I
AE
S In
t
er
na
t
io
na
l J
o
urna
l o
f
Art
if
icia
l In
t
ellig
ence
(
I
J
-
AI
)
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
,
p
p
.
810
~
823
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
1
.
p
p
8
1
0
-
8
2
3
810
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
Neuro
-
DA
Net
:
du
a
l attent
io
n
deep
neura
l net
wo
rk
lo
ng
sho
rt
-
term
me
mo
ry
f
o
r
a
utism spect
rum
diso
rder de
tect
io
n
Su
j
a
t
ha
H
a
nu
m
a
ntha
ra
y
a
p
pa
,
M
a
njula
Rudra
g
o
ud
a
B
ha
ra
m
a
g
o
ud
ra
S
c
h
o
o
l
o
f
E
l
e
c
t
r
o
n
i
c
s
a
n
d
C
o
mm
u
n
i
c
a
t
i
o
n
E
n
g
i
n
e
e
r
i
n
g
,
R
EV
A
U
n
i
v
e
r
si
t
y
,
B
e
n
g
a
l
u
r
u
,
I
n
d
i
a
Art
icle
I
nfo
AB
S
T
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
n
2
6
,
2
0
2
5
R
ev
is
ed
Dec
2
5
,
2
0
2
5
Acc
ep
ted
J
an
1
0
,
2
0
2
6
Au
ti
sm
sp
e
c
tru
m
d
iso
r
d
e
r
(ASD
)
is
n
e
u
r
o
lo
g
ica
l
il
l
n
e
ss
a
ffe
c
ts
a
b
il
it
y
o
f
in
d
i
v
id
u
a
ls
to
c
o
m
m
u
n
ica
te
a
n
d
i
n
tera
c
t
so
c
ially
,
a
n
d
i
t
is
d
iag
n
o
s
e
d
in
a
n
y
ti
m
e
.
Early
d
e
tec
ti
o
n
o
f
ASD
i
s
e
sp
e
c
ially
sig
n
ifi
c
a
n
t
d
u
e
to
it
s
su
b
tl
e
c
h
a
ra
c
teristics
a
n
d
h
i
g
h
c
o
sts
a
ss
o
c
iate
d
with
th
e
d
e
tec
ti
o
n
p
r
o
c
e
ss
.
Trad
it
io
n
a
l
d
e
e
p
lea
rn
in
g
(
DL
)
m
o
d
e
ls
stru
g
g
le
t
o
c
a
p
tu
re
in
tri
c
a
te
sp
a
ti
o
tem
p
o
ra
l
d
e
p
e
n
d
e
n
c
ies
in
fu
n
c
ti
o
n
a
l
m
a
g
n
e
ti
c
re
so
n
a
n
c
e
ima
g
in
g
(fM
RI)
d
a
ta,
re
su
lt
i
n
g
in
m
i
n
imiz
e
d
d
e
tec
ti
o
n
p
e
rfo
rm
a
n
c
e
a
n
d
p
o
o
r
g
e
n
e
ra
li
z
a
ti
o
n
.
T
o
a
d
d
re
ss
t
h
e
se
d
ra
wb
a
c
k
s,
th
e
p
ro
p
o
se
d
N
eu
ro
-
DA
Ne
t
c
o
m
b
in
e
s
a
d
u
a
l
-
a
tt
e
n
ti
o
n
d
e
e
p
n
e
u
ra
l
n
e
two
rk
(DA
-
DN
N)
with
l
o
n
g
s
h
o
rt
-
term
m
e
m
o
ry
(LS
TM
)
to
e
fficie
n
tl
y
lea
rn
sp
a
ti
a
l
a
n
d
tem
p
o
ra
l
fe
a
tu
re
s
fro
m
fM
RI
sc
a
n
s.
Th
e
c
o
n
ti
n
u
o
u
s
wa
v
e
let
tran
sfo
rm
(CWT
)
is
u
se
d
to
e
x
trac
t
m
u
lt
i
-
sc
a
le
fe
a
tu
re
s
a
n
d
th
e
p
ri
n
c
ip
a
l
c
o
m
p
o
n
e
n
t
a
n
a
l
y
sis
(P
CA)
is
u
ti
li
z
e
d
to
d
ime
n
sio
n
a
li
ty
re
d
u
c
ti
o
n
,
wh
i
c
h
e
n
h
a
n
c
e
s
ro
b
u
st
n
e
ss
a
n
d
e
ffi
c
a
c
y
.
Th
e
d
u
a
l
se
lf
-
a
tt
e
n
ti
o
n
m
e
c
h
a
n
ism
imp
ro
v
e
s
th
e
in
terp
re
tab
i
li
ty
o
f
th
e
m
o
d
e
l
b
y
fo
c
u
sin
g
o
n
c
rit
ica
l
b
ra
i
n
re
g
i
o
n
s
a
n
d
ti
m
e
ste
p
s
th
a
t
a
re
m
o
st
r
e
lev
a
n
t
to
ASD
se
v
e
rit
y
.
Th
e
d
e
v
e
l
o
p
e
d
Ne
u
ro
-
DA
Ne
t
o
b
tain
s
th
e
h
i
g
h
e
st
a
c
c
u
ra
c
y
o
f
9
8
.
5
1
%
o
n
a
u
ti
sm
b
ra
in
ima
g
i
n
g
d
a
ta
e
x
c
h
a
n
g
e
(
ABID
E
)
-
I
a
n
d
9
8
.
8
1
%
o
n
ABID
E
-
II
d
a
tas
e
ts wh
e
n
c
o
m
p
a
re
d
with
trad
it
io
n
a
l
a
lg
o
rit
h
m
s.
K
ey
w
o
r
d
s
:
Au
tis
m
s
p
ec
tr
u
m
d
is
o
r
d
er
C
o
n
tin
u
o
u
s
wav
elet
tr
an
s
f
o
r
m
s
D
u
a
l
-
a
t
t
e
n
t
i
o
n
d
e
e
p
n
e
u
r
a
l
n
e
t
w
o
r
k
Fu
n
ctio
n
al
m
ag
n
etic
r
eso
n
an
c
e
im
ag
in
g
L
o
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
Prin
cip
al
co
m
p
o
n
en
t a
n
al
y
s
is
T
h
is i
s
a
n
o
p
e
n
a
c
c
e
ss
a
rticle
u
n
d
e
r th
e
CC B
Y
-
SA
li
c
e
n
se
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
Su
jath
a
Han
u
m
an
th
a
r
ay
ap
p
a
Sch
o
o
l o
f
E
lectr
o
n
ics an
d
C
o
m
m
u
n
icatio
n
E
n
g
in
ee
r
in
g
,
R
E
VA
Un
iv
er
s
ity
B
en
g
alu
r
u
,
I
n
d
ia
E
m
ail: su
jag
1
2
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Au
tis
m
s
p
ec
tr
u
m
d
is
o
r
d
er
(
AS
D)
is
n
eu
r
o
d
ev
el
o
p
m
en
tal
illn
ess
co
n
s
id
er
ed
with
ess
en
tial d
if
f
icu
lties
with
s
o
cial
co
m
m
u
n
icatio
n
,
a
r
estricte
d
r
a
n
g
e
o
f
i
n
ter
ests
,
r
ep
etitiv
e
b
e
h
av
io
r
s
an
d
a
ty
p
ical
p
er
ce
p
tu
al
r
esp
o
n
s
es
[
1
]
–
[
3
]
.
Sy
m
p
to
m
s
o
f
ASD
g
en
er
ally
e
m
er
g
e
o
n
ea
r
ly
ch
ild
h
o
o
d
,
alth
o
u
g
h
s
o
c
ial
d
ef
icits
ar
e
n
o
t
n
o
ticea
b
le
in
d
if
f
icu
lt
s
o
cial
en
v
ir
o
n
m
en
ts
[
4
]
,
[
5
]
.
T
h
e
in
cr
ea
s
in
g
p
r
ev
alen
ce
o
f
A
SD
h
as
m
ad
e
it
an
ess
en
tial
p
u
b
lic
h
ea
lth
co
n
ce
r
n
[
6
]
.
E
ar
ly
a
n
d
ac
c
u
r
ate
d
ia
g
n
o
s
is
o
f
ASD
is
s
ig
n
if
ican
t
f
o
r
en
a
b
lin
g
tim
ely
in
ter
v
en
tio
n
s
,
wh
ic
h
im
p
r
o
v
e
q
u
ality
o
f
life
f
o
r
in
d
iv
id
u
als
with
th
e
d
is
o
r
d
er
[
7
]
.
T
h
o
u
g
h
th
e
h
eter
o
g
en
eo
u
s
n
atu
r
e
o
f
ASD,
with
its
wid
e
r
an
g
e
o
f
s
ev
er
ity
an
d
s
y
m
p
to
m
lev
els,
m
ak
es
d
iag
n
o
s
is
ch
allen
g
in
g
,
ca
u
s
e
m
is
d
iag
n
o
s
is
an
d
d
elay
s
[
8
]
.
Mo
r
eo
v
er
,
ASD
d
iag
n
o
s
is
is
ch
allen
g
in
g
p
r
o
ce
s
s
wh
ich
in
clu
d
es
s
er
ies
o
f
ca
r
ef
u
l
s
tep
s
th
at
in
v
o
lv
e
lo
n
g
-
ter
m
clin
ical
m
o
n
ito
r
in
g
,
ea
r
l
y
ass
es
s
m
en
t
th
r
o
u
g
h
ca
r
e
g
iv
er
an
d
p
r
o
f
ess
io
n
al
in
ter
v
iews
with
th
e
p
h
y
s
ician
[
9
]
,
[
1
0
]
.
R
ec
en
tly
,
m
ac
h
in
e
lear
n
in
g
(
ML
)
an
d
d
ee
p
lear
n
in
g
(
DL
)
m
o
d
els
h
av
e
b
ee
n
im
p
lem
en
ted
to
en
h
an
ce
ASD
d
iag
n
o
s
is
[
1
1
]
.
T
h
ese
alg
o
r
ith
m
s
h
a
v
e
b
ee
n
em
p
lo
y
e
d
f
o
r
n
eu
r
o
im
a
g
in
g
in
f
o
r
m
atio
n
,
esp
ec
ially
f
u
n
ctio
n
al
m
a
g
n
etic
r
eso
n
an
ce
im
ag
in
g
(
f
MRI)
an
d
s
tr
u
ctu
r
al
m
ag
n
etic
r
eso
n
an
ce
im
ag
in
g
(
s
MRI)
to
d
etec
tin
g
p
atter
n
s
wh
ich
d
if
f
e
r
en
tiate
in
d
iv
id
u
als
with
ASD
f
r
o
m
n
eu
r
o
ty
p
ica
l
co
n
tr
o
ls
[
1
2
]
,
[
1
3
]
.
R
ec
en
tly
,
DL
alg
o
r
ith
m
s
ar
e
p
r
ef
e
r
r
ed
o
v
er
co
n
v
en
tio
n
al
ML
alg
o
r
ith
m
s
b
ec
au
s
e
o
f
th
eir
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
N
eu
r
o
-
DA
N
et:
d
u
a
l a
tten
tio
n
d
ee
p
n
e
u
r
a
l n
etw
o
r
k
lo
n
g
s
h
o
r
t
-
term
…
(
S
u
ja
th
a
Ha
n
u
ma
n
th
a
r
a
ya
p
p
a
)
811
ca
p
ab
ilit
y
f
o
r
au
t
o
m
atica
lly
lear
n
in
g
d
if
f
ic
u
lt
h
ier
ar
c
h
ical
f
ea
tu
r
es
f
r
o
m
r
aw
d
at
a,
p
ar
ticu
lar
ly
in
h
ig
h
-
d
im
e
n
s
io
n
al
ar
ea
s
lik
e
n
eu
r
o
im
ag
in
g
[
1
4
]
.
Un
lik
e
ML
alg
o
r
ith
m
s
,
wh
ich
r
ely
o
n
h
a
n
d
cr
af
ted
f
ea
tu
r
es,
d
ee
p
m
o
d
els
ex
tr
ac
t
r
ich
,
m
u
lti
-
s
ca
le
r
ep
r
esen
tatio
n
s
d
ir
e
ctly
f
r
o
m
f
MRI
an
d
s
MRI
s
ca
n
s
[
1
5
]
.
