i
nf
r
a
s
tr
uc
tur
e
de
bt
that
lea
de
r
s
mor
e
r
e
a
dil
y
r
e
c
og
nize
a
nd
f
und.
T
he
pa
pe
r
's
main
c
ontr
ibut
ion
is
twof
old:
a
f
ive
-
indi
c
a
tor
diagnos
ti
c
f
r
a
mew
or
k,
de
r
ived
th
r
ough
the
qua
li
tative
s
ynthes
is
pr
oc
e
s
s
de
s
c
r
ibed
in
S
e
c
ti
on
3,
that
gives
lea
de
r
s
a
wa
y
to
r
e
c
ognize
a
nd
diag
nos
e
the
de
bt
be
f
or
e
model
s
e
lec
ti
on;
a
nd
a
c
a
pa
bil
it
y
c
e
il
ing
hypothes
is
that
r
e
f
r
a
mes
a
c
las
s
of
AI
o
utcome
s
c
ur
r
e
ntl
y
mi
s
a
tt
r
ibut
e
d
to
model
c
hoice
a
s
s
ubs
tr
a
te
-
bound
ins
tea
d.
T
he
thr
e
e
-
wa
v
e
s
e
que
n
c
e
buil
t
on
thi
s
diagnos
is
gives
lea
de
r
s
a
wa
y
to
r
e
media
te
the
de
bt
in
or
de
r
that
unblocks
c
a
pa
bil
it
y
f
a
s
tes
t,
w
hil
e
the
s
e
que
nc
ing
e
xc
e
pti
ons
identif
ied
in
S
e
c
ti
on
6
.
1
k
e
e
p
the
f
r
a
mew
or
k
a
ppli
c
a
ble
to
the
mor
e
s
e
ve
r
e
c
a
s
e
s
a
s
tr
ict
or
de
r
ing
would
mi
s
ha
ndle.
P
os
it
ioni
ng
the
d
e
bt
a
s
a
c
oor
dination
pr
oblem
r
a
ther
than
a
pur
e
ly
tec
hnica
l
one
c
lar
if
ies
both
why
it
pe
r
s
is
ts
a
nd
wha
t
kind
o
f
s
pons
or
s
hip
it
s
r
e
pa
ir
r
e
quir
e
s
.
T
he
f
utur
e
im
pa
c
t
of
thi
s
c
ontr
ibut
ion
de
pe
nds
on
the
e
mpi
r
ica
l
a
nd
qua
nti
tative
wor
k
it
is
de
s
igned
to
e
na
ble:
a
va
li
da
ted
s
e
ve
r
it
y
metr
ic
would
let
the
diagnos
ti
c
move
f
r
om
qua
li
ta
ti
ve
tr
iage
to
c
ompar
a
ble,
be
nc
hmar
ke
d
a
s
s
e
s
s
me
nt
a
c
r
os
s
platf
or
ms
,
a
nd
a
c
ontr
oll
e
d
o
r
longi
tudi
na
l
tes
t
o
f
the
c
a
pa
bil
it
y
c
e
il
ing
hypothes
is
would
e
s
tablis
h,
r
a
ther
than
pr
opos
e
,
the
s
tr
uc
tur
a
l
li
mi
t
thi
s
pa
pe
r
a
r
gue
s
f
or
.
AC
KNOWL
E
DGM
E
N
T
S
T
he
a
uthor
a
c
knowle
dge
s
the
pr
a
c
ti
ti
one
r
e
nga
g
e
ments
r
e
f
e
r
e
nc
e
d
thr
oughout
thi
s
pa
pe
r
,
whic
h
inf
or
med
the
qua
li
tative
s
ynthes
is
de
s
c
r
ibed
in
s
e
c
ti
on
3,
whi
le
noti
ng
that
no
c
li
e
nt
-
identif
ying
or
c
onf
idential
inf
o
r
mation
f
r
om
thos
e
e
nga
ge
ments
is
dis
c
los
e
d
he
r
e
in.
F
UN
DI
NG
I
NF
ORM
AT
I
ON
T
he
a
uthor
s
tate
s
that
no
r
e
s
e
a
r
c
h
gr
a
nt
or
c
ontr
a
c
t
f
unding
wa
s
r
e
c
e
ived
f
or
thi
s
wor
k
.
AU
T
HO
R
CONT
RI
B
U
T
I
ONS
S
T
AT
E
M
E
N
T
T
his
jour
na
l
us
e
s
the
C
ontr
ibut
o
r
R
oles
T
a
xo
nomy
(
C
R
e
diT
)
to
r
e
c
ognize
indi
vidual
a
uthor
c
ontr
ibut
ions
,
r
e
duc
e
a
utho
r
s
hip
dis
putes
,
a
nd
f
a
c
il
it
a
te
c
oll
a
bor
a
ti
on.
