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.
11
,
No
.
4
,
A
u
g
u
s
t
2021
,
p
p
.
3
5
2
9
~
3
5
3
8
I
SS
N:
2
0
8
8
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
1
1
i
4
.
pp
3
5
2
9
-
3
5
3
8
3529
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ij
ec
e.
ia
esco
r
e.
co
m
Alert
N
et
:
Dee
p co
nv
o
lutiona
l
-
recur
rent
n
eura
l ne
tw
o
rk
m
o
del
for driv
ing
alert
n
ess
det
ec
tion
P.
C.
Nis
s
i
m
a
g
o
ud
a
r,
A.
V.
Na
nd
i
,
Aa
k
a
n
ks
ha
P
a
t
il,
G
iree
s
ha
H
.
M.
S
c
h
o
o
l
o
f
El
e
c
tro
n
ics
a
n
d
C
o
m
m
u
n
ica
ti
o
n
E
n
g
in
e
e
rin
g
,
KL
E
T
e
c
h
n
o
l
o
g
ica
l
Un
iv
e
rsit
y
,
Hu
b
b
a
ll
i
,
K
a
rn
a
tak
a
,
In
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
J
u
l 1
0
,
2
0
2
0
R
ev
i
s
ed
Dec
2
3
,
2
0
2
0
A
cc
ep
ted
J
an
1
9
,
2
0
2
1
Dro
w
s
y
d
riv
in
g
is
o
n
e
o
f
th
e
m
a
jo
r
p
ro
b
lem
s
lea
d
in
g
to
m
a
n
y
ro
a
d
a
c
c
id
e
n
ts.
El
e
c
tro
e
n
c
e
p
h
a
lo
g
ra
p
h
y
(EE
G
)
i
s
o
n
e
o
f
th
e
m
o
st
re
li
a
b
le
so
u
rc
e
s
to
d
e
tec
t
sle
e
p
o
n
-
se
t
w
h
il
e
d
riv
i
n
g
a
s
th
e
re
is
th
e
d
irec
t
i
n
v
o
lv
e
m
e
n
t
o
f
b
io
lo
g
ica
l
sig
n
a
ls.
T
h
e
p
re
se
n
t
w
o
rk
f
o
c
u
se
s
o
n
u
sin
g
th
e
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
a
rc
h
it
e
c
tu
re
,
b
u
il
t
u
s
in
g
Re
sN
e
ts
a
n
d
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
b
a
se
d
se
q
u
e
n
c
e
to
se
q
u
e
n
c
e
m
o
d
e
ls
w
it
h
a
tt
e
n
ti
o
n
-
d
e
c
o
d
e
r.
T
h
e
m
o
d
e
l
is
b
u
il
t
to
re
d
u
c
e
th
e
c
o
m
p
lex
c
o
m
p
u
tatio
n
s
re
q
u
ired
f
o
r
f
e
a
tu
re
e
x
tra
c
ti
o
n
.
T
h
e
m
o
d
e
l
lea
rn
s
d
e
e
p
e
r
w
it
h
m
in
i
m
ize
d
lo
ss
a
n
d
train
in
g
e
rro
r
.
T
h
e
o
u
tp
u
t
o
f
R
e
sN
e
ts,
th
e
f
e
a
tu
re
s
a
re
in
p
u
t
to
e
n
c
o
d
e
r
-
d
e
c
o
d
e
r
b
a
se
d
se
q
u
e
n
c
e
to
se
q
u
e
n
c
e
m
o
d
e
ls,
b
u
il
t
u
sin
g
Bi
-
d
irec
ti
o
n
a
l
lo
n
g
-
sh
o
rt
m
e
m
o
rie
s.
S
e
q
u
e
n
c
e
to
S
e
q
u
e
n
c
e
m
o
d
e
l
lea
rn
s
th
e
c
o
m
p
lex
f
e
a
tu
re
s
o
f
th
e
sig
n
a
l
a
n
d
a
n
a
ly
z
e
th
e
o
u
tp
u
t
o
f
p
a
st
a
n
d
f
u
tu
re
sta
tes
si
m
u
l
tan
e
o
u
sly
f
o
r
c
las
si
f
ica
ti
o
n
o
f
d
ro
w
s
y
/slee
p
st
a
g
e
-
1
a
n
d
a
lert
sta
g
e
s.
A
lso
,
to
o
v
e
rc
o
m
e
th
e
u
n
e
q
u
a
l
d
istri
b
u
ti
o
n
(c
las
s
-
im
b
a
lan
c
e
)
d
a
ta
p
ro
b
lem
p
re
se
n
t
in
th
e
d
a
t
a
se
ts,
th
e
p
ro
p
o
se
d
l
o
ss
f
u
n
c
ti
o
n
s
h
e
lp
i
n
a
c
h
iev
in
g
th
e
id
e
n
ti
c
a
l
e
rro
r
f
o
r
b
o
th
m
a
jo
rit
y
a
n
d
m
in
o
rit
y
c
la
ss
e
s
d
u
ri
n
g
th
e
ra
in
in
g
o
f
th
e
n
e
tw
o
rk
f
o
r
e
a
c
h
s
lee
p
sta
g
e
.
T
h
e
m
o
d
e
l
p
ro
v
id
e
s
a
n
o
v
e
ra
ll
-
a
c
c
u
ra
c
y
o
f
8
7
.
9
2
%
a
n
d
8
7
.
0
5
%
,
a
m
a
c
ro
-
F1
-
c
o
re
o
f
7
8
.
0
6
%
,
a
n
d
7
9
.
6
6
%
a
n
d
Co
h
e
n
'
s
-
k
a
p
p
a
sc
o
re
o
f
0
.
7
8
a
n
d
0
.
7
9
f
o
r
th
e
S
lee
p
-
EDF
2
0
1
3
a
n
d
2
0
1
8
d
a
ta se
ts
re
sp
e
c
ti
v
e
l
y
.
K
ey
w
o
r
d
s
:
A
tte
n
tio
n
n
et
w
o
r
k
B
id
ir
ec
tio
n
al
L
ST
M
C
las
s
i
m
b
ala
n
ce
E
lectr
o
en
ce
p
h
alo
g
r
a
m
L
o
s
s
f
u
n
c
tio
n
s
R
esNet
s
Seq
u
en
ce
m
o
d
els
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
:
P
.
C
.
Nis
s
i
m
a
g
o
u
d
ar
Sch
o
o
l o
f
E
lectr
o
n
ics a
n
d
C
o
m
m
u
n
icatio
n
E
n
g
i
n
ee
r
in
g
KL
E
T
ec
h
n
o
lo
g
ica
l U
n
i
v
er
s
i
t
y
Hu
b
b
lli,
Kar
n
ata
k
a,
I
n
d
ia
-
580031
E
m
ail:
p
cn
g
o
u
d
ar
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
T
h
e
p
h
y
s
io
lo
g
ical
s
i
g
n
al
s
l
ik
e
E
E
G,
E
C
G,
E
MG
,
a
n
d
E
OG
ar
e
th
e
p
r
o
v
e
n
m
ea
s
u
r
es
f
o
r
t
h
e
a
n
al
y
s
i
s
an
d
d
etec
tio
n
o
f
ab
n
o
r
m
a
liti
e
s
in
t
h
e
ar
ea
o
f
cli
n
ical
d
iag
n
o
s
is
,
b
u
t
r
ec
en
tl
y
t
h
ese
s
ig
n
al
s
also
h
a
v
e
f
o
u
n
d
th
eir
u
s
a
g
e
i
n
m
a
n
y
o
th
er
a
p
p
licatio
n
s
.
Di
v
er
aler
t
n
es
s
d
etec
tio
n
is
o
n
e
s
u
ch
ap
p
licatio
n
,
w
h
er
e
t
h
ese
p
h
y
s
io
lo
g
ical
s
ig
n
al
s
ca
n
als
o
b
e
u
s
ed
.
Am
o
n
g
s
t
t
h
e
v
ar
i
o
u
s
p
h
y
s
io
lo
g
ical
s
ig
n
al
s
,
E
l
ec
tr
o
en
ce
p
h
alo
g
r
a
m
(
E
E
G)
w
h
ic
h
v
ar
ies
i
n
f
r
eq
u
en
c
y
a
n
d
ti
m
e
-
in
v
ar
ian
t
f
ea
t
u
r
es
is
f
o
u
n
d
to
b
e
a
d
ir
ec
t
in
d
icato
r
o
f
d
r
iv
er
’
s
aler
tn
es
s
lev
el.
T
h
e
s
tan
d
ar
d
1
0
-
2
0
s
y
s
te
m
is
u
s
ed
to
co
llect
E
E
G
s
ig
n
als
f
r
o
m
d
if
f
er
en
t
lo
ca
tio
n
s
o
f
t
h
e
s
ca
lp
u
s
i
n
g
th
e
elec
tr
o
d
es
[
1
].
T
h
e
d
if
f
er
en
t
f
r
eq
u
e
n
c
y
co
m
p
o
n
en
t
s
alo
n
g
w
it
h
th
e
r
elate
d
a
m
p
litu
d
e
le
v
els
o
b
s
er
v
ed
o
v
er
th
e
ti
m
e
r
ep
r
es
en
t
th
e
c
o
n
d
itio
n
o
f
t
h
e
b
r
ain
[
2
].
T
h
er
e
ar
e
5
s
tag
es
o
f
s
lee
p
in
w
h
ic
h
t
h
e
b
r
ain
p
r
o
d
u
ce
s
d
is
tin
g
u
i
s
h
ab
le
elec
tr
ic
p
atter
n
s
w
h
ic
h
h
elp
in
t
h
e
class
i
f
icat
io
n
o
f
s
ta
g
es
.
T
h
e
P
SG
s
ig
n
als
ar
e
co
llected
f
r
o
m
a
s
u
b
j
ec
t
d
u
r
in
g
th
e
e
n
tire
n
i
g
h
t
o
f
s
leep
an
d
ar
e
m
a
n
u
a
ll
y
s
co
r
ed
b
y
s
leep
ex
p
er
ts
i
n
t
o
d
if
f
er
e
n
t
s
leep
s
ta
g
e
s
b
y
v
i
s
u
all
y
a
n
al
y
zi
n
g
t
h
e
s
i
g
n
al
s
f
o
r
a
s
p
ec
if
ic
ti
m
e
f
r
a
m
e
[
3
]
.
T
h
e
cr
iter
ia
f
o
r
s
leep
s
tag
e
s
co
r
in
g
ar
e
p
r
o
p
o
s
ed
in
r
ec
h
ts
c
h
af
f
e
n
an
d
k
ales
(
R
K)
[
4
]
m
a
n
u
al
w
h
ic
h
w
a
s
f
u
r
t
h
e
r
d
ev
elo
p
ed
b
y
th
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
2
9
-
3538
3530
Am
er
ica
n
ac
ad
e
m
y
o
f
s
leep
m
ed
icin
e
(
AASM)
[
5
]
.