T
h
is
is
ess
en
tial
in
au
tis
m
d
iag
n
o
s
is
,
wh
er
e
th
e
s
u
b
tle
s
p
atial
an
d
tem
p
o
r
al
p
atter
n
s
ar
e
c
o
m
p
l
ex
to
ca
p
t
u
r
e
[
1
6
]
.
Self
-
atten
tio
n
m
ec
h
an
is
m
im
p
r
o
v
es
th
e
DL
m
o
d
els
th
r
o
u
g
h
f
o
cu
s
in
g
o
n
r
ele
v
an
t
b
r
ain
r
eg
io
n
s
,
en
h
a
n
cin
g
in
ter
p
r
etab
ilit
y
a
n
d
ac
cu
r
ac
y
[
1
7
]
.
T
h
is
en
a
b
les
a
n
etwo
r
k
f
o
r
ca
p
tu
r
in
g
lo
n
g
-
r
a
n
g
e
d
ep
e
n
d
en
cies
with
o
u
t
th
e
d
r
awb
ac
k
s
o
f
s
eq
u
en
ce
s
.
Featu
r
e
ex
tr
ac
tio
n
b
y
wav
elet
tr
a
n
s
f
o
r
m
s
(
W
T
)
en
h
an
ce
s
r
o
b
u
s
tn
ess
an
d
h
ig
h
lig
h
ts
th
e
ASD
s
p
ec
if
ic
ab
n
o
r
m
ali
ties
m
u
ch
ef
f
icien
tly
[
1
8
]
.
T
h
e
d
im
en
s
io
n
ality
r
ed
u
ctio
n
th
r
o
u
g
h
p
r
in
cip
al
co
m
p
o
n
en
t
an
al
y
s
is
(
P
C
A)
r
ef
in
es
th
ese
f
ea
tu
r
es
f
o
r
ef
f
ec
tiv
e
m
o
d
el
tr
ain
in
g
[
1
9
]
.
T
h
ese
m
o
d
els
im
p
r
o
v
e
a
m
u
ch
p
r
ec
is
e,
s
ca
lab
le
an
d
au
t
o
m
atic
alg
o
r
ith
m
f
o
r
ASD
f
r
o
m
b
r
ain
s
ca
n
s
.
S
o
n
g
e
t
a
l
.
[
2
0
]
p
r
e
s
en
t
e
d
a
n
o
v
e
l
d
i
ag
n
o
s
i
s
a
l
g
o
r
i
th
m
t
h
a
t
c
o
m
b
in
e
d
g
r
a
p
h
co
n
v
o
lu
t
io
n
a
l
n
e
t
w
o
r
k
s
(
G
C
N
)
w
i
t
h
d
u
a
l
t
r
a
n
s
f
o
r
m
e
r
a
r
ch
i
t
e
c
tu
r
e
s
,
o
p
t
i
m
iz
e
d
b
y
c
o
-
t
r
a
in
i
n
g
s
tr
a
t
e
g
y
.
I
n
i
t
i
a
ll
y
,
a
t
r
a
n
s
f
o
r
m
e
r
w
a
s
d
e
te
r
m
in
e
d
f
o
r
c
ap
t
u
r
i
n
g
i
n
t
r
ic
a
t
e
t
e
m
p
o
r
al
f
e
a
t
u
r
e
s
f
r
o
m
f
M
R
I
d
a
t
a
,
wh
i
c
h
w
er
e
e
s
s
e
n
t
i
a
l
t
o
u
n
d
er
s
t
a
n
d
i
n
g
b
r
a
i
n
a
c
t
i
v
i
ty
o
v
e
r
t
im
e
.
T
h
e
n
e
x
t
t
r
an
s
f
o
r
m
e
r
wa
s
a
p
p
l
i
e
d
to
i
m
p
r
o
v
e
f
u
s
i
o
n
o
f
t
e
m
p
o
r
a
l
f
e
a
t
u
r
e
s
w
i
th
s
p
a
t
i
a
l
f
e
a
t
u
r
e
s
l
e
ar
n
ed
t
h
r
o
u
g
h
G
C
N
,
ef
f
i
ci
e
n
t
l
y
in
t
e
g
r
a
t
i
n
g
d
i
m
e
n
s
i
o
n
s
o
f
n
eu
r
o
i
m
ag
i
n
g
d
a
t
a
.
A
co
-
tr
a
i
n
in
g
s
tr
a
t
e
g
y
w
a
s
i
n
t
r
o
d
u
ce
d
to
s
i
m
u
l
t
an
eo
u
s
l
y
u
s
e
f
M
R
I
an
d
s
M
R
I
d
a
t
a
,
e
n
h
an
c
i
n
g
t
h
e
c
a
p
a
c
i
ty
o
f
th
e
m
o
d
e
l
f
o
r
g
e
n
e
r
a
l
iz
a
t
i
o
n
a
c
r
o
s
s
v
a
r
io
u
s
d
a
t
a
s
e
t
s
.
T
h
e
f
M
R
I
i
m
ag
e
s
i
n
c
l
u
d
e
r
ed
u
n
d
an
t
f
e
a
t
u
r
e
s
t
h
a
t
h
i
n
d
e
r
l
e
ar
n
in
g
e
f
f
i
c
a
cy
a
n
d
l
o
s
e
cr
i
t
i
c
a
l
d
at
a
.
T
a
n
g
e
t
a
l.
[
2
1
]
s
u
g
g
e
s
t
e
d
t
h
e
g
r
ap
h
n
e
u
r
a
l
n
e
t
w
o
r
k
(
G
N
N)
a
n
d
l
o
n
g
s
h
o
r
t
-
t
e
r
m
m
em
o
r
y
(
L
ST
M
)
f
o
r
A
S
D
.
S
u
g
g
e
s
t
e
d
m
o
d
el
c
a
p
t
u
r
e
d
s
p
a
t
i
a
l
a
t
t
r
i
b
u
t
e
s
i
n
f
M
R
I
in
f
o
r
m
a
ti
o
n
th
r
o
u
g
h
G
N
N
a
n
d
a
g
g
r
e
g
a
t
ed
t
em
p
o
r
a
l
d
a
t
a
o
f
d
y
n
am
i
c
f
u
n
c
t
io
n
a
l
c
o
n
n
e
c
t
iv
i
t
y
b
y
L
S
T
M
f
o
r
g
en
e
r
a
t
in
g
m
u
ch
co
m
p
le
t
e
s
p
a
t
io
-
t
e
m
p
o
r
a
l
f
e
a
t
u
r
e
r
e
p
r
e
s
e
n
t
at
i
o
n
o
f
f
M
R
I
.
T
h
e
d
y
n
a
m
i
c
g
r
ap
h
p
o
o
l
in
g
a
lg
o
r
i
t
h
m
wa
s
d
ev
e
l
o
p
ed
f
o
r
ex
t
r
a
c
t
i
n
g
t
h
e
l
a
s
t
n
o
d
e
r
ep
r
e
s
e
n
t
a
t
i
o
n
f
r
o
m
t
h
e
d
y
n
a
m
i
c
g
r
a
p
h
r
e
p
r
e
s
e
n
t
a
t
io
n
.
T
o
a
d
d
r
e
s
s
v
a
r
i
ab
l
e
d
ep
e
n
d
e
n
c
ie
s
o
n
d
y
n
a
m
i
c
f
e
a
tu
r
e
co
n
n
e
c
t
i
v
i
ty
in
t
im
e
s
c
a
l
e
s
,
m
e
th
o
d
i
n
tr
o
d
u
c
e
d
j
u
m
p
co
n
n
e
c
t
io
n
m
e
ch
a
n
i
s
m
f
o
r
i
m
p
r
o
v
i
n
g
d
a
t
a
e
x
tr
a
c
t
i
o
n
am
o
n
g
in
t
e
r
n
a
l
u
n
i
t
s
a
n
d
c
ap
t
u
r
e
d
a
t
t
r
ib
u
te
s
i
n
v
a
r
i
o
u
s
t
im
e
s
ca
l
e
s
.
T
h
e
m
o
d
e
l
s
t
r
u
g
g
l
e
s
w
i
th
i
m
b
a
l
a
n
ce
d
d
a
ta
,
l
e
ad
i
n
g
t
o
b
i
a
s
ed
l
e
a
r
n
in
g
a
n
d
p
o
o
r
g
e
n
er
a
l
i
z
a
t
io
n
.
L
i
u
e
t
a
l.
[
2
2
]
d
e
v
e
lo
p
ed
m
u
l
t
i
-
a
t
la
s
d
e
ep
en
s
e
m
b
l
e
n
e
tw
o
r
k
(
M
A
D
E
)
f
o
r
A
S
D
,
w
h
i
c
h
co
m
b
i
n
e
d
m
u
l
ti
a
t
l
a
s
e
s
o
f
f
M
R
I
i
n
f
o
r
m
a
t
io
n
b
y
w
e
ig
h
t
ed
d
e
e
p
en
s
em
b
l
e
n
e
t
w
o
r
k
.
T
h
e
d
e
v
e
lo
p
e
d
m
o
d
e
l
co
m
b
in
ed
d
e
m
o
g
r
ap
h
i
c
d
a
t
a
i
n
to
a
p
r
ed
ic
t
i
o
n
wo
r
k
f
lo
w
t
h
a
t
i
m
p
r
o
v
e
s
d
i
a
g
n
o
s
i
s
o
f
A
S
D
p
e
r
f
o
r
m
a
n
c
e
an
d
p
r
o
v
id
e
s
m
u
c
h
h
o
l
i
s
t
i
c
p
e
r
s
p
e
c
t
i
v
e
i
n
p
a
t
ie
n
t
p
r
o
f
i
l
i
n
g
.
T
h
e
m
o
d
e
l
f
a
i
l
ed
t
o
e
x
tr
a
c
t
s
p
a
t
i
a
l
a
n
d
t
e
m
p
o
r
a
l
d
y
n
am
i
c
s
in
f
MR
I
d
a
t
a,
wh
i
ch
d
e
g
r
a
d
e
s
d
e
t
ec
t
i
o
n
p
er
f
o
r
m
a
n
ce
.
A
s
h
r
af
e
t
a
l
.
[
2
3
]
i
n
tr
o
d
u
c
e
d
a
5
7
-
l
ay
e
r
co
n
v
o
lu
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
wo
r
k
(
C
N
N
)
ar
c
h
i
te
c
t
u
r
e
n
am
e
d
N
eu
r
o
Ne
t
5
7
,
wh
i
ch
e
x
t
r
a
c
te
d
f
e
a
tu
r
e
s
f
r
o
m
f
a
c
tu
a
l
l
y
o
f
f
M
R
I
.
A
f
te
r
p
r
e
-
t
r
a
i
n
in
g
o
n
b
r
a
i
n
tu
m
o
r
d
a
ta
,
i
n
tr
o
d
u
ce
d
m
e
th
o
d
w
a
s
a
b
l
e
t
o
e
x
tr
a
c
t
f
e
m
a
l
e
p
h
e
n
o
t
y
p
i
c
f
e
a
t
u
r
e
s
f
r
o
m
th
e
au
t
i
s
m
b
r
a
in
i
m
ag
e
.
T
h
e
n
,
u
s
ed
an
an
t
c
o
lo
n
y
en
ab
le
d
s
y
s
t
e
m
t
o
s
e
le
c
t
a
f
ea
t
u
r
e
s
u
b
s
e
t
,
r
ed
u
c
i
n
g
ex
t
r
a
c
t
ed
f
e
a
tu
r
e
s
s
i
z
e
.
H
o
w
ev
e
r
,
in
t
r
o
d
u
c
e
d
C
N
N
m
o
d
e
l
t
e
n
d
s
to
lo
s
e
f
i
n
e
-
g
r
a
in
e
d
s
p
a
t
i
a
l
f
e
a
tu
r
e
s
in
i
t
s
i
n
i
t
i
a
l
l
ay
e
r
s
.
Kh
an
an
d
Kata
r
y
a
[
2
4
]
im
p
lem
en
ted
a
b
at
alg
o
r
ith
m
-
p
ar
ticle
s
war
m
o
p
tim
izat
io
n
-
L
STM
(
B
AT
-
PSO
-
L
STM
)
n
etwo
r
k
f
o
r
ASD
d
iag
n
o
s
is
.
