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.
15
,
No
.
2
,
A
p
r
il
2
0
2
6
,
p
p
.
1
4
1
8
~
1
4
2
7
I
SS
N:
2
2
5
2
-
8
9
3
8
,
DOI
: 1
0
.
1
1
5
9
1
/ijai.v
15
.i
2
.
p
p
1
4
1
8
-
1
4
2
7
1418
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
a
i
.
ia
esco
r
e.
co
m
No
v
el conv
o
lution neura
l n
etwork
mo
del f
o
r dys
g
ra
phia
-
a
ff
ecte
d han
dw
rit
ing
clas
sifica
tion
Nis
ha
Va
nja
ri
1
,2
,
P
ra
s
a
nn
a
J
.
Sh
et
e
1
1
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
K
.
J.
S
o
m
a
i
y
a
S
c
h
o
o
l
o
f
En
g
i
n
e
e
r
i
n
g
,
M
u
mb
a
i
,
I
n
d
i
a
2
D
e
p
a
r
t
me
n
t
o
f
C
o
m
p
u
t
e
r
En
g
i
n
e
e
r
i
n
g
,
K
.
J.
S
o
m
a
i
y
a
I
n
st
i
t
u
t
e
o
f
T
e
c
h
n
o
l
o
g
y
,
M
u
mb
a
i
,
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
May
16
,
2
0
2
5
R
ev
is
ed
J
an
8
,
2
0
2
6
Acc
ep
ted
J
an
25
,
2
0
2
6
It
is
e
stim
a
ted
th
a
t
1
0
%
o
f
th
e
p
o
p
u
lati
o
n
i
n
th
e
wo
rl
d
su
ffe
rs
fro
m
lea
rn
in
g
d
isa
b
il
it
ies
li
k
e
d
y
sle
x
ia,
d
y
sg
ra
p
h
ia,
a
n
d
d
y
sc
a
lcu
li
a
.
Lea
rn
i
n
g
d
isa
b
il
it
ie
s
a
re
n
e
u
ro
lo
g
ica
l
d
iso
r
d
e
rs
in
wh
ich
c
h
il
d
re
n
stru
g
g
le
wit
h
re
a
d
in
g
,
writi
n
g
a
n
d
m
a
th
e
m
a
ti
c
a
l
sk
i
ll
s.
D
y
sg
ra
p
h
ia
d
iso
rd
e
r
imp
a
c
ts
o
n
writi
n
g
a
b
il
it
ies
o
f
stu
d
e
n
ts
a
n
d
th
u
s
m
a
y
b
e
a
h
u
rd
l
e
in
th
e
ir
lea
rn
in
g
a
n
d
e
v
a
l
u
a
ti
o
n
o
f
su
b
jec
t
m
a
tt
e
r.
He
n
c
e
e
a
rly
d
e
tec
ti
o
n
/p
re
d
ictio
n
o
f
lea
rn
in
g
d
isa
b
il
it
y
(LD)
in
sc
h
o
o
l
g
o
i
n
g
c
h
il
d
re
n
wi
ll
g
re
a
tl
y
h
e
lp
i
n
p
ro
v
id
i
n
g
n
e
c
e
ss
a
ry
a
c
c
o
m
m
o
d
a
ti
o
n
s
s
o
a
s
to
e
a
se
th
e
ir
fu
tu
re
lea
rn
i
n
g
c
u
rv
e
.
I
n
re
c
e
n
t
y
e
a
rs
re
se
a
rc
h
e
rs
h
a
v
e
u
se
d
se
v
e
ra
l
d
e
e
p
lea
rn
in
g
a
lg
o
rit
h
m
s
th
a
t
p
r
o
d
u
c
e
a
u
to
m
a
ted
a
n
d
trai
n
e
d
m
o
d
e
ls
wh
ich
c
a
n
b
e
u
se
fu
l
in
t
h
e
h
a
n
d
writi
n
g
c
las
sifica
ti
o
n
.
T
o
p
r
o
p
e
rl
y
c
a
p
tu
re
th
e
d
isti
n
c
t
h
a
n
d
writi
n
g
i
n
c
o
n
si
ste
n
c
ies
li
n
k
e
d
to
d
y
sg
ra
p
h
ia,
t
h
is
st
u
d
y
c
o
n
tain
s
e
x
p
e
rime
n
ts
th
a
t
d
e
term
in
e
h
o
w
v
a
ri
o
u
s
c
o
n
v
o
lu
ti
o
n
n
e
u
ra
l
n
e
two
rk
(CNN
)
m
o
d
e
l
lay
e
rs
c
o
n
tri
b
u
te
t
o
p
e
rfo
rm
a
n
c
e
.
T
o
a
d
d
r
e
ss
it
,
th
is
re
se
a
rc
h
fo
c
u
se
d
o
n
t
h
e
imp
r
o
v
e
d
n
o
v
e
l
m
o
d
e
l
b
a
se
d
o
n
CNN
a
n
d
targ
e
ted
d
y
sg
ra
p
h
ia
E
n
g
li
s
h
h
a
n
d
writi
n
g
c
las
sifica
ti
o
n
with
9
8
%
a
c
c
u
r
a
c
y
with
1
0
2
,
6
9
1
trai
n
a
b
le
p
a
ra
m
e
ters
.
Th
e
m
o
d
e
l
is
train
e
d
o
n
b
o
t
h
n
o
rm
a
l
a
n
d
d
y
sg
ra
p
h
ia
-
a
ffe
c
ted
h
a
n
d
writi
n
g
,
in
c
re
a
sin
g
it
s
a
c
c
u
ra
c
y
in
i
d
e
n
ti
f
y
in
g
in
d
i
v
id
u
a
l
d
iffere
n
c
e
s.
K
ey
w
o
r
d
s
:
C
o
n
v
o
lu
tio
n
n
eu
r
al
n
etwo
r
k
Dee
p
lear
n
in
g
Dy
s
g
r
ap
h
ia
Han
d
wr
itin
g
class
if
icatio
n
L
ea
r
n
in
g
d
is
ab
ilit
y
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
:
Nis
h
a
Van
jar
i
Dep
ar
tm
en
t o
f
C
o
m
p
u
ter
E
n
g
i
n
ee
r
in
g
,
K.
J
.
So
m
aiy
a
I
n
s
titu
te
o
f
T
ec
h
n
o
lo
g
y
3
3
/A,
Nan
d
ad
ee
p
C
HS,
C
h
em
b
u
r
,
Mu
m
b
ai,
I
n
d
ia
E
m
ail: n
v
an
jar
i@
s
o
m
aiy
a.
ed
u
1.
I
NT
RO
D
UCT
I
O
N
I
t
is
esti
m
ated
th
at
1
0
%
o
f
th
e
p
o
p
u
latio
n
in
th
e
wo
r
ld
s
u
f
f
er
s
f
r
o
m
lear
n
i
n
g
d
is
ab
ilit
ies
lik
e
d
y
s
lex
ia,
d
y
s
g
r
ap
h
ia,
a
n
d
d
y
s
ca
lcu
lia
[
1
]
–
[
4
]
.
Prim
a
r
y
i
m
p
a
ct
o
f
d
y
s
g
r
ap
h
ia,
a
lear
n
in
g
d
i
s
ab
ilit
y
(
L
D)
,
is
o
n
an
in
d
iv
id
u
al'
s
ca
p
ac
ity
f
o
r
wr
itten
ex
p
r
ess
io
n
.
Dy
s
g
r
a
p
h
ia
an
d
p
ar
ticu
lar
lear
n
in
g
d
ef
icit
in
wr
itten
ex
p
r
ess
io
n
r
ef
er
to
p
e
o
p
le
wh
o
,
d
esp
ite
p
r
o
p
er
e
d
u
ca
ti
o
n
,
wr
ite
b
elo
w
t
h
eir
co
g
n
i
tiv
e
lev
el
an
d
ag
e.
Dy
s
g
r
ap
h
ia
is
a
s
p
ec
ialized
LD
th
at
m
ak
es
it
d
if
f
icu
lt
to
d
u
p
licate
b
o
th
letter
s
an
d
n
u
m
b
er
s
.
Dep
en
d
in
g
o
n
th
e
ch
ild
'
s
ag
e
an
d
d
ev
elo
p
m
en
tal
s
tag
e,
d
y
s
g
r
ap
h
ia
ca
n
p
r
esen
t
a
v
ar
iety
o
f
s
y
m
p
to
m
s
.
Mu
ltip
le
cu
es
m
u
s
t
b
e
tak
en
in
to
ac
c
o
u
n
t
f
o
r
an
ac
cu
r
ate
d
iag
n
o
s
is
o
f
d
y
s
g
r
ap
h
ia,
wh
ic
h
p
r
esen
ts
m
ajo
r
o
b
s
tacle
s
.
As
h
an
d
wr
itin
g
is
a
way
to
co
n
v
ey
th
o
u
g
h
ts
an
d
em
o
tio
n
s
,
m
an
y
r
esear
ch
er
s
h
av
e
wo
r
k
ed
o
n
h
an
d
wr
itin
g
ch
ar
ac
ter
r
ec
o
g
n
itio
n
s
y
s
tem
s
to
an
aly
ze
im
ag
es
o
f
wr
itten
m
ater
ial
an
d
tr
y
to
f
in
d
t
h
e
p
atter
n
f
r
o
m
it.
R
esear
ch
er
s
h
av
e
u
s
ed
s
ev
er
al
d
ee
p
lear
n
in
g
al
g
o
r
ith
m
s
lik
e
co
n
v
o
l
u
tio
n
n
eu
r
al
n
etwo
r
k
(
C
NN)
,
L
eNe
t5
,
an
d
cu
s
to
m
ized
C
NN
th
at
ca
n
b
e
u
s
ef
u
l
in
p
r
ed
ictin
g
th
e
s
y
m
p
to
m
s
o
f
lear
n
in
g
d
is
ab
ilit
ies
an
d
p
r
o
d
u
ce
au
to
m
ated
an
d
tr
ain
e
d
m
o
d
els th
at
s
ig
n
if
ican
tly
m
in
im
ize
th
e
n
ee
d
o
f
h
u
m
an
in
ter
v
en
tio
n
[
5
]
–
[
9
]
.
C
NNs
ar
e
b
u
ilt
f
r
o
m
a
m
u
lti
-
lay
er
ed
f
ee
d
-
f
o
r
war
d
n
eu
r
al
n
etwo
r
k
,
wh
ich
co
n
s
is
ts
o
f
s
tack
in
g
s
ev
er
al
h
id
d
en
lay
e
r
s
in
a
s
p
e
cif
ic
o
r
d
er
.
W
ith
its
s
eq
u
en
tial
ar
ch
itectu
r
e,
C
NN
is
ab
le
to
ac
q
u
ir
e
k
n
o
wled
g
e
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
o
ve
l c
o
n
vo
lu
tio
n
n
eu
r
a
l n
et
w
o
r
k
mo
d
el
fo
r
d
ysg
r
a
p
h
ia
-
a
ff
ec
ted
h
a
n
d
w
r
itin
g
cla
s
s
ifica
ti
o
n
(
N
is
h
a
V
a
n
j
a
r
i
)
1419
ab
o
u
t
h
ier
ar
ch
ical
ch
ar
ac
te
r
is
tics
.
C
NN
m
o
d
els
ar
e
b
ased
o
n
m
ath
em
atica
l
o
p
e
r
atio
n
s
o
n
ev
er
y
lay
er
an
d
th
e
m
o
d
el
p
e
r
f
o
r
m
an
ce
ca
n
b
e
i
m
p
r
o
v
e
d
b
y
tu
n
i
n
g
th
e
p
ar
a
m
eter
s
.
Fo
r
d
ev
el
o
p
in
g
th
e
c
u
s
to
m
m
o
d
el,
s
o
m
e
r
esear
ch
er
s
p
r
o
p
o
s
ed
a
r
ch
itec
tu
r
al
ch
an
g
es
to
im
p
r
o
v
e
t
h
e
p
er
f
o
r
m
an
ce
o
f
m
o
d
el.
W
ith
th
is
ex
p
er
ien
ce
,
i
n
o
r
d
er
t
o
class
if
y
d
y
s
g
r
a
p
h
ia
s
y
m
p
to
m
s
u
s
in
g
h
an
d
wr
itten
alp
h
ab
et
im
a
g
es,
th
is
s
tu
d
y
will
co
m
p
ar
e
b
asic
C
NN
m
o
d
els
with
o
u
r
p
r
o
p
o
s
ed
m
o
d
el
th
at
c
o
m
p
r
is
es
h
y
p
er
p
ar
am
eter
tu
n
in
g
a
n
d
a
r
c
h
itectu
r
al
ch
a
n
g
es.
Ad
d
itio
n
ally
,
th
e
ac
cu
r
ac
y
an
d
lo
s
s
an
aly
s
is
o
f
th
e
C
N
N
n
etwo
r
k
will
b
e
u
s
ed
to
v
alid
a
te
th
e
co
m
p
ar
is
o
n
s
.
T
h
is
p
ap
er
p
r
o
p
o
s
es,
d
esig
n
a
n
d
d
ev
el
o
p
a
n
o
v
el
n
ested
co
n
v
o
lu
tio
n
C
NN
m
o
d
el
n
am
e
d
Nested
_
C
o
n
v
_
C
NN
f
o
r
d
y
s
g
r
a
p
h
ia
s
y
m
p
to
m
s
clas
s
if
icatio
n
an
d
ev
alu
ate
its
p
er
f
o
r
m
an
ce
o
n
Dy
s
Data
h
a
n
d
wr
i
tin
g
d
ataset
(
b
ased
o
n
NI
ST
s
p
ec
ial
d
ata
b
ase1
9
a
n
d
a
v
ailab
le
at
Kag
g
le)
[
5
]
–
[
1
0
]
.
I
n
v
esti
g
atio
n
was
c
o
n
d
u
c
ted
to
co
m
p
ar
e
t
h
e
n
o
v
el
m
o
d
el
with
th
e
b
asic CNN m
o
d
el.
T
h
e
r
est
o
f
th
is
p
ap
er
is
o
r
g
a
n
ized
as
f
o
llo
ws.
Sectio
n
2
p
r
esen
ts
th
e
r
elate
d
wo
r
k
,
wh
i
ch
r
ev
iews
r
ec
en
t
s
tu
d
ies.
Sectio
n
s
3
an
d
4
d
escr
ib
e
th
e
d
ataset
an
d
th
e
p
r
o
p
o
s
ed
m
eth
o
d
s
.
Sectio
n
5
p
r
esen
ts
th
e
r
esu
lts
an
d
an
aly
s
is
.
Sectio
n
6
p
r
o
v
id
es th
e
ab
latio
n
s
tu
d
y
.
Fin
ally
,
s
ec
tio
n
7
co
n
clu
d
es th
e
p
a
p
er
.
2.
RE
L
AT
E
D
WO
RK
A
g
o
v
er
n
m
e
n
t
s
tu
d
y
n
ews
r
elea
s
e
f
r
o
m
2
0
1
5
s
tates
th
at
7
0
–
8
0
%
o
f
ch
ild
r
en
with
lear
n
in
g
im
p
air
m
en
ts
h
av
e
r
ea
d
in
g
d
if
f
icu
lties
.
T
h
e
m
o
s
t
p
r
ev
al
en
t
ty
p
e
o
f
L
D
ar
e
d
y
s
lex
i
a
an
d
d
y
s
g
r
a
p
h
ia.
W
o
r
ld
wid
e,
th
e
esti
m
ated
af
f
ec
ted
p
o
p
u
latio
n
r
a
n
g
es
f
r
o
m
5
–
2
0
%.
I
n
I
n
d
ia,
1
5
%
o
f
th
e
p
o
p
u
latio
n
is
esti
m
ated
to
b
e
s
u
f
f
er
in
g
f
r
o
m
LD
.
R
esear
ch
er
s
h
av
e
u
s
ed
s
e
v
er
al
d
ee
p
lear
n
i
n
g
alg
o
r
ith
m
s
th
at
ca
n
b
e
u
s
ef
u
l
in
p
r
ed
ictin
g
th
e
s
y
m
p
to
m
s
o
f
lear
n
in
g
d
is
ab
ilit
ies
lik
e
d
y
s
l
ex
ia
an
d
d
y
s
g
r
ap
h
ia.
T
h
e
m
ai
n
o
b
jectiv
e
o
f
d
ee
p
lear
n
in
g
is
to
p
r
o
d
u
ce
au
to
m
ated
an
d
tr
ai
n
ed
m
o
d
els
t
h
at
s
ig
n
if
ican
tly
m
in
im
ize
t
h
e
n
ee
d
f
o
r
h
u
m
an
in
ter
ac
tio
n
.