A
cc
o
r
d
in
g
to
R
K
m
an
u
al
s
leep
s
ta
g
es
ca
n
b
e
class
i
f
ied
in
to
th
e
f
o
llo
w
i
n
g
s
ta
g
es,
w
ak
e
(
W
)
s
tag
e,
f
o
u
r
n
o
n
-
r
ap
i
d
ey
e
m
o
v
e
m
e
n
t
s
ta
g
e
(
N
R
E
M)
,
an
d
r
ap
id
ey
e
m
o
v
e
m
e
n
t
(
R
E
M)
.
T
h
e
a
w
a
k
e
n
in
g
s
tag
e
i
s
w
a
k
e
s
ta
g
e
(
W
)
,
NR
E
M
is
t
h
e
f
ir
s
t
s
ta
g
e
o
f
s
le
ep
an
d
th
e
S2
clas
s
is
w
h
e
n
th
e
ac
t
u
al
s
leep
s
tag
e
b
eg
in
s
,
S3
is
th
e
d
ee
p
s
leep
p
h
ase
co
n
ti
n
u
ed
in
t
h
e
S4
s
ta
g
e,
an
d
in
R
E
M
th
e
e
y
es
ar
e
clo
s
ed
w
it
h
r
ap
id
m
o
v
e
m
e
n
t.
A
cc
o
r
d
in
g
to
th
e
AA
SM,
s
leep
s
tag
e
3
(
S3
)
an
d
s
leep
s
tag
e
4
(
S4
)
ar
e
co
n
s
id
er
ed
as
o
n
e
class
r
ep
r
e
s
en
ted
b
y
SW
S
(
s
lo
w
-
w
a
v
e
s
leep
)
[
6
,
7
]
.
E
E
G
an
aly
s
is
a
n
d
d
ec
is
io
n
m
a
k
i
n
g
in
cl
u
d
e
f
ea
t
u
r
e
ex
tr
ac
tio
n
[
8
]
u
s
i
n
g
f
r
eq
u
e
n
c
y
,
ti
m
e,
an
d
ti
m
e
-
f
r
eq
u
e
n
c
y
m
e
th
o
d
s
,
f
ea
tu
r
e
s
elec
tio
n
to
r
etain
o
n
l
y
s
i
g
n
i
f
ica
n
t
f
ea
t
u
r
es,
a
n
d
class
i
f
icat
io
n
u
s
i
n
g
d
ec
is
io
n
alg
o
r
ith
m
s
.
T
i
m
e
-
f
r
eq
u
e
n
c
y
tr
an
s
f
o
r
m
atio
n
s
o
r
n
o
n
-
li
n
ea
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
tech
n
iq
u
es
ar
e
al
s
o
b
ein
g
u
s
e
d
in
th
e
liter
at
u
r
e
f
o
r
t
h
e
e
x
t
r
ac
tio
n
o
f
r
elev
a
n
t
f
ea
t
u
r
es
f
o
r
clas
s
i
f
icatio
n
[
9
]
.
A
s
t
h
e
E
E
G
s
i
g
n
al
h
as
co
m
p
lex
b
eh
a
v
io
r
,
ad
v
a
n
ce
d
p
r
o
ce
s
s
i
n
g
,
an
d
m
ac
h
i
n
e
lear
n
in
g
al
g
o
r
ith
m
s
ar
e
r
eq
u
i
r
ed
,
w
h
ic
h
lear
n
t
h
e
co
m
p
le
x
it
y
o
f
t
h
e
s
i
g
n
als
a
n
d
ar
e
o
v
er
th
r
es
h
o
ld
b
ased
tech
n
iq
u
es
[
1
0
,
11]
.
R
ec
en
t
l
y
,
t
h
e
r
ec
u
r
r
en
t
n
e
u
r
al
n
et
w
o
r
k
s
(
R
NN
’
s
)
h
a
v
e
b
ee
n
p
r
o
v
en
to
h
a
v
e
b
e
t
t
e
r
p
e
r
f
o
r
m
a
n
c
e
t
h
a
n
t
h
e
m
e
t
h
o
d
s
u
s
e
d
i
n
“
c
l
a
s
s
i
c
”
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
a
l
g
o
r
i
t
h
m
s
o
n
d
a
t
a
s
e
t
s
w
i
t
h
t
e
m
p
o
r
a
l
i
n
f
o
r
m
a
t
i
o
n
(
M
i
k
o
l
o
v
et
a
l.
[
1
2
]
,
Gr
av
es
et
a
l.
[
13
]
,
Kar
p
ath
y
a
n
d
F
e
i
-
F
e
i
,
[1
4]
)
.
T
h
e
m
o
s
t
c
o
m
m
o
n
l
y
u
s
e
d
n
e
t
w
o
r
k
s
a
r
e
L
o
n
g
s
h
o
r
t
-
t
e
r
m
m
e
m
o
r
y
(
L
S
T
M
’
s
)
n
e
u
r
a
l
n
e
tw
o
r
k
s
,
t
h
e
s
e
n
e
tw
o
r
k
s
a
r
e
s
u
c
c
e
s
s
f
u
l
l
y
a
p
p
l
i
e
d
o
n
r
aw
E
E
G
s
i
g
n
a
l
s
(
D
a
v
i
d
s
o
n
et
a
l.
[
1
5
]
)
as
w
ell
a
s
to
s
leep
d
ata
(
Su
p
r
atak
et
a
l
.
[
16
]
)
.
T
h
e
liter
atu
r
e
s
h
o
w
s
b
o
th
m
ac
h
in
e
lear
n
i
n
g
an
d
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
b
ein
g
u
s
ed
f
o
r
s
le
ep
s
tag
e
s
co
r
in
g
,
u
s
i
n
g
E
E
G
s
ig
n
a
ls
.
T
h
e
s
tan
d
ar
d
m
ac
h
i
n
e
lear
n
in
g
ap
p
r
o
ac
h
es
u
s
e
h
an
d
-
en
g
i
n
ee
r
ed
f
ea
t
u
r
e
ex
tr
ac
tio
n
a
n
d
s
elec
tio
n
m
et
h
o
d
s
b
u
t
w
o
r
k
w
ell
w
it
h
a
m
o
d
e
r
atel
y
s
ized
d
ata
s
et
.
T
h
e
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
es
,
lik
e
C
N
N
[
17
]
a
n
d
R
N
N
,
l
e
a
r
n
t
h
e
f
e
a
t
u
r
e
s
a
u
t
o
m
a
t
i
c
a
l
l
y
w
i
t
h
o
u
t
r
e
q
u
i
r
i
n
g
d
a
t
a
p
r
e
p
r
o
c
e
s
s
i
n
g
a
n
d
f
e
a
t
u
r
e
e
x
t
r
a
c
t
i
o
n
s
t
e
p
s
b
u
t
r
e
q
u
i
r
e
l
a
r
g
e
d
a
t
a
t
o
t
r
a
i
n
t
h
e
n
e
t
w
o
r
k
.
T
h
e
p
ap
e
r
is
o
r
g
an
ized
as
f
o
llo
w
s
,
af
ter
d
is
cu
s
s
i
n
g
t
h
e
i
n
tr
o
d
u
ctio
n
a
n
d
th
e
r
elate
d
r
esear
ch
i
n
s
ec
tio
n
1
,
w
e
d
is
c
u
s
s
t
h
e
r
esea
r
ch
m
e
th
o
d
s
i
n
s
ec
tio
n
2
,
w
h
ic
h
elab
o
r
ates o
n
t
h
e
m
o
d
el
ar
ch
itectu
r
e.
Sectio
n
3
d
is
cu
s
s
es
i
m
p
le
m
e
n
tatio
n
m
eth
o
d
s
,
w
h
ich
d
escr
ib
es
th
e
d
etails
o
f
d
ata
s
ets
an
d
th
eir
u
s
a
g
e,
tr
ai
n
i
n
g
p
r
o
ce
d
u
r
e,
an
d
p
ar
am
eter
o
p
t
i
m
izatio
n
.
Sect
io
n
4
,
d
is
c
u
s
s
e
s
t
h
e
r
es
u
lts
an
d
d
i
s
cu
s
s
io
n
;
t
h
e
r
es
u
lt
s
f
o
r
t
w
o
d
if
f
er
e
n
t
t
y
p
es
o
f
d
ata
s
ets,
an
d
t
w
o
d
if
f
er
e
n
t
E
E
G
ch
an
n
els.
W
e
c
o
m
p
ar
e
o
u
r
r
esu
lt
s
w
i
th
th
e
p
er
f
o
r
m
an
ce
o
f
o
th
er
r
elate
d
s
tate
o
f
ar
t
m
et
h
o
d
s
in
th
is
s
ec
t
io
n
.
I
n
t
h
e
last
s
e
ctio
n
5
,
w
e
co
n
cl
u
d
e
o
n
o
u
r
r
esu
lt
s
.
2.
RE
S
E
ARCH
M
E
T
H
O
D
T
h
is
s
ec
tio
n
d
is
c
u
s
s
e
s
t
h
e
d
et
ailed
m
et
h
o
d
o
f
t
h
e
p
r
o
p
o
s
ed
m
o
d
el
d
ev
elo
p
ed
f
o
r
d
etec
ti
n
g
d
r
iv
er
’
s
aler
tn
es
s
au
to
m
atica
ll
y
u
s
in
g
E
E
G
s
ig
n
als.
T
h
e
m
o
d
el
is
d
esig
n
ed
u
s
in
g
d
ee
p
r
ec
u
r
r
en
t
n
eu
r
al
n
et
w
o
r
k
s
an
d
d
o
es n
o
t r
eq
u
ir
e
t
h
e
co
n
v
e
n
tio
n
al
ar
ti
f
ac
t
r
e
m
o
v
in
g
p
r
ep
r
o
ce
s
s
i
n
g
s
tep
s
a
n
d
f
ea
tu
r
e
ex
tr
ac
t
io
n
/
s
elec
t
io
n
s
tep
s
.
T
h
e
m
o
d
el
u
s
e
s
r
esid
u
al
n
et
w
o
r
k
s
,
w
h
ic
h
ar
e
tr
an
s
f
o
r
m
e
d
f
o
r
m
o
f
C
NN
s
to
lear
n
f
e
atu
r
es
an
d
en
co
d
er
-
d
ec
o
d
er
b
ased
s
eq
u
en
ce
m
o
d
e
ls
u
s
in
g
R
NN
to
d
ec
id
e
o
n
s
leep
s
tag
e
ca
teg
o
r
izatio
n
.
2
.
1
.
P
re
-
pro
ce
s
s
ing
P
r
ep
r
o
ce
s
s
in
g
o
f
E
E
G
s
i
g
n
a
ls
in
v
o
l
v
es
t
h
e
f
o
llo
w
in
g
t
h
r
ee
s
t
ep
s
,
T
h
e
in
p
u
t
i
s
ta
k
en
f
r
o
m
Slee
p
-
E
DF
2
0
1
3
/2
0
1
8
d
ata,
w
h
ic
h
is
s
e
g
m
e
n
ted
i
n
to
th
e
3
0
-
s
ep
o
ch
.