Her
e,
u
tili
ze
d
th
r
ee
d
if
f
e
r
en
t
d
is
tin
ct
d
ata
s
ets
s
u
ch
as
ad
u
lts
,
ch
ild
r
en
an
d
to
d
d
le
r
s
,
f
o
r
co
m
p
r
eh
en
s
iv
e
a
n
aly
s
is
o
f
alg
o
r
ith
m
s
.
B
AT
an
d
PS
O
s
elec
t
t
h
e
f
ea
tu
r
es
an
d
f
ee
d
th
em
to
th
e
a
d
ap
tiv
e
f
ea
t
u
r
e
f
u
s
io
n
tech
n
iq
u
e
a
n
d
L
ST
M
class
if
ier
.
T
h
e
im
p
lem
en
t
ed
m
o
d
el
m
itig
ated
ch
allen
g
es
lik
e
o
v
e
r
f
itti
n
g
,
s
lo
w
tr
ain
in
g
,
m
o
d
el
in
ter
p
r
e
tab
ilit
y
,
g
en
er
aliza
tio
n
a
b
ilit
y
,
an
d
m
in
im
ized
tr
ain
in
g
tim
e.
T
h
e
im
p
lem
en
t
ed
d
ee
p
m
o
d
el
ca
u
s
es
o
v
er
f
itt
in
g
in
s
m
all
n
eu
r
o
im
ag
in
g
d
ata,
wh
ich
m
in
im
izes
g
en
er
aliza
tio
n
ab
ilit
y
o
f
t
h
e
m
o
d
el.
Srir
am
ak
r
is
h
n
an
et
a
l.
[
2
5
]
d
e
v
elo
p
e
d
f
r
ac
tio
n
al
wh
ale
-
d
r
iv
in
g
tr
ai
n
in
g
-
b
ased
o
p
tim
izatio
n
(
FW
C
T
B
O
)
with
C
NN
-
en
ab
led
tr
a
n
s
f
er
lear
n
in
g
(
T
L
)
to
d
etec
t
ASD.
Dev
elo
p
ed
m
o
d
el
was
d
esig
n
ed
th
r
o
u
g
h
in
cl
u
d
i
n
g
f
r
ac
tio
n
al
ca
lcu
lu
s
(
FC
)
,
wh
ale
o
p
tim
izatio
n
alg
o
r
ith
m
(
W
OA)
,
an
d
d
r
iv
in
g
tr
ain
in
g
-
b
ased
o
p
tim
izatio
n
(
DT
B
O)
,
wh
ich
tr
ain
e
d
C
NN
-
T
L
h
y
p
er
p
ar
am
eter
s
.
A
d
d
itio
n
ally
,
C
NN
u
s
e
d
h
y
p
er
p
ar
am
eter
s
f
r
o
m
th
e
tr
ain
ed
m
eth
o
d
s
s
u
ch
as
Sh
u
f
f
leNe
t
an
d
Alex
Net.
Fo
r
en
h
an
cin
g
d
etec
tio
n
ef
f
icac
y
,
th
e
n
u
b
ar
ea
was
id
en
tifie
d
an
d
p
r
o
ce
s
s
ed
u
s
in
g
a
f
u
n
ctio
n
al
c
o
n
n
ec
tiv
ity
-
en
a
b
led
wh
ale
d
r
iv
i
n
g
tr
ain
in
g
o
p
tim
izatio
n
(
W
DT
B
O)
ap
p
r
o
ac
h
.
De
v
elo
p
ed
m
o
d
el
f
ailed
to
f
o
c
u
s
o
n
m
u
c
h
in
f
o
r
m
ativ
e
d
ata,
wh
ich
d
eg
r
a
d
es
th
e
d
etec
tio
n
p
er
f
o
r
m
an
ce
.
T
r
ad
itio
n
al
DL
-
b
ased
alg
o
r
it
h
m
s
s
tr
u
g
g
le
to
co
m
p
letely
e
x
tr
ac
t
co
m
p
lex
s
p
atio
-
tem
p
o
r
al
d
e
p
e
n
d
en
cies
th
at
e
x
is
t
in
f
MRI
d
ata,
lim
itin
g
th
eir
d
etec
tio
n
ac
cu
r
ac
y
f
o
r
ASD.
I
t
s
tr
u
g
g
les
with
f
ea
tu
r
e
r
ed
u
n
d
an
cy
;
lo
s
s
o
f
s
p
atial
s
tr
u
ctu
r
e
an
d
o
v
er
f
itti
n
g
is
s
u
es
in
teg
r
ated
with
h
ig
h
-
d
im
e
n
s
io
n
al
n
eu
r
o
im
ag
i
n
g
d
ata.
Mo
r
e
o
v
er
,
class
im
b
alan
ce
in
ASD
s
ev
er
ity
lev
el
m
in
im
izes
th
e
m
o
d
el’
s
r
o
b
u
s
tn
ess
an
d
g
en
er
aliza
tio
n
ab
ilit
y
.
T
h
e
ex
is
tin
g
alg
o
r
ith
m
s
d
o
n
’
t
co
m
b
in
e
g
lo
b
al
s
p
atial
p
atter
n
s
an
d
tem
p
o
r
al
d
y
n
am
ics.
T
h
e
m
ain
aim
o
f
th
is
m
an
u
s
cr
ip
t
is
to
d
ev
elo
p
a
DL
f
r
am
ewo
r
k
t
h
at
co
r
r
ec
tly
d
etec
ts
ASD
b
y
f
MRI
d
ata.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
aim
s
t
o
ef
f
icien
tly
ca
p
tu
r
e
s
p
atio
tem
p
o
r
al
p
atter
n
s
b
y
a
co
m
b
in
atio
n
o
f
a
s
elf
-
atten
tio
n
m
ec
h
an
is
m
an
d
an
L
STM
ar
ch
itectu
r
e.
T
h
is
p
r
o
ce
s
s
im
p
r
o
v
es f
ea
tu
r
e
lear
n
in
g
by
WT
f
o
r
m
u
lti
-
r
eso
lu
tio
n
a
n
aly
s
is
an
d
PC
A
f
o
r
d
im
en
s
io
n
ality
r
ed
u
cti
o
n
.
T
h
is
m
o
d
el
m
itig
ates
th
e
is
s
u
es
lik
e
f
ea
tu
r
e
r
ed
u
n
d
an
c
y
,
s
p
atial
d
ata
lo
s
s
an
d
clas
s
im
b
al
an
ce
,
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
an
d
g
en
er
aliza
tio
n
ab
ilit
y
.
T
h
e
p
r
im
ar
y
c
o
n
tr
ib
u
tio
n
s
o
f
th
e
m
an
u
s
cr
ip
t a
r
e
d
esc
r
ib
ed
as f
o
llo
ws
:
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
810
-
8
2
3
812
i)
Dev
elo
p
ed
a
n
o
v
el
n
e
u
r
o
d
e
v
elo
p
m
en
tal
d
u
al
atten
tio
n
L
STM
n
etwo
r
k
(
Neu
r
o
-
DANet
)
m
o
d
el
wh
ic
h
co
m
b
in
es
d
u
al
-
atten
tio
n
d
ee
p
n
eu
r
al
n
etwo
r
k
(
DNN)
with
L
STM
f
o
r
ca
p
tu
r
in
g
s
p
atial
an
d
tem
p
o
r
al
d
ep
en
d
e
n
ce
s
in
f
MRI
d
ata,
w
h
ich
en
h
a
n
ce
s
ASD
s
ev
er
ity
d
etec
tio
n
an
d
s
ev
er
ity
class
if
icatio
n
th
r
o
u
g
h
lear
n
in
g
d
ee
p
an
d
co
n
te
x
t
-
awa
r
e
f
ea
tu
r
es.
ii)
E
m
p
lo
y
ed
co
n
tin
u
o
u
s
wav
elet
tr
an
s
f
o
r
m
(
C
W
T
)
f
o
r
ca
p
tu
r
i
n
g
m
u
lti
-
s
ca
le
f
ea
tu
r
es
f
r
o
m
f
MRI
im
ag
es,
wh
ich
ca
p
tu
r
es
g
lo
b
al
an
d
f
in
e
-
g
r
ain
ed
p
atter
n
s
.
T
h
is
im
p
r
o
v
es
th
e
s
en
s
itiv
ity
o
f
th
e
m
o
d
el
f
o
r
s
u
b
tle
ASD
-
r
elate
d
ab
n
o
r
m
alities
.
iii)
I
n
clu
d
ed
th
e
s
elf
-
atten
tio
n
m
e
ch
an
is
m
f
o
r
f
o
cu
s
o
n
p
r
im
ar
y
b
r
ain
ar
e
as
an
d
tim
e
s
eq
u
en
c
es,
en
h
an
cin
g
f
ea
tu
r
e
r
elev
a
n
ce
an
d
d
etec
tio
n
clar
ity
.
T
h
is
atten
tio
n
m
ec
h
an
is
m
h
ig
h
lig
h
ts
m
an
y
in
f
o
r
m
ativ
e
p
atter
n
s
f
o
r
ASD
d
etec
tio
n
an
d
s
ev
er
it
y
class
if
icatio
n
.
T
h
is
r
esear
ch
m
an
u
s
cr
ip
t
is
s
y
s
tem
ized
as
f
o
llo
ws:
s
ec
tio
n
2
p
r
o
v
id
es
a
d
etailed
ex
p
l
an
atio
n
o
f
th
e
p
r
o
p
o
s
ed
m
o
d
el.
Sectio
n
3
g
iv
es
th
e
o
u
tco
m
es
an
d
co
m
p
ar
is
o
n
o
f
a
p
r
o
p
o
s
ed
m
o
d
el.
S
ec
tio
n
4
co
n
clu
d
es
a
m
an
u
s
cr
ip
t.
2.
RE
S
E
ARCH
M
E
T
H
O
D
I
n
th
is
ar
ticle
d
ev
elo
p
ed
th
e
Neu
r
o
-
DANe
t,
wh
ich
in
teg
r
a
tes
s
elf
-
atten
tio
n
DNN
an
d
L
STM
f
o
r
ASD
d
etec
tio
n
.
T
h
e
d
atasets
u
s
ed
f
o
r
th
is
m
an
u
s
cr
ip
t
ar
e
a
u
tis
m
b
r
ain
im
ag
in
g
d
ata
e
x
ch
a
n
g
e
(
AB
I
DE
)
-
I
a
n
d
AB
I
DE
-
I
I
,
th
en
th
e
im
ag
es
ar
e
p
r
e
-
p
r
o
ce
s
s
ed
b
y
u
s
in
g
im
ag
e
r
esizin
g
an
d
d
ata
au
g
m
en
tatio
n
.
T
h
e
m
ea
n
in
g
f
u
l
f
ea
t
u
r
es
ar
e
ex
tr
a
cted
b
y
u
s
in
g
W
T
a
n
d
d
im
en
s
io
n
ality
is
m
in
im
ized
b
y
u
s
i
n
g
PC
A.
At
last
,
t
h
e
s
elf
-
atten
tio
n
DNN
-
L
STM
n
e
two
r
k
is
u
s
ed
to
d
etec
t
th
e
d
i
f
f
er
en
t
class
es
o
f
ASD.
Fig
u
r
e
1
r
ep
r
esen
ts
t
h
e
p
r
o
ce
s
s
o
f
ASD
s
ev
er
ity
class
if
icatio
n
u
s
in
g
s
elf
-
atten
tio
n
DNN
-
L
STM
.
Fig
u
r
e
1
.
Pro
ce
s
s
o
f
ASD
s
ev
er
ity
class
if
icatio
n
u
s
in
g
s
elf
-
atten
tio
n
DNN
-
L
STM
2
.
1
.
Da
t
a
s
et
I
n
th
is
a
r
ticle,
u
s
ed
AB
I
DE
d
ataset
m
u
c
h
claim
ed
f
o
r
i
ts
ex
ten
s
iv
e
n
e
u
r
o
im
ag
i
n
g
d
ata,
wh
ich
co
n
tain
s
AB
I
DE
-
I
[
2
6
]
an
d
AB
I
DE
-
II
[
2
7
]
.
Data
s
et
in
clu
d
es
f
MRI
s
ca
n
s
f
r
o
m
v
ar
io
u
s
p
ar
ticip
an
ts
.
