I
nte
rna
t
io
na
l J
o
urna
l o
f
E
lect
rica
l a
nd
Co
m
p
ute
r
E
ng
in
ee
ring
(
I
J
E
CE
)
Vo
l.
6
,
No
.
6
,
Dec
em
b
er
201
6
,
p
p
.
31
3
1
~
31
41
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
6i
6
.
1
2
9
6
7
3131
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
jo
u
r
n
a
l.c
o
m/o
n
lin
e/in
d
ex
.
p
h
p
/I
JE
C
E
EE
G
Ba
sed Ey
e S
tate C
la
ss
ifica
tion usin
g
Deep
Belie
f
Net
w
o
rk
a
nd Stac
k
ed
Aut
o
Enco
der
Sa
na
m
Na
re
j
o
1
,
E
ro
s
P
a
s
er
o
2
,
F
a
rz
a
na
K
uls
o
o
m
3
1,
2
De
p
a
rt
m
e
n
t
o
f
El
e
c
tro
n
ics
&
Tele
c
o
m
m
u
n
ica
ti
o
n
s,
P
o
li
tec
n
ico
Di
T
o
rin
o
,
Italy
3
De
p
a
rt
m
e
n
t
o
f
El
e
c
tri
c
a
l,
Co
m
p
u
ter &
El
e
c
tro
n
ics
,
Un
iv
e
rsity
o
f
P
a
v
ia,
Italy
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Au
g
1
4
,
2
0
1
6
R
ev
i
s
ed
Oct
17
,
2
0
1
6
A
cc
ep
ted
No
v
1
,
2
0
1
6
A
Br
a
in
-
Co
m
p
u
ter
In
terf
a
c
e
(B
CI)
p
ro
v
id
e
s
a
n
a
lt
e
rn
a
ti
v
e
c
o
m
m
u
n
ica
ti
o
n
in
terf
a
c
e
b
e
t
we
e
n
th
e
h
u
m
a
n
b
ra
in
a
n
d
a
c
o
m
p
u
ter
.
T
h
e
El
e
c
tro
e
n
c
e
p
h
a
lo
g
ra
m
(EE
G
)
si
g
n
a
ls
a
re
a
c
q
u
ired
,
p
r
o
c
e
ss
e
d
a
n
d
m
a
c
h
in
e
lea
rn
in
g
a
lg
o
rit
h
m
s
a
r
e
f
u
rth
e
r
a
p
p
li
e
d
to
e
x
trac
t
u
se
f
u
l
in
f
o
r
m
a
ti
o
n
.
Du
r
in
g
EE
G
a
c
q
u
isit
io
n
,
a
rti
f
a
c
ts
a
r
e
in
d
u
c
e
d
d
u
e
to
i
n
v
o
lu
n
tary
e
y
e
m
o
v
e
m
e
n
ts
o
r
e
y
e
b
li
n
k
,
c
a
stin
g
a
d
v
e
rse
e
ff
e
c
t
s
o
n
sy
ste
m
p
e
r
f
o
r
m
a
n
c
e
.
T
h
e
a
im
o
f
th
is
re
se
a
rc
h
is
to
p
re
d
ict
e
y
e
sta
t
e
s
f
ro
m
EE
G
sig
n
a
ls
u
sin
g
De
e
p
lea
rn
in
g
a
rc
h
it
e
c
tu
re
s
a
n
d
p
re
se
n
t
im
p
ro
v
e
d
c
las
sif
ier
m
o
d
e
ls.
Re
c
e
n
t
stu
d
ies
re
f
le
c
t
th
a
t
De
e
p
Ne
u
ra
l
Ne
t
w
o
rk
s
a
re
tren
d
i
n
g
sta
te
o
f
th
e
a
rt
M
a
c
h
in
e
lea
rn
in
g
a
p
p
ro
a
c
h
e
s
.
T
h
e
re
f
o
re
,
t
h
e
c
u
rre
n
t
w
o
rk
p
re
se
n
ts
th
e
i
m
p
le
m
e
n
tatio
n
o
f
De
e
p
Be
li
e
f
Ne
t
w
o
rk
(DBN
)
a
n
d
S
tac
k
e
d
A
u
to
En
c
o
d
e
rs
(S
A
E)
a
s
Clas
s
if
iers
w
it
h
e
n
c
o
u
ra
g
in
g
p
e
rf
o
rm
a
n
c
e
a
c
c
u
ra
c
y
.
On
e
o
f
th
e
d
e
sig
n
e
d
S
AE
m
o
d
e
ls
o
u
t
p
e
rf
o
rm
s
th
e
p
e
r
f
o
r
m
a
n
c
e
o
f
DBN
a
n
d
th
e
m
o
d
e
ls
p
re
se
n
ted
in
e
x
isti
n
g
re
se
a
rc
h
b
y
a
n
im
p
re
s
siv
e
e
rro
r
ra
te
o
f
1
.
1
%
o
n
th
e
tes
t
se
t
b
e
a
rin
g
a
c
c
u
ra
c
y
o
f
9
8
.
9
%
.
T
h
e
f
in
d
in
g
s
i
n
t
h
is
st
u
d
y
,
m
a
y
p
ro
v
id
e
a
c
o
n
tri
b
u
ti
o
n
t
o
w
a
rd
s
th
e
sta
te
o
f
th
e
a
rt
p
e
r
f
o
r
m
a
n
c
e
o
n
th
e
p
ro
b
lem
o
f
EE
G
b
a
s
e
d
e
y
e
sta
t
e
c
las
si
f
ica
ti
o
n
.
K
ey
w
o
r
d
:
B
C
I
Dee
p
b
elief
n
et
w
o
r
k
s
Dee
p
lear
n
in
g
ar
ch
itect
u
r
es
E
lectr
o
en
ce
p
h
alo
g
r
a
m
E
y
e
s
tate
cla
s
s
i
f
icatio
n
Stack
ed
au
to
e
n
co
d
er
s
Co
p
y
rig
h
t
©
201
6
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
San
a
m
Nar
ej
o
,
Dep
ar
t
m
en
t o
f
E
lectr
o
n
ics a
n
d
T
elec
o
m
m
u
n
icatio
n
,
P
o
litecn
ico
Di
T
o
r
in
o
,
I
taly
.
E
m
ail: sa
n
a
m
.
n
ar
ej
o
@
p
o
lito
.
it
1.
I
NT
RO
D
UCT
I
O
N
B
r
ain
-
C
o
m
p
u
ter
I
n
ter
f
ac
e
(
B
C
I
)
is
o
n
e
o
f
th
e
e
m
er
g
i
n
g
f
iel
d
s
in
h
u
m
an
co
m
p
u
ter
in
ter
ac
t
io
n
(
HC
I
)
.
I
t
h
as
a
b
r
o
ad
s
p
ec
tr
u
m
o
f
s
o
licitu
d
e;
i
n
clu
d
i
n
g
s
a
f
et
y
-
cr
iti
ca
l,
s
ec
u
r
it
y
s
u
r
v
eilla
n
ce
,
in
d
u
s
tr
ial
a
n
d
m
ed
ical
ap
p
licatio
n
s
.
I
t
h
a
s
p
la
y
ed
a
p
iv
o
tal
r
o
le
i
n
m
ed
ical
ap
p
licat
io
n
s
,
f
o
r
in
s
ta
n
ce
,
p
atie
n
ts
w
i
t
h
m
o
to
r
d
is
ab
ilit
ie
s
ca
n
alle
v
iate
co
m
m
u
n
ica
tio
n
ab
ilit
ies
w
i
th
af
o
r
esaid
.
B
C
I
ac
h
iev
e
s
t
h
i
s
g
o
al
b
y
estab
li
s
h
in
g
a
co
n
n
ec
tio
n
p
ath
b
et
w
ee
n
s
ti
m
u
lated
b
r
ain
an
d
e
x
ter
n
a
l
d
ev
ice.
B
C
I
s
ar
e
f
r
eq
u
en
tl
y
ai
m
ed
f
o
r
r
esear
ch
,
m
ap
p
in
g
o
r
r
ep
air
in
g
h
u
m
a
n
co
g
n
i
tiv
e
o
r
s
en
s
o
r
y
-
m
o
to
r
f
u
n
ctio
n
s
.
I
n
1
9
2
4
B
er
g
er
w
as
t
h
e
f
ir
s
t
o
n
e
to
d
is
co
v
er
t
h
e
elec
tr
ic
ac
tiv
it
y
o
f
t
h
e
h
u
m
a
n
b
r
ain
.
T
h
e
elec
tr
ic
al
ac
tiv
it
y
o
f
t
h
e
b
r
ain
is
r
ec
o
r
d
ed
th
r
o
u
g
h
E
lect
r
o
en
ce
p
h
alo
g
r
ap
h
y
(
E
E
G)
in
t
h
e
f
o
r
m
a
s
ig
n
al.
I
t
ca
n
b
e
a
ls
o
r
eg
ar
d
ed
as
t
h
e
r
ec
o
r
d
in
g
o
f
th
e
b
r
ain
's
s
p
o
n
ta
n
eo
u
s
elec
tr
ical
ac
ti
v
i
t
y
o
v
er
a
p
er
io
d
o
f
ti
m
e.
Du
r
i
n
g
ac
q
u
i
s
itio
n
,
s
i
g
n
als
ar
e
o
b
tain
ed
b
y
ap
p
l
y
i
n
g
s
ev
er
al
elec
tr
o
d
es
o
v
er
s
ca
lp
s
u
r
f
ac
e.
Ho
w
e
v
er
t
h
e
p
o
s
itio
n
an
d
n
u
m
b
er
o
f
e
lec
tr
o
d
es
ar
e
a
p
p
licatio
n
an
d
g
o
al
d
ep
en
d
en
t.
T
h
is
elec
tr
ical
ac
tiv
it
y
is
th
e
n
r
ec
o
r
d
e
d
in
a
f
o
r
m
o
f
s
i
g
n
al.
T
h
e
s
ig
n
al
o
b
tai
n
ed
f
r
o
m
E
E
G
elec
tr
o
d
es
is
p
r
o
ce
s
s
ed
in
s
ep
ar
ate
ch
a
n
n
e
l
s
u
b
s
eq
u
en
t
l
y
a
m
p
li
f
ied
.
O
n
e
m
a
y
al
ter
n
ati
v
el
y
u
s
e
th
e
ter
m
ch
an
n
el
o
r
e
lectr
o
d
e
[
1
]
.
A
P
r
in
cip
al
asp
ec
t
in
t
h
e
in
v
esti
g
a
tio
n
o
f
E
E
G
s
i
g
n
a
l
is
,
th
e
a
m
o
u
n
t
o
f
in
f
o
r
m
atio
n
th
at
ca
n
b
e
ac
q
u
ir
ed
f
r
o
m
th
e
r
ea
l
E
E
G
r
ec
o
r
d
in
g
s
.
As
s
tated
ea
r
lier
th
at
i
n
1
9
2
0
s
,
th
e
s
en
s
iti
v
it
y
o
f
E
E
G
to
v
ar
io
u
s
ch
an
g
es
o
f
h
u
m
a
n
b
r
ain
’
s
f
u
n
ctio
n
al
s
tat
e
w
as
d
e
m
o
n
s
tr
a
ted
.
I
t
w
as
al
s
o
f
o
u
n
d
in
th
e
ab
o
v
e
m
e
n
tio
n
ed
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
6
,
No
.
6
,
Dec
em
b
er
2
0
1
6
:
31
3
1
–
31
41
3132
s
tu
d
ie
s
th
at
a
s
i
m
p
le
ac
tio
n
o
f
clo
s
in
g
t
h
e
e
y
e
s
,
g
i
v
es
r
i
s
e
t
o
r
eg
u
lar
o
s
cillat
io
n
s
i
n
t
h
e
E
E
G
w
a
v
e.
W
h
er
ea
s
d
u
r
in
g
th
e
co
n
d
u
ct
o
f
a
n
y
m
e
n
tal
ac
tiv
it
y
t
h
ese
o
s
cil
latio
n
s
b
ec
o
m
e
f
as
ter
an
d
less
r
e
g
u
lar
.
A
s
i
g
n
i
f
ica
n
t
c
h
alle
n
g
e
to
an
E
E
G
b
ased
s
y
s
te
m
,
i
s
t
h
e
in
ter
f
er
en
ce
o
f
ar
t
if
ac
t
s
i
n
t
h
e
s
i
g
n
als.
A
r
ti
f
ac
t
s
ar
e
h
i
g
h
f
r
eq
u
en
c
y
s
ig
n
al
s
b
elo
n
g
to
n
o
n
-
ce
r
eb
r
a
l
o
r
ig
in
an
d
ca
n
d
r
a
m
atica
ll
y
alter
th
e
r
ec
o
r
d
ed
s
ig
n
al.
T
h
e
ar
tif
ac
ts
ar
e
d
iv
id
ed
in
to
t
w
o
g
r
o
u
p
s
as
i
n
ter
n
a
l
an
d
ex
ter
n
al.
T
h
e
ex
ter
n
al
ar
t
if
ac
ts
ar
e
g
e
n
er
ated
f
r
o
m
t
h
e
en
v
ir
o
n
m
e
n
t
o
r
p
o
w
er
eq
u
ip
m
en
t
s
.
T
h
e
in
ter
n
a
l
ar
tif
ac
ts
ar
e
e
y
e
b
l
in
k
s
,
e
y
e
m
o
v
e
m
e
n
ts
,
m
u
s
cle
an
d
r
esp
ir
ato
r
y
ar
ti
f
ac
t
s
[
2
-
3
]
.
Du
r
in
g
t
h
e
E
E
G
ex
p
er
i
m
en
tal
p
r
o
ce
d
u
r
es,
th
e
s
u
b
j
ec
ts
ca
n
n
o
t
co
n
tr
o
l
s
p
o
n
tan
eo
u
s
e
y
e
m
o
v
e
m
e
n
t
s
o
r
b
lin
k
s
[
4
]
.
T
h
ese
ar
tif
ac
ts
ar
e
al
m
o
s
t
in
e
v
itab
le,
m
a
y
s
e
r
io
u
s
l
y
d
i
s
to
r
t
b
r
ain
ac
tiv
it
y
.
T
h
er
ef
o
r
e,
th
ese
o
cc
u
r
r
en
ce
s
estab
lis
h
t
h
e
p
r
o
m
in
e
n
ce
o
f
r
esear
ch
o
n
E
E
G,
e
y
e
s
ta
te
s
ig
n
al
a
n
al
y
s
is
.