Sp
o
o
n
et
al
.
[
4
]
a
ttem
p
ted
o
p
tical
ch
ar
ac
ter
r
ec
o
g
n
itio
n
(
OC
R
)
to
o
ls
lik
e
T
ess
er
ac
t
(
Go
o
g
le)
to
tr
an
s
cr
ib
e
th
e
wr
itin
g
in
t
h
e
p
h
o
to
g
r
ap
h
s
b
e
f
o
r
e
tu
r
n
in
g
to
m
ac
h
in
e
lear
n
in
g
to
d
etec
t
d
y
s
lex
ia
b
ased
o
n
s
p
ellin
g
er
r
o
r
s
an
d
o
th
er
te
x
tu
al
ch
ar
ac
ter
is
tics
an
d
also
cr
ea
ted
L
eNe
t
-
5
to
class
if
y
d
y
s
lex
ia
h
an
d
wr
itin
g
.
T
h
e
1
3
8
,
5
0
0
im
ag
es
in
th
e
d
ataset
wer
e
p
r
e
-
p
r
o
ce
s
s
ed
an
d
en
h
an
ce
d
with
d
ata
b
ef
o
r
e
b
ein
g
f
ed
in
t
o
th
e
n
etwo
r
k
.
I
s
a
et
al
.
[
8
]
c
o
m
p
ar
ed
v
a
r
io
u
s
C
NN
m
o
d
els
an
d
e
x
p
er
im
en
ted
C
NN1
,
C
NN2
,
C
NN3
,
an
d
L
eNe
t
-
5
an
d
g
o
t
8
7
%
ac
cu
r
ac
y
in
C
N
N1
m
o
d
el
to
d
etec
t
th
e
LD
s
y
m
p
to
m
.
R
ec
en
t
s
tu
d
ies
o
n
d
y
s
g
r
ap
h
ia
in
cr
ea
s
in
g
ly
s
u
p
p
o
r
t th
e
u
s
e
o
f
C
NN
-
b
ased
p
ip
elin
es a
n
d
f
ea
tu
r
e/cla
s
s
if
ier
f
u
s
io
n
[
1
1
]
–
[
1
6
]
; h
o
wev
e
r
,
th
ey
f
r
eq
u
en
tly
lack
co
m
p
r
eh
e
n
s
iv
e
ab
latio
n
s
an
d
clea
r
ju
s
tific
atio
n
s
f
o
r
a
r
ch
i
tectu
r
al
an
d
tr
ain
in
g
d
ec
is
io
n
s
.
R
ec
en
t
s
tu
d
ies
co
m
p
ar
in
g
ar
ch
itectu
r
es a
n
d
r
elea
s
in
g
p
ed
iatr
ic
d
atasets
h
ig
h
lig
h
t th
at
ac
cu
r
ac
y
d
ep
e
n
d
s
o
n
m
u
lti
-
s
ca
le
s
tr
o
k
e
m
o
d
ellin
g
a
n
d
ef
f
ec
tiv
e
tr
ain
i
n
g
co
n
f
ig
u
r
atio
n
s
,
r
ath
er
th
an
s
o
lely
o
n
d
ee
p
er
b
ac
k
b
o
n
es.
T
ab
le
1
will
g
iv
e
s
u
m
m
ar
y
o
f
r
elev
an
t
p
ap
e
r
s
th
at
u
s
ed
C
NN
tech
n
iq
u
e
to
class
if
y
th
e
h
a
n
d
wr
itin
g
to
ac
h
iev
e
d
y
s
g
r
ap
h
ia
h
an
d
w
r
itin
g
class
if
icatio
n
.
Ma
in
ly
h
an
d
wr
itin
g
class
if
icatio
n
is
p
er
f
o
r
m
ed
o
n
d
ata
co
llected
in
th
e
f
o
r
m
o
f
o
n
li
n
e
h
an
d
wr
itin
g
(
d
ata
co
llected
u
s
in
g
d
ig
ital
tab
let
with
d
ig
it
al
p
en
o
r
s
ty
lu
s
)
o
r
o
f
f
lin
e
h
an
d
wr
itin
g
(
s
ca
n
n
e
d
i
m
ag
es).
I
n
th
e
T
ab
le
1
,
f
o
cu
s
is
o
n
ly
th
o
s
e
p
ap
er
s
w
h
ich
u
s
ed
h
an
d
wr
itten
d
ata
in
o
f
f
lin
e
m
o
d
e
o
r
tak
e
n
v
ia
tab
let
an
d
d
ig
ital
p
en
.
R
esear
ch
er
u
s
ed
Dy
s
Data
d
ataset
h
as
7
8
,
2
7
5
s
am
p
les
f
o
r
th
e
n
o
r
m
al
class
,
5
2
,
1
9
6
s
am
p
les
f
o
r
th
e
r
ev
er
s
al,
an
d
8
,
0
2
9
s
am
p
les
f
o
r
th
e
co
r
r
ec
ted
class
.
I
n
o
r
d
er
to
d
is
tin
g
u
is
h
p
eo
p
le
with
m
ild
co
g
n
itiv
e
im
p
air
m
e
n
t
(
MCI)
f
r
o
m
h
ea
lth
y
co
n
tr
o
ls
,
Kaw
a
et
a
l.
[
1
0
]
f
o
cu
s
ed
o
n
s
p
atial
an
d
d
y
n
am
ic
h
an
d
wr
i
tin
g
an
aly
s
is
.
I
t
em
p
h
asizes
h
o
w
well
-
r
eliab
le
h
an
d
w
r
itin
g
m
ea
s
u
r
es
a
r
e
as
m
ar
k
er
s
o
f
c
o
g
n
itiv
e
im
p
air
m
en
t.
R
ib
eir
o
et
a
l.
[
1
7
]
p
u
b
lis
h
ed
a
s
tu
d
y
th
at
p
r
esen
ted
th
e
"
b
ag
o
f
s
am
p
lin
g
s
"
m
eth
o
d
,
wh
ic
h
u
s
es r
ec
u
r
r
e
n
t n
eu
r
al
n
etwo
r
k
s
to
h
elp
d
ia
g
n
o
s
e
Par
k
in
s
o
n
'
s
d
is
ea
s
e
b
y
an
a
ly
zin
g
h
an
d
wr
itin
g
.
T
h
e
tech
n
i
q
u
e
ca
p
tu
r
ed
m
in
u
t
e
m
o
to
r
ab
n
o
r
m
alities
,
h
ig
h
lig
h
tin
g
th
e
p
r
o
m
is
e
o
f
c
u
ttin
g
-
e
d
g
e
n
e
u
r
al
n
etwo
r
k
tech
n
iq
u
es f
o
r
m
ed
ical
d
ia
g
n
o
s
is
.
Z
an
u
y
et
a
l.
[
1
8
]
g
r
o
u
p
u
s
es
o
f
b
io
m
etr
ics
o
f
h
an
d
wr
itin
g
f
o
r
h
ea
lth
m
o
n
ito
r
in
g
ar
e
c
o
v
er
ed
in
t
h
is
s
tu
d
y
,
alo
n
g
with
p
o
ten
tial
f
u
t
u
r
e
d
ev
elo
p
m
en
ts
.
I
t
h
ig
h
lig
h
t
s
th
e
g
r
o
win
g
ap
p
licatio
n
o
f
h
an
d
wr
itin
g
an
aly
s
is
o
u
ts
id
e
o
f
m
ed
ical
d
iag
n
o
s
is
.
Usi
n
g
q
u
an
titativ
e
elec
tr
o
en
ce
p
h
alo
g
r
ap
h
y
(
q
E
E
G)
d
ata
an
d
h
an
d
wr
itin
g
d
y
n
am
ics,
C
h
ai
et
a
l.
[
1
9
]
d
e
s
cr
ib
e
a
n
o
v
el
m
eth
o
d
f
o
r
id
e
n
tify
in
g
MCI.
C
o
m
b
in
i
n
g
v
a
r
io
u
s
p
h
y
s
io
lo
g
ical
d
ata
ty
p
es
a
p
p
ea
r
s
to
h
av
e
a
d
v
an
tag
es,
as
t
h
is
m
u
ltimo
d
a
l
ap
p
r
o
ac
h
g
r
ea
tly
im
p
r
o
v
es
d
etec
tio
n
ac
cu
r
ac
y
.
Acc
o
r
d
in
g
to
Dev
i
a
n
d
Kav
y
a
[
2
0
]
,
d
y
s
g
r
a
p
h
ia
d
is
o
r
d
er
ca
n
b
e
p
r
ed
icted
an
d
class
if
ied
u
s
in
g
an
in
tellig
en
t
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
.
Dee
p
lear
n
in
g
a
p
p
r
o
ac
h
es
h
av
e
b
ee
n
s
h
o
wn
to
b
e
e
f
f
ec
tiv
e
in
m
an
ag
in
g
lear
n
in
g
p
r
o
b
lem
s
,
an
d
th
e
m
eth
o
d
h
a
s
s
h
o
wn
p
o
ten
tial
in
e
ar
ly
i
d
e
n
tific
atio
n
an
d
in
ter
v
en
tio
n
.
Ku
n
h
o
th
et
a
l.
[
2
1
]
wo
r
k
ed
to
d
ia
g
n
o
s
e
d
y
s
g
r
a
p
h
ia
in
ch
ild
r
en
b
y
u
tili
zin
g
a
n
o
v
el
m
o
d
if
ie
d
im
ag
e
d
ataset
in
co
n
ju
n
ctio
n
with
C
NN
f
ea
tu
r
e
ex
tr
ac
tio
n
an
d
class
if
ier
f
u
s
io
n
.
T
h
e
p
o
ten
tial
o
f
AI
-
d
r
iv
en
d
ia
g
n
o
s
tic
to
o
ls
to
in
cr
ea
s
e
th
e
ac
cu
r
ac
y
o
f
d
e
v
elo
p
m
e
n
tal
p
r
o
b
lem
d
iag
n
o
s
is
is
h
ig
h
lig
h
ted
b
y
th
is
m
eth
o
d
[
2
1
]
,
[
2
2
]
.
H
a
n
d
wr
i
t
i
n
g
h
a
s
l
o
n
g
b
e
e
n
s
t
u
d
i
ed
t
o
d
e
t
e
c
t
co
n
d
i
t
i
o
n
s
s
u
c
h
a
s
M
C
I
,
P
a
r
k
in
s
o
n
’
s
d
i
s
e
a
s
e
,
a
n
d
d
y
s
g
r
a
p
h
i
a
,
w
i
th
e
a
r
ly
a
p
p
r
o
a
ch
e
s
r
e
ly
i
n
g
o
n
h
a
n
d
cr
a
f
t
e
d
o
r
s
h
a
l
lo
w
f
e
a
tu
r
e
s
t
h
a
t
o
f
te
n
-
l
i
m
i
t
ed
g
e
n
er
a
l
i
z
a
t
io
n
a
n
d
l
a
c
k
e
d
c
li
n
i
c
a
l
v
a
l
i
d
a
t
io
n
.
M
o
r
e
r
ec
en
t
C
N
N
-
b
a
s
ed
m
o
d
e
l
s
d
em
o
n
s
t
r
a
te
d
s
t
r
o
n
g
e
r
p
e
r
f
o
r
m
a
n
c
e
,
b
u
t
m
an
y
s
t
i
l
l
r
e
l
i
e
d
o
n
a
d
u
l
t
d
a
ta
r
a
th
e
r
th
an
p
ed
i
a
t
r
i
c
s
am
p
l
e
s
,
r
e
d
u
c
i
n
g
d
a
t
a
s
e
t
d
iv
e
r
s
i
t
y
,
a
n
d
th
ey
w
e
r
e
r
ar
e
l
y
t
e
s
t
e
d
i
n
r
e
al
s
c
r
e
en
i
n
g
co
n
t
ex
t
s
.
T
o
a
d
d
r
e
s
s
th
e
s
e
l
i
m
i
ta
t
i
o
n
s
,
r
e
s
e
a
r
ch
e
r
s
b
e
g
a
n
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.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
1
8
-
1
4
2
7
1420
e
x
p
lo
r
in
g
e
n
s
e
m
b
l
e
a
n
d
m
u
l
t
i
m
o
d
a
l
a
p
p
r
o
a
c
h
e
s
.
F
o
r
i
n
s
t
a
n
c
e
,
o
n
e
s
t
u
d
y
co
m
b
i
n
ed
b
o
th
o
n
l
in
e
h
a
n
d
w
r
i
t
in
g
s
i
g
n
al
s
a
n
d
o
f
f
l
in
e
im
a
g
e
s
t
h
r
o
u
g
h
co
n
d
i
t
i
o
n
al
f
e
a
tu
r
e
f
u
s
i
o
n
,
a
c
h
i
ev
i
n
g
8
8
.
8
%
a
cc
u
r
a
c
y
[
2
3
]
,
w
h
i
l
e
an
o
th
e
r
u
s
ed
C
N
N
-
b
a
s
e
d
f
e
a
t
u
r
e
a
n
d
c
l
a
s
s
i
f
ie
r
f
u
s
io
n
to
p
u
s
h
p
e
r
f
o
r
m
an
c
e
a
s
h
i
g
h
a
s
9
7
.
3
%
[
2
4
]
.
A
r
e
c
en
t
r
e
v
i
e
w
f
u
r
t
h
er
h
i
g
h
li
g
h
t
ed
t
h
e
v
a
l
u
e
o
f
m
u
l
t
i
m
o
d
a
l
i
n
t
eg
r
a
t
io
n
f
o
r
d
e
v
e
lo
p
in
g
m
o
r
e
g
en
e
r
a
l
i
z
ab
l
e
s
y
s
t
e
m
s
,
th
o
u
g
h
e
x
p
l
i
c
i
t
e
th
i
c
a
l
c
o
n
c
er
n
s
s
u
ch
a
s
p
r
iv
a
c
y
an
d
in
f
o
r
m
ed
c
o
n
s
e
n
t
r
em
a
i
n
e
d
l
a
r
g
e
ly
u
n
ad
d
r
e
s
s
e
d
.
B
u
i
ld
i
n
g
o
n
th
is
t
r
en
d
,
r
e
s
e
a
r
ch
er
s
h
a
v
e
in
c
r
e
a
s
i
n
g
ly
tu
r
n
e
d
to
t
r
an
s
f
o
r
m
er
-
b
a
s
ed
m
o
d
e
l
s
f
o
r
d
y
s
g
r
a
p
h
i
a
d
e
t
e
c
t
io
n
.
A
v
i
s
i
o
n
t
r
an
s
f
o
r
m
er
(
V
i
T
)
o
u
t
p
er
f
o
r
m
e
d
tr
ad
i
t
i
o
n
a
l
C
N
N
s
s
u
c
h
a
s
v
i
s
u
a
l
g
eo
m
e
t
r
y
g
r
o
u
p
(
V
G
G
1
6
)
,
R
e
s
N
e
t
5
0
,
an
d
I
n
c
ep
t
i
o
n
V
3
,
r
e
a
ch
i
n
g
a
m
a
c
r
o
F
1
-
s
c
o
r
e
o
f
0
.
9
2
in
e
ar
l
y
d
y
s
g
r
a
p
h
i
a
t
a
s
k
s
.
T
h
e
s
e
s
t
u
d
i
e
s
c
l
e
ar
l
y
d
e
m
o
n
s
t
r
a
t
e
t
h
e
p
o
w
e
r
o
f
d
e
ep
e
r
an
d
m
o
r
e
c
o
m
p
l
ex
ar
c
h
i
te
c
t
u
r
e
s
,
b
u
t
t
h
e
y
a
l
s
o
h
i
g
h
l
i
g
h
t
r
ec
u
r
r
i
n
g
ch
a
l
l
en
g
e
s
:
t
r
a
n
s
f
o
r
m
e
r
s
ty
p
i
ca
l
l
y
r
e
q
u
ir
e
l
ar
g
e
a
n
n
o
t
a
te
d
d
a
t
as
e
t
s
,
w
h
i
c
h
r
a
i
s
e
s
q
u
e
s
t
i
o
n
s
o
f
d
a
t
a
s
e
t
d
i
v
er
s
i
t
y
a
n
d
r
ep
r
e
s
en
t
a
t
i
v
en
e
s
s
,
a
s
wel
l
a
s
e
t
h
i
ca
l
co
n
c
er
n
s
a
r
o
u
n
d
d
a
t
a
co
l
l
e
c
t
io
n
a
n
d
r
e
s
o
u
r
c
e
a
c
c
e
s
s
i
b
i
l
i
ty
.