T
h
e
AB
I
DE
-
I
d
ataset
co
n
tain
s
4
1
9
in
d
i
v
id
u
als
d
iag
n
o
s
ed
to
ASD
an
d
5
3
0
n
eu
r
o
ty
p
ical
co
n
tr
o
ls
.
AB
I
DE
-
II
d
ataset
co
n
tain
s
9
2
ASD
an
d
1
0
3
n
eu
r
o
ty
p
ical
co
n
tr
o
ls
.
T
h
is
s
u
b
s
tan
tial
p
ar
ticip
an
t
to
o
l,
s
o
u
r
ce
d
f
r
o
m
m
u
ltip
le
in
ter
n
atio
n
al
r
esear
c
h
f
ac
ilit
ies,
im
p
r
o
v
es
s
tatis
tical
r
o
b
u
s
tn
ess
an
d
s
u
p
p
o
r
ts
wid
e
g
en
e
r
aliza
tio
n
ab
ilit
y
.
T
ab
le
1
r
ep
r
esen
ts
th
e
d
ataset
d
escr
ip
tio
n
o
f
AB
I
DE
-
I
an
d
T
a
b
le
2
r
ep
r
esen
ts
th
e
d
ataset
d
escr
ip
tio
n
o
f
AB
I
DE
-
II
d
ataset,
an
d
Fig
u
r
e
s
2
an
d
3
s
h
o
w
th
e
s
am
p
le
i
m
ag
es o
f
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
.
T
ab
le
1
.
Data
s
et
d
escr
ip
tio
n
o
f
AB
I
DE
-
I
d
ataset
C
l
a
s
ses
A
S
D
TD
To
t
a
l
i
m
a
g
e
s
N
u
mb
e
r
o
f
sam
p
l
e
s
1
0
5
8
1
1
6
3
2
2
2
1
T
ab
le
2
.
Data
s
et
d
escr
ip
tio
n
o
f
AB
I
DE
-
II
d
ataset
C
l
a
s
ses
M
i
l
e
d
M
o
d
e
r
a
t
e
S
e
v
e
r
e
TD
To
t
a
l
i
m
a
g
e
s
N
u
mb
e
r
o
f
sam
p
l
e
s
1
6
0
2
7
7
2
2
8
4
4
5
1
1
1
0
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
N
eu
r
o
-
DA
N
et:
d
u
a
l a
tten
tio
n
d
ee
p
n
e
u
r
a
l n
etw
o
r
k
lo
n
g
s
h
o
r
t
-
term
…
(
S
u
ja
th
a
Ha
n
u
ma
n
th
a
r
a
ya
p
p
a
)
813
Fig
u
r
e
2
.
Sam
p
le
im
a
g
es o
f
A
B
I
DE
-
I
d
ataset
Fig
u
r
e
3
.
Sam
p
le
im
a
g
es
o
f
A
B
I
DE
-
II
d
ataset
2
.
2
.
P
re
-
pro
ce
s
s
ing
I
n
itially
,
f
o
r
AB
I
DE
-
II
d
ataset,
s
p
lit
th
e
d
ata
b
ased
o
n
th
e
s
o
cial
r
esp
o
n
s
iv
en
ess
s
ca
le
(
SR
S
)
T
-
v
alu
e
b
y
s
tan
d
ar
d
an
n
o
tatio
n
s
.
ASD
Sev
er
ity
is
clas
s
if
ied
in
to
4
g
r
o
u
p
s
f
o
r
ea
ch
d
o
m
ain
d
e
p
en
d
in
g
o
n
th
e
SR
S
v
alu
e.
SR
S
to
tal
T
-
s
co
r
e
p
r
o
v
id
ed
in
AB
I
DE
-
II
p
h
en
o
t
y
p
i
c
d
ata
is
ap
p
lied
to
f
MRI
s
u
b
jects,
d
u
e
to
SR
S
v
alu
es
ar
e
clin
ical
o
r
b
eh
a
v
i
o
r
al
m
ea
s
u
r
es,
n
o
t
b
ased
o
n
im
ag
in
g
m
o
d
ality
.
T
h
e
SR
S
v
alu
es
f
o
r
ev
e
r
y
in
d
iv
id
u
al
class
is
d
escr
ib
ed
a
s
f
o
llo
ws
:
–
T
y
p
ically
d
ev
el
o
p
in
g
(
T
D)
–
S
R
S v
alu
e
≤
59
–
Mild
ASD
–
SR
S v
alu
e
r
an
g
e
6
0
-
65
–
Mo
d
er
ate
ASD
–
SR
S v
alu
e
r
an
g
e
6
6
-
75
–
Sev
er
e
ASD
–
SR
S v
alu
e
≥
76
2
.
2
.
1
.
I
m
a
g
e
re
s
izing
I
n
th
is
ar
ticle,
th
e
im
a
g
es
f
r
o
m
f
MRI
s
ca
n
s
a
r
e
r
esized
to
a
f
ix
ed
d
im
en
s
io
n
o
f
2
2
4
×
2
2
4
.
T
h
is
r
esizin
g
is
ess
en
tial
b
ec
au
s
e
n
eu
r
o
im
a
g
in
g
d
ata
g
en
e
r
ally
v
ar
ies
in
r
eso
lu
tio
n
d
u
e
t
o
v
a
r
io
u
s
s
ca
n
n
er
s
an
d
p
o
s
itio
n
s
o
f
th
e
b
r
ain
s
lice.
T
h
e
DL
m
o
d
els
r
e
q
u
ir
e
u
n
if
o
r
m
in
p
u
t
d
im
e
n
s
io
n
s
f
o
r
ef
f
ec
tiv
e
b
atch
p
r
o
ce
s
s
in
g
an
d
tr
ain
in
g
s
tab
ilit
y
.
2
.
2
.
2
.
Da
t
a
a
ug
m
ent
a
t
io
n
I
n
th
is
ar
ticle,
d
ata
a
u
g
m
e
n
tatio
n
tech
n
iq
u
es
lik
e
f
lip
p
in
g
,
r
o
tatio
n
,
a
n
d
tr
a
n
s
latio
n
ar
e
u
t
ilized
f
o
r
s
im
u
latin
g
v
ar
iatio
n
in
MRI
a
cq
u
is
itio
n
with
o
u
t
c
h
an
g
i
n
g
t
h
e
d
iag
n
o
s
tic
v
alu
e
o
f
im
a
g
es.
T
h
ese
tech
n
i
q
u
es
ar
e
em
p
lo
y
ed
f
o
r
en
h
a
n
cin
g
g
en
er
aliza
tio
n
ab
ilit
y
o
f
th
e
m
o
d
el.
T
h
ese
tr
an
s
latio
n
s
s
im
u
late
g
en
er
al
v
ar
iatio
n
s
in
MRI
ac
q
u
is
itio
n
lik
e
h
em
is
p
h
er
ic
s
y
m
m
etr
y
,
h
ea
d
p
o
s
itio
n
in
g
th
at
alter
s
d
iag
n
o
s
tic
d
ata
i
n
f
MRI
s
lices.
Au
g
m
en
tatio
n
h
elp
s
th
e
m
o
d
el
to
lea
r
n
s
p
atia
lly
r
o
b
u
s
t
f
ea
tu
r
es
an
d
o
v
er
c
o
m
es
th
e
ef
f
ec
ts
o
f
class
im
b
alan
ce
in
ASD
s
ev
er
ity
d
ata.
–
Fli
p
p
in
g
:
it
r
ef
er
s
to
m
ir
r
o
r
in
g
im
ag
es
with
ax
is
lik
e
h
o
r
izo
n
tal
an
d
v
er
tical
f
lip
p
in
g
.
Fli
p
p
in
g
d
o
esn
’
t
d
is
to
r
t c
lin
ical
f
ea
tu
r
es a
n
d
h
e
lp
s
th
e
m
o
d
el
to
lea
r
n
in
v
ar
ian
t f
ea
tu
r
es o
n
b
o
th
s
id
es o
f
t
h
e
b
r
ain
.
–
R
o
t
a
t
i
o
n
:
i
t
h
e
l
p
s
t
h
e
m
o
d
e
l
to
le
a
r
n
r
o
t
at
i
o
n
-
i
n
v
ar
i
a
n
t
f
ea
t
u
r
e
s
an
d
m
i
n
i
m
iz
e
s
o
v
e
r
f
i
t
t
i
n
g
f
o
r
a
l
i
g
n
m
e
n
t
p
a
t
t
er
n
s
.
–
T
r
an
s
latio
n
:
it
s
h
if
ts
th
e
wh
o
le
im
ag
e
h
o
r
izo
n
tally
,
v
e
r
tically
b
y
a
f
ew
p
ix
els
o
r
m
illi
m
eter
s
.
T
h
is
en
s
u
r
es
a
m
o
d
el
f
o
r
r
ec
o
g
n
izin
g
f
ea
tu
r
es
r
e
g
ar
d
less
o
f
th
eir
ac
cu
r
ate
s
p
atial
p
o
s
itio
n
,
en
h
a
n
cin
g
r
o
b
u
s
tn
ess
f
o
r
m
i
n
o
r
s
p
atial
s
h
if
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
810
-
8
2
3
814
2
.
3
.
F
e
a
t
ure
ex
t
r
a
ct
io
n
W
T
p
r
o
v
id
e
s
t
h
e
ca
p
ab
i
l
i
t
y
f
o
r
d
ec
o
m
p
o
s
in
g
f
M
R
I
im
a
g
es
i
n
to
m
u
l
t
ip
l
e
f
r
e
q
u
e
n
c
y
an
d
r
e
s
o
l
u
t
io
n
l
e
v
e
l
s
,
en
a
b
l
in
g
t
h
a
t
t
o
ex
t
r
a
c
t
b
o
t
h
g
l
o
b
a
l
s
t
r
u
c
t
u
r
e
p
a
tt
e
r
n
s
an
d
f
i
n
e
-
g
r
a
i
n
e
d
in
f
o
r
m
a
t
i
o
n
.
T
r
a
d
i
t
io
n
a
l
C
N
N
m
o
d
e
l
s
l
i
k
e
M
o
b
i
l
e
Ne
t
,
R
e
s
N
e
t
l
e
ar
n
t
h
e
f
e
a
t
u
r
e
s
b
y
s
t
a
c
k
e
d
c
o
n
v
o
l
u
t
io
n
s
w
i
t
h
o
u
t
e
x
p
l
i
c
i
t
l
y
s
e
p
a
r
a
t
in
g
th
e
f
r
eq
u
en
cy
s
c
a
le
s
,
w
h
i
ch
l
i
m
i
t
s
i
t
s
s
en
s
i
t
i
v
i
ty
f
o
r
s
u
b
t
l
e
te
x
tu
r
e
s
,
e
s
p
e
c
i
a
ll
y
i
n
in
i
t
i
a
l
la
y
er
s
.
A
d
d
i
t
io
n
a
l
ly
,
W
T
i
s
in
h
e
r
en
tl
y
a
p
p
r
o
p
r
i
a
t
e
f
o
r
s
m
a
l
l
d
a
t
a
b
e
c
au
s
e
o
f
i
t
s
h
a
n
d
cr
a
f
t
ed
a
n
d
d
o
m
a
i
n
-
a
w
ar
e
d
e
v
e
lo
p
m
e
n
t,
w
h
er
e
d
e
e
p
C
N
N
m
o
d
e
l
s
g
e
n
er
a
l
ly
o
v
e
r
-
p
a
r
a
m
e
t
er
i
z
ed
f
o
r
l
i
m
i
te
d
f
MR
I
d
a
t
a
a
n
d
l
e
ad
s
t
o
o
v
er
f
i
t
t
in
g
.
I
n
i
t
ia
l
l
y
,
t
i
m
e
-
f
r
eq
u
en
c
y
e
l
e
m
en
t
s
a
r
e
e
x
tr
a
c
t
ed
i
n
ev
e
r
y
s
i
g
n
a
l
th
r
o
u
g
h
th
e
CWT
.