Nam
e
of
Au
t
h
or
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
M
ihi
r
S
ha
h
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
onc
e
pt
ua
li
z
a
ti
on
M
:
M
e
th
odol
ogy
So
:
So
f
twa
r
e
Va
:
Va
li
da
ti
on
Fo
:
Fo
r
ma
l
a
na
ly
s
is
I
:
I
nve
s
ti
ga
ti
on
R
:
R
e
s
our
c
e
s
D
:
D
a
ta
C
ur
a
ti
on
O
:
W
r
it
in
g
-
O
r
ig
in
a
l
D
r
a
f
t
E
:
W
r
it
in
g
-
R
e
vi
e
w
&
E
di
ti
ng
Vi
:
Vi
s
ua
li
z
a
ti
on
Su
:
Su
pe
r
vi
s
io
n
P
:
P
r
oj
e
c
t
a
dmi
ni
s
tr
a
ti
on
Fu
:
Fu
ndi
ng a
c
qui
s
it
io
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
nt
J
E
lec
&
C
omp
E
ng
I
S
S
N:
2088
-
8708
I
nfor
mation
ar
c
hit
e
c
tur
e
de
bt:
W
hy
legac
y
plat
for
m
s
c
he
ma
de
c
is
ions
c
ons
t
r
ain
…
(
M
ihi
r
Shah
)
2481
CONF
L
I
CT
OF
I
NT
E
RE
S
T
S
T
AT
E
M
E
N
T
T
he
a
uthor
de
c
lar
e
s
no
c
onf
li
c
t
of
int
e
r
e
s
t.
DA
T
A
AV
AI
L
A
B
I
L
I
T
Y
Da
ta
a
va
il
a
bil
it
y
is
not
a
ppli
c
a
ble
to
thi
s
pa
pe
r
a
s
no
ne
w
da
ta
we
r
e
c
r
e
a
ted
o
r
a
na
lyze
d
in
thi
s
s
tudy.
RE
F
E
RE
NC
E
S
[
1]
J
.
R
ys
e
f
f
,
B
.
D
e
B
r
uhl
,
a
nd
S
.
J
.
N
e
w
be
r
r
y,
“
T
he
r
oot
c
a
us
e
s
of
f
a
il
ur
e
f
or
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
pr
oj
e
c
ts
a
nd
how
th
e
y
c
a
n
s
uc
c
e
e
d:
A
voi
di
ng t
he
a
nt
i
-
pa
tt
e
r
ns
of
A
I
,”
2024. doi:
10.7249/r
r
a
2680
-
1.
[
2]
W
. C
unni
ngha
m, “
T
he
W
yC
a
s
h por
tf
ol
io
ma
na
ge
me
nt
s
ys
te
m,
”
A
C
M
SI
G
P
L
A
N
O
O
P
S
M
e
s
s
e
nge
r
, vol
. 4, no. 2, pp. 29
–
30, 1993,
doi
:
10.1145/157710.157
715.
[
3]
H
. K
le
in
w
a
ks
, A
. B
a
tc
h
e
lo
r
, a
nd T
. H
. B
r
a
dl
e
y, “
T
e
c
hni
c
a
l
de
b
t
in
s
ys
te
ms
e
ngi
ne
e
r
in
g
—
A
s
y
s
te
ma
ti
c
l
it
e
r
a
tu
r
e
r
e
vi
e
w
,”
Sy
s
t
e
m
s
E
ngi
ne
e
r
in
g
, vol
. 26, no. 5, pp. 675
–
687, 2023, doi:
10.1002/s
ys
.21681.
[
4]
D
.
S
c
ul
le
y
e
t
al
.
,
“
H
id
de
n
te
c
hni
c
a
l
de
bt
in
m
a
c
hi
ne
le
a
r
ni
ng
s
ys
te
ms
,
”
in
A
dv
anc
e
s
in
N
e
u
r
al
I
nf
or
m
at
io
n
P
r
oc
e
s
s
in
g
Sy
s
te
m
s
,
2015, vol. 2015
-
J
a
nua
r
y, pp. 2503
–
2511.
[
5]
D
.
A
lb
uque
r
que
e
t
al
.