R
ec
en
tl
y
,
t
h
e
a
r
ea
o
f
d
ee
p
lear
n
in
g
is
at
tr
ac
tin
g
w
id
esp
r
ea
d
in
ter
est
b
y
p
r
o
d
u
cin
g
r
e
m
ar
k
ab
le
r
esear
ch
i
n
al
m
o
s
t
e
v
er
y
asp
e
ct
o
f
ar
ti
f
icial
in
telli
g
en
ce
.
A
p
ar
t
f
r
o
m
ac
h
ie
v
i
n
g
e
m
p
ir
i
ca
l
s
u
cc
ess
in
t
h
e
en
o
r
m
o
u
s
n
u
m
b
er
o
f
p
r
ac
tical
ap
p
licatio
n
s
,
it
h
a
s
p
r
o
v
id
ed
s
tate
o
f
t
h
e
ar
t
p
er
f
o
r
m
a
n
ce
i
n
n
at
u
r
al
la
n
g
u
a
g
e
p
r
o
ce
s
s
in
g
,
s
p
ee
ch
r
ec
o
g
n
itio
n
,
o
b
j
ec
t
r
ec
o
g
n
itio
n
a
n
d
m
a
n
y
o
t
h
er
d
o
m
ai
n
s
[
5
]
.
Dee
p
lear
n
in
g
h
as
b
ec
o
m
e
o
n
e
o
f
t
h
e
s
i
g
n
if
ica
n
t p
ar
ts
o
f
th
e
m
ac
h
i
n
e
lear
n
in
g
f
a
m
il
y
.
I
t is b
ased
o
n
t
h
e
s
et
o
f
alg
o
r
it
h
m
s
t
h
at
atte
m
p
t
s
t
o
lear
n
h
ier
ar
ch
ica
l,
n
o
n
li
n
ea
r
r
ep
r
esen
tatio
n
s
o
f
d
ata.
I
n
a
b
r
o
ad
er
asp
ec
t
th
is
ap
p
r
o
ac
h
ca
n
also
b
e
ter
m
ed
as
R
ep
r
esen
tat
io
n
lear
n
i
n
g
.
L
ea
r
n
t
R
ep
r
ese
n
tatio
n
s
o
f
te
n
r
e
s
u
lt
s
i
n
m
u
ch
b
etter
p
er
f
o
r
m
an
ce
t
h
an
ca
n
b
e
o
b
tain
ed
w
it
h
h
a
n
d
-
d
esig
n
ed
o
r
h
an
d
-
en
g
i
n
ee
r
ed
r
ep
r
esen
tatio
n
s
f
o
r
i
n
s
ta
n
ce
m
a
th
e
m
atica
l
o
r
s
tati
s
tical
ca
lcu
latio
n
s
[
6
]
.
A
lt
h
o
u
g
h
al
l
th
e
d
ee
p
lear
n
i
n
g
ap
p
r
o
ac
h
es,
s
h
ar
e
th
e
id
ea
o
f
n
este
d
r
ep
r
esen
tatio
n
o
f
d
ata
[
7
]
.
On
th
e
o
th
er
h
an
d
,
it
is
n
o
t
al
w
a
y
s
tr
u
e
th
at
d
ee
p
lear
n
in
g
ar
ch
itect
u
r
es
m
a
y
p
er
f
o
r
m
b
etter
th
a
n
th
e
s
h
allo
w
o
n
es.
Dee
p
er
ar
ch
ite
ctu
r
es
m
a
y
lead
,
w
h
e
n
t
h
er
e
is
s
u
f
f
icie
n
t
a
m
o
u
n
t
o
f
d
ata
to
ca
p
tu
r
e
th
e
p
atter
n
s
an
d
th
e
tas
k
is
co
m
p
lex
e
n
o
u
g
h
to
b
e
lear
n
t th
r
o
u
g
h
h
ier
ar
ch
ical
m
u
lti
-
le
v
el
n
o
n
-
lin
ea
r
tr
a
n
s
f
o
r
m
a
tio
n
s
.
I
n
th
i
s
er
a,
en
o
r
m
o
u
s
r
esear
c
h
g
o
i
n
g
o
n
w
it
h
t
h
e
ex
p
er
ien
t
ial
an
d
th
eo
r
etica
l
f
i
n
d
in
g
s
h
as
b
r
o
u
g
h
t
Dee
p
L
ea
r
n
i
n
g
A
r
c
h
itect
u
r
es
(
DL
A
s
)
to
th
e
at
ten
t
io
n
o
f
m
ac
h
in
e
lear
n
i
n
g
co
m
m
u
n
it
y
.
T
h
is
p
ap
er
p
r
esen
ts
th
e
u
s
e
o
f
d
ee
p
lear
n
i
n
g
tec
h
n
iq
u
es
to
cla
s
s
i
f
y
t
h
e
s
tate
o
f
t
h
e
e
y
e
f
r
o
m
E
E
G
s
ig
n
al
s
.
T
h
is
r
e
s
u
l
ted
a
s
i
m
p
r
o
v
ed
ac
cu
r
ac
y
o
f
p
r
ed
icti
n
g
m
o
d
el
o
v
er
t
h
e
co
n
v
e
n
tio
n
al
m
ac
h
i
n
e
lear
n
i
n
g
m
eth
o
d
s
.
T
h
e
f
o
cu
s
o
f
o
u
r
r
esear
ch
is
t
o
i
m
p
le
m
e
n
t
d
e
ep
lear
n
in
g
ar
ch
itect
u
r
es
a
s
class
i
f
icatio
n
m
o
d
els;
s
p
ec
if
i
ca
ll
y
Dee
p
B
elief
Net
w
o
r
k
(
DB
N)
an
d
s
tack
ed
A
u
to
E
n
co
d
er
(
SA
E
)
an
d
p
r
o
v
id
e
co
m
p
ar
ativ
e
a
n
al
y
s
i
s
o
n
th
e
b
eh
a
v
io
r
an
d
p
er
f
o
r
m
a
n
ce
o
f
S
A
E
a
n
d
DB
F.
Su
b
s
eq
u
en
t
l
y
,
c
o
m
p
ar
i
n
g
t
h
e
o
b
tain
ed
r
esu
lts
w
i
th
co
n
v
en
t
io
n
al
Ma
c
h
i
n
e
lear
n
in
g
m
o
d
el
s
o
f
ea
r
lier
s
t
u
d
ies.
Dee
p
lear
n
i
n
g
is
ac
h
i
ev
in
g
s
tate
-
of
-
t
h
e
-
ar
t
r
es
u
lt
s
ac
r
o
s
s
a
r
an
g
e
o
f
d
if
f
ic
u
lt
p
r
o
b
le
m
d
o
m
ai
n
s
.
T
o
th
e
b
est
o
f
o
u
r
k
n
o
w
led
g
e,
th
is
is
th
e
f
ir
s
t
ap
p
licatio
n
o
f
E
y
e
s
tate
p
r
ed
ictio
n
o
f
E
E
G
s
ig
n
al
s
u
s
in
g
d
ee
p
ar
ch
itect
u
r
es.
T
h
e
r
est
o
f
th
e
m
an
u
s
cr
ip
t
is
s
t
r
u
ct
u
r
ed
as
f
o
llo
w
s
.
Sect
i
o
n
2
p
r
o
v
id
es
th
e
in
f
o
r
m
at
io
n
o
n
th
e
r
elate
d
w
o
r
k
f
o
r
e
y
e
s
ta
te
cl
ass
i
f
icatio
n
,
p
r
ed
ic
tio
n
o
r
id
en
ti
f
icatio
n
v
ia
E
E
G
s
ig
n
als.
Sectio
n
3
p
r
o
v
id
es
th
e
b
r
ief
o
v
er
v
ie
w
o
f
DL
As
i
m
p
le
m
en
ted
i
n
t
h
is
r
esear
c
h
w
o
r
k
.
T
h
e
d
etails
o
f
t
h
e
r
es
ea
r
ch
m
et
h
o
d
o
lo
g
y
f
o
llo
w
ed
is
e
x
p
lai
n
ed
in
Secti
o
n
4
.
Sectio
n
5
p
r
o
v
id
es
t
h
e
v
is
io
n
o
n
r
esu
lts
a
n
d
d
is
c
u
s
s
i
o
n
.
Sectio
n
6
d
ea
ls
w
it
h
t
h
e
co
n
cl
u
s
io
n
an
d
f
u
t
u
r
e
w
o
r
k
.
2.
RE
L
AT
E
D
WO
RK
E
y
e
s
tate
cla
s
s
i
f
icat
io
n
is
a
k
in
d
o
f
co
m
m
o
n
ti
m
e
s
er
ies
p
r
o
b
lem
f
o
r
d
etec
ti
n
g
h
u
m
a
n
co
g
n
iti
v
e
s
tate,
w
h
ic
h
ar
e
n
o
t
o
n
l
y
cr
u
c
ial
to
m
ed
ical
ca
r
e
b
u
t
also
s
i
g
n
i
f
ica
n
t
f
o
r
d
aily
li
f
e
ch
o
r
es.
T
h
er
e
ar
e
v
ar
io
u
s
ap
p
licatio
n
ar
ea
s
r
elate
d
to
t
h
e
id
en
ti
f
icatio
n
o
f
th
e
h
u
m
an
co
g
n
iti
v
e
s
tate
w
h
er
e
E
E
G
e
y
e
s
tate
cla
s
s
i
f
icat
io
n
task
is
th
e
ce
n
tr
al
ele
m
e
n
t,
s
u
ch
as
ep
ilep
tic
s
eiz
u
r
e
d
etec
tio
n
[
8
]
,
s
tr
es
s
f
ea
t
u
r
e
id
en
t
if
icatio
n
[
9
]
,
d
r
iv
i
n
g
d
r
o
w
s
i
n
ess
d
etec
tio
n
[
1
0
]
,
in
f
an
t
s
leep
-
w
ak
i
n
g
s
tate
id
en
tific
atio
n
[
1
1
]
.
A
lth
o
u
g
h
s
o
m
e
r
esear
ch
w
o
r
k
h
a
s
alr
ea
d
y
b
ee
n
d
o
n
e
f
o
r
e
y
e
s
ta
te
p
r
ed
ictio
n
,
id
en
t
if
ica
tio
n
o
r
d
ete
ctio
n
in
f
ac
ial
i
m
a
g
es
an
d
v
is
u
al
r
ec
o
r
d
in
g
s
to
o
[
1
2
]
.
Ho
w
e
v
er
,
as
illu
s
tr
ated
in
I
n
tr
o
d
u
ctio
n
o
u
r
ai
m
is
to
p
r
ed
ict
th
e
ey
e
s
tate
f
r
o
m
E
E
G
s
ig
n
al
s
as
it
ca
u
s
e
s
d
eter
io
r
atio
n
i
n
s
i
g
n
a
l.
T
h
er
ef
o
r
e,
to
f
o
s
ter
th
i
s
s
i
tu
atio
n
clas
s
i
f
icatio
n
i
s
d
o
n
e
b
y
d
ee
p
lea
r
n
in
g
ar
ch
itect
u
r
es.
I
n
th
e
liter
at
u
r
e,
r
esear
ch
er
s
h
av
e
atte
m
p
ted
to
s
u
cc
ess
f
u
ll
y
r
e
m
o
v
e
e
y
e
b
li
n
k
ar
ti
f
a
cts
b
y
d
e
-
v
elo
p
in
g
s
ev
e
r
al
m
et
h
o
d
s
f
o
r
in
s
ta
n
ce
[
1
3
-
1
5
]
.
T
h
e
au
th
o
r
s
in
[
1
6
]
p
r
o
v
id
ed
th
e
co
m
p
a
r
is
o
n
o
f
SVM
an
d
A
N
N
f
o
r
cla
s
s
i
f
icatio
n
o
f
E
E
G
e
y
e
s
tates
s
u
c
h
a
s
,
e
y
e
b
li
n
k
,
e
y
e
o
p
en
an
d
e
y
e
clo
s
ed
.
T
h
e
SVM
w
as
j
u
s
ti
f
ied
as
t
h
e
p
r
ef
er
r
ed
ch
o
ice
o
f
m
o
d
el
o
v
er
A
NN
o
n
th
e
b
asis
o
f
p
er
f
o
r
m
an
ce
ac
c
u
r
ac
y
g
i
v
e
n
b
y
b
o
th
m
o
d
els.
I
n
ad
d
itio
n
,
a
h
ier
ar
ch
ical
clas
s
if
ica
tio
n
al
g
o
r
ith
m
is
d
ev
el
o
p
ed
u
s
in
g
a
t
h
r
esh
o
ld
i
n
g
m
et
h
o
d
f
o
r
o
f
f
li
n
e
r
ec
o
g
n
itio
n
o
f
f
o
u
r
d
ir
ec
tio
n
s
o
f
e
y
e
m
o
v
e
m
en
ts
f
r
o
m
E
E
G
s
ig
n
al
s
[
1
7
]
.
T
h
e
r
esear
ch
er
s
in
[
1
8
]
p
r
o
p
o
s
ed
an
E
E
G
b
ased
ey
e
tr
ac
k
i
n
g
s
o
lu
t
io
n
b
y
u
s
in
g
t
h
e
in
f
o
r
m
atio
n
f
r
o
m
t
w
o
d
i
f
f
er
e
n
t
s
o
u
r
ce
s
,
i.e
.
,
Hea
d
m
o
u
n
ted
Vid
eo
-
Oc
u
lu
o
g
r
ap
h
y
a
n
d
1
6
-
ch
a
n
n
e
led
E
E
G
s
ig
n
al.
T
h
eir
p
r
o
p
o
s
ed
m
o
d
el
ac
h
ie
v
ed
th
e
ac
cu
r
ac
y
o
f
9
7
.
5
7
%
b
y
e
x
tr
ac
tin
g
th
e
f
ea
t
u
r
es
f
r
o
m
s
o
u
r
ce
b
y
u
s
i
n
g
I
C
A
r
ath
er
th
an
b
an
d
-
p
as
s
p
r
ep
r
o
ce
s
s
ed
E
E
G
s
ig
n
al
s
.