C
o
e
f
f
i
c
i
en
t
o
f
C
W
T
i
s
d
e
t
e
r
m
i
n
ed
a
s
s
i
g
n
a
l
co
n
v
o
lu
t
i
o
n
(
)
w
i
t
h
a
t
r
a
n
s
l
a
te
d
a
n
d
s
c
a
l
ed
v
e
r
s
io
n
o
f
w
a
v
e
le
t
,
(
)
w
h
i
c
h
i
s
g
iv
e
n
a
s
(
1
)
.
(
,
)
=
1
√
∫
(
)
.
∗
(
−
)
∞∞
−
∞
(
1
)
I
n
(
1
)
,
th
e
is
wav
elet
s
ca
le,
th
e
is
tim
e
s
h
if
t
lo
ca
tio
n
a
n
d
∗
is
d
if
f
icu
lt
co
n
ju
g
ate.
A
Mo
r
let
wav
elet
is
ch
o
s
en
as
a
m
o
th
er
wav
elet
wh
ich
h
as
a
b
etter
p
r
o
p
o
r
tio
n
(
1
.
0
3
)
am
o
n
g
f
r
eq
u
en
cy
b
an
d
a
n
d
wav
elet
s
ca
le,
th
at
s
u
p
p
o
r
ts
f
o
r
in
ter
p
r
etin
g
o
u
tco
m
es
in
a
f
r
eq
u
e
n
cy
d
o
m
ain
.
T
h
r
o
u
g
h
v
ar
y
in
g
a
wav
elet
s
ca
le
an
d
tr
an
s
latio
n
with
a
l
o
ca
lized
tim
e
in
d
ex
,
a
n
im
ag
e
is
g
en
er
ated
th
at
r
e
p
r
esen
ts
th
e
am
p
litu
d
e
o
f
f
ea
tu
r
es
ac
r
o
s
s
d
if
f
er
en
t
s
ca
l
es.
Am
p
litu
d
e
is
v
ar
ied
with
tim
e
f
o
r
g
e
n
er
atin
g
s
ca
lo
g
r
a
m
im
ag
es,
th
at
is
,
ac
cu
r
ate
v
alu
e
o
f
W
T
co
ef
f
ici
en
ts
.
W
T
ex
tr
ac
ts
s
p
atial
an
d
f
r
eq
u
e
n
cy
d
ata
in
m
u
ltip
le
r
es
o
lu
tio
n
s
,
allo
win
g
f
o
r
ca
p
tu
r
in
g
m
ea
n
in
g
f
u
l
p
att
er
n
s
f
r
o
m
f
MRI
d
ata.
T
h
is
h
e
lp
s
to
h
ig
h
lig
h
t
s
u
b
tle
b
r
ain
a
ctiv
ity
d
if
f
er
en
ce
s
lik
e
ASD
th
r
o
u
g
h
p
r
eser
v
in
g
ess
en
tial
s
tr
u
ctu
r
al
an
d
f
u
n
ctio
n
al
in
f
o
r
m
atio
n
wh
ile
m
in
im
izin
g
n
o
is
e.
T
r
ad
e
-
o
f
f
o
f
WT
:
–
L
o
wer
lev
els (
1
-
2
)
r
etain
h
ig
h
-
f
r
eq
u
e
n
cy
n
o
is
e
an
d
m
icr
o
-
p
a
tter
n
s
,
th
at
d
o
n
o
t g
e
n
er
alize
well.
–
Hig
h
er
lev
els (
5
-
6
)
o
v
er
-
s
m
o
o
th
in
g
im
ag
es,
lo
s
in
g
ess
en
tial a
n
ato
m
ical
in
f
o
r
m
atio
n
.
I
n
th
is
a
r
ticle,
lev
el
4
as
th
e
o
p
tim
u
m
b
alan
ce
,
wh
ich
p
r
eser
v
es
f
in
e
to
m
id
r
a
n
g
e
in
f
o
r
m
atio
n
a
n
d
g
lo
b
al
b
r
ain
s
tr
u
ctu
r
e.
I
t
ef
f
i
cien
tly
b
alan
ce
s
th
e
in
f
o
r
m
at
io
n
r
eten
tio
n
an
d
m
in
im
izes
n
o
is
e,
ess
en
tial
to
d
if
f
er
en
tiate
s
u
b
tle
ASD
v
ar
iatio
n
s
.
Fro
m
th
e
WT
,
a
to
t
al
5
5
,
9
0
6
f
ea
tu
r
es
ar
e
ex
tr
a
cted
an
d
g
i
v
en
t
o
th
e
d
im
en
s
io
n
ality
r
ed
u
ctio
n
p
h
ase.
Fig
u
r
es
4
a
n
d
5
r
ep
r
esen
t
th
e
ex
t
r
ac
ted
f
ea
t
u
r
es
f
o
r
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
.
Fig
u
r
e
4
.
E
x
tr
ac
ted
f
ea
t
u
r
es f
o
r
AB
I
DE
-
I
d
ataset
Fig
u
r
e
5
.
E
x
tr
ac
ted
f
ea
t
u
r
es f
o
r
AB
I
DE
-
II
d
ataset
2
.
4
.
P
rincipa
l c
o
m
po
nent
a
na
ly
s
is
f
o
r
dim
ens
io
na
lity
re
du
ct
io
n
PC
A
p
r
es
er
v
es
th
e
i
n
t
er
n
al
a
r
c
h
it
ec
t
u
r
e
o
f
w
av
el
et
-
tr
a
n
s
f
o
r
m
ed
f
MRI
f
e
at
u
r
es
t
h
r
o
u
g
h
ca
p
t
u
r
in
g
p
r
in
ci
p
a
l
a
x
es
o
f
v
a
r
i
a
n
ce
t
h
at
r
e
f
l
ec
t
m
u
c
h
in
f
o
r
m
a
ti
v
e
a
n
d
co
r
r
el
ate
d
s
p
a
tia
l
-
f
r
e
q
u
e
n
c
y
p
atte
r
n
s
.
T
h
is
m
o
d
el
en
s
u
r
es
t
h
e
p
r
i
m
a
r
y
s
ta
tis
ti
ca
l
s
tr
u
ct
u
r
e
o
f
b
r
a
in
a
b
n
o
r
m
ali
tie
s
i
n
t
eg
r
ate
d
wi
th
t
h
e
s
e
v
er
i
ty
o
f
ASD
t
h
at
r
e
m
ai
n
s
in
t
ac
t
in
a
m
i
n
i
m
i
ze
d
d
im
en
s
i
o
n
al
f
e
at
u
r
e
s
p
ac
e,
f
ac
ili
tati
n
g
e
f
f
ec
ti
v
e
d
et
ec
ti
o
n
.
PC
A
i
d
e
n
ti
f
i
es
th
e
p
r
i
n
ci
p
al
co
m
p
o
n
e
n
ts
a
n
d
th
e
d
i
r
e
cti
o
n
s
i
n
f
ea
tu
r
e
s
p
ac
e
th
at
e
x
t
r
ac
t
m
an
y
v
ar
ia
n
c
es.
H
o
we
v
e
r
,
C
h
i
-
s
q
u
a
r
e
a
n
d
ANOV
A
a
r
e
s
u
i
ta
b
le
f
o
r
ca
t
eg
o
r
ic
al
f
ea
t
u
r
es
,
w
h
e
r
e
as
h
er
e
t
h
e
f
e
at
u
r
es
a
r
e
c
o
n
ti
n
u
o
u
s
v
al
u
es
.
Fi
g
u
r
es
6
a
n
d
7
r
e
p
r
ese
n
t
t
h
e
t
r
a
n
s
f
o
r
m
e
d
f
ea
t
u
r
es
u
s
i
n
g
PC
A
o
n
AB
I
D
E
-
I
a
n
d
AB
I
D
E
-
I
I
d
a
tase
ts
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
N
eu
r
o
-
DA
N
et:
d
u
a
l a
tten
tio
n
d
ee
p
n
e
u
r
a
l n
etw
o
r
k
lo
n
g
s
h
o
r
t
-
term
…
(
S
u
ja
th
a
Ha
n
u
ma
n
th
a
r
a
ya
p
p
a
)
815
Fig
u
r
e
6
.
T
r
an
s
f
o
r
m
e
d
f
ea
tu
r
e
s
f
o
r
AB
I
DE
-
I
d
ataset
Fig
u
r
e
7
.
T
r
an
s
f
o
r
m
e
d
f
ea
tu
r
e
s
f
o
r
AB
I
DE
-
II
d
ataset
2
.
5
.
Cla
s
s
if
ica
t
io
n
Self
-
atten
tio
n
m
ec
h
an
is
m
is
ef
f
icien
t
f
o
r
an
al
y
s
is
o
f
f
MRI
b
ec
au
s
e
th
at
ex
tr
ac
ts
lo
n
g
-
r
a
n
g
e
s
p
atial
an
d
tem
p
o
r
al
r
elatio
n
s
h
ip
s
,
with
o
u
t
r
ely
in
g
o
n
a
s
eq
u
en
tial
p
r
o
ce
s
s
.
T
h
is
p
r
o
ce
s
s
allo
ws
f
ast
tr
ain
in
g
an
d
g
o
o
d
s
ca
lin
g
with
h
u
g
e
f
MRI
d
ata.
I
n
t
h
is
ar
ticle,
s
elf
-
atten
tio
n
in
d
u
al
atten
tio
n
b
lo
ck
s
f
o
cu
s
es
o
n
ess
en
tial
b
r
ain
a
r
ea
s
an
d
tim
e
p
o
in
ts
,
i
m
p
r
o
v
i
n
g
f
ea
tu
r
e
r
ep
r
esen
tatio
n
wh
ile
m
ain
tain
in
g
tr
ain
i
n
g
s
tab
ilit
y
b
y
r
esid
u
al
co
n
n
ec
tio
n
s
.
An
a
d
d
itio
n
al
p
o
s
t
-
ad
d
itio
n
s
elf
-
atten
tio
n
lay
er
r
ef
in
es
t
h
ese
f
ea
tu
r
es
b
ef
o
r
e
f
ee
d
in
g
th
em
in
t
o
L
STM
,
wh
ich
m
o
d
els
th
e
tem
p
o
r
al
d
y
n
a
m
ics.
T
h
is
in
teg
r
ated
alg
o
r
ith
m
en
h
a
n
ce
s
th
e
m
o
d
el’
s
ca
p
ab
ilit
y
f
o
r
id
en
tify
in
g
s
u
b
tle
an
d
c
o
m
p
l
ex
b
r
ai
n
p
atter
n
s
,
ca
u
s
in
g
m
u
ch
p
r
ec
is
e
f
MRI
d
etec
tio
n
w
h
en
co
m
p
ar
e
d
with
tr
ad
itio
n
al
atten
tio
n
m
ec
h
an
is
m
s
.
Fig
u
r
e
8
r
e
p
r
esen
ts
th
e
ar
c
h
itectu
r
e
o
f
Ne
u
r
o
-
DANe
t m
o
d
el
f
o
r
ASD.
Fig
u
r
e
8
.
Ar
c
h
itectu
r
e
o
f
Neu
r
o
-
DANe
t m
o
d
el
f
o
r
ASD
d
ete
ctio
n
2
.
5
.
1
.
Dee
p
neura
l net
wo
rk
DNN
is
cla
s
s
o
f
ar
tific
ial
n
eu
r
al
n
etwo
r
k
(
ANN)
in
clu
d
es
m
u
ltip
le
p
r
o
ce
s
s
in
g
lay
er
s
,
ab
le
to
lear
n
d
if
f
icu
lt
f
ea
tu
r
es
an
d
p
atter
n
s
f
r
o
m
d
ata.
L
a
y
er
s
in
a
DNN
ar
e
class
if
ied
in
to
in
p
u
t,
h
id
d
en
,
an
d
o
u
tp
u
t
lay
er
s
,
with
ev
er
y
lay
er
b
ein
g
f
u
lly
c
o
n
n
ec
ted
.
T
h
is
a
r
ch
itectu
r
e
en
ab
led
a
m
eth
o
d
f
o
r
p
r
o
ce
s
s
in
g
an
d
an
aly
zin
g
d
ata
at
d
if
f
er
en
t
ab
s
tr
ac
tio
n
lev
els,
ef
f
ec
tiv
ely
im
p
r
o
v
in
g
th
eir
p
r
ed
ictiv
e
ab
ilit
ies.