T
h
e
r
a
w
-
E
E
G
s
ig
n
als ar
e
u
s
ed
w
it
h
o
u
t
an
y
p
r
io
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
T
h
e
s
ig
n
al
is
lab
elled
w
it
h
th
e
h
elp
o
f
th
e
an
n
o
tat
io
n
f
i
le
th
e
H
y
p
n
o
g
r
a
m
,
w
h
ic
h
is
av
a
ilab
le
w
it
h
th
e
d
ataset
f
o
r
ea
ch
s
u
b
j
ec
t.
30
-
s
E
E
G
ep
o
ch
s
ar
e
n
o
r
m
aliz
ed
s
o
th
at
ea
ch
ep
o
ch
h
a
s
m
ea
n
v
al
u
e
as z
er
o
an
d
v
ar
ia
n
ce
a
s
o
n
e.
T
h
e
p
r
e
-
p
r
o
ce
s
s
in
g
s
tep
s
d
o
n
o
t
in
cl
u
d
e
an
y
n
o
is
e/ar
ti
f
ac
t
r
em
o
v
al
tec
h
n
iq
u
es
o
r
f
r
eq
u
en
c
y
b
a
n
d
s
ep
ar
atio
n
tec
h
n
iq
u
es.
As
w
e
u
s
e
n
e
u
r
al
n
et
w
o
r
k
s
t
h
er
e
i
s
n
o
n
ee
d
to
u
s
e
n
o
is
e
r
e
m
o
v
al
tech
n
iq
u
e
s
.
T
h
e
s
a
m
p
le
s
i
g
n
al
is
s
h
o
w
n
i
n
Fi
g
u
r
e
1.
2
.
2
.
T
he
m
o
del a
rc
hite
ct
ure
T
h
e
p
r
o
b
lem
o
f
d
etec
tin
g
s
le
ep
o
n
-
s
et
w
h
ile
d
r
iv
i
n
g
is
a
p
r
o
b
lem
o
f
class
if
icatio
n
o
f
s
leep
s
tag
e
s
w
it
h
t
h
e
s
eq
u
en
tial
i
n
n
a
tu
r
e.
Hen
ce
,
s
eq
u
e
n
ce
to
s
eq
u
en
c
e
m
o
d
el
p
r
o
p
o
s
ed
in
Fig
u
r
e
2
h
as
t
h
e
f
o
llo
w
i
n
g
co
m
p
o
n
e
n
t
s
,
i)
R
e
s
Net
s
w
it
h
s
k
ip
co
n
n
ec
tio
n
s
f
o
r
ex
tr
ac
ti
n
g
f
ea
tu
r
e
s
,
ii)
B
i
-
L
ST
Ms
w
h
i
ch
p
r
o
ce
s
s
p
ast
a
n
d
f
u
tu
r
e
i
n
f
o
r
m
atio
n
s
i
m
u
lta
n
eo
u
s
l
y
,
an
d
iii)
A
tte
n
tio
n
d
ec
o
d
er
,
w
h
ic
h
lear
n
s
o
n
l
y
s
i
g
n
if
ica
n
t
f
ea
tu
r
es.
W
e
ar
e
r
ef
er
r
in
g
to
t
h
r
ee
clas
s
es
f
o
r
class
i
f
icatio
n
,
i.e
.
aler
t/a
w
a
k
e
s
tate,
s
leep
o
n
-
s
et
s
ta
g
e/
s
lee
p
s
tag
e
1
,
an
d
s
leep
s
tag
e.
T
h
e
s
eq
u
e
n
ce
to
s
eq
u
e
n
ce
m
o
d
el
w
o
r
k
s
o
n
t
h
e
id
ea
o
f
en
co
d
er
an
d
d
ec
o
d
e
r
tech
n
iq
u
e.
Fo
r
ea
ch
3
0
-
s
E
E
G
ep
o
ch
,
th
e
in
p
u
t
s
eq
u
e
n
ce
is
en
co
d
ed
u
s
in
g
th
e
en
co
d
er
p
ar
t
o
f
th
e
m
o
d
el,
an
d
th
e
c
a
t
e
g
o
r
y
/
c
l
a
s
s
o
f
t
h
e
i
n
p
u
t
s
e
q
u
e
n
c
e
i
s
c
o
m
p
u
t
e
d
b
y
t
h
e
d
e
c
o
d
e
r
p
a
r
t
o
f
t
h
e
m
o
d
e
l
.
B
o
t
h
e
n
c
o
d
e
r
s
a
n
d
d
e
c
o
d
e
r
s
a
r
e
b
u
i
l
t
u
s
i
n
g
B
I
-
L
ST
Ms.
T
h
e
s
eq
u
en
ce
to
s
eq
u
e
n
ce
m
o
d
el
w
o
r
k
s
o
n
t
h
e
id
ea
o
f
en
co
d
er
an
d
d
ec
o
d
e
r
tech
n
iq
u
e.
Fo
r
ea
ch
3
0
-
s
E
E
G
ep
o
ch
,
th
e
in
p
u
t
s
eq
u
en
c
e
is
en
co
d
ed
u
s
in
g
t
h
e
en
co
d
er
p
ar
t
o
f
th
e
m
o
d
el,
an
d
th
e
ca
teg
o
r
y
/cla
s
s
o
f
t
h
e
in
p
u
t
s
eq
u
e
n
ce
i
s
co
m
p
u
ted
b
y
t
h
e
d
ec
o
d
er
p
ar
t
o
f
th
e
m
o
d
el.
B
o
th
en
co
d
er
s
a
n
d
d
ec
o
d
er
s
ar
e
b
u
il
t
u
s
i
n
g
B
I
-
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
lert
N
et
:
Dee
p
co
n
vo
lu
tio
n
a
l
-
r
ec
u
r
r
en
t n
eu
r
a
l n
etw
o
r
k
mo
d
el
fo
r
d
r
ivin
g
…
(
P
.
C
.
N
i
s
s
ima
g
o
u
d
a
r
)
3531
L
ST
Ms.
T
h
e
en
co
d
er
ca
p
tu
r
es
th
e
d
ep
en
d
en
cies
r
elate
d
to
lo
n
g
-
s
h
o
r
t
co
n
tex
t
s
b
et
w
ee
n
t
h
e
tar
g
et
class
e
s
an
d
in
p
u
t
s
.
T
h
e
in
p
u
t
f
o
r
th
e
en
co
d
er
s
is
ti
m
e
-
s
er
ies
f
ea
tu
r
e
s
o
b
tain
ed
f
r
o
m
R
e
s
Net
s
.
T
h
e
tim
e
s
er
ies
n
o
n
-
li
n
ea
r
d
ep
en
d
en
cies
ar
e
ca
p
tu
r
ed
f
o
r
d
etec
tin
g
t
h
e
tar
g
et
s
b
y
th
e
e
n
co
d
er
.
T
h
e
o
u
tp
u
t
o
f
th
e
en
co
d
ed
s
eq
u
en
ce
is
f
ed
to
th
e
atte
n
tio
n
n
et
w
o
r
k
a
n
d
f
u
r
th
er
,
t
h
e
y
ar
e
d
ec
o
d
ed
f
o
r
d
etec
tin
g
t
h
e
ca
teg
o
r
y
.
Ne
x
t,
we
w
il
l
d
is
c
u
s
s
ea
c
h
m
o
d
u
le
i
n
d
etail.
F
ig
u
r
e
1
.
Sa
m
p
le
E
E
G
Si
g
n
al
s
Fig
u
r
e
2
.
T
h
e
m
o
d
el
ar
ch
itect
u
r
e
w
ith
t
h
e
b
asic
R
esNet
b
lo
c
k
2
.
3
.
T
he
ResNet
s
R
esNet
a
r
esid
u
al
n
e
u
r
al
n
et
wo
r
k
is
a
d
ee
p
n
e
u
r
al
n
et
w
o
r
k
t
h
at
u
s
e
s
s
h
o
r
tcu
ts
to
j
u
m
p
s
o
m
e
la
y
er
s
,
ca
lled
s
k
ip
co
n
n
ec
tio
n
s
[1
8
]
.
T
h
e
b
asic
R
esNet
b
lo
c
k
is
s
h
o
w
n
in
Fi
g
u
r
e
2
.
I
t
w
as
d
ev
el
o
p
ed
in
th
e
v
ie
w
o
f
av
o
id
in
g
t
h
e
d
eg
r
ad
atio
n
p
r
o
b
le
m
w
h
ic
h
is
e
n
co
u
n
ter
ed
in
d
ee
p
er
n
eu
r
al
n
et
w
o
r
k
s
,
it
w
as
o
b
s
er
v
ed
th
at
as t
h
e
d
ep
th
o
f
th
e
n
e
u
r
al
n
et
w
o
r
k
s
in
cr
ea
s
es,
ac
cu
r
ac
y
g
et
s
s
at
u
r
ated
an
d
d
ec
r
ea
s
es
r
a
p
i
d
ly
.
T
h
e
R
esNe
t
m
o
d
el
is
i
m
p
le
m
en
ted
w
it
h
t
w
o
o
r
th
r
ee
s
k
ip
s
th
at
co
n
tai
n
R
e
L
U
ac
t
iv
atio
n
f
u
n
ctio
n
an
d
B
atch
No
r
m
in
b
et
w
ee
n
th
e
s
k
ip
la
y
er
s
,
th
i
s
w
ill
h
elp
to
av
o
id
v
an
i
s
h
in
g
g
r
ad
ien
t,
as
t
h
e
n
et
w
o
r
k
r
eu
s
e
s
ac
ti
v
atio
n
f
r
o
m
t
h
e
p
r
ev
io
u
s
la
y
er
h
a
n
d
till
th
e
ad
j
ac
en
t
lay
er
lear
n
s
t
h
e
w
eig
h
t
s
,
o
n
l
y
t
h
e
w
eig
h
t
s
o
f
ad
j
ac
en
t
la
y
er
s
ar
e
co
n
s
id
er
ed
,
th
is
p
r
o
v
id
es
b
est
r
esu
lts
w
h
e
n
a
n
o
n
li
n
ea
r
la
y
er
is
s
k
ip
p
ed
o
r
th
e
co
n
s
ec
u
ti
v
e
la
y
er
s
ar
e
lin
ea
r
.
T
h
e
s
k
ip
co
n
n
ec
tio
n
s
u
s
e
o
n
l
y
f
e
w
er
la
y
er
s
i
n
t
h
e
s
tar
ti
n
g
tr
ain
i
n
g
s
t
ag
es,
w
h
ic
h
s
i
m
p
li
f
ie
s
t
h
e
la
y
er
.
T
h
e
lear
n
in
g
i
s
th
u
s
f
aster
r
ed
u
ci
n
g
t
h
e
v
a
n
i
s
h
in
g
g
r
ad
ien
t
s
’
i
m
p
ac
t.
O
n
th
e
l
ater
p
ar
t
o
f
th
e
tr
ain
i
n
g
,
t
h
e
n
e
t
w
o
r
k
r
es
to
r
es
th
e
la
y
er
w
h
ic
h
w
a
s
s
k
ip
p
ed
to
l
ea
r
n
th
e
f
ea
t
u
r
e
s
p
ac
e.