T
ab
le
1
.
R
ev
iew
o
n
m
o
d
ality
u
s
ed
with
u
s
ed
class
if
ier
an
d
r
esear
ch
o
p
p
o
r
tu
n
ities
R
e
f
e
r
e
n
c
e
M
o
d
a
l
i
t
y
D
a
t
a
s
e
t
C
l
a
s
si
f
i
e
r
G
r
o
u
p
A
c
c
u
r
a
c
y
(
%)
R
e
se
a
r
c
h
o
p
p
o
r
t
u
n
i
t
i
e
s
[
6
]
O
f
f
l
i
n
e
N
I
S
T+
sp
e
c
i
a
l
d
a
t
a
se
t
1
9
C
N
N
C
h
i
l
d
r
e
n
84
N
o
f
e
a
t
u
r
e
f
u
si
o
n
,
m
i
n
i
ma
l
a
b
l
a
t
i
o
n
s
[
7
]
O
f
f
l
i
n
e
N
I
S
T+
sp
e
c
i
a
l
d
a
t
a
se
t
1
9
C
N
N
C
h
i
l
d
r
e
n
87
S
h
a
l
l
o
w
f
e
a
t
u
r
e
s,
l
i
mi
t
e
d
g
e
n
e
r
a
l
i
z
a
t
i
o
n
[
8
]
O
f
f
l
i
n
e
P
r
i
v
a
t
e
C
N
N
C
h
i
l
d
r
e
n
>
8
0
A
d
u
l
t
d
a
t
a
,
n
o
t
p
e
d
i
a
t
r
i
c
[
9
]
O
f
f
l
i
n
e
N
I
S
T+
sp
e
c
i
a
l
d
a
t
a
se
t
1
9
C
N
N
C
h
i
l
d
r
e
n
87
A
d
u
l
t
d
a
t
a
,
n
o
t
p
e
d
i
a
t
r
i
c
[
1
1
]
O
n
l
i
n
e
P
r
i
v
a
t
e
S
V
M
A
d
u
l
t
9
6
.
3
C
a
l
l
s f
o
r
r
i
c
h
e
r
,
c
l
i
n
i
c
a
l
d
a
t
a
se
t
s
[
1
2
]
O
n
l
i
n
e
P
r
i
v
a
t
e
U
sed
w
e
b
a
p
p
A
d
u
l
t
-
N
o
n
-
p
e
d
i
a
t
r
i
c
,
n
o
g
l
y
p
h
f
o
c
u
s
[
1
3
]
O
n
l
i
n
e
P
r
i
v
a
t
e
R
C
N
N
El
d
e
r
-
D
a
t
a
s
e
t
t
r
a
n
sp
a
r
e
n
c
y
i
ssu
e
s
[
1
4
]
O
f
f
l
i
n
e
P
r
i
v
a
t
e
A
p
p
A
d
u
l
t
-
H
i
g
h
a
c
c
u
r
a
c
y
,
l
i
t
t
l
e
a
b
l
a
t
i
o
n
[
1
5
]
O
f
f
l
i
n
e
P
r
i
v
a
t
e
K
-
D
C
T
-
D
T
L
6
t
o
1
2
>
8
0
S
u
p
p
o
r
t
s s
t
r
o
k
e
-
t
e
x
t
u
r
e
a
n
a
l
o
g
y
[
1
6
]
O
f
f
l
i
n
e
P
r
i
v
a
t
e
C
N
N
6
t
o
1
2
>
8
4
En
t
r
o
p
y
f
e
a
t
u
r
e
i
d
e
a
t
r
a
n
sf
e
r
a
b
l
e
[
2
0
]
O
f
f
l
i
n
e
i
m
a
g
e
s
D
y
sg
r
a
p
h
i
a
d
a
t
a
se
t
V
i
T
Tr
a
n
sf
o
r
mer
F1
=
0
.
9
2
H
i
g
h
c
o
m
p
u
t
e
,
n
e
e
d
s
l
a
r
g
e
d
a
t
a
[
2
1
]
O
n
l
i
n
e
+
o
f
f
l
i
n
e
H
a
n
d
w
r
i
t
i
n
g
si
g
n
a
l
s
C
N
N
+
R
N
N
(
f
u
si
o
n
)
M
u
l
t
i
m
o
d
a
l
8
8
.
8
0
Li
mi
t
e
d
g
e
n
e
r
a
l
i
z
a
t
i
o
n
[
2
5
]
O
f
f
l
i
n
e
P
r
i
v
a
t
e
S
V
M
,
R
F
6
t
o
1
2
~
9
0
C
a
n
a
p
p
l
y
m
u
l
t
i
mo
d
a
l
f
o
r
sev
e
r
i
t
y
l
e
v
e
l
s
I
n
co
n
tr
ast,
o
u
r
p
r
o
p
o
s
ed
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
is
d
esig
n
ed
with
p
r
ac
ticality
in
m
in
d
.
B
y
s
tack
in
g
s
m
all
k
er
n
els
ac
r
o
s
s
co
n
v
o
lu
tio
n
al
b
lo
ck
s
,
th
e
m
o
d
el
is
ab
le
to
ca
p
tu
r
e
s
u
b
tle
h
an
d
wr
itin
g
d
etails
s
u
ch
as
s
tr
o
k
e
th
ick
n
ess
,
s
lan
t,
an
d
r
ev
er
s
als
wh
ile
d
r
asti
ca
lly
cu
ttin
g
d
o
wn
o
n
tr
ain
a
b
le
p
ar
am
eter
s
an
d
co
m
p
u
tatio
n
.
T
h
is
lig
h
tweig
h
t
d
esig
n
en
ab
les
ac
cu
r
ac
y
lev
els
co
m
p
ar
ab
le
to
lar
g
er
m
o
d
els
,
b
u
t
in
a
f
o
r
m
th
at
is
b
o
th
ef
f
icien
t
an
d
ac
ce
s
s
ib
le,
d
ir
ec
tly
a
d
d
r
ess
in
g
r
ea
l
-
wo
r
ld
c
h
allen
g
es
s
u
ch
as
lim
ited
p
ed
iatr
ic
d
ataset
av
ailab
ilit
y
,
m
o
d
est
h
ar
d
war
e
in
ed
u
ca
tio
n
al
en
v
ir
o
n
m
en
ts
,
an
d
th
e
n
ee
d
f
o
r
clin
ically
p
r
ac
tical
s
o
lu
tio
n
s
.
Sh
o
wca
s
in
g
th
e
b
r
o
ad
ap
p
licab
ilit
y
o
f
d
ee
p
lear
n
in
g
in
b
o
th
o
cc
u
p
atio
n
al
s
af
ety
an
d
ed
u
ca
tio
n
al
h
ea
lth
co
n
tex
ts
,
o
u
r
m
o
d
el
o
f
f
e
r
s
a
s
tr
o
n
g
b
alan
ce
b
etwe
en
a
cc
u
r
ac
y
,
s
p
ee
d
,
an
d
p
r
ac
tica
lity
f
o
r
lar
g
e
-
s
ca
le
d
y
s
g
r
ap
h
ia
d
etec
tio
n
.
3.
DATAS
E
T
Fo
r
r
esear
ch
p
u
r
p
o
s
e,
E
n
g
li
s
h
h
an
d
wr
itten
d
ataset
is
c
o
n
s
id
er
ed
.
T
h
e
d
ataset
u
s
ed
f
o
r
th
e
ex
p
er
im
en
tatio
n
was
co
llected
b
y
I
s
a
et
a
l.
[
8
]
,
th
is
d
ataset
h
as
to
tal
1
3
8
,
5
0
0
im
ag
es
(
7
8
,
2
7
5
im
a
g
es
f
o
r
th
e
n
o
r
m
al
class
,
5
2
,
1
9
6
f
o
r
th
e
r
ev
er
s
al,
an
d
8
,
0
2
9
f
o
r
t
h
e
co
r
r
ec
ted
class
)
.
Fo
llo
win
g
im
ag
es
will
g
iv
e
m
o
r
e
d
etails
ab
o
u
t
it.
I
n
Fig
u
r
es
1
to
3
,
d
ataset
co
n
tain
s
b
lack
an
d
wh
ite
alp
h
ab
ets
im
ag
es.
No
r
m
al
h
an
d
wr
itten
ch
ar
ac
ter
m
ea
n
s
n
o
s
y
m
p
to
m
o
f
d
y
s
g
r
ap
h
ia
b
u
t
co
r
r
ec
ted
h
an
d
wr
itten
ch
ar
ac
ter
m
ea
n
s
it’
s
a
b
eg
in
n
in
g
s
tag
e
o
f
d
y
s
g
r
a
p
h
ia
as
s
tr
o
k
es
ar
e
n
o
t
f
ir
m
an
d
tr
y
i
n
g
to
r
ewr
ite
o
r
s
o
m
e
tim
e
h
eig
h
t
an
d
wid
t
h
is
n
o
t
ap
p
r
o
p
r
iate
lik
e
n
o
r
m
al
h
a
n
d
wr
itten
ch
a
r
ac
ter
s
.
R
ev
er
s
al/d
y
s
g
r
ap
h
ic
h
an
d
wr
itten
ch
ar
ac
ter
m
ea
n
s
s
tr
o
n
g
s
y
m
p
to
m
o
f
d
y
s
g
r
ap
h
ia
as
ch
ild
u
s
ed
to
wr
ite
m
ir
r
o
r
im
ag
e
ch
ar
ac
te
r
s
an
d
s
tr
o
k
e
p
atter
n
is
alwa
y
s
d
if
f
er
e
n
t
f
o
r
s
am
e
ch
ar
ac
ter
.
W
e
h
av
e
ap
p
lied
au
g
m
en
tatio
n
to
tr
y
to
b
alan
ce
th
e
d
ataset
an
d
also
p
r
ep
r
o
ce
s
s
ed
to
o
b
tain
n
o
is
eless
clea
n
im
ag
es.
Fo
r
p
r
o
ce
s
s
in
g
th
e
im
ag
e
f
r
o
m
d
at
aset,
th
e
s
ize
o
f
im
ag
e
i
s
m
ad
e
2
9
×2
9
p
ix
els
an
d
in
p
u
t to
th
e
p
r
o
p
o
s
ed
m
o
d
el.
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
o
ve
l c
o
n
vo
lu
tio
n
n
eu
r
a
l n
et
w
o
r
k
mo
d
el
fo
r
d
ysg
r
a
p
h
ia
-
a
ff
ec
ted
h
a
n
d
w
r
itin
g
cla
s
s
ifica
ti
o
n
(
N
is
h
a
V
a
n
j
a
r
i
)
1421
Fig
u
r
e
1
.
No
r
m
al
h
an
d
wr
itten
ch
ar
ac
ter
s
im
ag
e
Fig
u
r
e
2
.
C
o
r
r
ec
te
d
h
a
n
d
wr
itten
ch
ar
ac
ter
s
im
ag
e
Fig
u
r
e
3
.
R
ev
er
s
al/d
y
s
g
r
a
p
h
ic
h
an
d
wr
itten
ch
a
r
ac
ter
s
im
ag
e
T
o
h
a
n
d
le
t
h
e
im
b
alan
ce
i
n
o
u
r
h
a
n
d
wr
itin
g
d
ataset,
we
u
p
s
am
p
led
th
e
m
in
o
r
ity
class
es
an
d
a
p
p
lie
d
d
ata
au
g
m
e
n
tatio
n
to
in
c
r
ea
s
e
d
iv
er
s
ity
an
d
v
ar
iab
ilit
y
.
Sp
e
cif
ically
,
im
ag
es
wer
e
au
g
m
e
n
ted
with
r
o
tatio
n
s
u
p
to
2
0
°,
h
o
r
izo
n
tal
an
d
v
er
ti
ca
l
s
h
if
ts
u
p
to
2
0
%,
s
h
ea
r
in
g
u
p
to
2
0
%,
an
d
zo
o
m
in
g
u
p
to
2
0
%.
T
h
ese
s
tep
s
en
s
u
r
ed
th
at
all
class
es
h
ad
s
u
f
f
icien
t
r
ep
r
esen
tatio
n
a
n
d
h
el
p
ed
th
e
m
o
d
el
lear
n
m
o
r
e
r
o
b
u
s
t
an
d
g
en
er
alize
d
f
ea
tu
r
es,
im
p
r
o
v
in
g
o
v
er
all
cla
s
s
if
icatio
n
p
er
f
o
r
m
a
n
ce
an
d
to
m
ak
e
o
u
r
m
o
d
el
m
o
r
e
r
o
b
u
s
t
an
d
g
e
n
er
aliza
b
le.
4.
M
E
T
H
O
D
I
n
p
r
o
p
o
s
ed
m
o
d
el
o
f
Fig
u
r
e
4
s
h
o
ws
r
esear
ch
m
ajo
r
l
y
f
o
c
u
s
ed
o
n
co
n
v
o
l
u
tio
n
;
as
co
n
v
o
lu
tio
n
h
as
k
er
n
el/f
ilter
,
s
tr
id
e
to
s
lid
e
k
er
n
el
o
n
im
ag
e
an
d
ap
p
ly
in
g
ele
m
en
twis
e
m
u
ltip
licatio
n
an
d
th
en
s
u
m
m
atio
n
.
B
y
u
s
in
g
co
n
v
o
lu
tio
n
b
ac
k
-
to
-
b
a
ck
,
m
o
d
el
will
lear
n
m
o
r
e
co
m
p
lex
h
ie
r
ar
ch
ical
f
ea
t
u
r
es.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el
h
as
4
co
n
v
o
lu
ti
o
n
lay
er
s
,
2
m
ax
p
o
o
lin
g
lay
er
s
,
f
latten
lay
e
r
f
o
llo
wed
with
f
u
lly
co
n
n
ec
t
ed
o
r
d
en
s
e
lay
er
.
Ma
x
p
o
o
lin
g
will
b
e
u
s
ed
f
o
r
d
im
en
s
io
n
ality
r
e
d
u
ctio
n
a
n
d
it
ex
tr
ac
ts
m
ax
im
u
m
v
alu
es
f
r
o
m
p
atch
o
f
ea
c
h
f
ea
tu
r
e
m
a
p
a
n
d
elim
in
ates
u
n
wan
ted
v
alu
es
to
r
ed
u
c
e
th
e
co
m
p
u
tatio
n
al
s
tr
ess
o
f
m
o
d
el
an
d
also
av
o
id
s
o
v
er
f
itti
n
g
p
r
o
b
lem
.
I
n
th
e
m
o
d
el,
So
f
tMa
x
ac
tiv
atio
n
f
u
n
ctio
n
is
u
s
ed
o
n
o
u
tp
u
t
lay
er
an
d
r
ec
tifie
d
lin
ea
r
u
n
it
(
R
eL
U
)
o
n
in
ter
n
al
lay
er
as
R
eL
U
f
o
cu
s
es
o
n
0
to
m
ax
v
alu
e
an
d
it
h
elp
s
u
s
to
f
o
cu
s
e
d
r
elev
an
t
d
ata
an
d
s
en
d
it to
n
ex
t la
y
e
r
f
o
r
p
r
o
ce
s
s
in
g
.
Fig
u
r
e
4
.
Pro
p
o
s
ed
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
Af
ter
f
latten
lay
er
,
d
e
n
s
e
la
y
er
is
ad
d
e
d
with
R
eL
U
ac
t
iv
atio
n
f
u
n
ctio
n
an
d
o
u
tp
u
t
lay
er
with
So
f
tMa
x
ac
tiv
atio
n
f
u
n
ctio
n
.
R
eL
U
is
n
o
t
d
ee
p
ly
af
f
ec
ted
b
y
th
e
v
a
n
is
h
in
g
g
r
a
d
ien
t
is
s
u
e,
an
d
p
er
m
its
g
r
ad
ien
ts
to
s
tay
b
ig
,
en
s
u
r
in
g
ef
f
icien
t b
ac
k
p
r
o
p
ag
ati
o
n
an
d
q
u
ick
er
co
n
v
er
g
en
ce
.
So
f
tMa
x
is
ap
p
r
o
p
r
iate
f
o
r
m
u
lti
-
class
class
if
icatio
n
jo
b
s
s
in
ce
it
tr
an
s
f
o
r
m
s
a
n
e
u
r
al
n
etwo
r
k
'
s
r
aw
o
u
tp
u
t
s
co
r
es
in
t
o
p
r
o
b
ab
ilit
ies
th
at
ad
d
u
p
to
1
.