DNN
i
s
t
r
ain
ed
b
y
f
o
r
war
d
p
r
o
p
a
g
atio
n
an
d
b
ac
k
p
r
o
p
ag
a
tio
n
ap
p
r
o
ac
h
es,
an
d
g
r
a
d
ien
t
d
escen
t
is
u
tili
ze
d
f
o
r
o
p
tim
izin
g
th
e
weig
h
ts
.
I
n
p
r
o
ce
s
s
o
f
f
o
r
war
d
p
r
o
p
a
g
atio
n
,
ev
er
y
n
eu
r
o
n
r
ec
ei
v
es
in
p
u
t
s
f
r
o
m
th
e
p
r
io
r
lay
er
,
wh
ich
i
s
tr
an
s
m
itted
b
y
a
weig
h
ted
s
u
m
,
f
o
llo
we
d
b
y
n
o
n
-
lin
ea
r
ac
tiv
atio
n
f
u
n
ctio
n
f
o
r
g
en
er
atin
g
th
e
r
esu
lt.
T
h
e
m
ath
e
m
atica
l
ex
p
r
ess
io
n
f
o
r
th
is
p
r
o
ce
s
s
is
g
iv
en
in
(
2
)
.
[
]
=
(
[
]
[
−
1
]
+
[
]
)
(
2
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
810
-
8
2
3
816
Fo
r
a
g
iv
en
lay
er
,
th
eir
r
esu
lt is
r
ep
r
esen
ted
as
[
]
wh
ich
is
ex
ec
u
ted
b
y
em
p
lo
y
in
g
an
ac
tiv
atio
n
f
u
n
ctio
n
is
em
p
lo
y
e
d
o
n
th
e
w
eig
h
ted
s
u
m
in
p
u
t a
n
d
b
ias ter
m
f
o
r
th
at
lay
er
.
Her
e,
weig
h
ted
in
p
u
t is ac
q
u
ir
e
d
th
r
o
u
g
h
m
u
ltip
ly
i
n
g
th
e
r
esu
lt
f
r
o
m
p
r
io
r
la
y
er
[
−
1
]
with
weig
h
t
m
atr
ix
[
]
in
th
e
p
r
esen
t
lay
er
.
T
h
e
[
]
r
ep
r
esen
ts
a
b
ias
ter
m
,
wh
ic
h
is
in
tr
o
d
u
ce
d
f
o
r
ad
d
itio
n
a
l
ad
ju
s
tm
en
t,
an
d
f
ac
ilit
ates
m
o
d
el
p
er
f
o
r
m
an
c
e
o
p
tim
is
atio
n
.
Af
ter
ac
q
u
ir
in
g
th
e
g
r
ad
ien
t,
g
r
ad
ien
t
d
esce
n
t
is
u
tili
ze
d
f
o
r
u
p
d
atin
g
th
e
weig
h
ts
o
f
th
e
n
etwo
r
k
.
T
h
e
m
ath
em
atica
l
ex
p
r
ess
io
n
f
o
r
t
h
e
weig
h
t
u
p
d
ate
f
o
r
m
u
la
is
g
iv
en
in
(
3
)
.
I
n
th
e
(
3
)
,
th
e
r
ep
r
esen
ts
lear
n
in
g
r
ate.
=
−
(
3
)
2
.
5
.
2
.
Self
-
a
t
t
ent
i
o
n m
ec
ha
nis
m
Fo
r
g
lo
b
al
d
ep
en
d
en
cies
o
f
in
p
u
t
s
en
ten
ce
,
two
s
ig
n
if
ican
t
r
ea
s
o
n
s
th
at
n
ee
d
to
b
e
m
itig
ated
:
i)
m
ea
s
u
r
in
g
atten
tio
n
o
f
ev
e
r
y
wo
r
d
in
an
in
p
u
t
s
eq
u
en
c
e
an
d
ii)
ex
tr
ac
tin
g
s
en
ten
ce
s
eq
u
en
ce
d
ata.
Fo
r
ca
p
tu
r
in
g
g
lo
b
al
d
e
p
en
d
e
n
cies
o
f
im
ag
es
in
d
ata,
m
ea
s
u
r
e
th
e
atten
tio
n
o
f
ev
e
r
y
im
ag
e
in
th
e
d
ata
at
f
ir
s
t.
Self
-
atten
tio
n
is
m
u
ch
s
ca
lab
le
an
d
p
ar
allel
atten
tio
n
m
ea
s
u
r
in
g
tech
n
i
q
u
e.
Self
-
atten
ti
o
n
is
co
n
s
id
er
ed
as
co
n
ten
t
-
en
a
b
led
q
u
er
y
p
r
o
ce
s
s
,
th
at
ex
ec
u
tes
atten
tio
n
f
u
n
ctio
n
o
n
g
r
o
u
p
o
f
q
u
e
r
ies
an
d
p
ac
k
a
g
es
th
at
to
m
atr
ix
Q.
I
n
s
im
ilar
tim
e,
k
e
y
s
an
d
v
alu
es
a
r
e
f
illed
to
m
atr
ices
an
d
.
I
n
an
ar
ea
o
f
im
ag
e
p
r
o
ce
s
s
in
g
,
s
elf
-
atten
tio
n
m
ec
h
a
n
is
m
is
u
t
ilized
f
o
r
d
e
p
en
d
e
n
cy
o
f
ev
er
y
wo
r
d
with
e
n
tire
s
ig
n
als.
Fo
r
o
b
tain
in
g
m
o
d
ality
co
r
r
elatio
n
,
th
e
d
o
t
p
r
o
d
u
ct
is
co
m
p
u
ted
.
T
h
e
m
atr
ix
r
es
u
ltin
g
f
r
o
m
a
m
u
ltip
licatio
n
o
f
an
d
is
a
r
elatio
n
s
h
ip
am
o
n
g
ea
c
h
im
a
g
e
an
d
wh
o
le
o
th
er
im
ag
es.
C
o
r
r
elatio
n
v
alu
e
o
f
ev
er
y
m
o
d
ality
is
p
r
o
d
u
ce
d
th
r
o
u
g
h
s
o
f
tm
ax
f
u
n
cti
o
n
.
At
last
,
th
is
v
alu
e
is
m
u
ltip
lied
with
m
ap
p
in
g
m
atr
ix
V
o
f
S
f
o
r
ac
q
u
ir
in
g
wo
r
d
s
r
ep
r
esen
tatio
n
to
e
x
tr
ac
t
g
lo
b
al
d
ep
e
n
d
en
ce
d
ata.
Self
-
atten
tio
n
m
ec
h
an
is
m
is
e
x
ec
u
tin
g
t
h
e
atten
tio
n
weig
h
t
in
to
ev
er
y
m
o
d
ality
,
an
d
its
m
ath
em
atica
l
ex
p
r
ess
io
n
is
g
i
v
en
as
(
4
)
.
I
n
th
e
(
4
)
,
th
e
=
{
ℎ
1
,
ℎ
2
,
…
,
ℎ
}
∈
×
ℎ
an
d
=
{
ℎ
1
,
ℎ
2
,
…
,
ℎ
}
∈
×
ℎ
r
ep
r
esen
t
a
h
id
d
en
r
e
p
r
e
s
en
tatio
n
.
Nex
t,
,
,
an
d
V
r
ep
r
esen
ts
m
ap
p
ed
m
atr
ices,
th
at
is
in
iti
alize
d
with
m
u
ltip
ly
in
g
in
p
u
t
em
b
ed
d
in
g
an
d
r
esp
ec
tiv
e
weig
h
t
m
atr
ix
an
d
its
m
ath
em
atica
l
ex
p
r
ess
io
n
is
g
iv
en
as
(
5
)
to
(
7
)
.
I
n
(
5
)
to
(
7
)
,
th
e
,
,
∈
×
2
r
ep
r
esen
ts
m
ap
p
in
g
s
o
f
s
eg
m
en
t
r
ep
r
esen
tatio
n
with
d
ata
=
{
ℎ
̃
1
,
ℎ
̃
2
,
…
,
ℎ
̃
}
∈
×
ℎ
,
th
e
,
r
ep
r
esen
ts
lear
n
ab
le
p
ar
am
eter
m
atr
ices.
C
alcu
late
its
atten
tio
n
v
alu
es
f
o
r
ac
q
u
ir
in
g
s
elf
-
atten
tio
n
r
ep
r
esen
tatio
n
an
d
its
m
ath
em
atica
l
ex
p
r
ess
io
n
is
g
i
v
en
as
(
8
)
.
I
n
(
8
)
,
t
h
e
√
r
ep
r
es
en
ts
a
s
ca
lin
g
f
ac
to
r
a
n
d
th
at
d
im
en
s
io
n
is
s
etted
to
a
h
id
d
en
r
ep
r
esen
tati
o
n
.
ℎ
̃
=
(
ℎ
+
)
,
∈
[
1
,
]
(
4
)
=
̃
(
5
)
=
̃
(
6
)
=
̃
(
7
)
=
(
√
)
(
8
)
2
.
5
.
3
.
L
o
ng
s
ho
rt
-
t
er
m m
e
mo
ry
L
STM
is
v
ar
ian
t
o
f
r
ec
u
r
r
en
t
n
eu
r
al
n
etwo
r
k
(
R
NN
)
,
f
o
r
tim
e
s
er
ies
d
ata,
wh
ich
is
m
u
ch
s
u
itab
le
to
ca
p
tu
r
e
d
if
f
er
en
t d
ata
f
ea
tu
r
es b
ec
au
s
e
o
f
a
d
d
itio
n
al
s
to
r
ag
e
u
n
its
ab
le
to
s
to
r
e
h
is
to
r
ical
d
ata.
L
STM
n
etwo
r
k
ar
ch
itectu
r
e
em
p
lo
y
e
d
th
r
ee
g
ates
s
u
ch
as
f
o
r
g
et,
i
n
p
u
t
an
d
o
u
tp
u
t
g
ates
in
e
v
er
y
L
STM
n
eu
r
o
n
.
T
h
e
L
STM
n
etwo
r
k
p
r
o
ce
s
s
b
etter
wh
en
th
e
in
p
u
t
f
ea
tu
r
es
ar
e
i
n
d
ep
e
n
d
en
t
o
f
ea
c
h
o
th
er
.
T
h
e
u
n
c
o
r
r
elatio
n
n
atu
r
e
o
f
PC
A
-
tr
an
s
f
o
r
m
ed
f
ea
tu
r
es
h
el
p
s
th
e
L
STM
m
o
d
el
to
f
o
cu
s
o
n
lear
n
in
g
tem
p
o
r
al
r
elatio
n
s
h
ip
s
in
d
ata
in
s
tead
o
f
h
an
d
lin
g
th
e
co
r
r
elatio
n
am
o
n
g
f
ea
tu
r
es.
T
h
is
p
r
o
ce
s
s
m
ak
es
th
e
tr
ain
in
g
p
r
o
ce
s
s
m
u
ch
s
tab
le
an
d
ef
f
ec
tiv
e.
T
h
e
r
ep
r
esen
ts
th
e
r
est
o
f
th
e
in
f
o
r
m
atio
n
o
n
p
r
i
o
r
L
STM
n
eu
r
o
n
a
f
ter
f
ed
to
f
o
r
g
et
g
ate
,
th
e
r
ep
r
esen
ts
am
o
u
n
t
o
f
in
f
o
r
m
a
tio
n
f
o
r
in
p
u
t
o
n
p
r
esen
t
L
STM
n
eu
r
o
n
,
th
e
̃
r
ep
r
esen
ts
ca
n
d
id
ate
s
tate
ce
ll
v
alu
e
an
d
th
e
r
ep
r
esen
ts
r
esu
lt
am
o
u
n
t
o
f
d
ata
o
n
L
STM
n
eu
r
o
n
.
Ma
th
e
m
atica
l
ex
p
r
ess
io
n
f
o
r
e
v
er
y
g
ate
o
f
th
e
L
STM
ce
ll is
g
iv
en
as f
r
o
m
(
9
)
to
(
1
4
)
.