,
“
M
a
na
gi
ng
t
e
c
hni
c
a
l
de
bt
us
in
g
in
te
ll
ig
e
nt
te
c
hni
que
s
-
A
s
y
s
te
ma
ti
c
ma
ppi
ng
s
tu
dy,”
I
E
E
E
T
r
ans
ac
ti
ons
on Soft
w
ar
e
E
ngi
ne
e
r
in
g
, vol
. 49, no. 4, pp. 2202
–
2220, 2023,
doi
:
10.1109/T
S
E
.2022.3214764.
[
6]
S
.
S
c
he
r
z
in
ge
r
a
nd
S
.
S
id
or
ts
c
huc
k,
“
A
n
e
mpi
r
ic
a
l
s
tu
dy
on
t
he
de
s
ig
n
a
nd
e
vol
ut
io
n
of
N
oS
Q
L
da
ta
ba
s
e
s
c
he
ma
s
,”
in
L
e
c
tu
r
e
N
ot
e
s
in
C
om
put
e
r
Sc
ie
nc
e
(
in
c
lu
di
ng
s
ubs
e
r
ie
s
L
e
c
tu
r
e
N
ot
e
s
in
A
r
ti
fi
c
ia
l
I
nt
e
ll
ig
e
nc
e
and
L
e
c
tu
r
e
N
ot
e
s
in
B
io
in
fo
r
m
at
ic
s
)
,
2020, vol. 12400 L
N
C
S
, pp. 441
–
455. doi
:
10.1007/978
-
3
-
030
-
62522
-
1_33.
[
7]
S
.
S
in
gh
a
nd
J
.
S
in
gh,
“
A
s
ur
ve
y
on
ma
s
te
r
da
ta
ma
na
ge
me
nt
t
e
c
hni
que
s
f
or
bus
in
e
s
s
pe
r
s
pe
c
ti
ve
,
”
in
L
e
c
tu
r
e
N
ot
e
s
in
N
e
tw
o
r
k
s
and Sy
s
te
m
s
, 2022, vol. 291, pp.
609
–
617. doi:
10.1007/978
-
98
1
-
16
-
4284
-
5_54.
[
8]
J
.
Y
e
be
ne
s
S
e
r
r
a
no
a
nd
M
.
Z
or
r
il
la
,
“
A
da
ta
gove
r
na
nc
e
f
r
a
me
w
or
k
f
or
I
ndus
tr
y
4.0,”
I
E
E
E
L
at
in
A
m
e
r
ic
a
T
r
ans
ac
ti
ons
,
vol
.
19,
no. 12, pp. 2130
–
2138, 2021, doi:
10.1109/T
L
A
.2021.9480156.
[
9]
N
.
B
a
r
la
ug
a
nd
J
.
A
.
G
ul
la
,
“
N
e
ur
a
l
ne
twor
ks
f
or
e
nt
it
y
ma
tc
h
in
g:
A
s
ur
ve
y,”
A
C
M
T
r
ans
ac
ti
ons
on
K
now
le
dge
D
i
s
c
ov
e
r
y
f
r
om
D
at
a
, vol
. 15, no. 3, pp. 1
–
37, 2021, doi:
10.1145/3442200.
[
10]
N
.
B
a
r
la
ug,
“
L
E
M
O
N
:
E
xpl
a
in
a
bl
e
e
nt
it
y
ma
tc
hi
ng,”
I
E
E
E
T
r
ans
ac
ti
ons
on
K
now
le
dge
and
D
at
a
E
ngi
ne
e
r
in
g
,
vol
.
35,
no
.
8,
pp. 8171
–
8184, 2023, doi:
10.1109/T
K
D
E
.2022.3200644.
[
11]
R
.
S
.
P
it
a
ngue
ir
a
M
a
c
ie
l,
P
.
H
.
D
ia
s
V
a
ll
e
,
K
.
S
.
S
a
nt
os
,
a
nd
E
.
Y
.
N
a
ka
ga
w
a
,
“
S
ys
te
ms
in
te
r
ope
r
a
bi
li
ty
ty
pe
s
:
A
te
r
ti
a
r
y
s
tu
d
y,”
A
C
M
C
om
put
in
g Sur
v
e
y
s
, vol
. 56, no. 10, pp. 1
–
37, 2024, doi:
10.1145/3659098.
[
12]
B
.
P
a
n,
N
.
S
ta
kh
a
nova
,
a
nd
S
.
R
a
y,
“
D
a
t
a
pr
ove
na
nc
e
in
s
e
c
ur
it
y
a
nd
pr
iv
a
c
y,”
A
C
M
C
om
put
in
g
Sur
v
e
y
s
,
vol
.
55,
no.
14,
pp. 1
–
35, 2023, doi:
10.1145/3593294.