I
n
th
e
en
d
,
th
e
la
y
er
s
ar
e
u
s
u
all
y
e
x
p
an
d
ed
w
h
ic
h
s
ta
y
clo
s
e
to
m
an
i
f
o
ld
f
o
r
f
a
s
ter
le
ar
n
in
g
w
e
u
s
e
t
h
i
s
f
u
n
ctio
n
ali
t
y
o
f
R
esNet
s
to
ca
p
t
u
r
e
f
r
eq
u
en
c
y
in
f
o
r
m
atio
n
,
th
e
r
esid
u
al
co
n
n
ec
tio
n
h
elp
s
u
s
to
m
ai
n
tai
n
f
ea
tu
r
es
f
r
o
m
t
h
e
p
r
ev
io
u
s
la
y
er
.
T
h
ese
f
ea
t
u
r
es
ar
e
th
e
n
f
ed
to
th
e
R
NN
m
o
d
el
f
o
r
cl
as
s
i
f
icat
io
n
.
T
h
e
f
ea
t
u
r
es
o
b
tain
ed
f
r
o
m
R
esNet
s
lear
n
th
e
co
m
p
lex
f
ea
t
u
r
es
w
h
ic
h
h
elp
in
cla
s
s
i
f
icatio
n
.
I
n
t
h
e
ab
o
v
e
f
i
g
u
r
e,
t
h
e
s
h
o
r
tcu
t
s
ca
n
b
e
d
ir
ec
tl
y
u
s
ed
i
f
t
h
e
d
i
m
e
n
s
io
n
o
f
t
h
e
i
n
p
u
t
a
n
d
t
h
e
o
u
tp
u
t a
r
e
th
e
s
a
m
e
,
d
en
o
ted
b
y
ex
p
r
ess
io
n
(
1
)
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
2
9
-
3538
3532
=
(
,
)
+
(
1
)
W
h
er
e
x
is
i
n
p
u
t
a
n
d
y
is
o
u
t
v
ec
to
r
s
f
o
r
th
e
la
y
er
s
co
n
s
id
er
ed
f
o
r
s
k
ip
co
n
n
ec
t
io
n
.
T
h
e
(
,
)
is
th
e
f
u
n
ctio
n
r
ep
r
esen
ti
n
g
th
e
r
e
s
id
u
al
m
ap
.
I
f
th
e
n
u
m
b
er
o
f
la
y
e
r
s
o
f
th
e
r
e
s
id
u
al
b
lo
ck
is
t
w
o
,
th
e
f
u
n
ctio
n
F c
a
n
b
e
r
ep
r
esen
ted
as
=
2
(
1
)
w
i
th
σ
as
R
eL
U,
a
n
d
n
eg
lecti
n
g
b
iase
s
.
T
h
e
s
h
o
r
tcu
t
co
n
n
ec
tio
n
a
n
d
ele
m
e
n
t
-
w
is
e
ad
d
itio
n
ar
e
u
s
e
d
to
o
p
er
ate
+
.
I
f
t
h
e
d
i
m
e
n
s
io
n
o
f
t
h
e
in
p
u
t
a
n
d
o
u
tp
u
t
ar
e
n
o
t
s
a
m
e,
t
h
e
n
ze
r
o
p
ad
d
in
g
s
ar
e
d
o
n
e
an
d
th
e
s
h
o
r
tcu
t i
s
u
s
ed
to
m
atch
t
h
e
d
im
e
n
s
io
n
u
s
i
n
g
th
e
f
o
llo
w
in
g
f
o
r
m
u
la
,
=
(
,
)
+
+
(
2
)
I
f
th
e
m
atr
i
x
d
i
m
en
s
io
n
s
o
f
x
an
d
F
ar
e
m
a
tch
i
n
g
t
h
e
(
1
)
is
u
s
ed
,
f
o
r
ch
a
n
g
in
g
d
i
m
en
s
i
o
n
s
(
2
)
is
u
s
ed
.
T
h
e
W
s
ter
m
i
n
(
2
)
r
e
p
r
esen
ts
th
e
lin
ea
r
p
r
o
j
ec
tio
n
w
h
ic
h
i
s
p
er
f
o
r
m
ed
u
s
in
g
t
h
e
s
h
o
r
tcu
t
o
r
s
k
ip
co
n
n
ec
tio
n
s
.
T
h
e
R
esNet
is
u
s
ed
f
o
r
f
ea
t
u
r
e
e
x
tr
ac
tio
n
i
n
o
u
r
m
o
d
el.
T
h
e
r
esid
u
al
n
et
w
o
r
k
t
h
at
i
s
t
h
e
“
id
en
ti
f
icatio
n
o
f
s
h
o
r
tcu
t
co
n
n
ec
t
io
n
”
f
o
r
r
etain
i
n
g
t
h
e
f
e
atu
r
es
f
r
o
m
co
n
s
ec
u
tiv
e
la
y
er
s
is
i
m
p
o
r
tan
t.
T
h
e
f
ea
t
u
r
es
r
eq
u
ir
ed
f
o
r
th
e
id
en
tif
icat
io
n
o
f
t
h
r
ee
clas
s
es
ca
n
b
e
o
b
tain
ed
b
y
t
h
e
e
x
tr
ac
tio
n
o
f
f
ea
t
u
r
es
u
s
i
n
g
R
esNet
.
T
h
e
R
esNet
s
d
o
n
o
t
r
eq
u
ir
e
t
w
o
f
i
lter
s
to
ex
tr
ac
t
te
m
p
o
r
al
an
d
f
r
eq
u
en
c
y
-
b
a
s
ed
f
ea
tu
r
es,
r
ath
er
u
s
in
g
R
esNet
h
elp
s
to
r
etai
n
f
ea
t
u
r
es
i
n
t
h
e
co
n
s
ec
u
tiv
e
la
y
er
s
.
I
t
is
a
s
i
m
p
le
n
o
tio
n
to
i
n
cr
ea
s
e
f
ea
t
u
r
es
w
e
n
ee
d
t
o
in
cr
ea
s
e
t
h
e
n
u
m
b
er
o
f
la
y
er
s
,
th
at
s
i
m
p
l
y
s
tack
s
t
h
e
la
y
er
s
,
b
u
t
t
h
is
ca
n
ca
u
s
e
a
v
a
n
is
h
i
n
g
g
r
ad
ie
n
t
p
r
o
b
le
m
b
ec
au
s
e
t
h
e
in
cr
ea
s
e
in
la
y
er
s
w
il
l
also
i
n
cr
ea
s
e
t
h
e
b
ac
k
p
r
o
p
ag
atio
n
m
u
ltip
le
ti
m
es
ac
r
o
s
s
th
e
la
y
er
s
.
Du
e
to
m
u
ltip
licatio
n
,
th
e
g
r
ad
ien
t b
ec
o
m
e
s
i
n
f
i
n
ite
l
y
s
m
a
ll a
n
d
th
e
g
r
ad
ien
t sat
u
r
ates.
E
E
G
s
ig
n
al
i
s
f
ed
to
t
h
e
n
eu
r
al
n
et
w
o
r
k
as
a
n
ar
r
a
y
.
1
1
(
1
)
C
o
n
v
o
lu
tio
n
s
ar
e
ca
r
r
ied
o
u
t
th
r
o
u
g
h
o
u
t
t
h
e
n
et
w
o
r
k
f
o
r
f
e
atu
r
e
e
x
tr
ac
tio
n
.
T
h
e
f
ilter
s
iz
es
s
tar
t
f
r
o
m
6
4
an
d
g
o
u
p
to
5
1
2
.
A
s
th
er
e
ar
e
s
k
ip
co
n
n
ec
t
io
n
s
i
n
R
esNet
m
o
d
el
to
r
etai
n
f
ea
t
u
r
es
f
r
o
m
p
r
ev
io
u
s
la
y
er
s
,
w
h
e
n
ev
er
th
er
e
is
c
h
an
g
e
in
t
h
e
in
p
u
t
f
ilter
_
s
ize
a
n
d
o
u
tp
u
t
f
i
l
ter
_
s
ize
ze
r
o
p
ad
d
in
g
h
a
s
b
ee
n
d
o
n
e,
it
is
r
ep
r
esen
ted
as
f
il
t
er
s
ize/2
in
d
icati
n
g
th
e
ch
a
n
g
e
i
n
f
ilter
_
s
ize.
R
e
s
Nets
h
elp
s
i
n
r
etain
i
n
g
t
h
e
f
ea
tu
r
es
an
d
r
ed
u
ce
s
o
v
er
f
it
tin
g
w
h
ic
h
is
ca
u
s
ed
d
u
e
t
o
u
s
ag
e
o
f
a
f
u
ll
y
co
n
n
ec
te
d
lay
er
.
A
ls
o
,
t
h
er
e
is
n
o
m
a
x
-
p
o
o
lin
g
la
y
er
u
s
ed
d
u
e
to
t
h
e
u
s
e
o
f
a
g
lo
b
al
a
v
er
ag
e
p
o
o
li
ng
(
G
A
P
)
lay
er
.
T
h
e
d
etails o
f
R
esNet
ar
c
h
itec
tu
r
e
is
s
h
o
w
n
i
n
Fi
g
u
r
e
3
.
Fig
u
r
e
3
.
R
esNet
ar
ch
i
tectu
r
e
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
lert
N
et
:
Dee
p
co
n
vo
lu
tio
n
a
l
-
r
ec
u
r
r
en
t n
eu
r
a
l n
etw
o
r
k
mo
d
el
fo
r
d
r
ivin
g
…
(
P
.
C
.
N
i
s
s
ima
g
o
u
d
a
r
)
3533
2
.
4
.
Bi
-
L
ST
M
ba
s
ed
s
equen
ce
t
o
s
equence
m
o
dels
w
it
h
a
t
t
ent
io
n
Seq
u
en
ce
to
s
eq
u
e
n
ce
m
e
th
o
d
is
a
d
ee
p
lear
n
in
g
ap
p
r
o
ac
h
th
at
u
s
es
a
n
en
co
d
er
-
d
e
co
d
er
b
ased
m
ac
h
in
e
tr
a
n
s
la
tio
n
tec
h
n
iq
u
e
to
tr
an
s
late
t
h
e
g
iv
e
n
i
n
p
u
t
s
e
q
u
en
ce
i
n
to
an
al
ter
n
ati
v
e
o
u
t
p
u
t
s
eq
u
e
n
ce
w
it
h
a
tag
a
n
d
att
e
n
tio
n
w
ei
g
h
tag
e.
I
t
u
s
e
s
t
w
o
r
ec
u
r
r
en
t
n
e
u
r
al
n
et
w
o
r
k
s
(
R
NNs)
w
h
ic
h
w
o
r
k
to
g
eth
er
to
p
r
ed
ict
th
e
n
ex
t
o
u
tp
u
t
s
eq
u
en
ce
f
r
o
m
th
e
p
r
ev
io
u
s
in
p
u
t
s
eq
u
e
n
ce
w
it
h
a
s
p
ec
ial
to
k
e
.