C
NN
m
o
d
el
u
s
es
So
f
tMa
x
ac
tiv
atio
n
f
u
n
ctio
n
i
n
class
if
icatio
n
lay
er
an
d
s
p
ar
s
e
ca
teg
o
r
ical
cr
o
s
s
-
en
tr
o
p
y
(
SC
E
)
as th
e
lo
s
s
f
u
n
c
tio
n
.
So
f
tMa
x
f
u
n
ctio
n
is
u
s
in
g
(
1
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
2
5
2
-
8
9
3
8
I
n
t J Ar
tif
I
n
tell
,
Vo
l.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
1
8
-
1
4
2
7
1422
=
∑
3
=
1
(
1
)
zj
is
th
e
o
u
tp
u
t
f
r
o
m
d
en
s
e
la
y
er
,
Yj
is
class
j
p
r
ed
icted
p
r
o
b
ab
ilit
y
,
an
d
th
e
d
en
o
m
i
n
ato
r
is
th
e
ad
d
itio
n
o
f
p
r
o
b
a
b
ilit
ies o
f
all
3
class
es is
1
.
T
h
e
o
b
jectiv
e
f
u
n
ctio
n
o
r
lo
s
s
f
u
n
ctio
n
is
u
s
in
g
(
2
)
.
=
∑
l
og
=
1
(
2
)
S
is
n
u
m
b
er
o
f
t
r
ain
in
g
s
am
p
l
es,
ci
is
th
e
in
d
ex
o
f
c
o
r
r
ec
t
p
r
ed
icted
class
,
an
d
is
th
e
So
f
tMa
x
ca
lcu
lated
p
r
o
b
a
b
ilit
y
o
f
c
o
r
r
ec
t c
lass
.
B
y
s
u
b
s
titu
tin
g
(
1
)
an
d
(2
)
,
th
e
f
in
al
o
b
jectiv
e
f
u
n
ctio
n
is
o
b
tain
ed
as
(
3
)
.
=
∑
l
og
(
∑
3
=
1
)
=
1
(
3
)
I
t
is
th
e
f
in
al
eq
u
atio
n
o
f
p
r
o
p
o
s
ed
Nested
_
C
o
n
v
C
NN
ar
ch
itectu
r
e.
T
h
is
lo
s
s
f
u
n
ctio
n
v
alu
e
is
m
in
im
ized
d
u
r
in
g
tr
ain
in
g
to
r
ec
o
v
er
class
if
icatio
n
ac
cu
r
ac
y
.
O
u
r
a
i
m
i
s
t
o
m
i
n
i
m
i
z
e
t
h
e
l
o
s
s
a
s
i
t
w
i
l
l
b
o
o
s
t
t
o
c
l
a
s
s
i
f
y
t
h
e
c
o
r
r
e
c
t
c
l
as
s
.
T
a
b
l
e
2
wi
l
l
s
h
o
w
t
h
e
u
s
e
d
l
a
y
e
r
s
wi
t
h
f
u
n
c
t
i
o
n
al
i
ti
e
s
a
n
d
T
ab
l
e
3
s
h
o
ws
h
y
p
e
r
p
a
r
a
m
e
t
e
r
s
u
s
e
d
.
I
n
o
r
d
er
to
id
en
tify
ch
ar
ac
ter
is
tics
lik
e
ed
g
es,
tex
tu
r
es,
an
d
o
t
h
er
s
p
atial
p
atter
n
s
,
2
D
co
n
v
o
l
u
tio
n
ap
p
lies
a
co
llectio
n
o
f
k
er
n
els to
th
e
in
p
u
t im
ag
e
.
T
h
e
co
n
v
o
lu
tio
n
o
p
er
atio
n
'
s
o
u
tp
u
t is su
b
jecte
d
to
th
e
R
eL
U
f
u
n
ctio
n
.
R
eL
U
ad
d
s
n
o
n
-
lin
ea
r
ity
to
th
e
m
o
d
el
an
d
aid
s
in
its
lear
n
in
g
o
f
in
tr
icate
p
atter
n
s
b
y
s
ettin
g
all
n
eg
ativ
e
v
alu
es
in
t
h
e
o
u
t
p
u
t
f
ea
t
u
r
e
m
a
p
to
ze
r
o
.
T
h
e
lay
e
r
wo
u
l
d
ef
f
ec
tiv
ely
f
u
n
ctio
n
as
a
lin
ea
r
f
ilter
in
th
e
ab
s
en
ce
o
f
th
is
ac
tiv
atio
n
.
T
h
e
m
o
d
el
ca
n
lear
n
m
o
r
e
co
m
p
licated
p
att
er
n
s
th
an
a
s
in
g
le
co
n
v
o
l
u
tio
n
lay
er
co
u
ld
,
b
y
u
s
in
g
two
co
n
v
o
lu
ti
o
n
al
lay
er
s
b
ac
k
-
to
-
b
ac
k
.
So
,
o
u
r
m
o
d
el
will
lear
n
au
to
m
atica
lly
th
e
ef
f
ec
tiv
e
f
e
atu
r
es
an
d
it
d
esig
n
s
th
e
s
tr
u
c
tu
r
e
o
f
it
as
it
g
ain
s
in
f
o
r
m
ati
o
n
f
r
o
m
im
a
g
e
an
d
class
if
y
with
th
e
p
r
o
b
a
b
ilit
y
d
i
s
tr
ib
u
tio
n
.
T
ab
le
2
.
Pro
p
o
s
ed
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
lay
er
s
with
f
u
n
ctio
n
alities
S
r
N
o
B
l
o
c
k
La
y
e
r
s
B
e
n
e
f
i
t
s
1
B
l
o
c
k
1
–
2
-
c
o
n
v
o
l
u
t
i
o
n
f
u
n
c
t
i
o
n
s
f
o
l
l
o
w
e
d
b
y
m
a
x
p
o
o
l
i
n
g
C
o
n
v
2
D
(
3
2
f
i
l
t
e
r
s
,
3
×
3
,
R
e
LU
)
C
o
n
v
2
D
(
3
2
f
i
l
t
e
r
s
,
3
×
3
,
R
e
LU
)
M
a
x
P
o
o
l
i
n
g
2
D
(
2
×
2
)
Ex
t
r
a
c
t
l
o
w
-
l
e
v
e
l
h
a
n
d
w
r
i
t
i
n
g
f
e
a
t
u
r
e
s
a
n
d
r
e
f
i
n
e
a
n
d
r
e
d
u
c
e
f
e
a
t
u
r
e
m
a
p
s
2
B
l
o
c
k
2
–
1
-
c
o
n
v
o
l
u
t
i
o
n
f
u
n
c
t
i
o
n
f
o
l
l
o
w
e
d
b
y
m
a
x
p
o
o
l
i
n
g
C
o
n
v
2
D
(
6
4
f
i
l
t
e
r
s
,
3
×
3
,
R
e
LU
)
M
a
x
P
o
o
l
i
n
g
2
D
(
2
×
2
)
C
a
p
t
u
r
e
h
i
g
h
-
l
e
v
e
l
h
a
n
d
w
r
i
t
i
n
g
p
a
t
t
e
r
n
s
a
n
d
r
e
f
i
n
e
a
n
d
r
e
d
u
c
e
f
e
a
t
u
r
e
m
a
p
s
3
B
l
o
c
k
3
–
1
-
c
o
n
v
o
l
u
t
i
o
n
C
o
n
v
2
D
(
6
4
f
i
l
t
e
r
s
,
3
×
3
,
R
e
LU
)
C
a
p
t
u
r
e
h
i
g
h
-
l
e
v
e
l
h
a
n
d
w
r
i
t
i
n
g
p
a
t
t
e
r
n
s
4
F
l
a
t
t
e
n
i
n
g
F
l
a
t
t
e
n
(
)
C
o
n
v
e
r
t
f
e
a
t
u
r
e
ma
p
s
i
n
t
o
a
1
D
v
e
c
t
o
r
5
F
u
l
l
y
c
o
n
n
e
c
t
e
d
l
a
y
e
r
s
D
e
n
se
(
6
4
n
e
u
r
o
n
s,
R
e
LU
)
Le
a
r
n
h
i
g
h
-
l
e
v
e
l
h
a
n
d
w
r
i
t
i
n
g
f
e
a
t
u
r
e
s
6
O
u
t
p
u
t
l
a
y
e
r
D
e
n
se
(
3
n
e
u
r
o
n
s
,
s
o
f
t
m
a
x
)
F
o
r
m
u
l
t
i
-
c
l
a
ss
c
l
a
ssi
f
i
c
a
t
i
o
n
T
ab
le
3
.
Hy
p
er
p
a
r
am
eter
u
s
e
d
in
Nested
_
C
o
n
v
C
NN
P
a
r
a
me
t
e
r
V
a
l
u
e
O
p
t
i
mi
z
e
r
A
d
a
m
Le
a
r
n
i
n
g
r
a
t
e
0
.
0
0
1
(
d
e
f
a
u
l
t
)
Lo
ss f
u
n
c
t
i
o
n
S
C
E
M
e
t
r
i
c
s
A
c
c
u
r
a
c
y
N
u
mb
e
r
o
f
c
l
a
sse
s
3
I
n
p
u
t
si
z
e
2
9
×
2
9
×
1
4
.
1
.
E
x
perim
ent
a
l e
nv
ir
o
nm
ent
F
o
r
ex
p
e
r
i
m
e
n
t
a
t
io
n
p
u
r
p
o
s
e,
w
e
h
av
e
u
s
e
d
I
n
t
e
l
(
R
)
,
C
o
r
e
(
T
M
)
i
5
P
r
o
c
e
s
s
o
r
,
8
.
0
0
G
B
R
A
M
,
64
-
b
i
t
o
p
e
r
a
t
in
g
s
y
s
t
e
m
,
×
6
4
-
b
a
s
e
d
p
r
o
c
e
s
s
o
r
,
no
p
en
o
r
t
o
u
c
h
i
n
p
u
t
i
s
av
a
i
l
ab
l
e
;
r
a
th
e
r
in
p
u
t
i
s
ta
k
en
f
r
o
m
i
m
a
g
e
s
o
f
h
a
n
d
w
r
i
t
t
en
te
x
t
o
n
p
a
p
er
.
T
h
e
im
a
g
e
s
a
r
e
p
r
ep
r
o
c
e
s
s
e
d
to
s
ep
a
r
a
t
e
o
u
t
c
h
a
r
a
c
t
e
r
s
o
n
t
h
e
l
in
e
s
.
A
r
e
g
i
o
n
o
f
i
n
t
er
e
s
t
(
R
O
I
)
i
s
a
p
p
l
i
ed
t
o
ca
p
tu
r
e
p
o
r
t
i
o
n
o
f
a
n
i
m
ag
e
w
h
i
ch
c
o
n
t
a
in
s
a
l
p
h
a
b
e
t
th
a
t
w
e
w
an
t
t
o
f
i
l
t
e
r
an
d
s
a
v
e
i
t
a
s
an
i
m
ag
e
in
d
a
t
a
s
e
t.
So
,
w
e
h
av
e
r
ep
r
e
s
e
n
t
e
d
an
R
O
I
a
s
a
b
i
n
ar
y
m
a
s
k
im
a
g
e
f
o
r
th
e
e
x
p
er
i
m
en
t
a
t
i
o
n
we
r
u
n
th
e
c
o
d
e
o
n
G
o
o
g
l
e
C
o
l
ab
w
it
h
P
y
th
o
n
an
d
T
en
s
o
r
F
l
o
w
(
i
n
c
lu
d
in
g
K
e
r
a
s
)
,
N
u
m
P
y
,
M
a
tp
l
o
t
l
ib
(
f
o
r
v
i
s
u
a
l
i
z
a
t
i
o
n
)
,
O
p
en
C
V
(
f
o
r
i
m
ag
e
p
r
o
c
e
s
s
i
n
g
)
,
s
c
ik
i
t
-
l
ea
r
n
(
f
o
r
p
r
e
p
r
o
ce
s
s
i
n
g
)
,
a
n
d
P
an
d
a
s
(
f
o
r
d
a
ta
s
e
t
s
)
.
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
T
o
ex
tr
ac
t
in
f
o
r
m
atio
n
a
n
d
ca
teg
o
r
ize
h
an
d
wr
itten
ch
ar
a
cter
s
,
a
s
im
p
le
C
NN
f
o
r
h
an
d
wr
itin
g
class
if
icatio
n
co
n
s
is
ts
o
f
co
n
v
o
lu
tio
n
al
lay
er
s
f
o
llo
wed
b
y
d
en
s
e
lay
er
s
.
W
h
ile
th
e
m
ax
-
p
o
o
lin
g
lay
e
r
s
lo
wer
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
o
ve
l c
o
n
vo
lu
tio
n
n
eu
r
a
l n
et
w
o
r
k
mo
d
el
fo
r
d
ysg
r
a
p
h
ia
-
a
ff
ec
ted
h
a
n
d
w
r
itin
g
cla
s
s
ifica
ti
o
n
(
N
is
h
a
V
a
n
j
a
r
i
)
1423
d
im
en
s
io
n
ality
an
d
p
r
eser
v
e
s
ig
n
if
ican
t
f
ea
tu
r
es,
th
e
ea
r
lies
t
co
n
v
o
l
u
tio
n
al
lay
er
s
ca
p
tu
r
e
c
r
u
cial
p
atter
n
s
lik
e
ed
g
es
an
d
c
u
r
v
es.
Fo
llo
win
g
f
ea
tu
r
e
ex
tr
ac
tio
n
,
th
e
f
ea
tu
r
es
ar
e
r
u
n
th
r
o
u
g
h
d
en
s
e
la
y
er
s
th
at
ar
e
f
u
lly
co
n
n
ec
ted
.
T
h
is
aid
s
i
n
th
e
cr
ea
tio
n
o
f
f
in
al
p
r
e
d
ictio
n
s
u
s
i
n
g
lear
n
t
r
e
p
r
esen
tatio
n
s
.
Fo
r
s
im
p
le
h
an
d
wr
itin
g
r
ec
o
g
n
itio
n
ap
p
licatio
n
s
,
th
is
a
r
ch
itectu
r
e
wo
r
k
s
well
an
d
is
co
m
p
u
tatio
n
ally
ef
f
icien
t.
5
.
1
.
Scena
rio
1
:
ba
s
ic
co
nv
o
l
utio
n neura
l net
wo
rk
m
o
del
perf
o
rm
a
nce
a
na
ly
s
is
I
n
th
is
s
ce
n
ar
io
,
we
h
a
v
e
u
s
ed
b
asic
C
NN
m
o
d
el
with
3
co
n
v
o
lu
tio
n
,
3
m
ax
p
o
o
lin
g
,
1
d
e
n
s
e
lay
er
,
an
d
1
o
u
tp
u
t
lay
er
.
W
e
tr
ie
d
to
e
v
alu
ate
th
e
p
er
f
o
r
m
an
ce
o
f
th
is
m
o
d
el
o
n
h
an
d
wr
itin
g
d
ataset
f
o
r
its
class
if
icatio
n
.
T
ab
le
4
p
r
esen
t
s
th
e
p
er
f
o
r
m
a
n
ce
a
n
aly
s
is
o
f
b
asic
C
NN
m
o
d
el
f
o
r
v
ar
y
in
g
n
u
m
b
er
o
f
e
p
o
ch
s
.
Ov
er
ep
o
c
h
s
,
th
e
s
im
p
le
C
NN
m
o
d
el
co
n
s
is
ten
tly
im
p
r
o
v
es
its
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
,
d
em
o
n
s
tr
atin
g
s
u
cc
ess
f
u
l
lear
n
in
g
.
T
h
e
m
o
d
el
a
p
p
ea
r
s
t
o
g
en
er
alize
well
ev
en
in
i
ts
ea
r
ly
s
tag
es,
as
ev
id
en
ce
d
b
y
t
h
e
tr
ain
in
g
ac
cu
r
ac
y
c
o
m
m
en
cin
g
at
8
2
.
1
3
%
an
d
th
e
v
alid
atio
n
ac
c
u
r
ac
y
b
ein
g
m
ar
g
i
n
ally
h
ig
h
er
at
8
8
.
2
7
%.
B
o
th
m
ea
s
u
r
es
k
ee
p
r
is
in
g
as
tr
ain
in
g
g
o
es
o
n
,
cu
lm
in
atin
g
at
9
5
.
9
%
f
o
r
v
alid
atio
n
ac
cu
r
ac
y
an
d
9
6
.