[
13]
X
.
W
a
ng,
X
.
L
i,
a
nd
X
.
X
ia
,
“
R
e
s
e
a
r
c
h
on
da
ta
qua
li
ty
ma
na
g
e
me
nt
me
th
ods
a
nd
t
e
c
hnol
ogi
e
s
,”
in
2024
I
E
E
E
2nd
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on
I
m
age
P
r
o
c
e
s
s
in
g
and
C
o
m
put
e
r
A
ppl
ic
at
io
ns
,
I
C
I
P
C
A
2024
,
2024,
pp.
116
–
120.
doi
:
10.1109/I
C
I
P
C
A
61593.2024.10709151.
[
14]
D
.
Z
ha
e
t
al
.
,
“
D
a
ta
-
c
e
nt
r
ic
a
r
ti
f
ic
ia
l
in
te
ll
ig
e
nc
e
:
A
s
ur
ve
y,”
A
C
M
C
om
put
in
g
Sur
v
e
y
s
,
vol
.
57,
no.
5,
pp.
1
–
42,
M
a
y
2025,
doi
:
10.1145/3711118.
[
15]
L
.
H
ua
ng
e
t
al
.
,
“
A
s
ur
ve
y
on
ha
ll
uc
in
a
ti
on
in
la
r
ge
l
a
ngua
g
e
mode
ls
:
P
r
in
c
ip
le
s
,
ta
xonomy,
c
ha
ll
e
nge
s
,
a
nd
ope
n
que
s
ti
o
ns
,”
A
C
M
T
r
ans
ac
ti
ons
on I
nf
or
m
at
io
n Sy
s
te
m
s
, vol
. 43, no. 2, pp. 1
–
55, 2025, doi:
10.1145/3703155.
[
16]
Y
.
G
ong,
G
.
L
iu
,
Y
.
X
ue
,
R
.
L
i,
a
nd
L
.
M
e
ng,
“
A
s
ur
ve
y
on
da
ta
s
e
t
qua
li
ty
in
ma
c
hi
ne
le
a
r
ni
ng,”
I
nf
or
m
at
io
n
and
Sof
t
w
ar
e
T
e
c
hnol
ogy
, vol
. 162, p. 107268, 2023, doi
:
10.1016/j
.i
nf
s
of
.20
23.107268.
[
17]
M
.
P
r
ie
s
tl
e
y,
F
.
O
’
D
onne
ll
,
a
nd
E
.
S
im
pe
r
l,
“
A
s
ur
ve
y
of
da
t
a
qua
li
ty
r
e
qui
r
e
me
nt
s
th
a
t
ma
tt
e
r
in
M
L
de
ve
lo
pme
nt
pi
p
e
li
ne
s
,
”
J
our
nal
of
D
at
a and I
nf
or
m
at
io
n Q
ual
it
y
, vol
. 15, no. 2, pp. 1
–
39, 2023, doi:
10.1145/3592616.
[
18]
Y
.
L
i,
J
.
L
i,
Y
.
S
uha
r
a
,
A
.
D
oa
n,
a
nd
W
.
C
.
T
a
n,
“
D
e
e
p
e
nt
it
y
ma
tc
hi
ng
w
it
h
pr
e
-
tr
a
in
e
d
la
ngua
ge
mode
l
s
,”
P
r
oc
e
e
di
ngs
o
f
th
e
V
L
D
B
E
ndow
m
e
nt
, vol
. 14, no. 1, pp. 50
–
60, 2020, doi:
10.147
78/
3421424.3421431.
[
19]
R
.
P
e
e
te
r
s
a
nd
C
.
B
iz
e
r
,
“
D
ua
l
-
obj
e
c
ti
ve
f
in
e
-
tu
ni
ng
of
B
E
R
T
f
or
e
nt
it
y
ma
tc
hi
ng,”
P
r
oc
e
e
di
ngs
of
th
e
V
L
D
B
E
ndow
m
e
nt
,
vol
. 14, no. 10, pp. 1913
–
1921, 2021, doi:
10.14778/3467861.
3467878.
[
20]
C
. G
e
, P
. W
a
ng, L
.
C
he
n, X
. L
iu
, B
. Z
he
ng, a
nd Y
.
G
a
o, “
C
ol
l
a
bor
E
M
:
A
s
e
lf
-
s
upe
r
vi
s
e
d e
nt
it
y ma
tc
hi
ng f
r
a
me
w
or
k us
in
g m
ul
ti
-
f
e
a
tu
r
e
s
c
ol
la
bor
a
ti
on,”
I
E
E
E
T
r
ans
ac
ti
ons
on
K
now
le
dg
e
a
nd
D
at
a
E
ngi
ne
e
r
in
g
,
vol
.