T
h
e
ca
s
es
in
w
h
ich
w
e
n
ee
d
to
p
r
ed
ict
th
e
n
e
x
t
s
tate
b
ased
o
n
t
h
e
p
r
ev
io
u
s
s
tate,
l
ik
e
p
r
ed
ictin
g
d
r
iv
er
e
x
p
er
ien
ci
n
g
th
e
d
r
o
w
s
y
s
tate
w
h
ic
h
d
ep
en
d
s
o
n
t
h
e
b
e
h
av
io
r
o
f
p
ast
E
E
G
s
i
g
n
al
b
eh
a
v
io
r
,
s
eq
u
en
ce
m
o
d
els
ca
n
b
e
u
s
ed
.
As
co
m
p
ar
ed
to
t
h
e
co
n
v
e
n
tio
n
al
n
eu
r
al
n
et
w
o
r
k
s
w
h
er
e
all
th
e
i
n
p
u
t
s
to
th
e
n
et
w
o
r
k
a
n
d
th
e
co
r
r
esp
o
n
d
in
g
o
u
tp
u
t
s
ar
e
in
d
ep
en
d
en
t,
w
e
ar
e
ch
o
o
s
i
n
g
b
i
-
d
ir
ec
tio
n
al
L
ST
Ms
w
h
er
ei
n
th
e
n
e
x
t
s
tate
o
r
th
e
o
u
tp
u
t
is
p
r
ed
icted
f
r
o
m
t
h
e
cu
r
r
en
t
an
d
p
ast
in
p
u
t.
T
h
e
B
i
-
L
ST
Ms,
s
h
o
w
n
i
n
Fi
g
u
r
e
4
,
w
h
ic
h
h
a
v
e
t
w
o
s
er
ies
R
NNS
ca
n
r
em
e
m
b
er
th
e
p
r
ev
io
u
s
s
ta
te
an
d
w
it
h
t
h
at
i
n
f
o
r
m
atio
n
,
t
h
e
y
ca
n
p
r
ed
ict
th
e
n
e
x
t
s
tate.
L
ST
M
u
n
it
s
h
av
e
a
r
ich
i
n
ter
n
a
l
s
tr
u
ct
u
r
e.
T
h
e
v
ar
io
u
s
“
g
ate
s
”
d
eter
m
in
e
t
h
e
p
r
o
p
ag
atio
n
o
f
in
f
o
r
m
atio
n
a
n
d
ca
n
c
h
o
o
s
e
to
“
r
em
e
m
b
er
”
o
r
“
f
o
r
g
et”
i
n
f
o
r
m
atio
n
.
C
o
m
p
ar
ed
to
th
e
tr
ad
itio
n
a
l
u
n
id
ir
ec
ti
o
n
al
R
NNs,
t
h
e
o
p
er
atio
n
o
f
w
h
ic
h
o
n
l
y
d
ep
en
d
s
o
n
p
r
ev
io
u
s
i
n
p
u
t
s
tate,
B
i
-
d
ir
ec
tio
n
al
L
ST
Ms
p
r
o
ce
s
s
d
ata
b
o
th
i
n
f
o
r
w
ar
d
a
n
d
b
ac
k
w
ar
d
d
ir
ec
tio
n
s
s
i
m
u
lta
n
eo
u
s
l
y
.
He
n
ce
,
th
e
B
i
-
d
ir
ec
tio
n
al
L
ST
Ms
ar
e
u
s
ed
to
r
em
e
m
b
er
b
o
th
p
ast
an
d
f
u
t
u
r
e
d
ata
p
o
in
ts
,
an
d
th
e
i
n
p
u
ts
r
u
n
i
n
b
o
th
th
e
d
ir
ec
tio
n
s
,
o
n
e
f
r
o
m
f
u
t
u
r
e
to
p
ast
a
n
d
o
n
e
f
r
o
m
p
ast
to
f
u
t
u
r
e
u
s
i
n
g
t
w
o
h
id
d
en
s
tates.
I
n
B
id
ir
ec
tio
n
al
L
ST
M,
th
e
r
ep
lica
o
f
th
e
f
ir
s
t
r
ec
u
r
r
en
t
la
y
er
i
s
cr
ea
ted
an
d
th
e
i
n
p
u
t
is
g
iv
e
n
to
th
e
f
ir
s
t
la
y
er
in
th
e
n
o
r
m
al
ti
m
e
o
r
d
er
,
t=
1
,
.
.
.
T
,
w
h
ile
t
h
e
r
ev
er
s
e
d
in
p
u
t
in
t
h
e
ti
m
e
o
r
d
er
t=
T
,
…
1
,
is
p
r
o
v
id
ed
t
o
th
e
s
ec
o
n
d
o
r
b
ac
k
w
ar
d
la
y
er
[
1
9
]
.
T
h
e
o
u
tp
u
t
is
co
m
p
u
ted
as
th
e
w
ei
g
h
ted
s
u
m
o
f
th
e
t
wo
lay
er
s
.
T
h
e
s
a
m
e
is
r
ep
r
esen
ted
as (
3
)
-
(
5
)
.
ℎ
→
=
ℎ
(
→
+
→
ℎ
−
1
→
+
→
)
(
3
)
ℎ
←
=
ℎ
(
←
+
←
ℎ
+
1
←
+
←
)
(
4
)
=
(
[
ℎ
→
;
ℎ
←
]
+
)
(
5
)
T
h
e
h
id
d
en
s
tate
an
d
f
ee
d
-
f
o
r
w
ar
d
n
e
t
w
o
r
k
’
s
b
ias
ar
e
r
ep
r
esen
ted
as
(
ℎ
→
,
→
)
;
th
e
h
id
d
en
s
ta
te
an
d
b
ac
k
w
ar
d
n
et
w
o
r
k
’
s
b
ias
i
s
r
ep
r
esen
ted
as
(
ℎ
←
,
←
)
;
an
d
ar
e
t
h
e
i
n
p
u
t
a
n
d
t
h
e
o
u
tp
u
t
o
f
B
i
-
L
ST
M,
r
esp
ec
tiv
el
y
.
T
h
e
s
eq
u
en
ce
-
to
-
s
eq
u
en
ce
m
o
d
el
u
s
ed
in
o
u
r
m
o
d
el
co
n
s
is
t
s
o
f
an
en
co
d
er
an
d
a
d
ec
o
d
er
b
u
ilt
w
it
h
L
ST
Ms.
T
h
e
en
co
d
er
tak
es
th
e
in
p
u
t
a
s
o
n
e
s
eq
u
e
n
ce
a
t
a
ti
m
e
i
n
t
h
e
f
o
r
m
o
f
v
ec
to
r
r
ep
r
esen
tatio
n
,
a
n
d
th
e
d
ec
o
d
er
esti
m
ates
t
h
e
cl
ass
f
o
r
ea
ch
3
0
-
s
in
p
u
t
s
eq
u
en
ce
.
T
h
e
lo
n
g
-
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
u
n
it
s
o
f
t
h
e
en
co
d
er
ca
p
tu
r
e
th
e
co
n
te
x
t
d
ep
en
d
en
cies
o
f
th
e
i
n
p
u
t
an
d
t
h
e
o
u
tp
u
t
tar
g
et.
T
h
e
d
ec
o
d
er
th
u
s
co
m
p
u
tes
t
h
e
in
f
o
r
m
atio
n
o
f
h
id
d
en
s
tates
an
d
p
r
ed
ic
ts
th
e
o
u
tp
u
t
w
it
h
th
e
h
elp
o
f
S
o
f
t
m
a
x
[
2
0
]
.
Sin
ce
t
h
er
e
ar
e
th
r
ee
class
es to
b
e
class
if
ied
t
h
e
len
g
th
o
f
th
e
e
n
co
d
ed
v
ec
to
r
w
ill
b
e
th
r
ee
,
w
h
ich
i
s
e1
,
e2
,
an
d
e3
.
Fig
u
r
e
4
.
B
asic B
i d
ir
ec
tio
n
al
L
ST
M
2
.
5
.
T
he
a
t
t
e
ntio
n net
w
o
rk
T
h
e
en
co
d
ed
s
eq
u
en
ce
o
f
e
v
er
y
ep
o
ch
i
s
f
u
r
t
h
er
u
s
ed
to
g
et
t
h
e
tar
g
et
s
eq
u
e
n
ce
u
s
i
n
g
atten
t
io
n
n
et
w
o
r
k
,
w
h
ic
h
is
a
d
ec
o
d
er
p
ar
t
o
f
th
e
n
et
w
o
r
k
.
T
h
e
d
ec
o
d
er
is
also
b
u
ilt
u
s
i
n
g
L
ST
Ms.
I
n
th
e
s
tan
d
ar
d
d
ec
o
d
er
,
f
o
r
ev
er
y
s
eq
u
en
ce
o
f
i
n
p
u
ts
,
th
e
d
ec
o
d
er
g
en
er
a
tes
t
h
e
n
e
w
r
ep
r
esen
tatio
n
o
f
th
e
in
p
u
t
s
eq
u
e
n
ce
alo
n
g
w
i
th
a
tar
g
et
i
n
p
u
t
e
le
m
en
t.
T
h
e
last
i
n
p
u
t
co
m
in
g
to
th
e
d
ec
o
d
er
is
th
e
las
t
ef
f
e
ct
to
u
p
d
ate
f
o
r
th
e
h
id
d
en
s
tate
o
f
en
co
d
er
.
T
h
u
s
,
th
e
m
o
d
el
h
as
to
b
e
b
iased
ac
co
r
d
in
g
to
th
e
last
el
e
m
en
t.
So
,
th
e
u
s
e
o
f
atten
tio
n
m
ec
h
an
i
s
m
i
n
th
e
m
o
d
el
ca
n
ad
d
r
ess
s
u
ch
a
p
r
o
b
lem
.
T
h
e
atten
tio
n
n
e
t
w
o
r
k
lear
n
s
d
if
f
er
en
t
p
o
r
tio
n
s
of
th
e
o
u
tp
u
t
s
eq
u
e
n
ce
o
f
en
co
d
er
f
o
r
ea
ch
d
ec
o
d
in
g
s
tep
alo
n
g
w
it
h
co
n
s
id
er
in
g
th
e
e
n
tire
e
n
co
d
er
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
2
9
-
3538
3534
r
ep
r
esen
tatio
n
.
Hen
ce
t
h
e
d
ec
o
d
er
lear
n
s
o
n
ly
t
h
e
s
ig
n
i
f
ic
an
t
in
p
u
t
s
eq
u
e
n
ce
p
ar
ts
d
u
r
in
g
d
ec
o
d
in
g
s
ta
g
e.
W
ith
o
u
t
t
h
e
at
ten
t
io
n
m
ec
h
a
n
is
m
,
t
h
e
d
ec
o
d
er
o
p
er
atio
n
r
elies
o
n
t
h
e
h
id
d
en
v
ec
to
r
o
f
t
h
e
d
ec
o
d
er
’
s
B
i
-
L
ST
M.