3
9
%
f
o
r
tr
ai
n
in
g
ac
cu
r
ac
y
.
T
h
is
p
atter
n
s
h
o
ws
th
at
th
e
alg
o
r
ith
m
is
lear
n
in
g
h
an
d
wr
itin
g
p
atter
n
s
th
at
ar
e
r
elev
an
t
an
d
g
ettin
g
b
etter
at
m
ak
i
n
g
p
r
ed
ictio
n
s
o
v
e
r
tim
e.
L
o
s
s
es
f
o
r
tr
ain
in
g
an
d
v
alid
atio
n
b
o
t
h
s
tead
ily
d
ec
lin
e;
tr
ain
in
g
lo
s
s
f
alls
f
r
o
m
0
.
4
5
1
1
to
0
.
1
0
2
2
,
w
h
ile
v
alid
atio
n
lo
s
s
f
alls
f
r
o
m
0
.
3
1
0
4
to
0
.
1
1
7
5
.
As
th
e
d
is
cr
ep
an
cy
b
etwe
en
ac
tu
al
an
d
an
ticip
ated
v
alu
es
d
ec
r
ea
s
es,
th
is
s
h
o
ws
th
at
th
e
m
o
d
el
is
g
r
o
win
g
m
o
r
e
co
n
f
i
d
en
t
in
its
p
r
ed
ictio
n
s
.
B
u
t
th
er
e
is
a
s
m
all
o
s
cillatio
n
ar
o
u
n
d
th
e
f
if
th
ep
o
ch
,
wh
en
v
alid
atio
n
lo
s
s
m
ar
g
in
al
ly
r
is
e
an
d
v
alid
atio
n
ac
cu
r
ac
y
f
alls
f
r
o
m
9
4
.
0
2
%
to
9
3
.
6
7
%.
T
h
is
im
p
lies
th
at
th
e
m
o
d
el
m
ay
b
e
s
tar
tin
g
t
o
o
v
er
f
it
th
e
tr
ain
in
g
d
ata
a
litt
le
b
it
o
r
f
ac
in
g
a
co
m
p
lex
v
a
r
iety
in
h
an
d
wr
itin
g
s
ty
les.
Ho
wev
er
,
th
e
m
o
d
el
r
e
co
v
er
s
its
p
er
f
o
r
m
an
ce
a
n
d
ac
h
iev
es h
ig
h
v
alid
atio
n
ac
c
u
r
ac
y
in
later
ep
o
ch
s
.
T
ab
le
4
.
B
asic
co
n
v
o
l
u
tio
n
n
e
u
r
al
n
etwo
r
k
[
9
]
p
er
f
o
r
m
a
n
ce
an
aly
s
is
Ep
o
c
h
s
M
o
d
e
l
Tr
a
i
n
i
n
g
a
c
c
u
r
a
c
y
Tr
a
i
n
i
n
g
l
o
ss
V
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
V
a
l
i
d
a
t
i
o
n
l
o
ss
1
(
1
C
o
n
v
.
+
1
M
a
x
p
o
o
l
+
1
C
o
n
v
.
+
1
M
a
x
p
o
o
l
+
1
C
o
n
v
.
+
1
M
a
x
P
o
o
l
+
1
f
l
a
t
t
e
n
l
a
y
e
r
+
1
d
e
n
s
e
l
a
y
e
r
+
1
o
u
t
p
u
t
l
a
y
e
r
)
0
.
8
2
1
3
0
.
4
5
1
1
0
.
8
8
2
7
0
.
3
1
0
4
2
0
.
9
0
2
6
0
.
2
6
5
9
0
.
9
1
8
1
0
.
2
2
7
9
3
0
.
9
2
3
2
0
.
2
1
3
1
0
.
9
3
0
5
0
.
1
9
0
6
4
0
.
9
3
3
9
0
.
1
8
2
9
0
.
9
4
0
2
0
.
1
6
9
7
5
0
.
9
4
1
1
0
.
1
6
1
8
0
.
9
3
6
7
0
.
1
7
2
3
6
0
.
9
4
8
3
0
.
1
4
4
2
0
.
9
4
9
8
0
.
1
4
0
5
7
0
.
9
5
2
8
0
.
1
3
1
5
0
.
9
5
0
6
0
.
1
3
9
5
8
0
.
9
5
6
6
0
.
1
2
1
5
0
.
9
5
1
7
0
.
1
3
5
5
9
0
.
9
6
0
5
0
.
1
1
0
9
0
.
9
5
9
0
.
1
1
6
4
10
0
.
9
6
3
9
0
.
1
0
2
2
0
.
9
5
8
9
0
.
1
1
7
5
I
n
ter
n
al
la
y
er
s
u
s
e
R
eL
U
f
u
n
ctio
n
o
n
ea
ch
co
n
v
o
lu
tio
n
lay
er
an
d
So
f
tMa
x
o
n
last
lay
er
f
o
r
class
if
icatio
n
.
R
eL
U
co
n
s
id
er
ed
p
o
s
itiv
e
o
r
ze
r
o
a
n
d
So
f
tMa
x
u
s
ed
to
ca
lcu
late
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
tio
n
.
Fig
u
r
e
5
s
h
o
ws
th
e
tr
ain
i
n
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
a
n
d
lo
s
s
s
tatis
tics
.
As
p
er
o
b
s
er
v
atio
n
s
,
C
NN
m
o
d
el
p
er
f
o
r
m
s
well
o
n
t
r
ain
in
g
d
a
taset
with
9
6
.
3
9
%
ac
c
u
r
ac
y
an
d
1
0
%
lo
s
s
.
I
t
also
wo
r
k
s
well
o
n
v
alid
atio
n
d
ataset
with
9
5
.
8
9
%
ac
cu
r
ac
y
with
1
1
%
lo
s
s
.
W
e
h
a
v
e
ap
p
lied
ea
r
ly
s
to
p
in
c
o
d
e
a
n
d
it
o
b
s
er
v
e
th
at
th
is
is
th
e
b
est
ac
cu
r
ac
y
o
f
th
is
m
o
d
el.
W
e
tr
ied
t
o
test
it
o
n
r
ea
l
-
tim
e
im
ag
es
an
d
we
g
o
t
s
ig
n
i
f
ican
t
ch
a
n
g
e
as
it
u
s
ed
to
class
if
y
m
o
s
t o
f
s
am
p
l
es in
co
r
r
ec
ted
class
o
n
ly
an
d
as p
er
o
u
r
o
b
s
er
v
atio
n
,
it is
th
e
s
ig
n
o
f
o
v
er
f
itti
n
g
as d
ata
is
n
o
t c
lass
if
ied
p
r
o
p
er
ly
.
Fig
u
r
e
5
.
Plo
ts
o
f
C
NN
m
o
d
el
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
a
cy
an
d
l
o
s
s
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.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
1
8
-
1
4
2
7
1424
Alth
o
u
g
h
th
e
b
asic
C
NN
ac
h
iev
es
ex
ce
llen
t
ac
cu
r
ac
y
,
it
h
a
s
d
r
awb
ac
k
s
,
s
u
ch
as
th
e
p
o
s
s
ib
ilit
y
o
f
m
o
d
est
o
v
er
f
itti
n
g
a
n
d
p
o
s
s
ib
le
s
en
s
itiv
ity
to
ch
an
g
es
in
h
an
d
wr
itin
g
s
ty
les.
T
h
is
ca
n
b
e
f
ix
ed
b
y
p
r
o
p
o
s
in
g
a
cu
s
to
m
C
NN
with
an
ex
tr
a
f
ir
s
t
co
n
v
o
lu
tio
n
al
lay
er
.
T
h
e
n
e
two
r
k
wo
u
ld
b
e
a
b
le
to
ca
p
t
u
r
e
f
in
er
d
etails
an
d
in
cr
ea
s
e
class
if
icatio
n
r
o
b
u
s
tn
ess
if
th
is
ad
d
itio
n
al
lay
er
i
m
p
r
o
v
e
d
f
ea
tu
r
e
e
x
tr
ac
tio
n
e
ar
lier
.
T
h
e
p
r
o
p
o
s
ed
m
o
d
el'
s
ar
ch
itectu
r
e
ca
n
b
e
im
p
r
o
v
e
d
to
in
cr
ea
s
e
ac
cu
r
ac
y
,
r
ed
u
ce
o
v
er
f
itti
n
g
,
an
d
ef
f
icien
tly
h
an
d
le
a
g
r
ea
ter
v
ar
iety
o
f
h
an
d
wr
itin
g
s
ty
les.
5
.
2
.
Scena
rio
2
:
Nest
ed_
Co
nv
_
CNN
m
o
del per
f
o
r
m
a
nce
a
na
ly
s
is
T
ab
le
5
s
h
o
ws
th
e
m
o
d
el
is
ab
le
to
lear
n
a
h
ier
ar
ch
y
o
f
f
ea
tu
r
es
b
y
s
eq
u
en
tially
ap
p
ly
in
g
co
n
v
o
l
u
tio
n
al
lay
er
s
,
s
tar
tin
g
with
b
asic
ed
g
es
an
d
tex
tu
r
es
an
d
m
o
v
in
g
u
p
to
m
o
r
e
i
n
tr
icate
f
o
r
m
s
an
d
o
b
jects
in
th
e
d
ee
p
er
lay
er
s
.
Af
ter
e
x
p
er
im
e
n
tatio
n
,
we
o
b
tain
ed
ac
cu
r
ac
ies
a
n
d
lo
s
s
es
o
f
tr
ai
n
in
g
an
d
v
alid
atio
n
task
.
T
h
e
s
y
s
t
e
m
t
o
o
k
6
5
s
e
c
o
n
d
s
i
n
t
o
t
a
l t
o
f
i
n
is
h
t
h
e
e
p
o
c
h
s
.
S
i
n
c
e
m
i
n
i
m
al
l
o
s
s
is
a
m
e
a
s
u
r
e
o
f
t
h
e
m
o
d
e
l
'
s
e
f
f
e
c
ti
v
e
n
e
s
s
,
l
o
w
e
r
v
a
l
u
e
s
c
o
r
r
e
s
p
o
n
d
t
o
g
r
e
a
t
e
r
p
er
f
o
r
m
a
n
c
e
.
T
h
e
t
r
a
i
n
i
n
g
l
o
s
s
in
t
h
i
s
c
as
e
i
s
0
.
0
5
1
5
a
s
s
h
o
w
n
i
n
F
i
g
u
r
e
6
.
T
ab
le
5
.
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
p
er
f
o
r
m
a
n
ce
an
aly
s
is
Ep
o
c
h
s
M
o
d
e
l
Tr
a
i
n
i
n
g
a
c
c
u
r
a
c
y
Tr
a
i
n
i
n
g
l
o
ss
V
a
l
i
d
a
t
i
o
n
a
c
c
u
r
a
c
y
V
a
l
i
d
a
t
i
o
n
l
o
ss
1
(
2
C
o
n
v
.
+
1
M
a
x
p
o
o
l
+
1
C
o
n
v
.
+
1
M
a
x
p
o
o
l
+
1
C
o
n
v
.
+
1
f
l
a
t
t
e
n
l
a
y
e
r
+
1
D
e
n
s
e
l
a
y
e
r
+
1
O
u
t
p
u
t
l
a
y
e
r
)
0
.
8
6
1
1
0
.
3
6
3
9
0
.
9
1
7
7
0
.
2
3
3
9
2
0
.
9
3
0
4
0
.
1
9
4
8
0
.
9
3
9
1
0
.
1
6
8
9
3
0
.
9
4
6
7
0
.
1
4
7
7
0
.
9
5
1
6
0
.
1
3
6
6
4
0
.
9
5
7
6
0
.
1
1
8
8
0
.
9
5
3
0
0
.
1
3
1
3
5
0
.
9
6
3
9
0
.
1
0
1
4
0
.
9
6
1
4
0
.
1
0
9
3
6
0
.
9
6
8
6
0
.
0
8
6
9
0
.
9
6
2
3
0
.
1
0
8
0
7
0
.
9
7
3
2
0
.
0
7
5
5
0
.
9
6
6
8
0
.
0
9
3
4
8
0
.
9
7
6
6
0
.
0
6
6
1
0
.
9
6
8
7
0
.
0
8
6
7
9
0
.
9
7
9
7
0
.
0
5
8
0
0
.
9
6
9
7
0
.
0
8
3
8
10
0
.
9
8
1
7
0
.
0
5
1
5
0
.
9
7
2
6
0
.
0
7
7
2
Fig
u
r
e
6
.
Plo
ts
o
f
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
ac
cu
r
ac
y
an
d
l
o
s
s
W
e
h
av
e
s
et
d
ef
au
lt
lear
n
in
g
r
ate
with
b
atch
s
ize
8
an
d
Ad
am
o
p
tim
izer
.
T
h
e
m
o
d
el'
s
v
ali
d
atio
n
lo
s
s
is
co
m
p
u
ted
u
s
in
g
an
in
d
ep
e
n
d
en
t
v
alid
atio
n
d
ataset
an
d
u
n
u
s
ed
d
ata
s
am
p
les.
I
t
s
h
o
ws
t
h
e
lo
s
s
i
s
les
s
th
an
C
NN
[
9
]
an
d
ac
cu
r
ac
y
is
im
p
r
o
v
in
g
.
T
h
is
is
th
e
s
ig
n
o
f
r
o
b
u
s
t
m
o
d
el.
T
h
e
b
asic
C
NN
[
9
]
h
as
a
lo
w
tr
ain
in
g
lo
s
s
1
0
%
b
u
t
th
e
h
ig
h
v
alid
at
io
n
lo
s
s
1
1
%,
as
well
as
a
d
is
cr
ep
an
cy
b
etwe
en
tr
ain
in
g
ac
cu
r
ac
y
9
6
.
3
9
%
an
d
v
alid
atio
n
ac
cu
r
ac
y
9
5
.
8
9
%.
Giv
en
th
is
d
is
p
ar
ity
,
it
is
p
r
o
b
ab
ly
o
v
er
f
itti
n
g
th
e
tr
ain
i
n
g
s
e
t.
T
h
is
r
esu
lts
in
th
e
m
o
d
el
p
e
r
f
o
r
m
in
g
p
o
o
r
ly
o
n
th
e
v
alid
atio
n
s
et
b
y
ca
p
tu
r
in
g
n
o
is
e
an
d
p
a
r
ticu
lar
ch
a
r
ac
t
er
is
tics
th
at
d
o
n
o
t
ca
teg
o
r
ize
to
n
ew
d
ata.
T
ab
le
6
s
h
o
ws th
e
tr
ain
in
g
an
d
v
alid
atio
n
ac
cu
r
ac
y
an
d
l
o
s
s
s
tatis
ti
cs.
T
ab
le
6
.
C
o
m
p
a
r
is
o
n
o
f
p
er
f
o
r
m
an
ce
m
etr
ics
S
r
N
o
M
o
d
e
l
T.
A
c
c
.
T.
Lo
ss
V
.
A
c
c
.
V
.
Lo
ss
Tr
a
i
n
a
b
l
e
p
a
r
a
m
e
t
e
r
s
P
r
e
c
i
s
i
o
n
R
e
c
a
l
l
A
c
c
u
r
a
c
y
F1
-
sc
o
r
e
1
C
N
N
0
.
9
6
0
.
1
0
0
.
9
5
0
.
1
1
60
,
675
8
5
.
3
0
8
5
.
7
1
8
5
.
7
1
0
.
8
5
2
N
e
st
e
d
_
C
o
n
v
_
C
N
N
0
.
9
8
0
.
0
5
0
.
9
7
0
.
0
7
1
0
2
,
6
9
1
8
8
.
4
4
8
8
.
3
5
8
8
.
3
5
0
.
8
8
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
o
ve
l c
o
n
vo
lu
tio
n
n
eu
r
a
l n
et
w
o
r
k
mo
d
el
fo
r
d
ysg
r
a
p
h
ia
-
a
ff
ec
ted
h
a
n
d
w
r
itin
g
cla
s
s
ifica
ti
o
n
(
N
is
h
a
V
a
n
j
a
r
i
)
1425
T
h
e
p
r
o
p
o
s
ed
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
ac
h
iev
es
a
b
alan
ce
d
p
er
f
o
r
m
a
n
ce
,
with
b
o
th
h
i
g
h
tr
ain
i
n
g
ac
cu
r
ac
y
an
d
h
ig
h
v
alid
atio
n
ac
cu
r
ac
y
.
I
ts
tr
ain
in
g
l
o
s
s
0
.