A
n
o
v
el
m
ac
h
in
e
lear
n
in
g
ap
p
r
o
ac
h
,
in
cr
e
m
e
n
tal
attr
ib
u
te
le
ar
n
i
n
g
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
E
E
G
B
a
s
ed
E
ye
S
ta
te
C
la
s
s
ifica
t
io
n
u
s
in
g
Dee
p
B
elief
N
etw
o
r
k
a
n
d
S
ta
ck
ed
A
u
to
E
n
c
o
d
er
(
S
a
n
a
m
N
a
r
ejo
)
3133
(
I
AL
)
is
p
r
o
p
o
s
ed
in
[
1
9
]
f
o
r
E
E
G
ey
e
s
tate
ti
m
e
s
er
ies
class
i
f
icatio
n
.
T
h
e
I
A
L
al
g
o
r
ith
m
p
r
o
g
r
es
s
i
v
el
y
i
m
p
o
r
ts
an
d
tr
ain
s
f
ea
tu
r
es o
n
e
at
a
ti
m
e
to
p
r
ed
ict
th
e
class
lab
elin
g
.
I
n
[
2
0
]
,
th
e
r
esear
ch
er
s
h
av
e
s
tu
d
ied
th
e
d
i
f
f
er
e
n
ce
s
a
m
o
n
g
th
r
ee
s
tates,
i.e
.
E
y
e
-
c
lo
s
ed
,
E
y
e
-
o
p
en
an
d
A
tte
n
s
io
n
s
tate
s
o
f
h
u
m
a
n
u
s
in
g
E
E
G
s
ig
n
al
s
.
T
h
e
m
aj
o
r
g
o
al
o
f
th
eir
r
esear
ch
w
as
t
o
ad
d
r
ess
th
e
is
s
u
e
o
f
h
o
w
s
p
atial
-
te
m
p
o
r
al
p
r
o
p
er
ties
o
f
alp
h
a
r
h
y
m
e
s
w
er
e
af
f
ec
ted
b
y
t
h
e
c
h
an
g
e
o
f
h
u
m
a
n
b
r
ain
s
tate.
Ho
w
e
v
er
,
th
e
au
t
h
o
r
s
d
id
n
o
t
p
u
r
s
u
e
f
u
r
t
h
er
to
w
ar
d
s
th
e
p
r
ed
ictiv
e
m
o
d
el
f
o
r
th
e
clas
s
i
f
icatio
n
task
o
f
th
e
ab
o
v
e
m
en
t
io
n
ed
s
tate
s
.
T
h
e
r
esear
ch
er
s
d
ev
elo
p
ed
E
E
G
E
y
e
s
tate
co
r
p
u
s
a
n
d
test
ed
4
2
d
if
f
er
e
n
t
clas
s
i
f
ier
s
f
o
r
d
etec
tin
g
w
h
et
h
er
an
in
d
i
v
id
u
al
's
e
y
es
ar
e
o
p
en
o
r
clo
s
ed
b
as
ed
o
n
E
E
G
r
ec
o
r
d
in
g
s
[
2
1
]
.
I
t
w
a
s
e
x
p
lo
r
ed
th
at
k
*
al
g
o
r
ith
m
ac
h
ie
v
ed
th
e
h
i
g
h
e
s
t
ac
cu
r
ac
y
i
n
co
m
p
ar
is
o
n
w
it
h
o
th
er
s
,
ac
h
ie
v
i
n
g
t
h
e
co
r
r
ec
t
class
if
icatio
n
r
ate
o
f
9
7
.
3
%.
T
h
eir
d
ev
elo
p
ed
E
y
e
s
ta
te
co
r
p
u
s
is
n
o
w
s
to
r
ed
in
th
e
Ma
ch
in
e
lear
n
i
n
g
r
ep
o
s
ito
r
y
as
a
b
en
ch
m
ar
k
p
r
o
b
le
m
[
2
2
]
.
T
h
e
s
a
m
e
co
r
p
u
s
is
f
u
r
th
er
u
s
ed
in
[
2
3
]
as
tim
e
s
er
ies
clas
s
i
f
ic
atio
n
tas
k
b
ased
o
n
I
AL
ap
p
r
o
ac
h
,
w
it
h
t
h
e
Feat
u
r
e
o
r
d
er
in
g
b
ased
o
n
A
cc
u
m
u
la
tiv
e
Di
s
cr
i
m
i
n
ab
ilit
y
(
A
D)
.
T
h
e
s
tu
d
y
o
f
Ol
iv
er
R
o
e
s
ler
an
d
Dav
id
S
u
en
d
er
m
an
n
[
2
1
]
is
f
u
r
th
er
e
x
ten
d
ed
in
[
2
4
]
in
wh
ich
th
r
ee
d
if
f
er
e
n
t
en
s
e
m
b
le
lear
n
i
n
g
m
o
d
els
ar
e
d
ev
elo
p
ed
an
d
th
e
ac
cu
r
ac
y
ac
h
iev
ed
w
it
h
t
h
e
m
o
s
t
ac
cu
r
ate
m
o
d
el
co
n
s
tr
u
cted
f
r
o
m
R
e
g
u
lar
ized
R
an
d
o
m
Fo
r
est
R
R
F
an
d
K
*
is
9
7
.
4
%.
T
h
e
r
esear
ch
w
o
r
k
in
[
2
5
]
p
r
o
p
o
s
ed
a
n
o
v
el
s
y
s
te
m
b
ased
o
n
s
m
all
n
u
m
b
er
s
o
f
Ne
u
r
o
Fu
zz
y
r
u
les
f
o
r
s
i
m
ilar
cla
s
s
i
f
icat
io
n
p
r
o
b
le
m
.
I
n
th
e
ir
s
t
u
d
y
,
w
ei
g
h
t
p
ar
a
m
eter
t
u
n
n
i
n
g
w
a
s
d
o
n
e
b
y
ad
j
u
s
tin
g
t
h
e
s
ta
n
d
ar
d
d
ev
iatio
n
s
.
T
h
e
b
est
p
o
s
s
ib
le
r
es
u
lts
w
er
e
ac
h
iev
ed
b
y
t
u
n
in
g
t
h
e
p
ar
a
m
e
ter
s
as
y
m
m
etr
icall
y
t
h
r
o
u
g
h
te
s
t c
ases
w
i
th
t
h
e
av
er
a
g
e
er
r
o
r
r
ate
o
f
4
.
0
%.
3.
T
H
E
CO
M
P
RE
H
E
NS
I
VE
T
H
E
O
R
Y
T
h
e
DL
As
ar
e
co
m
p
r
is
ed
o
f
m
u
ltip
le
le
v
el
s
o
f
n
o
n
-
lin
ea
r
o
p
er
atio
n
s
,
lik
e
n
e
u
r
al
n
et
w
o
r
k
s
w
it
h
s
ev
er
al
h
id
d
en
la
y
er
s
.
T
h
e
s
ta
n
d
ar
d
tr
ain
in
g
s
tr
ateg
y
o
f
n
eu
r
al
n
et
w
o
r
k
a
n
d
t
h
e
g
r
ad
ie
n
t
b
ased
o
p
tim
izatio
n
,
o
f
ten
g
et
s
s
t
u
ck
in
lo
ca
l
m
i
n
i
m
a
an
d
f
i
n
d
s
t
h
e
p
o
o
r
s
o
lu
ti
o
n
f
o
r
th
e
n
et
w
o
r
k
w
it
h
m
u
lt
ip
le
h
id
d
en
la
y
er
s
.
I
n
tu
i
tiv
e
l
y
,
it
i
s
d
if
f
ic
u
lt
to
tr
ain
d
ee
p
n
eu
r
al
n
et
w
o
r
k
s
w
it
h
s
ta
n
d
ar
d
lear
n
in
g
al
g
o
r
ith
m
s
b
ec
au
s
e
s
ea
r
ch
i
n
g
th
e
p
ar
a
m
eter
s
p
ac
e
b
ec
o
m
es
d
if
f
ic
u
lt
i
n
d
ee
p
er
n
et
w
o
r
k
s
.
T
h
is
is
s
u
e
g
et
s
r
eso
l
v
ed
b
y
i
n
itializi
n
g
th
e
w
ei
g
h
t
pa
r
am
eter
s
o
f
d
ee
p
n
eu
r
al
n
et
w
o
r
k
t
h
r
o
u
g
h
g
r
ee
d
y
la
y
er
w
is
e
u
n
s
u
p
er
v
is
ed
tr
ain
in
g
a
s
in
tr
o
d
u
ce
d
in
[
2
6
-
2
8
]
.
3
.
1
.
D
eep
B
elief
N
et
w
o
rk
DB
N
h
o
ld
s
a
g
r
ea
t
p
r
o
m
i
s
e
a
s
a
p
r
in
cip
le
to
ad
d
r
ess
th
e
p
r
o
b
lem
o
f
tr
ain
in
g
d
ee
p
n
e
u
r
al
n
et
w
o
r
k
s
.
DB
N
ar
e
co
m
p
o
s
ed
o
f
m
u
lti
p
le
la
y
er
s
o
f
s
to
ch
a
s
tic,
u
n
s
u
p
er
v
is
ed
m
o
d
el
s
u
c
h
a
s
R
e
s
tr
icted
B
o
ltz
m
an
n
Ma
ch
i
n
es
(
R
B
Ms)
,
w
h
ich
ar
e
u
s
ed
to
i
n
it
ialize
t
h
e
n
et
w
o
r
k
in
t
h
e
r
e
g
io
n
o
f
p
ar
a
m
e
ter
s
p
ac
e
th
at
f
i
n
d
s
g
o
o
d
m
i
n
i
m
a
o
f
t
h
e
s
u
p
er
v
is
ed
o
b
j
e
ctiv
e.
R
B
Ms
ar
e
p
r
o
b
ab
ilis
tic
g
r
ap
h
ical
m
o
d
els
t
h
at
ar
e
co
n
n
ec
te
d
b
i
-
d
ir
ec
tio
n
all
y
a
n
d
ca
n
b
e
in
f
er
r
ed
as
s
to
ch
ast
ic
n
e
u
r
al
n
et
w
o
r
k
s
.
R
B
M
r
elies
o
n
t
w
o
la
y
er
s
tr
u
c
t
u
r
e
co
m
p
r
is
i
n
g
o
n
v
i
s
ib
le
an
d
h
id
d
en
n
o
d
es,
t
h
is
co
n
ce
p
t
is
s
h
o
w
n
in
Fi
g
u
r
e
1
.
T
h
e
v
is
ib
le
u
n
i
ts
co
n
s
tit
u
te
t
h
e
f
ir
s
t
la
y
er
a
n
d
co
r
r
esp
o
n
d
to
th
e
co
m
p
o
n
e
n
t
s
o
f
an
o
b
s
er
v
atio
n
w
h
er
e
a
s
th
e
h
id
d
en
u
n
its
m
o
d
el
d
ep
en
d
en
ci
es
b
et
w
ee
n
t
h
e
co
m
p
o
n
e
n
t
s
o
f
o
b
s
er
v
atio
n
s
.
T
h
e
R
B
M
is
tr
ain
ed
to
m
o
d
el
th
e
j
o
in
t
p
r
o
b
ab
ilit
y
d
is
tr
ib
u
t
io
n
o
f
in
p
u
t
s
o
r
ex
p
lan
ato
r
y
v
ar
iab
les
an
d
th
e
co
r
r
esp
o
n
d
in
g
lab
els
o
r
r
es
p
o
n
s
e
v
ar
iab
les,
b
o
th
r
ep
r
es
en
ted
b
y
t
h
e
v
i
s
ib
le
u
n
it
s
o
f
th
e
R
B
M.
T
h
e
p
r
o
b
a
b
ilit
ies o
f
b
o
th
h
id
d
en
an
d
v
is
ib
le
n
o
d
es a
r
e
ex
p
r
ess
ed
i
n
eq
u
atio
n
(
1
)
.
(
|
)
∏
(
|
)
(
|
)
∏
(
|
)
(
1
)
T
h
e
co
n
d
itio
n
al
p
r
o
b
ab
ilit
y
o
f
a
s
i
n
g
le
v
ar
iab
le
to
b
e
1
,
ca
n
b
e
th
e
n
i
n
ter
p
r
eted
as
t
h
e
f
ir
i
n
g
r
ate
o
f
a
s
to
ch
ast
ic
n
e
u
r
o
n
w
it
h
s
i
g
m
o
i
d
ac
tiv
atio
n
f
u
n
c
tio
n
σ
(
x
)
=
1
/(
1
+
e^(
-
x
)
)
.
T
h
er
ef
o
r
e
th
e
b
in
ar
y
s
tates o
f
h
id
d
e
n
u
n
i
ts
an
d
r
ec
o
n
s
tr
u
ctio
n
f
r
o
m
th
o
s
e
h
id
d
en
u
n
it
s
as
v
i
s
ib
le
n
o
d
es
in
ab
o
v
e
eq
u
ati
o
n
s
ca
n
b
e
f
u
r
th
e
r
in
ter
p
r
eted
in
f
o
r
m
o
f
eq
u
atio
n
s
(
2
)
an
d
(
3
)
as
(
|
)
(
∑
)
(
2
)
(
|
)
(
∑
)
(
3
)
W
h
er
e,
W
ij
is
th
e
w
eig
h
t
as
s
o
ciate
d
b
et
w
ee
n
t
h
e
u
n
it
s
v
j
an
d
h
i
w
h
er
ea
s
b
j
an
d
ci
ar
e
th
e
b
ias
ter
m
s
.
T
h
e
ch
an
g
e
i
n
a
w
ei
g
h
t p
ar
a
m
eter
is
th
e
n
g
i
v
e
n
b
y
(
4
)
.
(
)
(
4
)
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
6
,
No
.
6
,
Dec
em
b
er
2
0
1
6
:
31
3
1
–
31
41
3134
T
h
e
id
ea
is
th
at
t
h
e
h
id
d
en
n
eu
r
o
n
s
e
x
tr
ac
t
r
ele
v
an
t
f
ea
t
u
r
es
f
r
o
m
th
e
o
b
s
er
v
atio
n
s
.
T
h
ese
f
ea
t
u
r
es
s
er
v
e
as
in
p
u
t
to
a
n
o
th
er
R
B
M.
B
y
s
tac
k
i
n
g
R
B
Ms
i
n
t
h
i
s
w
a
y
,
o
n
e
ca
n
ac
h
ie
v
e
h
ig
h
le
v
el
r
ep
r
esen
tatio
n
s
b
y
lear
n
in
g
t
h
e
f
ea
t
u
r
es
f
r
o
m
f
ea
t
u
r
es.