T
h
e
s
eq
u
en
ce
to
s
eq
u
en
ce
m
o
d
el
i
n
cl
u
d
in
g
atte
n
tio
n
m
ec
h
a
n
is
m
i
s
m
o
r
e
e
f
f
ec
tiv
e
as
it
in
c
lu
d
es
b
o
th
en
co
d
er
’
s
r
ep
r
esen
tat
io
n
a
n
d
d
ec
o
d
er
w
it
h
h
id
d
en
v
ec
to
r
c
allin
g
t
h
e
co
n
te
x
t
o
r
atten
tio
n
v
ec
to
r
,
r
ep
r
esen
ted
as
(
ct)
.
A
tten
tio
n
w
ei
g
h
t
s
ar
e
co
m
p
u
ted
as
a
f
u
n
ctio
n
o
f
f
(
.
)
,
b
ef
o
r
e
co
m
p
u
ti
n
g
atte
n
ti
o
n
v
ec
to
r
(
ct)
.
T
h
e
co
n
tex
t
o
r
atten
tio
n
v
ec
to
r
(
ct)
is
p
r
o
b
a
b
ilit
ies
(
α
i)
,
r
elatin
g
to
th
e
s
i
g
n
i
f
ica
n
ce
o
f
ea
ch
h
id
d
en
s
ta
te,
m
u
ltip
lied
b
y
a
h
id
d
en
s
tate
(
).
(
ℎ
−
1
,
)
=
ℎ
(
ℎ
ℎ
−
1
+
)
(
6
)
∝
=
(
(
ℎ
−
1
,
)
)
≈
(
(
ℎ
−
1
,
)
)
∑
(
(
ℎ
,
=
1
)
)
∈
1
,
2
,
3
…
…
.
(
7
)
=
∑
∝
=
0
(
8
)
w
h
er
e
α
i i
s
t
h
e
s
i
g
n
if
ica
nt
o
f
p
ar
t i
o
f
t
h
e
h
id
d
en
s
tate.
T
h
e
f
(
.
)
,
is
a
co
m
b
in
at
io
n
o
f
t
h
e
e
n
c
o
d
er
’
s
h
id
d
en
s
tate
(
)
,
an
d
d
ec
o
d
er
’
s
h
id
d
en
s
tate
(
ℎ
−
1
)
,
w
it
h
th
e
ℎ
lay
er
f
o
llo
w
ed
b
y
.
L
ater
,
f
(
.
)
,
is
g
iv
e
n
to
th
e
s
o
f
t
m
a
x
m
o
d
u
le
to
ca
lcu
late
α
i
f
o
r
n
p
a
r
ts
.
T
h
en
th
e
co
m
p
u
tat
io
n
o
f
,
is
p
er
f
o
r
m
ed
b
y
th
e
atten
t
io
n
m
o
d
u
le,
w
h
ic
h
i
s
a
w
eig
h
ted
s
u
m
o
f
all
(
)
an
d
∝
v
ec
to
r
s
.
Hen
ce
,
w
h
ile
d
ec
o
d
in
g
,
th
e
m
o
d
el
ca
n
o
n
l
y
co
n
s
id
er
th
e
i
m
p
o
r
tan
t
r
eg
io
n
s
o
f
th
e
i
n
p
u
t
v
ec
to
r
s
eq
u
en
ce
.
3.
I
M
P
L
E
M
E
NT
AT
I
O
N
T
h
is
s
ec
tio
n
e
lab
o
r
ates
th
e
d
e
tails
o
f
th
e
i
m
p
le
m
e
n
tatio
n
o
f
th
e
al
g
o
r
ith
m
;
t
h
e
d
ata
s
e
t
p
r
ep
ar
atio
n
,
tr
ain
i
n
g
p
r
o
ce
d
u
r
e,
lo
s
s
ca
lcu
l
atio
n
,
an
d
th
e
e
v
al
u
atio
n
o
f
t
h
e
m
o
d
el
u
s
i
n
g
v
ar
io
u
s
m
e
tr
ics.
3
.
1
.
T
he
da
t
a
s
et
prepa
ra
t
io
n
T
h
e
d
atasets
u
s
ed
f
o
r
th
i
s
s
t
u
d
y
ar
e
co
m
m
o
n
p
u
b
lic
d
atase
ts
o
f
s
leep
-
ed
f
2
0
1
3
an
d
2
0
1
8
v
er
s
io
n
s
w
h
ic
h
co
n
s
i
s
t
o
f
6
1
an
d
1
9
7
p
o
ly
s
o
m
n
o
g
r
a
m
s
(
P
SGs
)
r
esp
ec
tiv
el
y
.
T
ab
le
1
s
h
o
w
s
th
e
d
at
a
co
r
r
esp
o
n
d
in
g
to
d
if
f
e
r
e
n
t
s
leep
clas
s
es.
W
e
co
n
s
id
er
t
h
e
d
ata
f
r
o
m
Fp
z
-
C
z
/P
z
-
Oz
E
E
G
c
h
a
n
n
el
s
f
o
r
o
u
r
all
an
al
y
s
i
s
.
T
h
e
d
ata
s
et
u
s
ed
h
er
e
d
o
es
n
o
t
h
av
e
an
eq
u
al
d
is
tr
ib
u
tio
n
o
f
all
s
l
ee
p
class
es,
th
e
s
leep
s
tag
e
s
W
an
d
o
th
er
s
leep
s
tag
e
s
ar
e
g
r
ea
ter
i
n
n
u
m
b
er
co
m
p
ar
ed
to
N1
-
st
ate.
S
u
ch
a
c
lass
i
m
b
ala
n
ce
p
r
o
b
lem
is
b
ett
er
ad
d
r
ess
ed
u
s
in
g
d
ee
p
lear
n
in
g
m
et
h
o
d
s
co
m
p
a
r
ed
to
co
n
v
e
n
tio
n
al
m
ac
h
in
e
l
ea
r
n
in
g
tech
n
iq
u
e
s
.
T
h
e
lo
s
s
ca
lcu
latio
n
m
et
h
o
d
u
s
ed
i
n
th
i
s
p
ap
er
also
h
elp
s
i
n
d
ea
li
n
g
w
it
h
th
e
clas
s
i
m
b
al
an
ce
p
r
o
b
le
m
.
I
n
ad
d
itio
n
to
t
h
is
,
th
e
d
ata
s
e
t
i
s
o
v
er
s
a
m
p
led
w
h
er
e
v
er
r
eq
u
ir
ed
to
b
alan
ce
th
e
n
u
m
b
er
o
f
all
s
leep
s
tag
e
cla
s
s
e
s
.
T
ab
le
1
.
E
E
G
d
ata
s
ets
D
a
t
a
S
e
t
W
a
k
e
N1
S
l
e
e
p
T
o
t
a
l
S
l
e
e
p
-
ED
F
-
13
8
,
0
5
5
6
0
4
6
,
5
2
9
1
5
,
1
8
8
S
l
e
e
p
-
ED
F
-
18
5
3
,
6
4
1
2
0
,
2
1
5
1
,
1
5
.
5
6
2
1
,
8
9
,
4
1
8
3
.
2
.
T
ra
ini
ng
pro
ce
du
re
w
it
h o
pti
m
izing
pa
ra
m
et
er
s
a
nd
hy
p
er
-
pa
ra
m
et
er
s
W
e
f
ee
d
3
0
s
-
ep
o
ch
to
th
e
R
es
Net
f
o
r
ex
tr
ac
tio
n
o
f
f
r
eq
u
e
n
c
y
co
m
p
o
n
e
n
t
r
elate
d
to
th
e
s
le
ep
s
tag
es
,
w
h
ic
h
is
f
u
r
th
er
co
n
n
ec
ted
to
s
eq
u
en
ce
-
to
-
s
eq
u
e
n
ce
m
o
d
els.
Fo
r
ea
ch
f
o
ld
,
o
n
e
p
ar
t
is
ta
k
e
n
f
o
r
test
in
g
r
est
an
d
is
u
s
ed
f
o
r
tr
ain
i
n
g
.
F
in
al
l
y
,
all
th
e
e
v
al
u
atio
n
r
es
u
lt
s
ar
e
co
m
b
i
n
ed
.
T
h
e
m
o
d
el
is
e
v
alu
ated
u
s
i
n
g
k
-
f
o
ld
cr
o
s
s
-
v
alid
atio
n
.
T
h
e
Sleep
-
E
D
F
2
0
1
3
d
ataset
i
s
tr
ain
ed
b
y
s
etti
n
g
k
v
al
u
e
to
2
0
an
d
th
e
s
leep
e
d
f
2
0
1
8
d
ata
s
et
is
tr
ai
n
ed
b
y
s
e
tti
n
g
k
v
al
u
e
to
1
0
.
C
r
o
s
s
-
v
al
id
atio
n
i
s
u
s
ed
to
ev
al
u
ate
t
h
e
m
ac
h
i
n
e
l
ea
r
n
in
g
m
o
d
el
s
w
it
h
th
e
h
elp
o
f
a
r
esa
m
p
li
n
g
p
r
o
ce
d
u
r
e.
T
h
e
s
in
g
le
p
ar
a
m
eter
ca
lled
‘
k
’
is
u
s
ed
an
d
i
t
d
en
o
t
es
to
t
h
e
n
u
m
b
er
o
f
g
r
o
u
p
s
to
b
e
s
p
ilt
o
n
th
e
av
ai
lab
le
d
atas
et.
C
r
o
s
s
-
va
lid
atio
n
is
ap
p
lied
to
ch
ec
k
th
e
b
eh
av
i
o
r
o
f
th
e
m
ac
h
i
n
e
lear
n
i
n
g
m
o
d
el
o
n
th
e
u
n
s
ee
n
d
atase
t.
So
,
w
h
e
n
v
alid
atio
n
i
s
d
o
n
e
it
i
s
to
c
h
ec
k
w
h
et
h
er
t
h
e
in
cr
ea
s
e
i
n
t
h
e
ac
c
u
r
ac
y
o
f
th
e
tr
ai
n
i
n
g
d
ata
al
s
o
lead
s
to
an
in
cr
ea
s
e
i
n
t
h
e
ac
cu
r
ac
y
o
f
th
e
d
atase
t
w
h
i
ch
is
n
o
t
p
r
ev
io
u
s
l
y
s
ee
n
b
y
t
h
e
n
et
w
o
r
k
.
I
t
is
d
o
n
e
to
m
i
n
i
m
i
ze
o
v
er
f
itti
n
g
.
T
h
is
m
et
h
o
d
is
less
b
iased
an
d
o
p
ti
m
ized
f
o
r
a
s
i
m
p
le
tr
ain
/te
s
t s
p
lit.