0
5
1
5
an
d
v
alid
atio
n
lo
s
s
0
.
0
7
7
2
ar
e
b
o
t
h
lo
w
an
d
clo
s
e
to
ea
ch
o
th
er
,
in
d
icatin
g
th
at
th
e
m
o
d
el
h
as
lear
n
ed
th
e
p
atter
n
s
in
th
e
d
ata
well
with
o
u
t
o
v
er
f
itti
n
g
.
T
h
u
s
,
th
e
p
r
o
p
o
s
ed
m
o
d
el
d
em
o
n
s
tr
ates
s
tr
o
n
g
g
en
e
r
aliza
tio
n
,
m
ea
n
in
g
it
p
er
f
o
r
m
s
well
o
n
n
ew
d
ata.
L
o
s
s
-
v
alid
atio
n
is
an
im
p
o
r
ta
n
t
tech
n
iq
u
e
f
o
r
p
r
o
d
u
cin
g
a
r
o
b
u
s
t
an
d
tr
u
s
two
r
th
y
ass
ess
m
en
t
o
f
a
m
o
d
el'
s
p
er
f
o
r
m
an
ce
,
en
s
u
r
in
g
it
g
e
n
e
r
alize
s
well
to
n
ew
d
ata,
an
d
a
s
s
is
tin
g
with
m
o
d
el
s
elec
tio
n
an
d
h
y
p
er
p
ar
am
eter
tu
n
in
g
.
T
esti
n
g
ac
cu
r
ac
y
is
also
0
.
9
5
a
n
d
lo
s
s
is
0
.
1
0
.
W
e
p
e
r
f
o
r
m
e
d
cr
o
s
s
v
alid
atio
n
an
d
t
h
e
m
ea
n
ac
cu
r
ac
y
:
0
.
9
6
,
s
tan
d
ar
d
d
ev
iatio
n
:
0
.
0
0
2
o
f
b
asic
m
o
d
el
an
d
n
o
v
e
l
C
NN
m
o
d
el
is
h
av
in
g
m
ea
n
ac
cu
r
ac
y
:
0
.
9
7
,
s
tan
d
ar
d
d
e
v
iatio
n
:
0
.
0
0
1
.
As
s
h
o
wn
in
T
ab
le
6
,
th
e
b
asic
C
NN
m
o
d
el
is
to
o
s
im
p
lis
tic
an
d
u
n
ab
le
t
o
g
en
er
alize
ad
eq
u
ately
to
n
e
w
d
ata,
o
v
er
f
itti
n
g
m
a
y
h
ap
p
en
.
T
h
e
n
u
m
b
e
r
o
f
tr
ain
a
b
le
p
ar
am
eter
s
in
th
e
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
lo
s
s
to
1
0
2
,
6
9
1
b
y
in
clu
d
in
g
a
n
ex
tr
a
c
o
n
v
o
lu
tio
n
al
lay
e
r
.
W
ith
th
e
ad
d
itio
n
o
f
ad
d
itio
n
al
f
ilter
s
an
d
weig
h
ts
f
r
o
m
th
is
n
ew
co
n
v
o
lu
tio
n
al
l
ay
er
,
t
h
e
m
o
d
el
was
ab
le
to
e
x
tr
ac
t
m
o
r
e
i
n
tr
icate
an
d
ab
s
tr
ac
t
f
ea
tu
r
es
f
r
o
m
th
e
in
p
u
t
d
ata.
I
n
o
r
d
e
r
to
b
alan
ce
o
v
er
f
itti
n
g
,
th
e
m
o
d
el'
s
co
m
p
lex
ity
is
also
in
cr
ea
s
ed
.
T
h
is
en
h
a
n
ce
m
en
t
em
p
h
asizes
h
o
w
o
p
tim
izi
n
g
th
e
m
o
d
el
ar
ch
itectu
r
e
m
ay
g
r
ea
tly
a
f
f
ec
t
p
er
f
o
r
m
an
ce
b
y
f
in
d
in
g
t
h
e
id
ea
l r
atio
o
f
lay
e
r
s
.
5
.
3
.
P
er
f
o
r
m
a
nce
m
e
t
rics a
nd
s
t
a
t
is
t
ica
l t
esting
Fig
u
r
e
7
o
f
co
n
f
u
s
io
n
m
atr
ix
s
h
o
ws
th
at
th
e
m
o
d
el
p
e
r
f
o
r
m
s
well
o
v
er
all,
with
h
ig
h
tr
u
e
p
o
s
itiv
e
co
u
n
ts
f
o
r
ea
c
h
class
:
1
7
,
8
7
9
f
o
r
class
0
,
1
6
,
8
2
5
f
o
r
class
1
,
an
d
1
5
,
4
1
1
f
o
r
class
2
.
T
h
e
co
n
f
u
s
io
n
m
at
r
ix
s
h
o
ws
th
at
th
e
m
o
d
el
p
e
r
f
o
r
m
s
well
o
v
er
all.
T
h
ese
n
u
m
b
er
s
in
d
icate
th
at
th
e
m
o
d
el
h
as
b
ee
n
ad
eq
u
ately
tr
ain
ed
an
d
is
ca
p
a
b
le
o
f
co
r
r
ec
tly
p
r
ed
ictin
g
th
e
v
ar
io
u
s
cl
ass
es
in
th
e
d
ataset.
T
h
e
h
i
g
h
t
r
u
e
p
o
s
i
t
i
v
e
c
o
u
n
ts
f
o
r
e
a
c
h
c
l
ass
d
e
m
o
n
s
t
r
a
t
e
t
h
e
m
o
d
e
l
'
s
r
o
b
u
s
t
n
e
s
s
a
n
d
a
b
i
l
it
y
to
h
a
n
d
l
e
s
e
v
e
r
a
l
c
l
as
s
es
w
i
t
h
r
e
a
s
o
n
a
b
l
e
a
c
c
u
r
a
c
y
.
T
h
i
s
p
e
r
f
o
r
m
a
n
c
e
is
a
g
o
o
d
i
n
d
i
c
a
t
o
r
o
f
t
h
e
m
o
d
e
l'
s
o
v
e
r
a
l
l
r
eli
a
b
i
l
it
y
a
n
d
e
f
f
i
c
a
c
y
a
t
c
l
as
s
i
f
y
in
g
t
a
s
k
s
.
Fig
u
r
e
7
.
C
o
n
f
u
s
io
n
m
atr
i
x
b
a
s
ic
C
NN
an
d
Nested
_
C
o
n
v
_
C
NN
T
h
er
e
is
a
n
o
ticea
b
le
p
er
f
o
r
m
an
ce
d
if
f
er
en
ce
b
etwe
e
n
th
e
b
asic
C
N
N
m
o
d
el
an
d
o
u
r
Nested
_
C
o
n
v
_
C
NN
m
o
d
el
in
ter
m
s
o
f
class
if
icatio
n
m
ea
s
u
r
es,
am
o
n
g
o
th
er
im
p
o
r
tan
t
p
er
f
o
r
m
an
ce
in
d
icato
r
s
.
W
ith
an
ac
cu
r
ac
y
o
f
9
6
.
3
9
%,
ex
is
tin
g
m
o
d
el
is
ab
le
to
class
if
y
s
lig
h
tly
m
o
r
e
th
an
h
alf
o
f
th
e
ca
s
es
co
r
r
ec
tly
.
O
u
r
m
o
d
el,
o
n
th
e
o
th
er
h
an
d
,
p
er
f
o
r
m
s
n
o
ticea
b
l
y
b
etter
,
p
r
o
p
er
ly
class
if
y
in
g
t
h
e
g
r
ea
t m
ajo
r
ity
o
f
ca
s
es
with
an
ac
cu
r
ac
y
o
f
8
8
.
3
5
%.
W
ith
a
r
ec
all
o
f
8
8
.
3
5
%,
it
co
r
r
ec
tly
d
etec
ts
th
e
m
ajo
r
ity
o
f
th
e
r
ea
l
p
o
s
itiv
e
ca
s
es,
an
d
it
s
p
r
ec
is
i
o
n
o
f
8
8
.
4
4
%
s
h
o
ws
th
at
al
m
o
s
t
all
o
f
it
s
p
o
s
itiv
e
p
r
ed
ictio
n
s
ar
e
ac
cu
r
ate.
T
h
e
F1
-
s
co
r
e
o
f
0
.
8
8
in
d
ica
tes
th
at
o
u
r
m
o
d
el
is
m
o
r
e
ad
ep
t
at
s
tr
ik
in
g
a
b
alan
ce
b
etwe
en
r
ec
all
a
n
d
p
r
ec
is
io
n
.
I
n
g
en
er
al,
o
u
r
m
o
d
el
p
er
f
o
r
m
s
f
ar
b
etter
th
an
e
x
is
tin
g
o
n
e,
m
a
k
in
g
it
a
lo
t
m
o
r
e
d
ep
e
n
d
ab
le
a
n
d
ef
f
icien
t
class
if
ier
.
T
h
e
p
er
f
o
r
m
an
ce
o
f
th
e
p
r
o
p
o
s
ed
n
ested
ca
s
ca
d
ed
C
NN
was
ev
alu
ated
u
s
in
g
5
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
an
d
co
m
p
ar
e
d
with
a
b
aselin
e
C
NN
[
9
]
.
Acr
o
s
s
th
e
f
o
ld
s
,
th
e
n
ested
C
N
N
ac
h
iev
ed
a
m
ea
n
v
alid
atio
n
ac
cu
r
ac
y
o
f
0
.
9
7
2
6
(
r
an
g
e:
0
.
9
7
1
–
0
.
9
7
3
)
with
a
co
r
r
esp
o
n
d
in
g
m
ea
n
v
alid
at
io
n
lo
s
s
o
f
0
.
0
7
7
2
,
o
u
tp
er
f
o
r
m
in
g
th
e
b
aselin
e
C
NN,
wh
ich
r
ec
o
r
d
ed
a
m
ea
n
v
alid
atio
n
ac
cu
r
ac
y
o
f
0
.
9
5
8
9
(
r
an
g
e:
0
.
9
5
6
–
0
.
9
6
0
)
an
d
m
ea
n
v
alid
atio
n
lo
s
s
o
f
0
.
1
1
7
5
.
I
t
d
esp
ite
h
av
in
g
o
n
ly
1
0
2
,
6
9
1
tr
ain
ab
le
p
ar
am
eter
s
,
d
em
o
n
s
tr
ated
s
u
p
er
io
r
p
er
f
o
r
m
an
ce
wh
ile
m
ain
tain
in
g
a
lig
h
tweig
h
t
d
esig
n
co
m
p
ar
ed
to
t
h
e
b
aselin
e’
s
6
0
,
6
7
5
p
ar
am
eter
s
.
Statis
t
ical
an
aly
s
is
u
s
in
g
a
p
air
ed
t
-
test
ac
r
o
s
s
th
e
5
f
o
ld
s
co
n
f
ir
m
e
d
th
at
t
h
e
im
p
r
o
v
e
m
en
t
in
ac
cu
r
ac
y
is
s
tatis
t
ically
s
ig
n
if
ican
t
(
p
<0
.
0
5
)
.
I
t
also
p
r
o
v
en
with
u
s
e
o
f
R
OC
cu
r
v
es
an
d
AUC
v
alu
e
s
,
wh
ich
s
h
o
w
h
o
w
ef
f
ec
tiv
ely
th
e
m
o
d
el
ca
n
d
i
s
tin
g
u
is
h
b
etwe
en
d
if
f
er
e
n
t
h
an
d
wr
itin
g
class
es.
E
n
co
u
r
a
g
in
g
ly
,
th
e
m
o
d
e
l
m
ain
tain
ed
h
ig
h
AUC
s
co
r
e
s
ev
en
u
n
d
er
ch
allen
g
in
g
c
o
n
d
itio
n
s
,
s
u
g
g
esti
n
g
it
ca
n
r
eliab
ly
r
ec
o
g
n
ize
h
an
d
wr
itin
g
p
atter
n
s
d
esp
ite
s
m
all
d
is
to
r
tio
n
s
.
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.
15
,
No
.
2
,
Ap
r
il
20
26
:
1
4
1
8
-
1
4
2
7
1426
6.
AB
L
AT
I
O
N
ST
UDY
W
h
en
we
tr
ied
s
im
p
lif
y
in
g
o
u
r
m
o
d
el
to
j
u
s
t
a
s
in
g
le
c
o
n
v
o
lu
tio
n
al
lay
er
,
we
n
o
ticed
a
cle
ar
d
r
o
p
i
n
p
er
f
o
r
m
an
ce
.
Acc
u
r
ac
y
f
ell
b
y
n
ea
r
ly
5
%,
an
d
t
h
e
lo
s
s
in
c
r
ea
s
ed
b
y
3
–
5
%,
s
h
o
win
g
t
h
at
th
e
m
o
d
el
s
tr
u
g
g
led
to
tr
u
ly
u
n
d
er
s
tan
d
th
e
h
an
d
wr
itin
g
p
atter
n
s
.
I
t
was
as
if
th
e
n
etwo
r
k
c
o
u
ld
o
n
ly
s
ee
th
e
r
o
u
g
h
o
u
tlin
es
b
u
t
m
is
s
ed
th
e
f
in
er
d
etails
—
th
e
cu
r
v
es,
lo
o
p
s
,
a
n
d
s
tr
o
k
es
th
at
m
ak
e
ea
ch
letter
u
n
iq
u
e.
Mu
l
tip
le
co
n
v
o
lu
tio
n
al
lay
er
s
ac
t
lik
e
lay
er
s
o
f
u
n
d
e
r
s
tan
d
in
g
:
th
e
f
ir
s
t
s
ee
s
ed
g
es
,
th
e
n
ex
t
s
tar
ts
to
r
ec
o
g
n
ize
s
h
ap
es,
an
d
d
ee
p
er
lay
er
s
ca
p
tu
r
e
th
e
f
u
ll
s
tr
u
ctu
r
e
o
f
letter
s
.
W
ith
o
u
t
th
is
h
ie
r
ar
ch
y
,
th
e
m
o
d
el
co
u
ld
n
’
t
g
en
er
alize
well,
a
n
d
lear
n
in
g
b
ec
am
e
u
n
s
tab
le.
T
h
i
s
e
x
p
e
r
i
e
n
c
e
r
e
i
n
f
o
r
c
e
d
w
h
y
s
ta
c
k
i
n
g
c
o
n
v
o
l
u
t
i
o
n
a
l
b
l
o
c
k
s
is
c
r
u
c
i
a
l
,
t
h
e
y
a
l
l
o
w
t
h
e
n
e
t
w
o
r
k
t
o
g
r
a
d
u
a
l
l
y
g
i
v
es
d
e
t
a
i
l
e
d
u
n
d
e
r
s
t
a
n
d
i
n
g
a
n
d
h
e
lp
i
n
g
i
t
a
c
h
i
e
v
e
h
i
g
h
e
r
a
c
c
u
r
ac
y
a
n
d
l
o
w
e
r
l
o
s
s
.
7.
CO
NCLU
SI
O
N
T
h
is
p
ap
er
p
r
esen
ted
im
p
r
o
v
ed
n
o
v
el
C
NN
m
o
d
el
f
o
r
d
y
s
g
r
ap
h
ia
class
if
icatio
n
b
ased
o
n
E
n
g
lis
h
h
an
d
wr
itin
g
.
W
e
h
av
e
e
x
p
er
i
m
en
ted
with
two
m
o
d
els
lik
e
b
asic
C
NN
an
d
p
r
o
p
o
s
ed
Nested
_
C
o
n
v
_
C
NN
an
d
p
r
o
p
o
s
ed
m
o
d
el
wo
r
k
e
d
well
as
th
e
tr
ain
in
g
ac
c
u
r
ac
y
is
9
8
.
1
7
%.
T
h
e
m
o
d
el'
s
v
alid
atio
n
lo
s
s
is
co
m
p
u
ted
u
s
in
g
an
in
d
e
p
en
d
e
n
t
v
alid
a
tio
n
d
ataset
an
d
u
n
u
s
ed
d
at
a
s
am
p
les.
I
t
m
ea
s
u
r
es
th
e
m
o
d
el'
s
ab
ilit
y
to
g
en
er
alize
to
n
ew
u
n
test
ed
d
ata
with
v
alid
atio
n
lo
s
s
is
0
.
0
7
7
2
an
d
v
alid
atio
n
ac
cu
r
ac
y
is
9
7
.
2
6
%.