DB
Ns
ar
e
ca
p
ab
le
o
f
e
x
tr
ac
t
in
g
a
d
ee
p
h
ier
ar
ch
ical
r
ep
r
e
s
en
tat
io
n
o
f
t
h
e
tr
ai
n
i
n
g
d
ata
.
T
h
e
to
p
la
y
er
s
o
f
DB
N
ar
e
r
esp
o
n
s
ib
le
as
a
s
o
u
r
ce
o
f
m
o
r
e
“
ab
s
tr
ac
t
an
d
m
ea
n
in
g
f
u
l”
r
ep
r
esen
tatio
n
s
t
h
at
ex
p
lai
n
th
e
in
p
u
t o
b
s
er
v
at
io
n
x
,
w
h
er
ea
s
l
o
w
er
la
y
er
ex
tr
ac
ts
“
lo
w
lev
e
l f
ea
t
u
r
es”.
Fig
u
r
e.
1
.
T
h
e
s
tr
u
ctu
r
e
o
f
a
s
i
n
g
le
R
B
M
co
m
p
o
s
ed
o
f
v
i
s
ib
l
e
la
y
er
an
d
h
id
d
en
la
y
er
3
.
2
.
Sta
ck
ed
A
uto
E
nco
der
An
alter
n
at
iv
e
ap
p
r
o
ac
h
to
d
e
ep
lear
n
in
g
i
s
Stac
k
ed
Au
to
e
n
co
d
er
,
w
h
er
e
th
e
h
id
d
en
u
n
it
s
g
e
n
er
ate
ex
p
lo
itab
le
n
u
m
er
ic
f
ea
tu
r
e
v
alu
es.
As
d
e
m
o
n
s
tr
ated
in
[
2
9
-
3
0
]
Au
to
E
n
co
d
er
s
h
av
e
b
e
co
m
e
t
h
e
d
o
m
i
n
a
n
t
f
o
ca
l p
o
i
nt
in
th
e
ar
ea
o
f
“
d
ee
p
ar
ch
itectu
r
e
s
”.
An
Au
to
E
n
c
o
d
er
is
a
n
e
u
r
al
n
et
w
o
r
k
t
h
at
c
o
n
s
is
ts
o
f
t
w
o
p
ar
ts
:
en
co
d
er
an
d
d
ec
o
d
er
.
I
t
is
tr
ain
ed
in
s
u
ch
a
w
a
y
to
r
ec
o
n
s
tr
u
ct
t
h
e
g
iv
e
n
i
n
p
u
t
at
t
h
e
o
u
t
p
u
t
la
y
er
.
T
h
e
f
ir
s
t
p
ar
t,
th
e
e
n
co
d
er
is
r
esp
o
n
s
i
b
le
to
co
m
p
u
te
th
e
h
id
d
en
r
ep
r
esen
tatio
n
s
h
o
f
i
n
p
u
t
s
p
a
ce
v
th
r
o
u
g
h
s
o
m
e
n
o
n
li
n
ea
r
tr
an
s
f
er
f
u
n
ct
io
n
f
(
v
)
.
T
h
e
d
ec
o
d
er
tak
es
th
ese
l
aten
ts
p
r
o
d
u
ce
d
b
y
en
co
d
er
s
an
d
f
u
r
th
er
p
as
s
es
i
t
th
r
o
u
g
h
s
o
m
e
n
o
n
-
lin
ea
r
i
t
y
o
r
lin
ea
r
it
y
g
’
(
h
)
t
o
g
e
n
e
r
a
t
e
t
h
e
r
e
c
o
n
s
t
r
u
c
t
i
o
n
o
u
t
p
u
t
v
.
h
e
s
e
t
o
f
p
a
r
a
m
e
t
e
r
s
an
d
’
a
r
e
l
e
a
r
n
t
s
i
m
u
l
t
a
n
e
o
u
s
l
y
o
n
t
h
e
t
a
s
k
o
f
m
i
n
i
m
i
z
i
n
g
r
e
c
o
n
s
t
r
u
c
t
i
o
n
e
r
r
o
r
(
v
,
v
)
w
h
i
c
h
i
s
a
m
e
a
s
u
r
e
o
f
th
e
d
is
cr
ep
an
c
y
b
et
w
ee
n
ac
t
u
a
l a
n
d
r
ec
o
n
s
tr
u
cted
.
T
h
e
er
r
o
r
f
u
n
ctio
n
is
e
x
p
r
ess
ed
in
(
5
)
.
∑
(
(
(
)
)
)
(
5
)
T
h
e
A
u
to
E
n
co
d
er
b
asicall
y
lear
n
s
to
m
ap
t
h
e
in
p
u
t
b
y
lear
n
in
g
t
h
e
n
o
n
-
lin
ea
r
o
r
s
o
m
e
r
ep
r
esen
tatio
n
s
o
f
i
n
p
u
t
s
p
ac
e
in
an
u
n
s
u
p
er
v
i
s
ed
m
a
n
n
er
.
T
h
e
s
tr
u
ctu
r
e
o
f
an
Au
to
E
n
co
d
er
w
i
th
a
s
in
g
l
e
h
id
d
en
la
y
er
is
elab
o
r
ated
in
Fig
u
r
e
2
.
T
h
e
o
u
tp
u
t
la
y
er
in
an
Au
to
n
co
d
er
s
h
o
u
ld
b
e
o
f
t
h
e
s
a
m
e
s
ize
as
th
e
in
p
u
t
la
y
er
,
b
ec
au
s
e
t
h
e
Au
to
E
n
co
d
er
is
n
o
t
cla
s
s
i
f
y
i
n
g
th
e
lab
eled
o
u
tp
u
t,
in
s
tead
it
is
r
ep
r
o
d
u
cin
g
th
e
in
p
u
t.
E
ac
h
Au
to
e
n
co
d
er
is
s
ep
ar
atel
y
tr
ain
ed
li
k
e
an
R
B
M
i
n
lay
er
w
i
s
e
m
a
n
n
er
.
Du
r
i
n
g
t
h
e
tr
ain
in
g
p
h
ase
t
h
e
r
ec
o
n
s
tr
u
ctio
n
s
a
t
t
h
e
o
u
tp
u
t
l
a
y
er
ar
e
co
m
p
ar
ed
w
it
h
t
h
e
ac
tu
al
g
i
v
en
in
p
u
t
to
r
ed
u
ce
th
e
d
if
f
er
e
n
ce
b
et
w
ee
n
th
e
m
.
S
tack
i
n
g
th
e
tr
ai
n
ed
u
n
s
u
p
er
v
is
ed
Au
to
E
n
co
d
er
o
v
er
a
n
o
th
e
r
A
u
to
E
n
co
d
er
an
d
th
en
tr
ain
i
n
g
t
h
e
w
h
o
le
n
et
w
o
r
k
g
lo
b
all
y
b
y
ad
d
in
g
an
o
u
t
p
u
t la
y
er
ca
te
g
o
r
ize
th
e
m
o
d
el
in
th
e
ca
teg
o
r
y
o
f
D
L
A
.
Fig
u
r
e
2
.
I
llu
s
tr
atio
n
o
f
an
Au
t
o
E
n
co
d
er
4.
R
E
S
E
ARCH
M
E
T
H
O
D
I
n
th
i
s
w
o
r
k
,
ab
o
v
e
m
en
tio
n
e
d
DL
A
m
o
d
els
ar
e
d
ev
elo
p
ed
an
d
tr
ain
ed
as
class
i
f
icatio
n
m
o
d
el
s
f
o
r
i
m
p
r
o
v
ed
E
E
G
b
ased
e
y
e
s
t
ate
id
en
ti
f
icatio
n
.
Am
o
n
g
th
e
v
ar
io
u
s
a
v
ailab
le
d
ee
p
ar
c
h
itect
u
r
es
o
f
n
eu
r
al
n
et
w
o
r
k
s
,
th
e
c
u
r
r
en
t
r
esear
c
h
f
o
c
u
s
e
s
o
n
t
h
e
i
m
p
le
m
e
n
ta
tio
n
o
f
DB
N
a
n
d
S
A
E
f
o
r
th
e
p
r
esen
t
s
t
u
d
y
.
I
n
o
r
d
er
to
tr
ain
th
e
s
e
d
ee
p
ar
c
h
itect
u
r
es
th
e
lear
n
i
n
g
al
g
o
r
ith
m
f
o
r
th
e
n
eu
r
al
n
et
w
o
r
k
tr
ain
i
n
g
is
in
li
n
e
w
it
h
[
2
7
-
2
8
]
.
W
e
f
u
r
t
h
er
i
n
v
e
s
tig
a
te
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
s
e
i
m
p
le
m
e
n
ted
d
ee
p
n
et
w
o
r
k
s
b
y
p
r
o
v
id
i
n
g
t
h
e
W
W
T
D
e
c
o
d
e
r
E
n
c
o
d
e
r
H
idd
e
n
I
n
p
u
t
Ou
t
p
u
t
h
v
v
̂
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
E
E
G
B
a
s
ed
E
ye
S
ta
te
C
la
s
s
ifica
t
io
n
u
s
in
g
Dee
p
B
elief
N
etw
o
r
k
a
n
d
S
ta
ck
ed
A
u
to
E
n
c
o
d
er
(
S
a
n
a
m
N
a
r
ejo
)
3135
co
m
p
ar
ati
v
e
an
al
y
s
is
w
it
h
ea
ch
o
th
er
an
d
w
ith
s
o
m
e
o
f
o
th
er
co
n
v
e
n
tio
n
al
m
ac
h
i
n
e
le
ar
n
in
g
ap
p
r
o
ac
h
es
m
en
tio
n
ed
in
t
h
e
s
u
b
s
eq
u
e
n
t
s
ec
tio
n
.
A
d
is
tin
g
u
is
h
ab
le
f
ac
to
r
o
f
th
e
d
ee
p
lear
n
in
g
ar
ch
itect
u
r
es
f
r
o
m
co
n
v
e
n
tio
n
al
cla
s
s
i
f
ier
s
is
t
h
e
au
to
m
at
ic
f
ea
t
u
r
e
lear
n
in
g
f
r
o
m
d
ata
w
h
ich
m
ain
l
y
co
n
tr
ib
u
tes
to
i
m
p
r
o
v
e
m
e
n
t
in
t
h
e
ac
c
u
r
ac
y
o
f
th
e
m
o
d
el.
A
p
ar
t
f
r
o
m
t
h
i
s
t
h
e
d
ee
p
le
ar
n
in
g
al
g
o
r
ith
m
s
i
n
tr
o
d
u
ce
d
r
ec
en
tl
y
,
s
o
l
v
es
t
h
e
h
ar
d
o
p
tim
izatio
n
p
r
o
b
le
m
in
t
o
s
ev
er
al
g
r
ee
d
y
s
tep
s
,
th
o
u
g
h
th
e
s
i
m
p
le
o
n
es,
to
tr
ain
t
h
e
d
ee
p
er
n
et
w
o
r
k
s
.
T
h
e
r
esear
ch
s
tr
ateg
y
f
o
llo
w
e
d
in
th
is
w
o
r
k
is
d
e
m
o
n
s
tr
ate
d
in
Fig
u
r
e
3
.
I
n
itiall
y
,
th
e
d
ata
is
p
r
e
-
p
r
o
ce
s
s
ed
b
y
r
e
m
o
v
in
g
th
e
n
o
is
e
o
r
an
y
k
i
n
d
o
f
an
o
u
tl
ier
.
Af
ter
w
ar
d
s
,
t
h
e
m
ea
n
i
n
g
f
u
l
f
e
atu
r
e
s
e
t
is
f
o
r
m
ed
b
y
ap
p
l
y
i
n
g
th
e
DW
T
tr
an
s
f
o
r
m
at
io
n
o
n
E
E
G
s
ig
n
al
as
e
x
p
lain
ed
i
n
th
e
later
p
ar
t
o
f
t
h
is
s
ec
tio
n
.
B
ef
o
r
e
p
r
esen
tin
g
th
e
e
x
tr
ac
ted
f
ea
t
u
r
e
s
et
to
t
h
e
cla
s
s
i
f
icatio
n
m
o
d
els
f
o
r
tr
ai
n
in
g
,
t
h
e
f
ea
t
u
r
e
s
et
w
as
n
o
r
m
alize
d
in
th
e
r
an
g
e
o
f
(
0
,
1
)
s
o
th
at
th
e
n
et
w
o
r
k
lea
r
n
s
ef
f
icie
n
tl
y
.
F
in
all
y
t
h
e
p
r
ep
r
o
ce
s
s
ed
d
atase
t
w
as
u
s
ed
f
o
r
th
e
tr
ain
i
n
g
o
f
cla
s
s
i
f
icat
io
n
m
o
d
els.
Fo
r
clas
s
i
f
icatio
n
m
o
d
els
,
as
w
e
e
x
p
lai
n
ed
ea
r
lier
,
o
u
r
m
aj
o
r
f
o
cu
s
is
to
ex
p
lo
r
e
th
e
p
er
f
o
r
m
a
n
ce
o
f
DB
N
an
d
S
A
E
.
T
h
e
m
aj
o
r
o
b
j
ec
tiv
e
w
as,
t
h
e
lo
w
est
p
o
s
s
ib
le
er
r
o
r
an
d
h
ig
h
ac
cu
r
ac
y
to
b
e
ac
h
iev
ed
b
y
i
m
p
le
m
e
n
ted
m
o
d
els.
He
n
ce
,
t
h
e
tr
ain
i
n
g
o
f
Dee
p
A
r
ch
itect
u
r
es
to
o
k
a
co
u
p
le
o
f
d
ay
s
,
w
h
ic
h
w
a
s
n
o
t
th
e
m
o
s
t
im
p
o
r
tan
t
tar
g
et
f
o
r
o
u
r
ca
s
e.
T
h
e
d
etail
ex
p
lan
atio
n
o
f
th
e
s
tep
s
f
o
llo
w
ed
ar
e
s
p
ec
if
ied
b
elo
w
.
T
h
e
E
E
G
d
ata
s
et
w
as
ta
k
en
f
r
o
m
t
h
e
UC
I
m
ac
h
i
n
e
L
ea
r
n
i
n
g
r
ep
o
s
ito
r
y
,
w
h
ic
h
i
s
a
b
en
c
h
m
ar
k
f
o
r
th
e
E
y
e
s
tate
clas
s
i
f
icatio
n
p
r
o
b
lem
.