T
h
e
n
et
w
o
r
k
is
tr
ain
ed
f
o
r
1
2
0
ep
o
ch
s
,
w
it
h
R
M
S
p
r
o
p
as
t
h
e
o
p
ti
m
izer
,
t
h
is
is
s
i
m
ilar
t
o
Gr
ad
ien
t
d
escen
t b
u
t t
h
e
o
s
cillatio
n
s
in
th
e
v
er
tical
d
ir
ec
tio
n
ar
e
r
es
tr
i
cted
,
h
elp
in
g
th
e
m
o
d
el
to
m
o
v
e
i
n
t
h
e
h
o
r
izo
n
tal
d
ir
ec
tio
n
to
co
n
v
er
g
e
f
a
s
ter
w
ith
th
e
i
n
cr
ea
s
e
i
n
th
e
lear
n
in
g
r
ate.
T
h
e
m
i
n
i
-
b
atch
e
s
o
f
s
ize
2
0
a
r
e
u
s
e
d
w
i
t
h
a
l
e
a
r
n
i
n
g
r
a
t
e
s
e
t
t
o
=
0
.
00001
a
n
d
t
h
e
L
2
r
e
g
u
l
a
r
i
z
a
t
i
o
n
e
l
e
m
e
n
t
w
i
t
h
=
0
.
001
t
o
m
i
n
i
m
i
z
e
ove
r
fit
tin
g
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
lert
N
et
:
Dee
p
co
n
vo
lu
tio
n
a
l
-
r
ec
u
r
r
en
t n
eu
r
a
l n
etw
o
r
k
mo
d
el
fo
r
d
r
ivin
g
…
(
P
.
C
.
N
i
s
s
ima
g
o
u
d
a
r
)
3535
3
.
3
.
L
o
s
s
ca
lcula
t
io
n
T
h
er
e
is
a
p
r
o
b
lem
of
d
ata
i
m
b
a
lan
ce
in
th
e
Sleep
-
E
D
F
d
ataset;
to
r
ed
u
ce
t
h
e
e
f
f
ec
t
of
th
i
s
is
s
u
e
we
u
s
e
M
SE
a
n
d
MSF
E
f
o
r
m
u
l
ti
class
cla
s
s
i
f
icat
io
n
.
T
h
e
m
ea
n
s
q
u
ar
ed
er
r
o
r
(
MSE
)
is
a
v
er
y
ef
f
ec
tiv
e
m
ea
n
s
to
d
eter
m
in
e
lo
s
s
f
u
n
ctio
n
s
in
d
ee
p
lear
n
in
g
m
o
d
els.
It
p
er
f
o
r
m
s
w
ell
f
o
r
a
b
ala
n
ce
d
d
ata
class
,
b
u
t
f
o
r
an
i
m
b
alan
ce
d
d
ataset
it
f
ails
,
t
h
i
s
is
d
u
e
to
t
h
e
f
ac
t
t
h
at
it
a
v
e
r
ag
es
t
h
e
lo
s
s
by
es
s
en
tia
ll
y
s
u
m
m
i
n
g
u
p
all
th
e
er
r
o
r
s
in
th
e
w
h
o
le
d
ataset.
T
h
is
can
ef
f
ec
ti
v
el
y
e
s
ti
m
ate
th
e
er
r
o
r
s
if
th
e
d
ataset
of
b
o
th
m
i
n
o
r
it
y
an
d
m
aj
o
r
ity
clas
s
is
th
e
s
a
m
e.
W
h
en
th
e
d
ataset
is
i
m
b
ala
n
ce
d
th
e
lo
s
s
ten
d
s
to
g
et
b
iased
to
t
h
e
m
aj
o
r
ity
clas
s
as
it
co
n
tr
ib
u
te
s
m
o
r
e
to
t
h
e
lo
s
s
w
h
e
n
co
m
p
ar
ed
to
th
e
m
i
n
o
r
it
y
cla
s
s
.
T
h
is
r
e
s
u
l
ts
in
t
h
e
l
o
s
s
w
h
ich
ca
p
tu
r
es
th
e
er
r
o
r
of
th
e
m
aj
o
r
it
y
cla
s
s
o
n
l
y
.
T
h
e
MSE
w
it
h
M
SF
E
can
be
u
s
ed
w
h
er
ei
n
t
h
e
m
ea
n
s
q
u
ar
ed
f
alse
er
r
o
r
(
MSF
E
),
f
ir
s
tl
y
it
av
er
a
g
es
th
e
er
r
o
r
s
ep
ar
ately
in
ea
c
h
class
an
d
t
h
en
ad
d
s
t
h
e
m
u
p
.
(
)
=
1
∑
(
−
^
)
=
(
9
)
(
)
=
∑
(
(
)
)
=
1
(
1
0
)
(
)
=
∑
(
(
)
2
)
=
1
(
1
1
)
w
h
er
e
ci
is
th
e
clas
s
lab
el,
C
i
is
t
h
e
n
u
m
b
er
o
f
s
a
m
p
le
s
,
N
is
th
e
n
u
m
b
er
o
f
av
ai
lab
le
clas
s
es,
l(
ci)
i
s
t
h
e
er
r
o
r
ca
lcu
lated
f
o
r
class
ci.
W
it
h
t
h
e
h
elp
o
f
MSE
a
n
d
MFSE,
t
h
e
lo
s
s
o
f
b
o
th
m
i
n
o
r
it
y
an
d
m
aj
o
r
ity
cla
s
s
e
s
is
co
n
s
id
er
ed
[
2
1
,
2
2
]
.
3
.
4
.
E
v
a
lua
t
io
n
m
et
ric
s
We
ev
alu
ate
t
h
e
m
o
d
el
u
s
in
g
o
v
er
all
-
ac
c
u
r
ac
y
,
r
ec
all,
p
r
ec
is
io
n
,
C
o
h
e
n
’
s
Kap
p
a
an
d
F1
-
s
co
r
e
.
T
h
e
o
v
er
all
-
ac
c
u
r
ac
y
is
r
ep
r
ese
n
te
d
by
th
e
r
atio
of
t
h
e
co
r
r
ec
t
n
u
m
b
er
of
p
r
ed
ictio
n
s
to
th
e
n
u
m
b
er
of
co
m
p
lete
in
p
u
t
d
ata
s
a
m
p
les.
P
r
ec
is
io
n
r
ep
r
esen
t
s
t
h
e
r
atio
of
co
r
r
ec
tl
y
cla
s
s
i
f
ied
s
a
m
p
les
(
tr
u
e
p
o
s
itiv
e
s
)
to
th
e
s
u
m
m
atio
n
of
tr
u
e
p
o
s
iti
v
es
a
n
d
clas
s
es
w
h
ic
h
ar
e
w
r
o
n
g
l
y
cla
s
s
i
f
ied
as
p
o
s
iti
v
e
(
f
a
ls
e
p
o
s
itiv
e)
.
It
c
h
ec
k
s
o
u
t
of
t
h
e
o
n
es
w
h
ic
h
t
h
e
m
o
d
el
p
r
ed
icts
as
p
o
s
iti
v
e,
h
o
w
m
an
y
ar
e
ac
tu
all
y
p
o
s
it
iv
e.
R
ec
all
is
r
ep
r
esen
ted
as
th
e
r
atio
o
f
co
r
r
ec
tly
cla
s
s
i
f
i
ed
s
a
m
p
les
to
th
e
s
u
m
m
atio
n
o
f
tr
u
e
p
o
s
itiv
e
s
an
d
f
al
s
e
n
eg
ati
v
es
(
clas
s
if
ied
w
r
o
n
g
l
y
as
n
e
g
ati
v
e)
.
I
t
ch
ec
k
s
o
u
t
o
f
th
e
o
n
e
s
w
h
ic
h
ar
e
ac
tu
all
y
p
o
s
it
iv
e
i
n
th
e
i
n
p
u
t
d
ata
s
a
m
p
les
h
o
w
m
an
y
d
id
th
e
m
o
d
el
p
r
ed
ict
p
o
s
itiv
e.
C
o
h
en
’
s
Kap
p
a
is
a
s
t
atis
tical
ap
p
r
o
ac
h
th
at
m
ea
s
u
r
es
in
tr
a(
i
n
ter
)
r
ater
r
eliab
ilit
y
f
o
r
t
h
e
ca
te
g
o
r
i
ca
l
(
q
u
alitati
v
e)
o
b
j
ec
ts
.
T
h
is
m
e
th
o
d
is
p
r
e
f
er
r
ed
o
v
er
d
ir
ec
t
p
er
ce
n
t
ag
r
ee
m
en
t
ca
lcu
latio
n
s
,
as
it
co
n
s
id
er
s
t
h
e
p
r
o
b
ab
ilit
y
o
f
t
h
e
ag
r
ee
m
e
n
t
o
cc
u
r
r
in
g
b
y
c
h
a
n
ce
.
I
t
is
ex
p
ec
ted
to
b
e
a
m
o
r
e
r
o
b
u
s
t
m
ea
s
u
r
e
[
2
3
,
2
4
]
.
F1
-
s
co
r
e
is
th
e
ca
lcu
lat
io
n
o
f
h
ar
m
o
n
ic
m
ea
n
b
et
w
ee
n
p
r
ec
is
io
n
a
n
d
th
e
r
ec
all.
4.
RE
SU
L
T
S
A
ND
D
I
SCU
SS
I
O
N
T
h
e
d
ata
s
ets
u
s
ed
ar
e
2
0
1
3
an
d
2
0
1
8
s
leep
ed
f
s
as
r
ef
er
r
ed
in
p
r
ev
io
u
s
s
ec
t
io
n
s
a
s
s
h
o
w
n
in
T
ab
le
1
.
T
h
e
r
esu
lt
s
co
n
s
is
t
in
g
o
f
co
n
f
u
s
io
n
m
atr
ix
a
n
d
th
e
p
er
-
clas
s
p
er
f
o
r
m
a
n
ce
f
o
r
b
o
th
v
er
s
io
n
s
o
f
d
at
a
an
d
b
o
th
FP
z
-
C
z
an
d
P
z
-
Oz,
E
E
G
ch
an
n
el
i
s
g
iv
e
n
i
n
T
ab
le
s
2
an
d
3
r
esp
ec
tiv
el
y
.
A
cc
o
r
d
in
g
to
liter
atu
r
e,
t
h
e
m
o
d
el
ca
n
b
e
e
v
alu
ated
b
y
t
wo
m
eth
o
d
s
,
o
n
e
m
e
t
h
o
d
is
to
u
s
e
ep
o
ch
s
f
r
o
m
t
h
e
s
a
m
e
s
u
b
j
ec
t
f
o
r
b
o
th
tr
ain
in
g
an
d
v
al
id
atio
n
w
h
ic
h
i
s
ca
lle
d
as
t
h
e
i
n
tr
a
-
s
u
b
j
ec
t
p
ar
ad
ig
m
an
d
th
e
o
th
er
i
s
t
h
e
i
n
te
r
-
s
u
b
j
ec
t
p
ar
ad
ig
m
,
w
h
er
ein
w
e
u
s
e
t
h
e
ep
o
ch
s
f
r
o
m
d
if
f
er
en
t
s
u
b
j
ec
ts
f
o
r
tr
ain
i
n
g
an
d
test
i
n
g
.