=
(
∙
+
ℎ
⋅
ℎ
−
1
+
)
(
9
)
=
(
∙
+
ℎ
∙
ℎ
−
1
+
)
(
1
0
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
N
eu
r
o
-
DA
N
et:
d
u
a
l a
tten
tio
n
d
ee
p
n
e
u
r
a
l n
etw
o
r
k
lo
n
g
s
h
o
r
t
-
term
…
(
S
u
ja
th
a
Ha
n
u
ma
n
th
a
r
a
ya
p
p
a
)
817
=
⊙
−
1
+
⊙
̃
(
1
1
)
̃
=
ℎ
(
⋅
+
ℎ
⨀
ℎ
−
1
+
)
(
1
2
)
=
(
∙
+
ℎ
∙
ℎ
−
1
+
)
(
1
3
)
ℎ
=
⊙
ℎ
(
)
(
1
4
)
I
n
(
9
)
t
o
(
1
4
)
,
th
e
r
ep
r
esen
ts
s
ig
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
,
a
⊙
r
ep
r
esen
ts
Had
am
ar
d
p
r
o
d
u
ct
an
d
r
ep
r
esen
ts
p
r
esen
t
in
p
u
t
v
ec
t
o
r
.
T
h
e
,
,
,
r
ep
r
esen
ts
r
esp
ec
tiv
e
weig
h
t
v
ec
to
r
f
o
r
in
p
u
t,
th
e
ℎ
−
1
r
ep
r
esen
ts
a
s
tate
v
ec
to
r
o
f
a
p
r
io
r
ce
ll
u
n
it’s
h
id
d
e
n
lay
er
.
T
h
e
ℎ
,
ℎ
,
ℎ
,
ℎ
r
ep
r
esen
ts
h
id
d
e
n
lay
er
weig
h
ts
o
n
s
tate
v
ec
to
r
,
th
e
,
,
,
r
ep
r
esen
ts
b
ias
v
ec
to
r
.
T
h
e
p
r
o
p
o
s
ed
m
eth
o
d
ef
f
icien
tly
ca
p
tu
r
es
s
p
atio
tem
p
o
r
al
p
atter
n
s
b
y
a
co
m
b
in
atio
n
o
f
a
s
elf
-
atten
tio
n
m
ec
h
an
is
m
an
d
an
L
STM
ar
ch
itectu
r
e.
T
h
is
p
r
o
ce
s
s
im
p
r
o
v
es
f
ea
tu
r
e
lear
n
in
g
b
y
W
T
f
o
r
m
u
lti
-
r
eso
lu
tio
n
an
aly
s
is
an
d
PC
A
f
o
r
d
im
en
s
io
n
ality
r
ed
u
ctio
n
.
T
h
is
m
o
d
el
m
itig
ates
th
e
i
s
s
u
es
lik
e
f
ea
tu
r
e
r
ed
u
n
d
an
c
y
,
s
p
atial
d
ata
lo
s
s
an
d
class
im
b
alan
ce
,
d
etec
tio
n
p
er
f
o
r
m
a
n
ce
an
d
g
e
n
er
aliza
tio
n
ab
ilit
y
.
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
h
e
d
ev
elo
p
e
d
m
o
d
el
is
s
i
m
u
lated
o
n
MA
T
L
AB
2
0
2
0
R
an
d
u
s
ed
s
y
s
tem
co
n
f
ig
u
r
atio
n
ar
e
i5
p
r
o
ce
s
s
o
r
,
8
GB
R
AM
,
an
d
W
in
d
o
ws
1
0
(
6
4
b
it).
T
h
e
v
alid
atio
n
m
etr
ics
s
u
ch
as
ac
cu
r
ac
y
,
s
en
s
itiv
ity
,
s
p
ec
if
icity
,
p
r
ec
is
io
n
,
F
1
-
s
co
r
e
,
an
d
Ma
tth
ew’
s
co
r
r
elatio
n
co
ef
f
icien
t
(
MCC
)
ar
e
co
n
s
id
e
r
ed
in
th
is
ar
ticle
to
v
alid
ate
th
e
p
er
f
o
r
m
an
ce
.
T
h
e
p
ar
am
eter
s
o
f
th
e
m
o
d
el
ar
e
d
escr
ib
ed
in
T
ab
le
3
.
Fig
u
r
es
9
an
d
1
0
r
ep
r
esen
t
th
e
test
ed
s
am
p
le
im
ag
es u
s
in
g
AB
I
DE
-
I
an
d
AB
I
DE
-
I
I
d
at
asets
.
T
ab
le
3
.
Mo
d
el
p
ar
am
ete
r
s
P
a
r
a
me
t
e
r
s
V
a
l
u
e
M
a
x
e
p
o
c
h
s
1
0
0
I
n
i
t
i
a
l
l
e
a
r
n
i
n
g
r
a
t
e
0
.
0
1
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
1
.
0
0
E
-
04
Le
a
r
n
i
n
g
r
a
t
e
d
r
o
p
f
a
c
t
o
r
0
.
2
B
a
t
c
h
si
z
e
16
Le
a
r
n
r
a
t
e
s
c
h
e
d
u
l
e
P
i
e
c
e
w
i
se
Le
a
r
n
r
a
t
e
d
r
o
p
p
e
r
i
o
d
1
0
0
G
r
a
d
i
e
n
t
d
e
c
a
y
f
a
c
t
o
r
0
.
9
1
Fig
u
r
e
9
.
T
ested
s
am
p
le
im
a
g
es o
f
AB
I
DE
–
I
d
ataset
T
a
b
l
e
4
r
e
p
r
e
s
e
n
t
s
a
n
ev
a
l
u
a
t
i
o
n
o
f
t
h
e
d
e
v
e
lo
p
e
d
N
e
u
r
o
-
D
A
N
e
t
w
i
t
h
d
i
f
f
e
r
en
t
c
l
a
s
s
i
f
i
e
r
s
o
n
A
B
I
D
E
-
I
a
n
d
A
B
I
D
E
-
II
d
a
t
as
e
t
s
.
D
i
f
f
er
e
n
t
c
l
a
s
s
i
f
i
er
s
,
s
u
c
h
a
s
g
a
t
ed
r
e
c
u
r
r
en
t
u
n
i
t
(
GR
U
)
,
b
id
i
r
e
ct
i
o
n
a
l
l
o
n
g
s
h
o
r
t
-
t
er
m
m
e
m
o
r
y
(
B
i
-
L
S
T
M
)
,
D
N
N
,
a
n
d
co
n
v
e
n
t
io
n
a
l
L
S
T
M,
a
r
e
t
a
k
en
t
o
v
a
l
id
a
t
e
a
p
er
f
o
r
m
a
n
ce
o
f
d
ev
e
l
o
p
ed
m
o
d
e
l
.
T
h
e
p
r
o
p
o
s
e
d
m
e
th
o
d
a
i
m
s
to
e
f
f
i
c
i
e
n
t
ly
c
a
p
tu
r
e
s
p
a
t
io
t
e
m
p
o
r
a
l
p
a
t
te
r
n
s
b
y
a
c
o
m
b
i
n
a
t
io
n
o
f
a
s
e
l
f
-
a
t
t
e
n
t
io
n
m
e
ch
a
n
i
s
m
an
d
a
n
L
S
T
M
a
r
ch
i
t
e
c
tu
r
e.
T
h
i
s
p
r
o
c
e
s
s
i
m
p
r
o
v
e
s
f
e
a
t
u
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
1
5
,
No
.
1
,
Feb
r
u
ar
y
2
0
2
6
:
810
-
8
2
3
818
l
e
a
r
n
in
g
b
y
W
T
f
o
r
m
u
l
t
i
-
r
e
s
o
l
u
t
io
n
an
a
l
y
s
i
s
a
n
d
P
C
A
f
o
r
d
i
m
e
n
s
io
n
a
l
i
ty
r
ed
u
c
t
i
o
n
.
T
h
i
s
m
o
d
e
l
m
i
t
i
g
a
t
e
s
t
h
e
i
s
s
u
e
s
l
i
k
e
f
e
a
tu
r
e
r
ed
u
n
d
an
c
y
,
s
p
a
t
i
a
l
d
a
t
a
l
o
s
s
a
n
d
c
l
a
s
s
i
m
b
a
l
a
n
ce
,
d
e
t
e
c
t
io
n
p
e
r
f
o
r
m
an
c
e
an
d
g
e
n
er
a
l
i
z
a
t
io
n
ab
i
l
i
t
y
.
De
v
e
lo
p
ed
Ne
u
r
o
-
D
A
Ne
t
o
b
ta
i
n
s
th
e
h
i
g
h
e
s
t
a
c
cu
r
a
cy
o
f
9
8
.
5
1
%
o
n
A
B
I
DE
-
I
an
d
9
8
.
8
1
%
o
n
A
B
I
D
E
-
II
d
a
t
a
s
e
t.
Fig
u
r
e
1
0
.
T
ested
s
am
p
le
im
a
g
es o
f
AB
I
DE
-
II
d
ataset
T
a
b
l
e
4
.
E
v
a
l
u
a
t
i
o
n
o
f
d
e
v
e
l
o
p
e
d
N
e
u
r
o
-
D
A
N
et
w
it
h
d
i
f
f
e
r
e
n
t
c
l
as
s
i
f
i
e
r
s
o
n
AB
I
DE
-
I
a
n
d
AB
I
D
E
-
II
d
a
t
as
e
ts
M
o
d
e
l
s
A
c
c
u
r
a
c
y
(
%)
S
e
n
s
i
t
i
v
i
t
y
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
M
C
C
(
%)
A
B
I
D
E
-
I
G
R
U
7
6
.
6
4
7
4
.
3
5
8
1
.
0
7
7
9
.
7
0
7
6
.
9
3
7
1
.
5
9
Bi
-
LST
M
8
1
.
6
5
7
9
.
0
3
8
3
.
1
9
8
2
.
6
3
8
0
.
7
9
7
8
.
1
9
DNN
7
9
.
2
6
7
9
.
2
1
7
5
.
6
9
7
6
.
8
3
7
8
.
0
0
7
4
.
6
6
LSTM
9
5
.
2
8
9
4
.
4
4
9
3
.
4
8
9
2
.
7
7
9
3
.
6
0
9
1
.
8
4
N
e
u
r
o
-
D
A
N
e
t
9
8
.
5
1
9
8
.
1
1
9
8
.
1
1
9
8
.
7
3
9
8
.
4
2
9
6
.
2
4
A
B
I
D
E
-
II
G
R
U
9
6
.
4
3
9
5
.
9
9
9
6
.
8
6
9
5
.
1
2
9
5
.
5
5
9
3
.
6
4
Bi
-
LST
M
9
4
.
2
4
9
3
.
8
4
9
3
.
2
0
9
3
.
0
9
9
3
.
4
7
9
1
.
0
0
DNN
9
7
.
1
9
9
6
.
4
4
9
7
.
0
6
9
7
.
1
3
9
6
.
7
8
9
5
.
7
1
LSTM
9
8
.
0
5
9
7
.
3
1
9
7
.
7
3
9
7
.
4
0
9
7
.
3
6
9
6
.
6
0
N
e
u
r
o
-
D
A
N
e
t
9
8
.
8
1
9
8
.
8
1
9
9
.
6
0
9
8
.
8
6
9
8
.
8
3
9
8
.
4
3
T
ab
le
5
r
ep
r
esen
ts
th
e
ev
alu
a
tio
n
o
f
5
-
f
o
ld
c
r
o
s
s
-
v
alid
atio
n
f
o
r
th
e
d
ev
elo
p
ed
Neu
r
o
-
D
ANe
t
with
d
if
f
er
en
t
class
if
ier
s
o
n
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
.
T
h
e
d
if
f
er
e
n
t
class
if
ier
s
s
u
ch
as
GR
U,
B
i
-
L
STM
,
DNN
,
an
d
co
n
v
en
tio
n
al
L
STM
,
ar
e
tak
en
to
e
v
alu
ate
a
p
er
f
o
r
m
a
n
ce
o
f
a
d
e
v
elo
p
e
d
m
o
d
el.
T
h
e
p
r
o
p
o
s
e
d
m
eth
o
d
aim
s
to
ef
f
icien
tly
ca
p
tu
r
e
s
p
atio
tem
p
o
r
al
p
atter
n
s
b
y
a
co
m
b
in
atio
n
o
f
a
s
elf
-
atten
tio
n
m
ec
h
an
is
m
an
d
an
L
STM
ar
ch
itectu
r
e
.