As
C
NN
is
m
ain
ly
u
s
ed
t
o
s
o
lv
e
th
e
is
s
u
es
o
f
f
ea
tu
r
e
lear
n
i
n
g
in
n
o
n
-
lin
ea
r
ity
-
b
ased
s
am
p
les
a
n
d
ef
f
ec
tiv
e
d
o
w
n
s
am
p
lin
g
,
th
e
n
etwo
r
k
p
er
f
o
r
m
s
b
etter
o
n
class
if
icatio
n
tas
k
s
.
T
h
ese
f
ea
tu
r
es
b
o
o
s
t
th
e
n
etwo
r
k
'
s
p
o
wer
an
d
ca
p
ac
ity
to
id
en
tify
in
tr
icate
p
atter
n
s
.
W
e
h
av
e
tr
ied
ar
ch
itectu
r
al
ch
an
g
es
an
d
it
wo
r
k
ed
well.
W
e
h
av
e
u
s
ed
d
ef
au
lt
lear
n
in
g
r
ate
an
d
b
atc
h
s
ize
o
f
8
with
Ad
am
o
p
tim
izer
.
W
h
ile
th
e
co
m
p
u
tatio
n
al
co
m
p
lex
ity
s
lig
h
tly
in
cr
ea
s
es,
th
e
b
en
ef
its
o
f
im
p
r
o
v
ed
r
ec
o
g
n
itio
n
p
er
f
o
r
m
an
ce
m
ak
e
th
e
Nested
_
C
o
n
v
_
C
NN
m
o
r
e
ap
p
r
o
p
r
iate
f
o
r
co
m
p
le
x
h
an
d
wr
itin
g
d
atasets
.
T
h
is
h
elp
s
ex
tr
ac
t
lo
w
er
-
lev
el
f
ea
tu
r
es
lik
e
ed
g
es
an
d
tex
tu
r
es
b
ef
o
r
e
m
o
v
in
g
to
d
ee
p
er
f
ea
tu
r
e
r
ep
r
esen
tatio
n
s
.
F
UNDING
I
NF
O
R
M
A
T
I
O
N
T
h
e
au
th
o
r
s
d
ec
lar
ed
t
h
at
th
ey
h
av
e
n
o
t r
ec
eiv
ed
an
y
f
u
n
d
in
g
f
o
r
t
h
is
r
esear
ch
.
AUTHO
R
CO
NT
RI
B
UT
I
O
NS ST
A
T
E
M
E
N
T
T
h
is
jo
u
r
n
al
u
s
es
th
e
C
o
n
t
r
ib
u
to
r
R
o
les
T
a
x
o
n
o
m
y
(
C
R
ed
iT)
to
r
ec
o
g
n
ize
in
d
iv
i
d
u
al
au
th
o
r
co
n
tr
ib
u
tio
n
s
,
r
ed
u
ce
au
th
o
r
s
h
ip
d
is
p
u
tes,
an
d
f
ac
ilit
ate
co
llab
o
r
atio
n
.
Na
m
e
o
f
Aut
ho
r
C
M
So
Va
Fo
I
R
D
O
E
Vi
Su
P
Fu
Nis
h
a
Van
jar
i
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
✓
Pra
s
an
n
a
J
.
Sh
ete
✓
✓
✓
✓
✓
✓
✓
✓
C
:
C
o
n
c
e
p
t
u
a
l
i
z
a
t
i
o
n
M
:
M
e
t
h
o
d
o
l
o
g
y
So
:
So
f
t
w
a
r
e
Va
:
Va
l
i
d
a
t
i
o
n
Fo
:
Fo
r
mal
a
n
a
l
y
s
i
s
I
:
I
n
v
e
s
t
i
g
a
t
i
o
n
R
:
R
e
so
u
r
c
e
s
D
:
D
a
t
a
C
u
r
a
t
i
o
n
O
:
W
r
i
t
i
n
g
-
O
r
i
g
i
n
a
l
D
r
a
f
t
E
:
W
r
i
t
i
n
g
-
R
e
v
i
e
w
&
E
d
i
t
i
n
g
Vi
:
Vi
su
a
l
i
z
a
t
i
o
n
Su
:
Su
p
e
r
v
i
s
i
o
n
P
:
P
r
o
j
e
c
t
a
d
mi
n
i
st
r
a
t
i
o
n
Fu
:
Fu
n
d
i
n
g
a
c
q
u
i
si
t
i
o
n
CO
NF
L
I
C
T
O
F
I
N
T
E
R
E
S
T
ST
A
T
E
M
E
NT
T
h
e
au
th
o
r
s
d
ec
lar
ed
t
h
at
th
ey
h
av
e
n
o
k
n
o
wn
co
m
p
etin
g
in
t
er
est.
DATA AV
AI
L
AB
I
L
I
T
Y
T
h
e
d
atasets
an
aly
ze
d
d
u
r
in
g
th
e
cu
r
r
e
n
t
s
tu
d
y
a
r
e
av
ailab
le
f
r
o
m
th
e
co
r
r
esp
o
n
d
in
g
au
th
o
r
u
p
o
n
r
ea
s
o
n
ab
le
r
eq
u
est.
RE
F
E
R
E
NC
E
S
[
1
]
E.
S
.
S
h
a
p
i
r
o
,
“
P
r
e
v
e
n
t
i
n
g
a
c
a
d
e
mi
c
f
a
i
l
u
r
e
,
”
S
c
h
o
o
l
Ps
y
c
h
o
l
o
g
y
Re
v
i
e
w
,
v
o
l
.
1
7
,
n
o
.
4
,
p
p
.
6
0
1
–
6
1
3
,
1
9
8
8
,
d
o
i
:
1
0
.
1
0
8
0
/
0
2
7
9
6
0
1
5
.
1
9
8
8
.
1
2
0
8
5
3
7
8
.
[
2
]
W
.
W
.
L
i
m,
K
.
J
.
Y
e
o
,
a
n
d
L
.
H
a
n
d
a
y
a
n
i
,
“
A
s
y
st
e
ma
t
i
c
r
e
v
i
e
w
o
n
i
n
t
e
r
v
e
n
t
i
o
n
s
f
o
r
c
h
i
l
d
r
e
n
w
i
t
h
d
y
sl
e
x
i
a
,
”
I
n
t
e
r
n
a
t
i
o
n
a
l
J
o
u
rn
a
l
o
f
Ev
a
l
u
a
t
i
o
n
a
n
d
Re
s
e
a
r
c
h
i
n
E
d
u
c
a
t
i
o
n
,
v
o
l
.
1
2
,
n
o
.
3
,
p
p
.
1
6
7
4
–
1
6
8
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
1
5
9
1
/
i
j
e
r
e
.
v
1
2
i
3
.
2
5
0
9
9
.
[
3
]
P
.
W
e
st
w
o
o
d
a
n
d
L
.
G
r
a
h
a
m,
“
H
o
w
man
y
c
h
i
l
d
r
e
n
w
i
t
h
s
p
e
c
i
a
l
n
e
e
d
s
i
n
r
e
g
u
l
a
r
c
l
a
sses
?
o
f
f
i
c
i
a
l
p
r
e
d
i
c
t
i
o
n
s
V
S
t
e
a
c
h
e
r
s’
p
e
r
c
e
p
t
i
o
n
s i
n
S
o
u
t
h
A
u
st
r
a
l
i
a
a
n
d
N
e
w
S
o
u
t
h
W
a
l
e
s,”
Au
s
t
r
a
l
i
a
n
J
o
u
rn
a
l
o
f
L
e
a
r
n
i
n
g
D
i
s
a
b
i
l
i
t
i
e
s
,
v
o
l
.
5
,
n
o
.
3
,
p
p
.
2
4
–
3
5
,
2
0
0
0
,
d
o
i
:
1
0
.
1
0
8
0
/
1
9
4
0
4
1
5
0
0
0
9
5
4
6
6
3
2
.
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
o
ve
l c
o
n
vo
lu
tio
n
n
eu
r
a
l n
et
w
o
r
k
mo
d
el
fo
r
d
ysg
r
a
p
h
ia
-
a
ff
ec
ted
h
a
n
d
w
r
itin
g
cla
s
s
ifica
ti
o
n
(
N
is
h
a
V
a
n
j
a
r
i
)
1427
[
4
]
K
.
S
p
o
o
n
,
D
.
C
r
a
n
d
a
l
l
,
a
n
d
K
.
S
i
e
k
,
“
T
o
w
a
r
d
s
d
e
t
e
c
t
i
n
g
d
y
sl
e
x
i
a
i
n
c
h
i
l
d
r
e
n
’
s
h
a
n
d
w
r
i
t
i
n
g
u
s
i
n
g
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
re
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
AI
f
o
r
S
o
c
i
a
l
G
o
o
d
W
o
rks
h
o
p
,
2
0
1
9
,
p
p
.
1
–
5.
[
5
]
M
.
S
.
A
.
B
.
R
o
s
l
i
,
I
.
S
.
I
sa,
S
.
A
.
R
a
ml
a
n
,
S
.
N
.
S
u
l
a
i
m
a
n
,
a
n
d
M
.
I
.
F
.
M
a
r
u
z
u
k
i
,
“
D
e
v
e
l
o
p
me
n
t
o
f
C
N
N
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
f
o
r
d
y
s
l
e
x
i
a
h
a
n
d
w
r
i
t
i
n
g
r
e
c
o
g
n
i
t
i
o
n
,
”
i
n
2
0
2
1
1
1
t
h
I
E
EE
I
n
t
e
rn
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
n
t
r
o
l
S
y
s
t
e
m
,
C
o
m
p
u
t
i
n
g
a
n
d
En
g
i
n
e
e
r
i
n
g
,
I
C
C
S
C
E
2
0
2
1
,
2
0
2
1
,
p
p
.
1
9
4
–
1
9
9
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
S
C
E5
2
1
8
9
.
2
0
2
1
.
9
5
3
0
9
7
1
.
[
6
]
N
.
S
.
L.
S
e
ma
n
,
I
.
S
.
I
sa,
S
.
A
.
R
a
ml
a
n
,
W
.
L
.
-
C
h
i
h
,
a
n
d
M
.
I
.
F
.
M
a
r
u
z
u
k
i
,
“
N
o
t
i
c
e
o
f
r
e
mo
v
a
l
:
c
l
a
ssi
f
i
c
a
t
i
o
n
o
f
h
a
n
d
w
r
i
t
i
n
g
i
mp
a
i
r
m
e
n
t
u
s
i
n
g
C
N
N
f
o
r
p
o
t
e
n
t
i
a
l
d
y
sl
e
x
i
a
s
y
m
p
t
o
m,
”
i
n
2
0
2
1
1
1
t
h
I
EEE
I
n
t
e
r
n
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
C
o
n
t
r
o
l
S
y
st
e
m
,
C
o
m
p
u
t
i
n
g
a
n
d
En
g
i
n
e
e
ri
n
g
,
I
C
C
S
C
E
2
0
2
1
,
2
0
2
1
,
p
p
.
1
8
8
–
193
,
d
o
i
:
1
0
.
1
1
0
9
/
I
C
C
S
C
E5
2
1
8
9
.
2
0
2
1
.
9
5
3
0
9
8
9
.
[
7
]
I
.
S
.
I
sa,
M
.
A
.
Za
h
i
r
,
S
.
A
.
R
a
ml
a
n
,
W
.
L
.
-
C
h
i
h
,
a
n
d
S
.
N
.
S
u
l
a
i
m
a
n
,
“
C
N
N
c
o
m
p
a
r
i
so
n
s
mo
d
e
l
s
o
n
d
y
s
l
e
x
i
a
h
a
n
d
w
r
i
t
i
n
g
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
ES
T
E
EM
Ac
a
d
e
m
i
c
J
o
u
rn
a
l
,
v
o
l
.
1
7
,
p
p
.
1
2
–
2
5
,
2
0
2
1
.
[
8
]
I
.
S
.
I
sa,
W
.
N
.
S
.
R
a
h
i
mi
,
S
.
A
.
R
a
ml
a
n
,
a
n
d
S
.
N
.
S
u
l
a
i
ma
n
,
“
A
u
t
o
ma
t
e
d
d
e
t
e
c
t
i
o
n
o
f
d
y
s
l
e
x
i
a
s
y
m
p
t
o
m
b
a
s
e
d
o
n
h
a
n
d
w
r
i
t
i
n
g
i
ma
g
e
f
o
r
p
r
i
m
a
r
y
s
c
h
o
o
l
c
h
i
l
d
r
e
n
,
”
P
ro
c
e
d
i
a
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
1
6
3
,
p
p
.
4
4
0
–
4
4
9
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
r
o
c
s.2
0
1
9
.
1
2
.
1
2
7
.
[
9
]
P
.
J.
G
r
o
t
h
e
r
a
n
d
K
.
K
.
H
a
n
a
o
k
a
,
“
N
I
S
T
sp
e
c
i
a
l
d
a
t
a
b
a
s
e
1
9
,
”
N
a
t
i
o
n
a
l
I
n
st
i
t
u
t
e
o
f
S
t
a
n
d
a
rd
s a
n
d
T
e
c
h
n
o
l
o
g
y
.
A
c
c
e
ss
e
d
:
A
p
r
.
2
0
,
2
0
2
4
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
n
i
s
t
.
g
o
v
/
sr
d
/
n
i
st
-
s
p
e
c
i
a
l
-
d
a
t
a
b
a
se
-
19
[
1
0
]
J.
K
a
w
a
,
A
.
B
e
d
n
o
r
z
,
P
.
S
t
ę
p
i
e
ń
,
J.
D
e
r
e
j
c
z
y
k
,
a
n
d
M
.
B
u
g
d
o
l
,
“
S
p
a
t
i
a
l
a
n
d
d
y
n
a
mi
c
a
l
h
a
n
d
w
r
i
t
i
n
g
a
n
a
l
y
si
s
i
n
mi
l
d
c
o
g
n
i
t
i
v
e
i
mp
a
i
r
m
e
n
t
,
”
C
o
m
p
u
t
e
rs
i
n
B
i
o
l
o
g
y
a
n
d
M
e
d
i
c
i
n
e
,
v
o
l
.
8
2
,
p
p
.
2
1
–
2
8
,
M
a
r
.
2
0
1
7
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
b
i
o
me
d
.
2
0
1
7
.
0
1
.
0
0
4
.
[
1
1
]
K
.
S
p
o
o
n
,
K
.
S
i
e
k
,
D
.
C
r
a
n
d
a
l
l
,
a
n
d
M
.
F
i
l
l
m
o
r
e
,
“
C
a
n
w
e
(
a
n
d
s
h
o
u
l
d
w
e
)
u
s
e
A
I
t
o
d
e
t
e
c
t
d
y
sl
e
x
i
a
i
n
c
h
i
l
d
r
e
n
’
s
h
a
n
d
w
r
i
t
i
n
g
?
,”
i
n
Pro
c
e
e
d
i
n
g
s
o
f
Ar
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
f
o
r
S
o
c
i
a
l
G
o
o
d
(
N
e
u
rI
P
S
2
0
1
9
)
,
2
0
1
9
,
p
p
.
1
–
6.
[
1
2
]
S.
P
a
t
e
l
,
“
A
-
Z
h
a
n
d
w
r
i
t
t
e
n
a
l
p
h
a
b
e
t
s
i
n
.
c
sv
f
o
r
ma
t
,
”
K
a
g
g
l
e
.
A
c
c
e
s
se
d
:
A
p
r
.
2
0
,
2
0
2
4
.
[
O
n
l
i
n
e
]
.
A
v
a
i
l
a
b
l
e
:
h
t
t
p
s
:
/
/
w
w
w
.
k
a
g
g
l
e
.
c
o
m
/
sa
c
h
i
n
p
a
t
e
l
2
1
/
a
z
-
h
a
n
d
w
r
i
t
t
e
n
-
a
l
p
h
a
b
e
t
s
-
in
-
c
sv
-
f
o
r
mat
[
1
3
]
I
.
S
.
H
u
ssei
n
a
n
d
N
.
A
.
Jas
i
m,
“
I
r
i
s
r
e
c
o
g
n
i
t
i
o
n
sc
h
e
m
e
b
a
s
e
d
o
n
e
n
t
r
o
p
y
a
n
d
c
o
n
v
o
l
u
t
i
o
n
a
l
n
e
u
r
a
l
n
e
t
w
o
r
k
,
”
Ad
v
a
n
c
e
s
i
n
T
e
c
h
n
o
l
o
g
y
I
n
n
o
v
a
t
i
o
n
,
v
o
l
.
9
,
n
o
.
3
,
p
p
.
1
9
7
–
2
0
9
,
2
0
2
4
,
d
o
i
:
1
0
.