T
h
e
E
E
G
s
ig
n
al
w
a
s
r
ec
o
r
d
e
d
f
r
o
m
1
4
d
if
f
er
en
t
elec
tr
o
d
es
w
it
h
th
e
E
m
o
tiv
E
E
G
Neu
r
o
h
ea
d
s
e
t.
T
h
e
e
y
e
s
tate
eit
h
er
o
p
en
o
r
clo
s
ed
w
a
s
ca
p
tu
r
ed
m
a
n
u
all
y
b
y
m
ea
n
s
o
f
a
ca
m
er
a
d
u
r
in
g
th
e
E
E
G
m
ea
s
u
r
e
m
e
n
t
a
n
d
th
e
ca
p
t
u
r
ed
f
r
a
m
e
s
w
er
e
i
n
ter
p
r
eted
as
1
an
d
0
in
co
r
r
esp
o
n
d
to
E
ye
-
clo
s
ed
an
d
E
y
e
-
o
p
en
s
tat
es.
T
h
e
d
u
r
atio
n
o
f
t
h
e
m
ea
s
u
r
e
m
en
t
w
as
1
1
7
s
ec
o
n
d
s
w
it
h
1
4
9
8
0
s
am
p
les.
T
h
e
E
E
G
s
ig
n
al
w
as
f
u
r
th
er
p
r
ep
r
o
ce
s
s
ed
f
o
r
o
u
tlier
s
a
n
d
n
o
is
e.
Fig
u
r
e
3
.
Flo
w
o
f
R
esear
c
h
M
eth
o
d
o
lo
g
y
I
n
o
r
d
er
to
ex
tr
ac
t
u
s
ef
u
l
f
ea
t
u
r
es
f
r
o
m
E
E
G
ti
m
e
s
er
ies
s
i
g
n
al
Dis
cr
ete
W
av
elet
T
r
an
s
f
o
r
m
(
DW
T
)
w
a
s
ap
p
lied
.
I
n
th
e
d
o
m
ai
n
o
f
b
io
m
ed
ical
s
i
g
n
als
D
W
T
h
as
b
ee
n
w
id
el
y
u
s
ed
f
o
r
ti
m
e
-
f
r
eq
u
e
n
c
y
an
al
y
s
is
[
3
1
]
.
T
r
ad
itio
n
all
y
u
s
ed
Sh
o
r
t
ter
m
Fo
u
r
ier
tr
an
s
f
o
r
m
s
p
er
f
o
r
m
te
m
p
o
r
al
r
eso
lu
tio
n
a
n
d
h
ig
h
li
g
h
t
ch
an
g
e
i
n
s
p
ec
tr
u
m
w
i
th
r
e
s
p
ec
t
to
ti
m
e.
O
n
t
h
e
o
th
er
h
an
d
it
g
iv
e
s
a
u
n
i
f
o
r
m
r
e
s
o
lu
tio
n
in
th
e
f
r
eq
u
e
n
c
y
d
o
m
ai
n
.
T
h
e
w
av
ele
t
tr
an
s
f
o
r
m
p
r
o
v
id
es
a
w
i
n
d
o
w
th
a
t
h
a
s
a
co
n
s
ta
n
t
r
elati
v
e
er
r
o
r
in
th
e
f
r
eq
u
e
n
c
y
d
o
m
a
in
,
r
ath
er
th
an
co
n
s
ta
n
t
ab
s
o
lu
te
er
r
o
r
,
at
th
e
ex
p
en
s
e
o
f
ti
m
e
r
eso
lu
tio
n
,
in
ac
co
r
d
an
ce
w
it
h
Heise
n
b
er
g
's
u
n
ce
r
tai
n
t
y
p
r
in
cip
le.
M
u
lti
le
v
el
W
av
elet
d
ec
o
m
p
o
s
itio
n
r
e
s
o
lv
es
th
e
m
aj
o
r
w
av
e
let
i
n
to
d
if
f
er
e
n
t
f
r
eq
u
en
c
y
b
an
d
s
alp
h
a,
b
eta,
g
a
m
m
a,
d
el
ta,
th
eta
an
d
s
i
g
m
a.
As
w
e
h
a
v
e
s
p
ec
if
ied
ea
r
lier
,
th
e
E
E
G
d
ata
s
et
u
s
ed
i
n
t
h
is
s
tu
d
y
co
n
tai
n
s
s
i
g
n
al
s
f
r
o
m
1
4
d
if
f
er
en
t
elec
tr
o
d
es.
T
h
e
1
4
elec
tr
o
d
es
ar
e
co
n
s
id
er
ed
as
1
4
d
if
f
er
en
t
ch
an
n
els
as
s
h
o
w
n
i
n
t
h
e
Fig
u
r
e
4
an
d
D
W
T
is
ap
p
lied
to
th
e
s
ig
n
a
l o
f
ea
ch
c
h
an
n
el.
S
i
g
na
l
P
r
e
pr
o
c
e
ss
i
ng
C
l
as
si
f
i
c
at
i
o
n
M
o
de
l
s
E
E
G
R
a
w
D
a
t
a
Fea
t
u
r
e
E
x
t
r
a
c
t
ion
Ou
t
li
e
r
s
a
n
d
N
o
i
s
e
R
e
m
o
v
a
l
D
e
e
p
L
e
a
r
n
ing
A
r
c
h
i
t
e
c
t
u
r
e
s
D
B
N
S
A
E
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
6
,
No
.
6
,
Dec
em
b
er
2
0
1
6
:
31
3
1
–
31
41
3136
Fig
u
r
e
4
.
E
E
G
Sig
n
a
ls
Ge
n
er
ate
d
f
r
o
m
1
4
El
ec
tr
o
d
es
T
h
e
DW
T
d
ec
o
m
p
o
s
itio
n
s
o
f
s
ig
n
al
s
ar
e
ac
tu
a
ll
y
th
e
o
u
tco
m
es
o
f
h
i
g
h
p
as
s
a
n
d
lo
w
p
as
s
f
ilter
in
g
.
T
h
e
ap
p
r
o
x
im
a
tio
n
s
ar
e
t
h
e
h
i
g
h
-
s
ca
le,
lo
w
-
f
r
eq
u
en
c
y
co
m
p
o
n
e
n
t
s
o
f
t
h
e
s
i
g
n
al.
T
h
e
d
etails
ar
e
t
h
e
lo
w
-
s
ca
le,
h
i
g
h
-
f
r
eq
u
e
n
c
y
co
m
p
o
n
en
ts
.
T
h
e
E
E
G
s
i
g
n
al
is
f
u
r
t
h
er
d
ec
o
m
p
o
s
ed
in
to
le
v
e
l
8
A
p
p
r
o
x
i
m
atio
n
s
a8
an
d
Deta
ils
d
1
-
d
8
.
T
h
e
lev
el
o
f
d
ec
o
m
p
o
s
itio
n
d
ep
en
d
s
o
n
th
e
p
r
in
cip
al
f
r
eq
u
e
n
c
y
co
m
p
o
n
en
ts
.
Fi
g
u
r
e
5
in
d
icate
s
t
h
e
p
h
e
n
o
m
e
n
a
o
f
4
lev
el
DW
T
d
ec
o
m
p
o
s
i
tio
n
tr
ee
.
T
h
ese
d
ec
o
m
p
o
s
itio
n
s
ter
m
ed
as
s
u
b
b
an
d
f
ea
t
u
r
es a
r
e
o
f
ten
u
s
ed
as a
n
i
n
p
u
t
f
ea
t
u
r
e
s
et
o
f
attr
ib
u
tes
f
o
r
v
ar
io
u
s
class
if
ica
tio
n
a
n
d
p
r
ed
ictio
n
task
s
[
3
2
]
.
Fo
r
ea
ch
c
h
an
n
el
s
i
g
n
al
s
ep
ar
ate
d
ec
o
m
p
o
s
it
io
n
s
w
er
e
c
o
m
p
u
ted
.
S
u
b
s
eq
u
e
n
tl
y
,
t
h
e
f
r
e
q
u
en
c
y
b
an
d
s
alp
h
a,
b
eta,
g
a
m
m
a,
th
eta
an
d
d
elta
ar
e
esti
m
a
ted
f
r
o
m
t
h
o
s
e
co
m
p
u
ted
d
ec
o
m
p
o
s
itio
n
s
.
I
n
th
i
s
w
a
y
,
ea
ch
s
ig
n
al
f
r
o
m
ea
ch
ch
a
n
n
e
l
is
p
r
esen
ted
in
ter
m
s
o
f
th
e
a
b
o
v
e
m
e
n
tio
n
ed
f
iv
e
b
an
d
s
.
T
h
er
ef
o
r
e
th
e
f
ea
t
u
r
e
s
et
co
n
s
is
t
s
o
f
to
tal
1
4
m
u
lt
i
p
lied
b
y
5
as
to
tal
attr
ib
u
te
s
.
T
h
e
n
o
r
m
aliza
tio
n
p
r
o
ce
d
u
r
e
w
a
s
f
u
r
t
h
er
ap
p
lied
o
n
th
e
f
ea
tu
r
e
s
et
to
b
r
in
g
th
e
v
al
u
es i
n
th
e
r
a
n
g
e
b
et
w
ee
n
0
an
d
1
.
Fig
u
r
e
5
.
P
r
esen
tatio
n
o
f
4
-
le
v
el
DW
T
Dec
o
m
p
o
s
itio
n
T
r
e
e
I
n
th
is
p
ar
t
o
f
w
o
r
k
,
e
x
p
er
i
m
en
ts
ar
e
p
r
ese
n
ted
to
an
a
l
y
ze
a
n
d
ex
p
lo
r
e
th
e
d
ee
p
lear
n
i
n
g
ar
ch
itect
u
r
es
th
at
f
all
s
in
t
w
o
b
r
o
a
d
an
d
m
aj
o
r
ca
teg
o
r
ies
i.e
.
DB
N
an
d
SA
E
an
d
to
f
u
r
t
h
er
in
v
esti
g
ate
t
h
e
p
er
f
o
r
m
a
n
ce
o
f
b
o
th
m
o
d
els o
n
th
e
b
asi
s
o
f
co
m
p
ar
ati
v
e
an
a
l
y
s
i
s
.
4
.
1
.
E
x
peri
m
e
nta
l
Set
up
1
I
n
th
is
s
tep
,
DB
N
m
o
d
el
w
as
d
ev
elo
p
ed
an
d
tr
ain
ed
.
T
h
e
n
e
t
w
o
r
k
m
o
d
el
w
as
f
o
r
m
ed
b
y
s
t
ac
k
in
g
th
e
R
B
Ms.
A
DB
N
w
as
d
ev
elo
p
ed
w
it
h
a
n
i
n
p
u
t
la
y
er
co
n
s
is
ti
n
g
o
f
7
0
u
n
its
,
th
r
ee
h
i
d
d
en
la
y
er
s
o
f
s
ize
(
6
0
0
-
400
-
1
0
0
)
an
d
an
o
u
tp
u
t
la
y
er
w
ith
o
n
e
u
n
it
to
p
r
ed
ict
t
h
e
class
o
f
an
E
E
G
s
i
g
n
a
l
i.e
.
o
p
en
o
r
clo
s
e
.
T
h
e
u
lti
m
ate
ar
ch
itec
tu
r
e
o
f
t
h
e
tr
ain
ed
DB
N
m
o
d
el
is
s
h
o
w
n
i
n
Fi
g
u
r
e
6
.
T
o
s
elec
t
th
e
to
p
o
lo
g
ical
s
tr
u
c
tu
r
e
o
f
th
e
h
id
d
en
la
y
er
s
o
f
DB
N
th
e
d
e
m
o
n
s
tr
atio
n
s
g
i
v
en
i
n
[
3
3
]
w
er
e
f
o
llo
w
ed
.
E
ac
h
lay
er
o
f
th
e
DB
N
is
in
d
ep
en
d
en
tl
y
tr
ai
n
ed
as
a
n
R
B
M
w
it
h
t
h
e
Si
g
m
o
id
ac
tiv
atio
n
f
u
n
ctio
n
.
T
h
e
s
tate
s
o
f
h
id
d
en
n
o
d
es
d
eter
m
in
ed
b
y
tr
ai
n
ed
R
B
M
w
er
e
u
s
ed
as
in
p
u
t
to
th
e
n
ex
t
la
y
er
.
Af
ter
u
n
s
u
p
er
v
is
ed
tr
ain
i
n
g
,
t
h
e
lab
els
o
f
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
E
E
G
B
a
s
ed
E
ye
S
ta
te
C
la
s
s
ifica
t
io
n
u
s
in
g
Dee
p
B
elief
N
etw
o
r
k
a
n
d
S
ta
ck
ed
A
u
to
E
n
c
o
d
er
(
S
a
n
a
m
N
a
r
ejo
)
3137
E
E
G
E
y
e
e
v
en
ts
w
er
e
p
r
o
v
id
ed
at
th
e
o
u
tp
u
t
la
y
er
f
o
r
lin
ea
r
m
ap
p
in
g
.
T
h
e
p
r
e
f
er
r
ed
p
ar
a
m
eter
s
elec
tio
n
s
f
o
r
th
e
p
r
etr
ain
i
n
g
o
f
DB
N
ar
e
r
e
v
ea
led
in
T
ab
le
1
.
Fig
u
r
e
6
.
DB
N
m
o
d
el
f
o
r
clas
s
if
ica
tio
n
T
ab
le
1
.
P
ar
am
eter
Setti
n
g
s
f
o
r
P
r
etr
ain
in
g
P
h
ase
P
a
r
a
me
t
e
r
V
a
l
u
e
N
o
.
o
f
I
t
e
r
a
t
i
o
n
s
10
L
e
a
r
n
i
n
g
R
a
t
e
0
.
0
1
B
a
t
c
h
S
i
z
e
2
T
r
a
n
sf
e
r
F
u
n
c
t
i
o
n
S
i
g
mo
i
d
On
ce
th
e
la
y
er
s
o
f
R
B
M
in
D
B
N
w
er
e
tr
ain
ed
,
an
o
u
tp
u
t
la
y
er
w
i
th
C
r
o
s
s
E
n
tr
o
p
y
co
s
t
f
u
n
ct
io
n
w
as
ad
d
ed
.
I
n
th
is
w
a
y
,
th
e
n
et
w
o
r
k
w
as
g
lo
b
all
y
f
i
n
e
t
u
n
ed
w
i
th
a
s
u
p
er
v
i
s
ed
alg
o
r
it
h
m
to
p
r
ed
ict
th
e
tar
g
ets
.