35,
no.
12,
pp.
12139
–
12152,
2
023,
doi
:
10.1109/T
K
D
E
.2021.3134806.
[
21]
L
.
G
a
z
z
a
r
r
i
a
nd
M
.
H
e
r
s
c
he
l,
“
E
nd
-
to
-
e
nd
ta
s
k
b
a
s
e
d
pa
r
a
ll
e
li
z
a
ti
on
f
or
e
nt
it
y
r
e
s
ol
ut
io
n
on
dyna
mi
c
d
a
ta
,”
in
P
r
oc
e
e
di
n
gs
-
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on Data E
ngi
ne
e
r
in
g
, 2021, vol. 202
1
-
A
pr
il
, pp. 1248
–
1259. doi:
10.1109/I
C
D
E
51399.2021.00112.
[
22]
M
.
J
.
E
c
c
le
s
,
D
.
J
.
E
va
ns
,
a
nd
A
.
J
.
B
e
a
umont
,
“
T
r
ue
r
e
a
l
-
ti
me
c
ha
nge
d
a
ta
c
a
pt
ur
e
w
it
h
w
e
b
s
e
r
vi
c
e
da
ta
ba
s
e
e
nc
a
ps
ul
a
ti
on,”
in
P
r
oc
e
e
di
ngs
-
2010 6th W
or
ld
C
ongr
e
s
s
on S
e
r
v
i
c
e
s
, Se
r
v
ic
e
s
-
1 2010
, 2010, pp. 128
–
131. doi:
10.1109/S
E
R
V
I
C
E
S
.2010.59.
[
23]
I
.
G
r
a
nge
l
-
G
onz
á
le
z
,
F
.
L
ös
c
h,
a
nd
A
.
U
l
M
e
hdi
,
“
K
now
le
dge
gr
a
phs
f
or
e
f
f
ic
ie
nt
in
te
gr
a
ti
on
a
nd
a
c
c
e
s
s
of
ma
nuf
a
c
tu
r
in
g
da
ta
,”
in
I
E
E
E
I
nt
e
r
nat
io
nal
C
onf
e
r
e
nc
e
on E
m
e
r
gi
ng T
e
c
hnol
ogi
e
s
a
nd F
ac
to
r
y
A
ut
om
at
io
n,
E
T
F
A
, 2020, vol.
2020
-
S
e
pt
e
, pp. 93
–
100.
doi
:
10.1109/E
T
F
A
46521.2020.9212156.
[
24]
A
.
G
oe
de
ge
buur
e
e
t
al
.
,
“
D
a
ta
me
s
h:
A
s
ys
te
m
a
ti
c
gr
a
y
li
te
r
a
tu
r
e
r
e
vi
e
w
,”
A
C
M
C
om
put
in
g
Sur
v
e
y
s
,
vol
.
57,
no.
1,
p.
11,
2024,
doi
:
10.1145/3687301.
[
25]
A
.
M
e
lo
,
R
.
F
a
gunde
s
,
V
.
L
e
na
r
duz
z
i,
a
nd
W
.
B
.
S
a
nt
os
,
“
I
de
nt
if
ic
a
ti
on
a
nd
me
a
s
ur
e
me
nt
of
r
e
qui
r
e
me
nt
s
te
c
hni
c
a
l
de
bt
in
s
of
twa
r
e
de
ve
lo
pme
nt
:
A
s
y
s
te
ma
ti
c
li
te
r
a
tu
r
e
r
e
vi
e
w
,”
J
ou
r
nal
of
Sy
s
te
m
s
and
Sof
tw
ar
e
,
vol
.
194,
p.
111483,
2022,
doi
:
10.1016/j
.j
s
s
.2022.111483.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
S
S
N
:
2088
-
8708
I
nt
J
E
lec
&
C
omp
E
ng
,
Vol
.
16
,
No.
5
,
Oc
tober
20
26
:
2473
-
2482
2482
B
I
OG
RA
P
H
Y
OF
AU
T
HO
R
M
i
hi
r
Sha
h
H
e
i
s
i
n
d
e
p
en
d
en
t
res
earc
h
er
i
n
U
S
A
.
H
e
can
b
e
co
n
t
act
e
d
at
emai
l
:
mi
h
i
rs
h
ah
c
o
n
n
ect
@
g
ma
i
l
.
co
m
.
Evaluation Warning : The document was created with Spire.PDF for Python.