4
6
6
0
4
/
a
i
t
i
.
2
0
2
4
.
1
3
5
1
6
.
[
1
4
]
S
.
R
a
n
g
a
sr
i
n
i
v
a
sa
n
,
M
.
S
.
S
.
S
u
r
e
s
h
,
A
.
O
l
sz
e
w
sk
i
,
S
.
S
e
t
l
u
r
,
B
.
Ja
y
a
r
a
ma
n
,
a
n
d
V
.
G
o
v
i
n
d
a
r
a
j
u
,
“
A
I
-
e
n
h
a
n
c
e
d
c
h
i
l
d
h
a
n
d
w
r
i
t
i
n
g
a
n
a
l
y
si
s
:
a
f
r
a
mew
o
r
k
f
o
r
t
h
e
e
a
r
l
y
scre
e
n
i
n
g
o
f
d
y
sl
e
x
i
a
a
n
d
d
y
s
g
r
a
p
h
i
a
,
”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
6
,
n
o
.
5
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
2
9
7
9
-
025
-
0
3
9
2
7
-
0.
[
1
5
]
B
.
M
a
n
i
me
k
a
l
a
,
D
.
U
m
a
ma
h
e
sw
a
r
i
,
J
.
R
o
z
a
r
i
o
,
M
.
K
a
n
n
a
n
,
a
n
d
P
.
M
.
S
a
v
i
t
h
a
,
“
D
y
sg
r
a
p
h
i
a
d
i
so
r
d
e
r
d
e
t
e
c
t
i
o
n
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
u
si
n
g
d
e
e
p
l
e
a
r
n
i
n
g
t
e
c
h
n
i
q
u
e
,
”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
6
,
n
o
.
3
,
M
a
r
.
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
2
9
7
9
-
0
2
5
-
0
3
8
2
5
-
5.
[
1
6
]
A
.
F
a
l
l
a
h
,
Y
.
Z
.
V
a
k
i
l
i
,
a
n
d
H
.
S
a
j
e
d
i
,
“
A
I
-
d
r
i
v
e
n
a
p
p
r
o
a
c
h
e
s
f
o
r
d
y
s
g
r
a
p
h
i
a
d
i
a
g
n
o
si
s
u
si
n
g
o
n
l
i
n
e
a
n
d
o
f
f
l
i
n
e
h
a
n
d
w
r
i
t
i
n
g
d
a
t
a
:
a
c
o
m
p
r
e
h
e
n
si
v
e
sc
o
p
i
n
g
r
e
v
i
e
w
,
”
PL
O
S
O
n
e
,
v
o
l
.
2
0
,
n
o
.
1
2
,
D
e
c
.
2
0
2
5
,
d
o
i
:
1
0
.
1
3
7
1
/
j
o
u
r
n
a
l
.
p
o
n
e
.
0
3
2
8
7
2
2
.
[
1
7
]
L.
C
.
R
i
b
e
i
r
o
,
L
.
C
.
A
f
o
n
s
o
,
a
n
d
J.
P
.
P
a
p
a
,
“
B
a
g
o
f
sam
p
l
i
n
g
s
f
o
r
c
o
m
p
u
t
e
r
-
a
ss
i
st
e
d
P
a
r
k
i
n
s
o
n
’
s
d
i
s
e
a
s
e
d
i
a
g
n
o
s
i
s
b
a
se
d
o
n
r
e
c
u
r
r
e
n
t
n
e
u
r
a
l
n
e
t
w
o
r
k
s,
”
C
o
m
p
u
t
e
r
s i
n
B
i
o
l
o
g
y
a
n
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
1
5
,
2
0
1
9
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
b
i
o
me
d
.
2
0
1
9
.
1
0
3
4
7
7
.
[
1
8
]
M
.
F
.
-
Z
a
n
u
y
,
J
.
F
i
e
r
r
e
z
,
M
.
A
.
F
e
r
r
e
r
,
M
.
D
i
a
z
,
R
.
T
o
l
o
s
a
n
a
,
a
n
d
R
.
P
l
a
m
o
n
d
o
n
,
“
H
a
n
d
w
r
i
t
i
n
g
b
i
o
m
e
t
r
i
c
s
:
a
p
p
l
i
c
a
t
i
o
n
s
a
n
d
f
u
t
u
r
e
t
r
e
n
d
s
i
n
e
-
se
c
u
r
i
t
y
a
n
d
e
-
h
e
a
l
t
h
,
”
C
o
g
n
i
t
i
v
e
C
o
m
p
u
t
a
t
i
o
n
,
v
o
l
.
1
2
,
n
o
.
5
,
p
p
.
9
4
0
–
9
5
3
,
2
0
2
0
,
d
o
i
:
1
0
.
1
0
0
7
/
s
1
2
5
5
9
-
020
-
09755
-
z.
[
1
9
]
J.
C
h
a
i
e
t
a
l
.
,
“
C
l
a
ss
i
f
i
c
a
t
i
o
n
o
f
m
i
l
d
c
o
g
n
i
t
i
v
e
i
mp
a
i
r
me
n
t
b
a
s
e
d
o
n
h
a
n
d
w
r
i
t
i
n
g
d
y
n
a
m
i
c
s
a
n
d
q
EEG
,
”
C
o
m
p
u
t
e
rs
i
n
B
i
o
l
o
g
y
a
n
d
Me
d
i
c
i
n
e
,
v
o
l
.
1
5
2
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
c
o
m
p
b
i
o
me
d
.
2
0
2
2
.
1
0
6
4
1
8
.
[
2
0
]
A
.
D
e
v
i
a
n
d
G
.
K
a
v
y
a
,
“
D
y
s
g
r
a
p
h
i
a
d
i
s
o
r
d
e
r
f
o
r
e
c
a
st
i
n
g
a
n
d
c
l
a
ssi
f
i
c
a
t
i
o
n
t
e
c
h
n
i
q
u
e
u
si
n
g
i
n
t
e
l
l
i
g
e
n
t
d
e
e
p
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
e
s
,
”
Pro
g
ress
i
n
N
e
u
ro
-
Ps
y
c
h
o
p
h
a
rm
a
c
o
l
o
g
y
a
n
d
Bi
o
l
o
g
i
c
a
l
Ps
y
c
h
i
a
t
r
y
,
v
o
l
.
1
2
0
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
p
n
p
b
p
.
2
0
2
2
.
1
1
0
6
4
7
.
[
2
1
]
J.
K
u
n
h
o
t
h
,
S
.
A
.
M
a
a
d
e
e
d
,
M
.
S
a
l
e
h
,
a
n
d
Y
.
A
k
b
a
r
i
,
“
C
N
N
f
e
a
t
u
r
e
a
n
d
c
l
a
ssi
f
i
e
r
f
u
si
o
n
o
n
n
o
v
e
l
t
r
a
n
sf
o
r
m
e
d
i
mag
e
d
a
t
a
s
e
t
f
o
r
d
y
s
g
r
a
p
h
i
a
d
i
a
g
n
o
s
i
s
i
n
c
h
i
l
d
r
e
n
,
”
Ex
p
e
r
t
S
y
s
t
e
m
s
w
i
t
h
A
p
p
l
i
c
a
t
i
o
n
s
,
v
o
l
.
2
3
1
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
1
6
/
j
.
e
sw
a
.
2
0
2
3
.
1
2
0
7
4
0
.
[
2
2
]
S
.
M
.
S
.
I
sl
a
m,
E.
K
.
D
e
y
,
M
.
N
.
A
.
Ta
w
h
i
d
,
a
n
d
B
.
M
.
M
.
H
o
ss
a
i
n
,
“
A
C
N
N
b
a
s
e
d
a
p
p
r
o
a
c
h
f
o
r
g
a
r
me
n
t
s
t
e
x
t
u
r
e
d
e
si
g
n
c
l
a
ss
i
f
i
c
a
t
i
o
n
,
”
Ad
v
a
n
c
e
s
i
n
T
e
c
h
n
o
l
o
g
y
I
n
n
o
v
a
t
i
o
n
,
v
o
l
.
2
,
n
o
.
4
,
p
p
.
1
1
9
–
1
2
5
,
2
0
1
7
.
[
2
3
]
R
.
S
h
a
r
m
a
a
n
d
A
.
S
i
n
g
h
,
“
A
n
i
n
t
e
g
r
a
t
e
d
a
p
p
r
o
a
c
h
t
o
w
a
r
d
s
e
f
f
i
c
i
e
n
t
i
ma
g
e
c
l
a
ssi
f
i
c
a
t
i
o
n
u
s
i
n
g
d
e
e
p
C
N
N
w
i
t
h
t
r
a
n
sf
e
r
l
e
a
r
n
i
n
g
a
n
d
P
C
A
,
”
Ad
v
a
n
c
e
s
i
n
T
e
c
h
n
o
l
o
g
y
I
n
n
o
v
a
t
i
o
n
,
v
o
l
.
7
,
n
o
.
2
,
p
p
.
1
0
5
–
1
1
7
,
2
0
2
2
,
d
o
i
:
1
0
.
4
6
6
0
4
/
a
i
t
i
.
2
0
2
2
.
8
5
3
8
.
[
2
4
]
P
.
S
h
a
r
ma,
B
.
A
g
a
r
w
a
l
,
G
.
S
.
Y
a
d
a
v
,
a
n
d
S
.
Ja
i
n
,
“
V
i
si
o
n
t
r
a
n
sf
o
r
m
e
r
-
b
a
s
e
d
m
o
d
e
l
f
o
r
e
a
r
l
y
d
e
t
e
c
t
i
o
n
o
f
d
y
sg
r
a
p
h
i
a
a
m
o
n
g
sch
o
o
l
s
t
u
d
e
n
t
s,”
M
i
c
r
o
sys
t
e
m
T
e
c
h
n
o
l
o
g
i
e
s
,
v
o
l
.
3
1
,
n
o
.
3
,
p
p
.
7
7
5
–
7
8
5
,
2
0
2
5
,
d
o
i
:
1
0
.
1
0
0
7
/
s
0
0
5
4
2
-
0
2
4
-
0
5
7
4
1
-
9.
[
2
5
]
B
.
A
g
a
r
w
a
l
,
S
.
J
a
i
n
,
K
.
B
e
l
a
d
i
y
a
,
Y
.
G
u
p
t
a
,
A
.
S
.
Y
a
d
a
v
,
a
n
d
N
.
J.
A
h
u
j
a
,
“
E
a
r
l
y
a
n
d
a
u
t
o
ma
t
e
d
d
i
a
g
n
o
si
s
o
f
d
y
sg
r
a
p
h
i
a
u
si
n
g
mac
h
i
n
e
l
e
a
r
n
i
n
g
a
p
p
r
o
a
c
h
,”
S
N
C
o
m
p
u
t
e
r
S
c
i
e
n
c
e
,
v
o
l
.
4
,
n
o
.
5
,
2
0
2
3
,
d
o
i
:
1
0
.
1
0
0
7
/
s
4
2
9
7
9
-
0
2
3
-
0
1
8
8
4
-
0
.
B
I
O
G
RAP
H
I
E
S O
F
AUTH
O
RS
Nisha
Va
n
ja
r
i
is
a
ss
istan
t
p
ro
fe
ss
o
r
in
De
p
a
rtme
n
t
o
f
C
o
m
p
u
ter
En
g
in
e
e
rin
g
,
KJ
S
IT,
M
u
m
b
a
i,
sin
c
e
Ja
n
u
a
ry
2
0
0
9
.
S
h
e
is
p
u
rsu
i
n
g
h
e
r
P
h
.
D.
i
n
Co
m
p
u
ter
E
n
g
i
n
e
e
rin
g
fro
m
KJ
S
S
E.
S
h
e
is
M
.
E.
i
n
I
n
fo
r
m
a
ti
o
n
Tec
h
n
o
lo
g
y
,
P
.
G.
D.
B.
M
.
fr
o
m
NMI
M
S
De
e
m
e
d
Un
iv
e
rsity
,
B
.
E.
in
C
o
m
p
u
ter
En
g
in
e
e
rin
g
f
ro
m
N
o
rth
M
a
h
a
ra
sh
t
ra
Un
iv
e
rsit
y
,
a
n
d
Di
p
lo
m
a
in
Co
m
p
u
ter
Tec
h
n
o
l
o
g
y
fro
m
BTE
.
S
h
e
h
a
s
d
o
n
e
‘
so
ftwa
re
q
u
a
l
it
y
t
e
stin
g
’
c
e
rti
fica
ti
o
n
–
CS
T
fr
o
m
S
QTL.
S
h
e
h
a
s
p
u
b
li
sh
e
d
m
o
re
t
h
a
n
2
0
p
a
p
e
rs
in
c
o
n
fe
re
n
c
e
s,
b
o
o
k
c
h
a
p
ters
,
jo
u
rn
a
ls.
F
o
r
t
h
e
las
t
fiv
e
y
e
a
rs,
h
e
r
m
a
in
re
se
a
r
c
h
wa
s
in
th
e
field
o
f
n
e
two
r
k
i
n
g
,
d
a
ta
sc
ien
c
e
a
n
d
a
rti
ficia
l
in
telli
g
e
n
c
e
.
Th
is
h
a
s
led
h
e
r
in
te
re
st
in
th
e
d
e
e
p
lea
rn
in
g
a
n
d
c
lo
u
d
c
o
m
p
u
ti
n
g
a
s
we
ll
.
S
h
e
h
a
s
p
u
b
li
sh
e
d
2
b
o
o
k
s
f
o
r
Dista
n
c
e
a
n
d
Op
e
n
Lea
rn
i
n
g
Co
u
rse
M
CA.
S
h
e
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
n
v
a
n
jari@s
o
m
a
i
y
a
.
e
d
u
.
Pra
sa
n
n
a
J
.
S
h
e
te
is
a
ss
o
c
iate
p
ro
fe
ss
o
r
in
C
o
m
p
u
ter
En
g
in
e
e
rin
g
,
KJ
S
S
E
.
He
is
P
h
.
D.
(S
c
ien
c
e
a
n
d
Tec
h
n
o
l
o
g
y
)
V.
J.
T.
I.
,
Un
iv
e
rsity
o
f
M
u
m
b
a
i,
M
.
Tec
h
.
(El
e
c
tro
n
ics
a
n
d
Tele
c
o
m
m
u
n
ica
ti
o
n
),
V.
J.
T
.
I.
,
Un
iv
e
rsity
o
f
M
u
m
b
a
i
a
n
d
B.
E
.
(
El
e
c
tro
n
ics
En
g
in
e
e
rin
g
),
Dr.
B.
A.
M
a
ra
th
wa
d
a
Un
iv
e
rsit
y
,
A
u
ra
n
g
a
b
a
d
.
He
is
wo
rk
in
g
i
n
K.
J.
S
o
m
a
iy
a
S
c
h
o
o
l
o
f
En
g
i
n
e
e
rin
g
sin
c
e
Ju
n
e
2
0
0
3
.
He
is
t
h
e
c
o
n
v
e
n
e
r
o
f
t
h
e
Eq
u
a
l
Op
p
o
rt
u
n
it
ies
Ce
ll
,
S
o
m
a
iy
a
Vi
d
y
a
v
i
h
a
r
Un
iv
e
rsity
,
c
o
o
r
d
i
n
a
to
r,
KJ
S
S
E
r
e
so
u
rc
e
c
e
n
ter
fo
r
p
wd
stu
d
e
n
ts,
F
o
rm
e
r
in
-
c
h
a
rg
e
,
KJ
S
S
E
d
a
ta
c
e
n
ter
.
He
h
a
s
p
u
b
li
sh
e
d
m
o
re
th
a
n
2
5
p
a
p
e
rs
in
c
o
n
fe
re
n
c
e
s,
b
o
o
k
c
h
a
p
ter,
a
n
d
jo
u
rn
a
ls.
His
re
se
a
rc
h
sp
e
c
ializa
ti
o
n
in
m
o
b
i
le
c
o
m
p
u
ti
n
g
,
Io
T
,
e
m
e
rg
in
g
wire
le
ss
n
e
two
rk
s
,
WS
Ns
,
M
AN
ET
s,
VA
NETs,
a
c
c
e
ss
ib
il
it
y
,
a
n
d
a
ss
isti
v
e
tec
h
n
o
lo
g
ies
.
He
c
a
n
b
e
c
o
n
tac
ted
a
t
e
m
a
il
:
p
ra
sa
n
n
a
sh
e
te@
so
m
a
iy
a
.
e
d
u
.
Evaluation Warning : The document was created with Spire.PDF for Python.