T
h
e
to
tal
6
0
0
iter
atio
n
s
w
er
e
i
m
p
le
m
e
n
ted
to
tr
ain
t
h
e
n
et
w
o
r
k
w
ith
t
h
e
ad
d
ed
lin
ea
r
o
u
tp
u
t la
y
er
.
4
.
2
.
E
x
peri
m
e
nta
l
Set
up
2
Stack
ed
Au
to
en
co
d
er
s
as
D
L
A
e
x
is
t
i
n
s
e
v
er
al
f
o
r
m
s
o
f
v
a
r
iatio
n
s
[
7
]
.
I
n
th
is
s
t
u
d
y
,
w
e
at
te
m
p
t
to
i
m
p
le
m
en
t
t
h
e
s
p
ar
s
e
A
u
to
E
n
co
d
er
s
.
I
n
th
e
p
r
i
m
ar
y
s
tep
,
to
f
i
n
d
th
e
ac
cu
r
ate
m
o
d
el,
s
ev
er
al
n
u
m
b
er
s
o
f
m
o
d
el
s
as
s
p
ar
s
e
S
A
E
w
er
e
d
ev
elo
p
ed
an
d
tr
ain
ed
.
I
n
i
tiall
y
,
th
ese
n
e
u
r
al
n
e
t
w
o
r
k
s
w
er
e
d
esig
n
ed
f
o
r
th
e
g
o
al
o
f
d
i
m
en
s
io
n
ali
t
y
r
ed
u
ctio
n
,
b
u
t,
w
it
h
th
e
g
r
o
w
i
n
g
f
a
m
e
o
f
R
ep
r
esen
ta
tio
n
L
ea
r
n
i
n
g
t
h
e
d
i
m
e
n
s
io
n
s
d
o
n
o
t
n
ec
es
s
ar
il
y
n
ee
d
to
b
e
lo
w
er
t
h
an
t
h
e
i
n
p
u
t
s
ize.
Ho
w
e
v
er
,
th
er
e
ex
i
s
ts
a
p
r
o
b
le
m
if
t
h
e
h
id
d
en
la
y
er
s
o
f
a
n
Au
to
E
n
co
d
er
ar
e
eit
h
er
allo
w
ed
to
o
m
u
ch
ca
p
ac
it
y
o
r
h
o
ld
in
s
u
f
f
i
cie
n
t
d
i
m
e
n
s
io
n
s
.
T
h
e
A
u
to
E
n
co
d
er
atte
m
p
ts
to
r
ep
licatio
n
ta
s
k
,
in
s
tead
o
f
lear
n
in
g
t
h
e
i
n
ter
n
al
r
ep
r
esen
tatio
n
o
f
i
n
p
u
t
d
is
tr
ib
u
t
io
n
a
n
d
ex
tr
ac
ti
n
g
th
e
u
s
e
f
u
l
i
n
f
o
r
m
atio
n
,
it
wo
r
k
s
as
an
id
en
tit
y
f
u
n
c
tio
n
.
T
h
e
p
r
o
v
is
io
n
f
o
r
t
h
is
is
s
u
e
is
a
R
e
g
u
lar
ize
Au
t
o
en
co
d
er
.
R
at
h
er
th
a
n
co
n
s
tr
ain
in
g
t
h
e
h
id
d
en
la
y
er
d
im
e
n
s
io
n
s
o
r
m
o
d
el
c
ap
ac
it
y
r
eg
u
lar
ized
Au
to
E
n
co
d
er
ar
e
s
p
ar
s
e
an
d
u
s
es a
lo
s
s
f
u
n
ctio
n
w
it
h
s
o
m
e
s
p
ar
s
it
y
p
en
a
lt
y
a
s
ex
p
r
ess
ed
i
n
(
6
)
.
∑
(
(
(
)
)
)
(
)
(
)
(
6
)
Sin
ce
o
u
r
f
o
cu
s
w
a
s
to
ch
o
o
s
e
th
e
m
o
d
els
w
it
h
t
h
e
b
est
clas
s
if
ica
tio
n
ab
ilit
y
,
th
er
e
f
o
r
e
o
n
ly
s
elec
ted
S
A
E
m
o
d
els
ar
e
in
c
lu
d
ed
h
er
e
f
o
r
f
u
r
t
h
er
d
is
cu
s
s
io
n
.
T
ab
le
2
d
e
m
o
n
s
tr
ates
t
h
e
s
elec
ted
o
n
es
f
r
o
m
v
ar
io
u
s
i
m
p
l
e
m
en
ted
S
AE
m
o
d
els a
lo
n
g
w
it
h
th
e
ir
p
ar
a
m
eter
s
p
ec
i
f
icatio
n
s
.
T
h
e
s
p
ar
s
it
y
p
r
o
p
o
r
tio
n
is
a
p
ar
a
m
eter
o
f
s
p
ar
s
it
y
r
eg
u
ler
izer
w
h
ic
h
m
u
s
t
b
e
s
elec
ted
i
n
t
h
e
r
a
n
g
e
o
f
0
to
1
,
w
h
er
ea
s
λ
an
d
β
ar
e
th
e
co
ef
f
icie
n
ts
to
2
w
ei
g
h
t
r
eg
u
lar
izatio
n
an
d
s
p
ar
s
it
y
r
eg
u
la
r
izatio
n
.
Deta
iled
d
is
cu
s
s
io
n
o
n
m
o
d
el
ac
cu
r
ac
y
an
d
p
er
f
o
r
m
a
n
ce
is
p
r
o
v
id
ed
in
th
e
n
e
x
t sectio
n
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
6
,
No
.
6
,
Dec
em
b
er
2
0
1
6
:
31
3
1
–
31
41
3138
T
ab
le
2
.
P
ar
am
eter
Setti
n
g
s
f
o
r
P
r
etr
ain
in
g
P
h
ase
S
A
E
mo
d
e
l
s
A
r
c
h
i
t
e
c
t
u
r
e
H
i
d
d
e
n
L
a
y
e
r
P
a
r
a
me
t
e
r
s
O
u
t
p
u
t
l
a
y
e
r
β
λ
S
p
a
r
si
t
y
P
r
o
p
o
r
t
i
o
n
A
c
t
i
v
a
t
i
o
n
F
u
n
c
t
i
o
n
A
c
t
i
v
a
t
i
o
n
F
u
n
c
t
i
o
n
S
A
E
1
70
-
1
0
0
-
2
0
0
-
2
3
0
.
0
0
1
0
.
0
5
En
c
o
d
e
r
=
l
o
g
si
g
D
e
c
o
d
e
r
=
p
u
r
e
l
i
n
S
o
f
t
max
S
A
E
2
70
-
2
0
0
-
1
0
0
-
2
3
0
.
0
0
1
0
.
0
5
S
A
E
3
70
-
5
0
0
-
2
0
0
-
2
3
0
.
0
0
1
0
.
0
5
S
A
E
4
70
-
6
0
0
-
2
0
0
-
2
3
0
.
0
0
1
0
.
0
5
S
A
E
5
70
-
1
0
0
-
1
0
0
-
2
1
0
.
0
1
0
.
0
5
S
A
E
6
70
-
2
0
0
-
1
0
0
-
2
4
0
.
0
0
1
0
.
0
6
5.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
As
in
ti
m
ated
ea
r
lier
,
th
e
e
m
p
h
asi
s
o
f
th
i
s
r
esear
ch
s
t
u
d
y
was
to
f
o
r
m
u
late
t
h
e
i
m
p
r
o
v
ed
E
E
G
b
ased
e
y
e
s
tate
clas
s
i
f
icatio
n
tas
k
u
s
in
g
t
w
o
d
i
f
f
er
e
n
t
p
ar
ad
ig
m
s
o
f
d
ee
p
n
eu
r
al
n
et
w
o
r
k
s
a
n
d
to
f
u
r
th
er
i
m
p
r
o
v
e
t
h
e
s
tate
o
f
t
h
e
ar
t p
er
f
o
r
m
an
ce
.
I
n
r
esp
o
n
s
e
to
co
n
d
u
c
t th
i
s
,
th
e
DB
N
an
d
SA
E
w
er
e
p
r
ef
er
r
e
d
ca
r
d
in
al
m
o
d
els as
d
ee
p
lear
n
in
g
ar
ch
itect
u
r
es.
F
o
r
th
is
p
u
r
p
o
s
e,
th
e
d
ata
s
o
u
r
ce
w
as
t
h
e
b
en
c
h
m
ar
k
d
ata
s
e
t
o
f
E
E
G
e
y
e
s
tate
class
i
f
icati
o
n
ti
m
e
s
er
ies
a
v
ai
lab
le
o
n
li
n
e
[
2
2
]
.
T
h
e
r
esu
lt
s
ac
h
ie
v
ed
h
er
e
ar
e
n
o
ticea
b
l
y
e
n
co
u
r
a
g
in
g
a
n
d
ev
id
en
ce
d
to
b
e
b
etter
in
m
o
s
t
o
f
th
e
ca
s
es
th
a
n
th
e
ea
r
lier
r
esu
lt
s
r
ep
o
r
ted
f
o
r
E
E
G
ey
e
s
tate
class
i
f
icat
io
n
p
r
o
b
lem
.
T
h
e
b
o
x
p
lo
t
is
d
r
a
w
n
s
ep
ar
ate
l
y
f
o
r
ea
c
h
cla
s
s
as
p
r
esen
ted
in
Fig
u
r
e
s
7
(
a)
an
d
(
b
)
i
n
o
r
d
er
to
s
t
u
d
y
th
e
d
is
tr
ib
u
t
io
n
al
ch
ar
ac
ter
i
s
ti
cs
o
f
a
1
4
-
ch
a
n
n
el
led
E
E
G
s
ig
n
al.
T
h
e
b
o
x
p
lo
t
is
a
u
s
e
f
u
l
w
a
y
to
v
is
u
alize
t
h
e
r
an
g
e
a
n
d
o
th
er
s
tatis
tical
c
h
a
r
ac
ter
is
tics
o
f
r
esp
o
n
s
e
v
ar
iab
les.
I
t
is
o
b
v
io
u
s
f
r
o
m
th
e
f
i
g
u
r
es
t
h
at
t
h
er
e
is
a
h
ig
h
er
v
a
lu
e
o
f
d
i
f
f
er
en
ce
w
h
e
n
th
e
s
ig
n
al
a
m
p
lit
u
d
e
f
r
o
m
ce
r
tain
s
e
n
s
o
r
s
is
co
m
p
ar
ed
.
I
t
is
ap
p
ar
en
t
f
r
o
m
v
ar
io
u
s
s
t
u
d
ies
t
h
at
a
d
ee
p
lear
n
i
n
g
a
lg
o
r
ith
m
g
r
ea
tl
y
r
ed
u
ce
s
t
h
e
er
r
o
r
o
n
test
s
ets
d
u
e
to
it
s
s
tr
o
n
g
u
n
s
u
p
er
v
is
ed
l
ea
r
n
i
n
g
p
h
ase.
A
p
ar
t
f
r
o
m
th
i
s
,
u
n
s
u
p
er
v
is
ed
p
r
etr
ain
i
n
g
b
eh
a
v
e
s
a
s
a
k
in
d
o
f
d
ata
d
ep
en
d
en
t
r
eg
u
l
ar
izer
to
av
o
id
o
v
er
f
itti
n
g
cr
i
ter
ia.
I
t
is
an
u
n
d
en
iab
le
f
ac
t
th
at
u
n
s
u
p
er
v
is
ed
p
r
etr
ain
in
g
b
ein
g
i
n
j
ec
ted
as
lo
ca
l
la
y
er
w
i
s
e
tr
ain
i
n
g
,
ex
tr
ac
ts
th
e
s
alie
n
t
i
n
f
o
r
m
atio
n
ab
o
u
t
t
h
e
i
n
p
u
t
d
is
tr
ib
u
tio
n
an
d
t
h
e
e
s
s
e
n
ce
o
f
n
o
n
li
n
ea
r
s
tr
u
ctu
r
e
p
r
ese
n
t
in
t
h
e
d
ata.
T
h
e
i
n
ter
p
r
etati
o
n
s
attai
n
ed
i
n
t
h
is
s
tu
d
y
ar
e
co
n
s
is
ten
t a
n
d
co
r
r
elate
d
w
it
h
th
e
ea
r
lier
s
t
u
d
ies co
n
d
u
cted
in
t
h
e
r
ea
l
m
o
f
d
ee
p
er
NN
ar
ch
itect
u
r
es.
Fig
u
r
e
7
(
a)
.
Statis
tical
R
a
n
g
e
o
f
E
lectr
o
d
e
S
ig
n
al
s
f
o
r
E
y
e
Op
en
,
(
b
)
Statis
tical
R
a
n
g
e
o
f
E
lectr
o
d
e
S
ig
n
al
s
f
o
r
E
y
e
C
lo
s
e
I
n
o
r
d
er
t
o
d
is
tin
g
u
i
s
h
th
e
b
es
t
d
ee
p
lear
n
in
g
m
o
d
el
s
f
r
o
m
a
m
o
n
g
t
h
o
s
e
i
m
p
le
m
en
ted
f
o
r
th
e
EEG
b
ased
ey
e
s
ta
te
class
i
f
icat
io
n
t
ask
,
th
e
co
m
p
ar
ati
v
e
an
a
l
y
s
is
w
a
s
p
er
f
o
r
m
ed
f
o
r
a
f
e
w
o
f
th
e
tr
ain
ed
m
o
d
els
,
s
u
m
m
ar
ized
in
T
ab
le
3
.
T
h
is
w
a
s
b
ased
o
n
th
e
clas
s
i
f
ier
’
s
p
er
f
o
r
m
an
ce
a
n
d
m
i
s
clas
s
i
f
icati
o
n
r
ate
as
av
er
ag
ed
er
r
o
r
o
n
test
s
ets.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
ea
ch
clas
s
i
f
ier
w
a
s
co
m
p
u
ted
u
s
i
n
g
t
h
e
m
o
s
t
co
m
m
o
n
l
y
u
s
ed
p
ar
am
eter
s
:
ac
c
u
r
ac
y
,
s
e
n
s
it
i
v
it
y
a
n
d
s
p
ec
i
v
it
y
.
A
cc
u
r
ac
y
was
ca
lcu
lated
as
e
x
p
r
ess
ed
b
y
th
e
(
7
)
.
Sen
s
iti
v
it
y
Evaluation Warning : The document was created with Spire.PDF for Python.