T
h
is
p
r
o
ce
s
s
im
p
r
o
v
es
f
ea
tu
r
e
l
ea
r
n
in
g
b
y
W
T
f
o
r
m
u
lti
-
r
eso
lu
tio
n
an
aly
s
is
an
d
PC
A
f
o
r
d
im
en
s
io
n
ality
r
ed
u
c
tio
n
.
T
h
is
m
o
d
el
m
itig
ates
th
e
is
s
u
es
lik
e
f
ea
tu
r
e
r
ed
u
n
d
a
n
c
y
,
s
p
atial
d
ata
lo
s
s
an
d
class
im
b
alan
ce
,
d
etec
tio
n
p
er
f
o
r
m
an
ce
an
d
g
en
er
al
izatio
n
ab
ilit
y
.
On
5
-
f
o
ld
cr
o
s
s
v
alid
atio
n
,
th
e
d
ev
elo
p
e
d
m
o
d
el
o
b
tain
s
a
h
i
g
h
ac
cu
r
ac
y
o
f
9
8
.
5
1
% o
n
AB
I
DE
-
I
an
d
9
8
.
8
1
% o
n
AB
I
DE
-
II
d
atasets
.
T
ab
le
6
r
ep
r
esen
ts
th
e
e
v
alu
atio
n
o
f
d
if
f
er
en
t
k
-
f
o
ld
v
ali
d
atio
n
s
ets
o
n
AB
I
DE
-
I
a
n
d
AB
I
DE
-
II
d
atasets
.
K
-
f
o
ld
v
alid
atio
n
s
ets
s
u
ch
as
K=
3
,
K=
5
,
K
=7
an
d
K=
8
is
ev
al
u
ated
f
o
r
t
h
e
d
ev
elo
p
e
d
Neu
r
o
-
DANe
t.
I
n
th
at,
t
h
e
d
e
v
elo
p
ed
m
o
d
el
s
h
o
ws
th
e
h
ig
h
est
ac
cu
r
ac
y
o
n
K=
5
wh
en
c
o
m
p
ar
ed
with
o
th
e
r
k
-
f
o
ld
v
alid
atio
n
s
ets.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J Ar
tif
I
n
tell
I
SS
N:
2252
-
8
9
3
8
N
eu
r
o
-
DA
N
et:
d
u
a
l a
tten
tio
n
d
ee
p
n
e
u
r
a
l n
etw
o
r
k
lo
n
g
s
h
o
r
t
-
term
…
(
S
u
ja
th
a
Ha
n
u
ma
n
th
a
r
a
ya
p
p
a
)
819
T
ab
le
5
.
E
v
alu
atio
n
o
f
5
-
f
o
ld
cr
o
s
s
v
alid
atio
n
f
o
r
d
ev
el
o
p
ed
Neu
r
o
-
DANe
t w
ith
d
if
f
e
r
en
t
class
if
ier
s
u
s
in
g
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
M
o
d
e
l
s
A
c
c
u
r
a
c
y
(
%)
S
e
n
s
i
t
i
v
i
t
y
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
M
C
C
(
%)
A
B
I
D
E
-
I
G
R
U
7
6
.
0
7
7
3
.
3
8
8
0
.
1
0
7
9
.
1
4
7
6
.
1
5
7
1
.
0
2
Bi
-
LST
M
8
0
.
6
8
7
8
.
0
6
8
2
.
6
2
8
1
.
6
6
7
9
.
8
2
7
7
.
6
2
DNN
7
8
.
2
9
7
8
.
6
4
7
5
.
1
2
7
5
.
0
1
7
6
.
7
8
7
4
.
0
9
LSTM
9
4
.
3
2
9
3
.
8
7
9
2
.
5
1
9
1
.
8
0
9
2
.
8
2
9
1
.
2
8
N
e
u
r
o
-
D
A
N
e
t
9
8
.
5
1
9
8
.
1
1
9
8
.
1
1
9
8
.
7
3
9
8
.
4
2
9
6
.
2
4
A
B
I
D
E
-
II
G
R
U
9
5
.
4
3
9
4
.
9
9
9
5
.
8
6
9
4
.
2
0
9
4
.
5
9
9
3
.
0
4
Bi
-
LST
M
9
3
.
2
4
9
2
.
8
4
9
3
.
9
8
9
3
.
0
0
9
2
.
9
2
9
0
.
0
0
DNN
9
6
.
1
9
9
5
.
4
4
9
6
.
9
6
9
6
.
9
3
9
6
.
1
8
9
5
.
0
1
LSTM
9
7
.
0
5
9
8
.
3
1
9
7
.
0
3
9
7
.
0
0
9
7
.
6
5
9
6
.
0
0
N
e
u
r
o
-
D
A
N
e
t
9
8
.
8
1
9
8
.
8
1
9
9
.
6
0
9
8
.
8
6
9
8
.
8
3
9
8
.
4
3
T
ab
le
6
.
E
v
alu
atio
n
o
f
d
if
f
er
e
n
t
k
-
f
o
l
d
v
alid
atio
n
s
ets f
o
r
th
e
d
ev
elo
p
e
d
Neu
r
o
-
DANe
t u
s
in
g
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
K
-
f
o
l
d
v
a
l
i
d
a
t
i
o
n
A
c
c
u
r
a
c
y
(
%)
S
e
n
s
i
t
i
v
i
t
y
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
M
C
C
(
%)
A
B
I
D
E
-
I
K
=
3
9
6
.
9
0
9
5
.
6
4
9
5
.
0
8
9
6
.
0
0
9
5
.
8
2
9
4
.
0
8
K
=
5
9
8
.
5
1
9
8
.
1
1
9
8
.
1
1
9
8
.
7
3
9
8
.
4
2
9
6
.
2
4
K
=
7
9
7
.
5
0
9
6
.
4
6
9
6
.
5
4
9
6
.
6
2
9
6
.
5
4
9
5
.
4
0
K
=
8
9
5
.
2
0
9
2
.
5
5
9
2
.
1
3
9
3
.
9
5
9
3
.
2
5
9
2
.
7
0
A
B
I
D
E
-
II
K
=
3
9
6
.
6
1
9
6
.
3
6
9
5
.
4
5
9
5
.
4
2
9
5
.
8
9
9
4
.
5
8
K
=
5
9
8
.
8
1
9
8
.
8
1
9
9
.
6
0
9
8
.
8
6
9
8
.
8
3
9
8
.
4
3
K
=
7
9
5
.
5
6
9
4
.
6
3
9
5
.
0
9
9
4
.
6
5
9
4
.
6
4
9
3
.
4
3
K
=
8
9
4
.
1
2
9
3
.
6
4
9
3
.
4
2
9
4
.
8
8
9
4
.
2
6
9
3
.
4
5
T
ab
le
7
r
ep
r
esen
ts
th
e
ev
a
lu
atio
n
o
f
d
i
f
f
er
en
t
L
2
r
eg
u
lar
izatio
n
v
alu
es
f
o
r
th
e
d
ev
elo
p
e
d
Neu
r
o
-
DANe
t
o
n
AB
I
DE
-
I
a
n
d
AB
I
DE
-
II
d
atasets
.
On
L
2
r
eg
u
lar
izatio
n
v
alu
es
o
f
1
.
0
0
E
-
0
4
,
th
e
d
ev
elo
p
ed
Neu
r
o
-
DANe
t
o
b
tain
ed
th
e
h
ig
h
est
ac
cu
r
ac
y
o
f
9
8
.
5
1
%
o
n
AB
I
DE
-
I
an
d
9
8
.
8
1
%
o
n
AB
I
DE
-
II
d
atasets
.
T
ab
le
8
r
ep
r
esen
ts
th
e
ev
alu
ati
o
n
o
f
d
i
f
f
er
en
t
b
atch
s
izes f
o
r
th
e
d
ev
elo
p
e
d
Neu
r
o
-
DANe
t u
s
in
g
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
.
On
B
atch
s
ize
o
f
1
6
,
d
ev
elo
p
ed
Neu
r
o
-
DANe
t
o
b
tain
ed
th
e
h
ig
h
est
a
cc
u
r
ac
y
o
f
9
8
.
5
1
%
on
AB
I
DE
-
I
an
d
9
8
.
8
1
% o
n
A
B
I
DE
-
II
d
atasets
.
T
ab
le
7
.
E
v
alu
atio
n
o
f
d
if
f
er
e
n
t L
2
r
e
g
u
lar
izatio
n
v
alu
es o
f
th
e
d
ev
elo
p
ed
Neu
r
o
-
DANe
t o
n
AB
I
DE
-
I
an
d
AB
I
DE
-
II
d
atasets
L2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
A
c
c
u
r
a
c
y
(
%)
S
e
n
s
i
t
i
v
i
t
y
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
M
C
C
(
%)
A
B
I
D
E
-
I
4
.
0
0
E
-
04
9
2
.
9
4
9
2
.
6
3
9
1
.
3
3
9
2
.
7
6
9
2
.
7
0
9
2
.
2
8
3
.
0
0
E
-
04
9
6
.
5
2
9
3
.
6
7
9
5
.
9
4
9
4
.
0
3
9
3
.
8
5
9
4
.
8
6
2
.
0
0
E
-
04
9
7
.
6
2
9
7
.
2
5
9
8
.
8
4
9
7
.
0
2
9
7
.
1
3
9
5
.
0
5
1
.
0
0
E
-
04
9
8
.
5
1
9
8
.
1
1
9
8
.
1
1
9
8
.
7
3
9
8
.
4
2
9
6
.
2
4
A
B
I
D
E
-
II
4
.
0
0
E
-
04
9
4
.
3
8
9
3
.
4
2
9
3
.
3
6
9
2
.
5
0
9
2
.
9
6
9
1
.
2
6
3
.
0
0
E
-
04
9
5
.
5
8
9
4
.
0
0
9
4
.
6
8
9
3
.
2
7
9
3
.
6
3
9
2
.
7
8
2
.
0
0
E
-
04
9
7
.
2
4
9
5
.
1
7
9
6
.
2
1
9
5
.
2
9
9
5
.
2
3
9
5
.
6
9
1
.
0
0
E
-
04
9
8
.
8
1
9
8
.
8
1
9
9
.
6
0
9
8
.
8
6
9
8
.
8
3
9
8
.
4
3
T
a
b
l
e
8
.
E
v
a
l
u
a
t
i
o
n
o
f
d
i
f
f
e
r
e
n
t
b
a
t
c
h
s
i
z
e
s
o
f
d
e
v
el
o
p
e
d
N
e
u
r
o
-
D
A
N
e
t
u
s
i
n
g
AB
I
D
E
-
I
a
n
d
AB
I
D
E
-
II
d
a
t
as
e
ts
B
a
t
c
h
si
z
e
A
c
c
u
r
a
c
y
(
%)
S
e
n
s
i
t
i
v
i
t
y
(
%)
S
p
e
c
i
f
i
c
i
t
y
(
%)
P
r
e
c
i
s
i
o
n
(
%)
F1
-
sc
o
r
e
(
%)
M
C
C
(
%)
A
B
I
D
E
-
I
1
2
8
9
2
.
6
8
9
1
.
5
6
9
0
.
0
9
9
1
.
7
1
9
1
.
6
3
9
0
.
9
2
64
9
5
.
6
1
9
5
.
6
8
9
4
.
3
6
9
4
.
4
1
9
5
.
0
4
9
3
.
1
3
32
9
6
.
9
7
9
6
.
6
5
9
4
.
0
7
9
5
.
7
6
9
6
.
2
0
9
4
.
3
2
16
9
8
.
5
1
9
8
.
1
1
9
8
.
1
1
9
8
.
7
3
9
8
.
4
2
9
6
.
2
4
A
B
I
D
E
-
II
1
2
8
9
5
.
2
5
9
4
.
7
4
9
4
.
2
2
9
3
.
6
5
9
4
.
2
0
9
2
.
2
1
64
9
4
.
0
1
9
3
.
7
4
9
4
.
8
7
9
3
.
5
0
9
3
.
6
2
9
1
.
8
5
32
9
7
.
1
0
9
6
.
8
4
9
5
.
5
6
9
5
.
7
7
9
6
.
3
1
9
4
.
5
5
16
9
8
.
8
1
9
8
.
8
1
9
9
.
6
0
9
8
.
8
6
9
8
.
8
3
9
8
.
4
3
Evaluation Warning : The document was created with Spire.PDF for Python.