I
J
E
C
E
I
SS
N:
2
0
8
8
-
8708
E
E
G
B
a
s
ed
E
ye
S
ta
te
C
la
s
s
ifica
t
io
n
u
s
in
g
Dee
p
B
elief
N
etw
o
r
k
a
n
d
S
ta
ck
ed
A
u
to
E
n
c
o
d
er
(
S
a
n
a
m
N
a
r
ejo
)
3139
an
d
s
p
ec
if
ic
it
y
ar
e
s
ta
tis
tica
l
m
ea
s
u
r
es
o
f
t
h
e
p
er
f
o
r
m
a
n
ce
as
ex
p
r
ess
ed
b
y
t
h
e
eq
u
atio
n
s
(
8
)
an
d
(
9
)
r
esp
ec
tiv
el
y
.
A
cc
u
r
ac
y
=T
o
tal
No
.
o
f
Sa
m
p
l
es C
las
s
i
f
ied
C
o
r
r
ec
tl
y
/T
o
tal
No
.
o
f
Sa
m
p
les
(
7
)
Sen
s
iti
v
it
y
=T
P
/
T
P
+FN
(
8
)
Sp
ec
if
icit
y
=T
N/T
N+
FP
(
9
)
T
ab
le
3
.
P
ar
am
eter
Setti
n
g
s
f
o
r
P
r
etr
ain
in
g
P
h
ase
I
t
is
e
v
id
en
t
f
r
o
m
t
h
e
T
ab
le
3
th
at
b
o
th
t
h
e
DB
N
a
n
d
S
A
E
m
o
d
el
s
ac
co
m
p
li
s
h
ed
h
i
g
h
er
p
r
ed
ictio
n
ac
cu
r
ac
y
a
n
d
ar
e
al
m
o
s
t
co
m
p
atib
le.
I
n
co
m
p
ar
is
o
n
w
i
th
c
o
n
v
e
n
tio
n
al
Ma
c
h
in
e
lear
n
in
g
m
o
d
els
,
th
e
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
b
ased
o
n
u
n
s
u
p
er
v
i
s
ed
p
r
e
tr
ain
in
g
an
d
s
u
p
er
v
i
s
ed
f
i
n
e
tu
n
i
n
g
p
r
o
v
id
es
a
b
etter
g
en
er
aliza
tio
n
t
h
an
t
h
e
s
u
p
er
v
is
ed
lear
n
in
g
m
o
d
el
s
w
it
h
r
an
d
o
m
i
n
itial
izatio
n
s
.
T
h
e
r
an
d
o
m
i
n
itializat
io
n
s
f
o
r
th
e
p
ar
a
m
eter
s
o
f
a
n
e
u
r
al
n
et
w
o
r
k
,
d
ec
lar
e
th
e
p
ar
a
m
eter
s
in
r
eg
io
n
s
o
f
t
h
e
p
ar
am
e
ter
s
p
ac
e
th
at
g
e
n
er
alize
p
o
o
r
ly
.
As
m
e
n
tio
n
ed
ea
r
lier
,
o
n
l
y
a
f
e
w
d
ev
elo
p
ed
m
o
d
els
ar
e
s
elec
ted
an
d
in
clu
d
ed
in
th
is
ar
ticle
f
o
r
d
is
cu
s
s
io
n
.
I
t
is
o
b
v
io
u
s
f
r
o
m
th
e
e
x
p
er
i
m
e
n
tal
co
n
s
eq
u
e
n
c
es
t
h
at
t
h
e
q
u
a
n
ti
tati
v
e
a
n
al
y
s
is
is
al
m
o
s
t
s
i
m
ilar
w
it
h
o
n
l
y
s
lig
h
t d
i
f
f
er
e
n
ce
s
.
On
e
o
f
t
h
e
s
u
p
er
lati
v
e
S
A
E
m
o
d
el
s
i.e
S
A
E
2
in
cl
in
ed
to
p
r
o
v
id
e
th
e
ac
c
u
r
ac
y
o
f
9
8
.
9
%
,
w
h
ich
is
ev
id
en
tl
y
h
i
g
h
er
t
h
a
n
t
h
e
r
est
o
f
t
h
e
o
th
er
m
o
d
els
p
er
f
o
r
m
an
ce
.
I
t
also
ap
p
ea
r
s
t
h
at
t
h
e
ap
p
lied
ap
p
r
o
ac
h
o
f
d
ev
elo
p
in
g
S
A
E
2
o
u
tp
er
f
o
r
m
s
t
h
e
DB
N
m
o
d
el
w
it
h
t
h
e
d
if
f
er
e
n
ce
o
f
1
.
1
%
e
x
ce
p
t
o
f
th
e
t
w
o
ca
s
e
s
.
T
h
e
co
n
s
tr
ain
ts
i
n
S
A
E
m
o
d
els
ex
p
er
im
e
n
t
ap
p
lied
to
th
e
o
p
ti
m
izatio
n
p
r
o
ce
d
u
r
e
en
f
o
r
ce
d
t
h
e
m
o
d
el
to
ca
p
tu
r
e
n
o
t
o
n
l
y
t
h
e
i
n
p
u
t
to
tar
g
et
d
ep
en
d
en
c
y
,
b
u
t
also
t
h
e
s
tati
s
tical
r
eg
u
lar
ities
o
f
t
h
e
i
n
p
u
t
d
is
tr
ib
u
tio
n
w
h
ic
h
r
esu
lted
in
p
r
o
m
i
n
e
n
tl
y
b
etter
p
er
f
o
r
m
a
n
ce
t
h
an
DB
N.
T
h
e
tw
o
e
x
ce
p
tio
n
al
ca
s
e
s
as
m
e
n
tio
n
ed
ea
r
lier
in
cl
u
d
e,
m
o
d
el
S
A
E
5
an
d
S
A
E
6
.
T
h
e
tu
n
i
n
g
p
ar
am
eter
s
in
th
e
s
e
m
o
d
els,
co
n
s
is
tin
g
o
f
co
ef
f
icie
n
ts
f
o
r
s
p
ar
s
it
y
an
d
L
2
r
eg
u
lar
iza
tio
n
,
ar
e
d
if
f
er
en
t
f
r
o
m
th
e
o
t
h
er
S
A
E
m
o
d
els
as
p
r
esen
ted
in
T
ab
le
2
.
C
o
n
tex
t
u
a
ll
y
,
th
is
r
esu
lted
in
co
n
s
id
er
ab
ly
lo
w
er
p
er
f
o
r
m
a
n
ce
as
co
m
p
ar
ed
to
th
e
o
th
er
d
ev
elo
p
ed
SA
E
m
o
d
els.
O
n
th
e
o
t
h
er
h
a
n
d
,
to
en
f
o
r
ce
th
e
clas
s
i
f
ier
to
p
r
ed
ict
w
ell,
th
e
s
elec
tio
n
o
f
s
u
f
f
icie
n
t
h
id
d
en
la
y
er
d
i
m
en
s
io
n
i
n
d
ee
p
er
ar
ch
itectu
r
e
is
an
o
t
h
er
cr
itical
p
o
in
t
to
co
n
s
id
er
.
T
o
av
o
id
elec
tin
g
t
h
e
h
id
d
en
la
y
er
s
izes
ar
b
itra
r
il
y
,
t
h
e
co
m
p
r
e
h
e
n
s
i
v
e
s
tr
ate
g
y
to
f
o
llo
w
i
s
w
ell
e
x
p
lain
ed
in
[
3
3
]
.
I
n
ad
d
itio
n
to
th
is
s
o
m
e
ef
f
o
r
t
m
u
s
t
b
e
lo
g
icall
y
d
o
n
e
w
h
ile
in
v
es
tig
a
tin
g
t
h
e
s
e
v
er
al
p
o
s
s
ib
ilit
ies
f
o
r
th
e
d
i
m
e
n
s
io
n
o
f
h
id
d
en
la
y
er
s
.
D
u
r
i
n
g
t
h
e
tr
ai
n
i
n
g
p
r
o
ce
d
u
r
e,
it
w
as
also
f
o
u
n
d
th
at
in
t
h
e
d
ee
p
er
h
ier
ar
ch
ical
s
tr
u
ct
u
r
e
o
f
t
h
e
n
et
w
o
r
k
,
w
h
e
n
p
er
f
o
r
m
i
n
g
t
h
e
R
ep
r
esen
tat
io
n
L
ea
r
n
i
n
g
o
r
f
ea
tu
r
e
ex
tr
ac
tio
n
,
th
e
m
o
d
el
lear
n
s
t
h
e
s
i
m
p
le
co
n
c
ep
ts
at
lo
w
er
la
y
er
lev
el
s
an
d
m
o
r
e
ab
s
tr
ac
t
co
n
ce
p
t
ar
e
lear
n
t
b
y
co
m
p
o
s
i
n
g
th
e
m
a
t h
i
g
h
er
to
p
lay
er
s
o
f
th
e
n
et
w
o
r
k
.
T
ab
le
4
.
C
o
m
p
ar
is
o
n
w
i
th
P
r
ev
io
u
s
R
esear
c
h
M
o
d
e
l
s
A
c
c
u
r
a
c
y
Er
r
o
r
R
a
t
e
P
r
o
p
o
se
d
D
L
A
mo
d
e
l
,
S
A
E
2
9
8
.
9
%
1
.
1
%
K
*
+
R
R
F
[
2
4
]
9
7
.
4
%
2
.
6
%
K
*
[
2
1
]
9
7
.
3
%
2
.
7
%
N
e
u
r
o
-
F
u
z
z
y
[
2
5
]
9
6
%
4
.
0
%
T
ab
le
4
h
ig
h
li
g
h
ts
th
e
ac
co
m
p
lis
h
m
e
n
t
o
f
o
u
r
b
est
class
i
f
i
er
i.e
SA
E
2
.
T
h
e
r
esear
ch
er
s
in
[
2
1
],
d
em
o
n
s
tr
ated
t
h
e
p
er
f
o
r
m
a
n
c
e
an
al
y
s
i
s
o
f
4
2
clas
s
i
f
ier
s
o
n
t
h
e
p
r
o
b
le
m
o
f
E
y
e
s
tate
cl
ass
i
f
icatio
n
t
h
r
o
u
g
h
E
E
G
s
ig
n
al.
Am
o
n
g
all
o
f
th
o
s
e
tr
ain
ed
clas
s
i
f
ier
s
,
K
*
was
f
o
u
n
d
to
b
e
t
h
e
m
o
s
t
p
r
o
m
is
in
g
o
n
e
w
it
h
a
n
ac
cu
r
ac
y
o
f
9
7
.
3
0
%.
Ho
w
e
v
e
r
,
af
ter
w
ar
d
s
t
h
e
au
t
h
o
r
s
in
[
2
4
]
ex
ten
d
ed
th
e
a
f
o
r
em
en
tio
n
ed
w
o
r
k
b
y
p
r
o
p
o
s
in
g
a
n
en
s
e
m
b
le
cla
s
s
i
f
ier
w
it
h
a
b
etter
p
er
f
o
r
m
a
n
c
e
o
f
ar
o
u
n
d
9
7
.
4
0
%
ac
cu
r
a
c
y
.
Me
an
w
h
ile
t
h
e
D
e
e
p
A
r
c
h
i
t
e
c
t
u
r
e
s
S
p
e
c
i
t
i
v
i
t
y
S
e
n
si
t
i
v
i
t
y
A
c
c
u
r
a
c
y
Er
r
o
r
r
a
t
e
D
B
N
0
.
9
9
0
.
8
9
9
7
.
1
%
2
.
9
%
S
A
E
1
0
.
9
5
0
.
9
8
9
7
.
6
%
2
.
4
%
S
A
E
2
0
.
9
6
0
.
9
9
9
8
.
9
%
1
.
1
%
S
A
E
3
0
.
9
4
0
.
9
9
9
7
.
8
%
2
.
2
%
S
A
E
4
0
.
9
4
0
.
9
9
9
7
.
8
%
2
.
2
%
S
A
E
5
0
.
8
0
0
.
9
7
9
2
.
9
%
7
.
1
%
S
A
E
6
0
.
7
4
0
.
9
9
9
1
.
9
%
8
.
8
%
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
J
E
C
E
Vo
l.
6
,
No
.
6
,
Dec
em
b
er
2
0
1
6
:
31
3
1
–
31
41
3140
r
esear
ch
er
s
tr
ai
n
ed
a
Neu
r
o
-
F
u
zz
y
c
lass
if
ier
w
h
ic
h
g
av
e
th
e
p
er
f
o
r
m
an
ce
ac
c
u
r
ac
y
ar
o
u
n
d
9
6
%.
I
n
c
o
n
tr
ast
to
th
e
ab
o
v
e
m
e
n
tio
n
ed
s
t
u
d
ies,
th
e
p
r
o
p
o
s
ed
SA
E
2
m
o
d
el
tr
ain
ed
w
it
h
th
e
s
p
ec
if
ied
p
r
o
ce
d
u
r
e
an
d
p
ar
am
e
ter
s
etti
n
g
s
o
u
tp
er
f
o
r
m
s
o
th
er
m
o
d
els
,
w
it
h
t
h
e
o
b
tain
ed
er
r
o
r
r
ate
o
f
o
n
l
y
1
.
1
%
o
n
t
h
e
te
s
t
s
e
t
.
Hig
h
er
ac
cu
r
ac
y
,
ar
o
u
n
d
9
8
.
9
%
,
is
ac
co
m
p
li
s
h
ed
o
n
th
e
test
s
a
m
p
les
w
h
ic
h
d
em
o
n
s
tr
ate
s
i
m
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
o
v
er
[
2
1
]
,
[
2
4
]
an
d
[
2
5
]
.
W
e
h
y
p
o
th
esi
ze
th
at
th
er
e
i
s
a
p
o
s
s
ib
ilit
y
o
f
ac
h
iev
i
n
g
m
u
ch
b
etter
r
e
s
u
lt
s
t
h
r
o
u
g
h
d
ee
p
ar
ch
itect
u
r
es
b
y
i
m
p
le
m
en
tin
g
De
n
o
is
i
n
g
a
n
d
C
o
n
tr
asti
v
e
Au
to
en
co
d
er
s
.
T
h
is
i
n
d
icat
es
a
d
ir
ec
tio
n
f
o
r
p
o
s
s
ib
le
f
u
tu
r
e
w
o
r
k
to
f
u
r
t
h
er
ex
ten
d
t
h
is
s
tu
d
y
.
6.
CO
NCLU
SI
O
N
T
h
e
d
ir
ec
te
d
r
esear
ch
f
o
cu
s
es
o
n
t
w
o
ap
p
r
o
ac
h
es
o
f
DL
A
f
o
r
E
E
G
E
y
e
s
tate
clas
s
i
f
icati
o
n
task
as
e
y
e
o
p
en
o
r
clo
s
e.
A
lt
h
o
u
g
h
t
h
is
h
as
b
ee
n
d
o
n
e
ea
r
lier
i
n
a
n
u
m
b
er
o
f
w
a
y
s
w
it
h
d
if
f
er
e
n
t
Ma
ch
i
n
e
lear
n
i
n
g
class
i
f
ier
s
,
alb
eit
o
u
r
en
d
ea
v
o
r
i
m
p
r
o
v
es
o
n
t
h
e
clas
s
if
ica
tio
n
ac
c
u
r
ac
y
b
y
ap
p
l
y
in
g
d
e
ep
ar
ch
itect
u
r
es
a
s
class
i
f
ier
s
.
Sev
er
al
n
u
m
b
er
s
o
f
m
o
d
els
w
er
e
tr
ai
n
ed
f
o
r
t
h
e
s
tu
d
y
a
n
d
s
o
m
e
o
f
th
e
m
a
r
e
p
r
esen
ted
i
n
t
h
i
s
ar
ticle
f
o
r
s
ig
n
if
ican
t
d
is
c
u
s
s
i
o
n
.
A
s
tr
ai
g
h
tf
o
r
w
ar
d
an
al
y
s
i
s
p
r
o
ce
d
u
r
e
h
as
b
ee
n
co
n
d
u
cted
o
n
th
e
b
asis
o
f
d
if
f
er
e
n
t
p
e
r
f
o
r
m
a
n
ce
m
ea
s
u
r
e
m
en
t
m
et
r
ic
s
.
Ou
r
tr
ai
n
ed
m
o
d
els
w
it
h
th
e
h
y
p
er
p
ar
a
m
eter
s
ett
in
g
s
a
n
d
ap
p
lied
h
id
d
en
la
y
er
d
i
m
en
s
i
o
n
s
ac
h
ie
v
ed
s
tr
i
k
i
n
g
p
er
f
o
r
m
an
ce
s
as
co
m
p
ar
ed
to
th
e
ex
is
t
in
g
o
n
es.
Mo
s
t
n
o
tab
l
y
,
to
o
u
r
k
n
o
w
led
g
e,
t
h
i
s
is
t
h
e
f
ir
s
t st
u
d
y
o
f
E
E
G
b
as
ed
ey
e
s
tate
clas
s
i
f
icatio
n
ta
s
k
s
u
s
in
g
d
ee
p
Neu
r
al
Net
w
o
r
k
m
o
d
els,
i.e
.
,
DB
N
an
d
S
A
E
.
T
h
e
s
tu
d
y
co
n
d
u
ct
ed
h
er
e
ca
n
b
e
ap
p
lied
in
d
if
f
er
en
t
ap
p
licatio
n
d
o
m
ai
n
s
o
f
B
C
I
an
d
f
o
r
th
e
id
en
tif
icat
io
n
o
f
h
u
m
a
n
c
o
g
n
itio
n
.
Mo
s
t
i
m
p
er
ativ
e
l
y
w
h
er
e
th
e
ar
tif
ac
t
s
g
en
er
ated
b
y
e
y
e
m
o
v
e
m
e
n
ts
o
r
ey
e
b
lin
k
s
ar
e
cr
u
cial,
th
er
e
is
a
n
ee
d
f
o
r
th
e
m
to
b
e
id
en
ti
f
ied
an
d
f
u
r
th
er
r
e
m
o
v
ed
.
ACK
NO
WL
E
D
G
E
M
E
NT
S
T
h
is
r
esear
ch
ac
ti
v
it
y
w
a
s
p
a
r
tl
y
f
u
n
d
ed
b
y
I
talia
n
MI
U
R
OP
L
ON
p
r
o
j
ec
t
an
d
s
u
p
p
o
r
ted
b
y
t
h
e
P
o
litecn
ico
o
f
T
u
r
in
NE
C
NE
UR
ONI
C
A
lab
o
r
ato
r
y
.
C
o
m
p
u
tatio
n
al
r
e
s
o
u
r
ce
s
w
er
e
p
ar
tl
y
p
r
o
v
id
ed
b
y
HP
C
@
P
OL
I
T
O,
(
h
ttp
://
www
.
h
p
c.
p
o
lito
.
it
).
RE
F
E
R
E
NC
E
S
[1
]
Ka
p
lan
A
.
Y
.
a
n
d
S
h
ish
k
in
S
.
L.
,
“
A
p
p
li
c
a
ti
o
n
o
f
th
e
c
h
a
n
g
e
-
p
o
i
n
t
a
n
a
ly
sis
to
th
e
in
v
e
stig
a
ti
o
n
o
f
th
e
b
ra
in
’s
e
lec
tri
c
a
l
a
c
ti
v
it
y
,”
i
n
No
n
-
p
a
ra
m
e
tric sta
ti
stica
l
d
ia
g
n
o
sis
,
S
p
ri
n
g
e
r
Ne
th
e
rlan
d
s
,
p
p
.
3
3
3
-
3
8
8
,
2
0
0
0
.
[2
]
T
e
p
lan
M
.
,
“
F
u
n
d
a
m
e
n
tals o
f
EE
G
m
e
a
su
re
m
e
n
t
,”
M
e
a
su
re
me
n
t
s
c
ien
c
e
re
v
iew
,
v
o
l/
issu
e
:
2
(2
)
,
p
p
.
1
-
1
,
2
0
0
2
.
[3
]
J.
Zh
a
n
g
a
n
d
V
.
L
.
P
a
tel
,
“
Distrib
u
te
d
c
o
g
n
it
io
n
,
re
p
re
se
n
tati
o
n
,
a
n
d
a
ff
o
rd
a
n
c
e
,”
Pra
g
ma
ti
c
s
&
Co
g
n
it
io
n
,
v
o
l/
issu
e
:
14
(
2
),
p
p
.
3
3
3
-
3
4
1
,
2
0
0
6
.
[4
]
Ho
ffm
a
n
n
S
.
a
n
d
F
a
lk
e
n
ste
in
M
.
,
“
T
h
e
c
o
rre
c
ti
o
n
o
f
e
y
e
b
li
n
k
a
rte
fa
c
ts
in
th
e
EE
G
:
a
c
o
m
p
a
riso
n
o
f
tw
o
p
ro
m
in
e
n
t
m
e
th
o
d
s.
PL
o
S
On
e
,
v
o
l/
issu
e
:
3
(
8
)
,
p
p
.
e
3
0
0
4
,
2
0
0
8
.
[5
]
L
.
Cu
n
,
e
t
a
l.
,
“
De
e
p
lea
rn
in
g
,”
N
a
tu
re
,
v
o
l/
issu
e
:
5
2
1
(
7
5
5
3
),
p
p
.
4
3
6
-
4
4
4
,
2
0
1
5
.
[6
]
Be
n
g
io
Y
.
,
e
t
a
l.
,
“
Re
p
re
se
n
tatio
n
lea
rn
in
g
:
A
re
v
ie
w
a
n
d
n
e
w p
e
rs
p
e
c
ti
v
e
s
,”
IEE
E
tra
n
sa
c
ti
o
n
s o
n
p
a
tt
e
rn
a
n
a
lys
is
a
n
d
ma
c
h
i
n
e
i
n
telli
g
e
n
c
e
,
v
o
l/
iss
u
e
:
3
5
(8
)
,
p
p
.
1
7
9
8
-
8
2
8
,
2
0
1
3
.
[7
]
I
.
G
o
o
d
f
e
ll
o
w
,
e
t
a
l.
,
“
De
e
p
L
e
a
rn
in
g
,
”
Bo
o
k
in
p
re
p
a
ra
ti
o
n
f
o
r
M
I
T
P
re
ss
,
2
0
1
6
.
[8
]
P
o
lat
K
.
a
n
d
G
ü
n
e
ş
S
.
,
“
Clas
sif
ic
a
ti
o
n
o
f
e
p
il
e
p
t
if
o
rm
EE
G
u
sin
g
a
h
y
b
rid
s
y
ste
m
b
a
se
d
o
n
d
e
c
isio
n
tree
c
las
si
f
ier
a
n
d
f
a
st F
o
u
rier
tran
sf
o
rm
,”
Ap
p
li
e
d
M
a
th
e
ma
ti
c
s
a
n
d
Co
m
p
u
t
a
ti
o
n
,
v
o
l/
issu
e
:
1
8
7
(2
)
,
p
p
.
1
0
1
7
-
26
,
2
0
0
7
.
[9
]
S
u
laim
a
n
N
.
,
e
t
a
l
.
,
“
No
v
e
l
m
e
th
o
d
s
f
o
r
stre
ss
f
e
a
tu
re
s
id
e
n
ti
f
ica
ti
o
n
u
sin
g
EE
G
sig
n
a
ls
,”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
S
imu
l
a
ti
o
n
:
S
y
ste
ms
,
S
c
ien
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
,
v
o
l/
issu
e
:
1
2
(
1
)
,
p
p
.
27
-
33
,
2
0
1
1
.
[1
0
]
Ye
o
M
.
V
.
,
e
t
a
l
.
,
“
Ca
n
S
VM
b
e
u
se
d
f
o
r
a
u
to
m
a
ti
c
EE
G
d
e
tec
ti
o
n
o
f
d
ro
w
sin
e
ss
d
u
ri
n
g
c
a
r
d
r
iv
in
g
?
”
S
a
fety
S
c
ien
c
e
,
v
o
l/
issu
e
:
4
7
(
1
)
,
p
p
.
1
1
5
-
24
,
2
0
0
9
.
[1
1
]
Estév
e
z
P
.
A
.
,
e
t
a
l.
,
“
P
o
ly
so
m
n
o
g
ra
p
h
ic
p
a
tt
e
rn
re
c
o
g
n
it
i
o
n
f
o
r
a
u
to
m
a
ted
c
las
si
f
ica
ti
o
n
o
f
sle
e
p
-
w
a
k
in
g
sta
tes
in
in
f
a
n
ts
,”
M
e
d
ica
l
a
n
d
B
io
l
o
g
ic
a
l
En
g
i
n
e
e
rin
g
a
n
d
C
o
mp
u
ti
n
g
,
v
o
l
/
issu
e
:
4
0
(
1
)
,
p
p
.
1
0
5
-
13
,
2
0
0
2
.
[1
2
]
Ch
e
n
g
E
.
,
e
t
a
l
.
,
“
Ey
e
sta
te
d
e
te
c
ti
o
n
i
n
f
a
c
ial
im
a
g
e
b
a
se
d
o
n
li
n
e
a
r
p
re
d
icti
o
n
e
rr
o
r
o
f
wa
v
e
l
e
t
c
o
e
ff
icie
n
ts
,”
i
n
Ro
b
o
ti
c
s a
n
d
Bi
o
mime
ti
c
s,
2
0
0
8
.
ROBI
O 2
0
0
8
.
IEE
E
In
ter
n
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
,
p
p
.
1
3
8
8
-
1
3
9
2
,
2
0
0
9
.
[1
3
]
G
.
G
r
a
tt
o
n
,
e
t
a
l.
,
“
A
n
e
w
m
e
th
o
d
f
o
r
o
ff
-
li
n
e
re
m
o
v
a
l
o
f
o
c
u
la
r
a
rti
fa
c
t
,”
El
e
c
tro
e
n
c
e
p
h
a
lo
g
r
a
p
h
y
a
n
d
c
li
n
ica
l
n
e
u
ro
p
h
y
sio
l
o
g
y
,
v
o
l/
issu
e
:
55
(
4
),
p
p
.
4
6
8
-
4
8
4
,
1
9
8
3
.
[1
4
]
S
h
o
k
e
r
L
.
,
e
t
a
l.
,
“
Re
m
o
v
a
l
o
f
e
y
e
b
li
n
k
in
g
a
rti
f
a
c
ts
f
ro
m
EE
G
in
c
o
rp
o
ra
ti
n
g
a
n
e
w
c
o
n
stra
in
e
d
BS
S
a
lg
o
rit
h
m
,”
i
n
S
e
n
so
r A
rr
a
y
a
n
d
M
u
lt
ich
a
n
n
e
l
S
ig
n
a
l
Pr
o
c
e
ss
in
g
W
o
rk
sh
o
p
Pr
o
c
e
e
d
in
g
s
,
p
p
.
1
7
7
-
1
8
1
,
2
0
0
4
.
[1
5
]
Krish
n
a
v
e
n
i
V
.
,
e
t
a
l
.
,
“
Re
m
o
v
a
l
o
f
o
c
u
lar
a
rti
f
a
c
ts
f
ro
m
EE
G
u
sin
g
a
d
a
p
t
iv
e
th
re
s
h
o
ld
i
n
g
o
f
w
a
v
e
le
t
c
o
e
ff
icie
n
ts
,”
J
o
u
rn
a
l
o
f
Ne
u
ra
l
En
g
i
n
e
e
rin
g
,
v
o
l/
issu
e
:
3
(4
)
,
p
p
.
3
3
8
,
2
0
0
6
.
[1
6
]
S
in
g
la
R
.
,
e
t
a
l.
,
“
Co
m
p
a
riso
n
o
f
S
VM
a
n
d
A
NN
f
o
r
c
las
si
f
ica
ti
o
n
o
f
e
y
e
e
v
e
n
ts
in
EE
G
,”
J
o
u
rn
a
l
o
f
B
io
me
d
ica
l
S
c
ien
c
e
a
n
d
E
n
g
i
n
e
e
rin
g
,
v
o
l/
issu
e
:
4
(0
1
)
,
p
p
.
62
,
2
0
1
1
.
[1
7
]
Be
lk
a
c
e
m
A
.
N
.
,
e
t
a
l.
,
“
Clas
sif
ica
ti
o
n
o
f
f
o
u
r
e
y
e
d
irec
ti
o
n
s
f
ro
m
EE
G
si
g
n
a
ls
f
o
r
e
y
e
-
m
o
v
e
m
e
n
t
-
b
a
se
d
c
o
m
m
u
n
ica
ti
o
n
sy
ste
m
s
,”
lif
e,
v
o
l.
1
,
p
p
.
3
,
2
0
1
4
.
Evaluation Warning : The document was created with Spire.PDF for Python.