I
n
o
u
r
s
tu
d
y
u
s
ed
t
h
e
s
ec
o
n
d
ap
p
r
o
ac
h
th
at
i
s
b
o
th
tr
y
i
n
g
a
n
d
test
in
g
ep
o
ch
s
co
m
e
f
r
o
m
d
if
f
er
e
n
t
s
u
b
j
ec
ts
.
T
ab
les
2
an
d
3
,
r
ep
r
esen
t
t
h
e
co
n
f
u
s
io
n
m
atr
ice
s
an
d
th
e
r
elate
d
p
er
f
o
r
m
a
n
ce
p
ar
a
m
eter
s
f
o
r
2
0
1
3
an
d
2
0
1
8
s
leep
e
d
f
d
ata;
also
,
b
o
th
tab
les
in
cl
u
d
e
th
e
r
es
u
lts
o
f
b
o
th
FP
z
-
C
z
an
d
P
z
-
Oz
ch
a
n
n
els.
T
r
u
e
p
o
s
itiv
e
v
al
u
es
ar
e
r
ep
r
esen
ted
in
th
e
m
ai
n
d
iag
o
n
al
o
f
th
e
co
n
f
u
s
io
n
m
atr
i
x
.
Fo
r
all
co
lu
m
n
s
,
tr
u
e
p
o
s
itiv
e
n
u
m
b
er
s
ar
e
h
i
g
h
er
co
m
p
ar
ed
to
o
th
er
n
u
m
b
er
s
.
T
h
e
p
r
ed
ictio
n
p
er
f
o
r
m
a
n
ce
p
ar
a
m
eter
s
,
p
r
ec
is
io
n
,
F1
s
co
r
e,
r
e
ca
ll,
an
d
s
p
ec
i
f
icit
y
al
s
o
s
h
o
w
n
i
n
th
e
tab
le.
P
er
f
o
r
m
a
n
ce
i
s
s
li
g
h
t
l
y
lo
w
f
o
r
N1
clas
s
co
m
p
ar
ed
to
o
th
er
class
e
s
,
b
u
t
r
ec
all
s
i
g
n
if
ican
t
l
y
co
n
v
i
n
ci
n
g
co
m
p
ar
ed
to
th
e
ex
is
t
in
g
liter
at
u
r
e.
T
h
e
liter
atu
r
e
av
ailab
le
h
as
m
o
s
tl
y
r
ec
o
r
d
ed
th
e
an
al
y
s
i
s
p
er
f
o
r
m
ed
f
o
r
s
leep
s
ta
g
es,
wh
er
e
th
e
N1
s
tag
e
i
s
p
ar
t
o
f
it.
T
h
e
r
esu
lts
o
f
th
e
N1
class
i
f
i
ca
tio
n
co
m
p
ar
ed
to
o
th
er
s
leep
class
e
s
ar
e
n
o
t
co
n
v
i
n
ci
n
g
.
O
u
r
r
esu
l
ts
s
h
o
w
th
at
t
h
e
m
o
d
el
w
e
p
r
o
p
o
s
ed
w
o
r
k
s
s
i
g
n
if
ican
tl
y
b
etter
f
o
r
th
e
N1
s
leep
s
tag
e
al
s
o
.
T
h
e
p
er
f
o
r
m
a
n
ce
is
v
er
if
ied
f
o
r
b
o
th
E
E
G
ch
an
n
el
s
an
d
f
o
r
b
o
th
d
ata
s
ets.
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
t
h
e
m
o
d
el
i
s
i
m
p
r
o
v
ed
f
o
r
t
h
e
f
o
l
lo
w
i
n
g
r
ea
s
o
n
s
:
i)
T
h
e
s
leep
s
tag
e
s
A
ler
t,
d
r
o
w
s
y
,
an
d
s
lee
p
ar
e
s
eq
u
en
tial
i
n
n
atu
r
e
a
n
d
ev
er
y
n
e
x
t sta
g
e
is
th
e
tr
an
s
itio
n
f
r
o
m
t
h
e
p
r
ev
io
u
s
s
t
a
g
e
an
d
is
r
elate
d
to
th
e
p
r
ev
io
u
s
s
tate.
He
n
ce
th
e
ap
p
licatio
n
o
f
s
eq
u
en
ce
t
o
s
eq
u
en
ce
lear
n
in
g
ap
p
r
o
ac
h
is
a
p
r
ef
er
r
ed
ch
o
ice
;
ii)
T
h
e
u
s
e
o
f
atte
n
tio
n
d
ec
o
d
er
an
d
B
i
-
L
ST
M
h
as
i
m
p
r
o
v
ed
th
e
p
er
f
o
r
m
an
ce
;
ii
i)
Use
o
f
R
esNe
ts
allo
w
s
u
s
to
h
av
e
a
d
ee
p
er
n
et
w
o
r
k
,
w
it
h
o
u
t
co
m
p
r
o
m
is
i
n
g
o
n
tr
ain
i
n
g
er
r
o
r
an
d
al
s
o
allo
w
s
lear
n
in
g
te
m
p
o
r
al
a
n
d
f
r
eq
u
e
n
c
y
d
o
m
ai
n
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
2
9
-
3538
3536
f
ea
t
u
r
es
w
i
th
o
u
t
h
av
in
g
e
x
tr
a
la
y
er
s
;
i
v
)
T
h
e
lo
s
s
ca
lcu
l
atio
n
p
r
o
ce
d
u
r
e
u
s
ed
h
elp
s
t
o
d
ea
l
w
i
th
clas
s
i
m
b
alan
ce
p
r
o
b
le
m
s
e
x
is
ti
n
g
i
n
th
e
d
ata
s
ets
,
i
v
)
A
s
i
m
i
lar
ap
p
r
o
ac
h
w
i
th
m
i
n
i
m
u
m
ch
a
n
g
es
ca
n
b
e
ad
ap
ted
f
o
r
class
i
f
icatio
n
p
r
o
b
lem
s
w
h
ich
h
a
v
e
s
eq
u
en
t
ial
b
eh
a
v
io
r
an
d
also
class
i
m
b
ala
n
ce
p
r
o
b
le
m
.
T
ab
le
2
.
C
o
n
f
u
s
io
n
m
atr
i
x
an
d
th
e
p
er
-
clas
s
p
er
f
o
r
m
a
n
ce
f
o
r
2
0
1
3
s
leep
e
d
f
d
ata
D
a
t
a
S
e
t
s
El
e
c
t
r
o
d
e
s
P
r
e
d
i
c
t
e
d
P
e
r
f
o
r
man
c
e
ma
t
r
i
x
f
o
r
e
a
c
h
c
l
a
ss
(
%
)
C
l
a
sse
s
W
a
k
e
N1
S
l
e
e
p
PR
RE
F1
-
sco
r
e
SP
2
0
1
3
S
l
e
e
p
ED
F
F
p
z
-
Cz
W
a
k
e
7
2
1
2
5
2
1
3
2
2
8
9
.
5
3
9
4
.
7
0
9
1
.
7
4
8
8
.
9
7
N1
86
4
0
2
1
1
4
6
6
.
7
7
4
0
.
2
0
5
1
.
5
8
9
8
.
5
5
S
l
e
e
p
3
1
7
4
1
0
5
8
0
2
8
8
.
8
6
9
3
.
0
1
9
0
.
8
8
9
1
.
8
7
2
0
1
3
S
l
e
e
p
ED
F
Pz
-
Oz
W
a
k
e
7
1
9
8
5
2
0
3
4
8
8
9
.
9
2
9
3
.
7
2
9
1
.
7
8
8
8
.
4
5
N1
90
4
1
0
1
0
2
6
8
.
1
0
4
1
.
8
8
5
3
.
4
2
9
8
.
6
2
S
l
e
e
p
3
9
2
3
5
6
5
7
8
1
8
8
.
5
4
9
2
.
7
7
9
0
.
6
0
9
1
.
6
5
T
ab
le
3.
C
o
n
f
u
s
io
n
m
atr
i
x
an
d
th
e
p
er
class
p
er
f
o
r
m
a
n
ce
f
o
r
2
0
1
8
s
leep
e
d
f
d
ata
D
a
t
a
S
e
t
s
El
e
c
t
r
o
d
e
s
P
r
e
d
i
c
t
e
d
P
e
r
f
o
r
man
c
e
ma
t
r
i
x
f
o
r
e
a
c
h
c
l
a
ss
(
%
)
C
l
a
sse
s
W
a
k
e
N1
S
l
e
e
p
PR
RE
F1
-
sco
r
e
SP
2
0
1
8
S
l
e
e
p
ED
F
F
p
z
-
Cz
A
c
t
u
a
l
W
a
k
e
4
8
,
6
7
6
1
9
6
4
3
0
0
1
9
0
.
7
4
7
9
.
6
7
8
4
.
8
8
9
6
.
1
3
N1
6
0
8
3
1
2
1
1
2
2
0
2
0
5
9
.
9
2
6
3
.
0
9
6
1
.
4
6
9
5
.
2
3
S
l
e
e
p
6
3
3
6
5
1
2
1
1
0
4
1
0
6
9
0
.
0
8
9
5
.
3
9
9
2
.
6
5
8
5
.
7
3
2
0
1
8
S
l
e
e
p
ED
F
Pz
-
Oz
W
a
k
e
4
8
4
9
2
1
9
8
8
3
1
6
1
9
0
.
4
0
7
9
.
7
8
8
4
.
7
5
9
5
.
9
9
N1
5
9
9
8
1
2
0
8
2
2
1
3
5
5
9
.
7
6
6
2
.
6
7
6
1
.
1
8
9
5
.
2
1
S
l
e
e
p
6
2
9
2
5
2
0
4
1
0
4
0
6
7
9
0
.
0
5
9
5
.
1
5
9
2
.
5
2
8
5
.
6
4
T
ab
le
4.
P
er
f
o
r
m
a
n
ce
co
m
p
ar
i
s
o
n
w
it
h
o
th
er
r
elate
d
alg
o
r
ith
m
s
A
u
t
h
o
r
s
D
a
t
a
se
t
El
e
c
t
r
o
d
e
s
N
o
.
o
f
C
l
a
sse
s
CV
O
v
e
r
a
l
l
A
c
c
u
r
a
c
y
O
v
e
r
a
l
l
F1
sco
r
e
C
o
h
e
n
K
a
p
p
a
A.
R.
H
a
ssa
n
et
a
l
.
[8
]
S
l
e
e
p
-
e
d
f
d
a
t
a
Pz
-
Oz
05
20
9
0
.
8
8
0
.
0
--
A.
S
u
p
r
a
t
a
k
et
a
l
.
[
1
6
]
S
l
e
e
p
-
e
d
f
d
a
t
a
F
p
z
-
Cz
05
20
8
2
.
0
7
6
.
9
0
.
7
6
O.
T
si
n
a
l
i
s
et
al
.
[2
5
]
S
l
e
e
p
e
d
f
d
a
t
a
F
p
z
-
Cz
05
20
7
8
.
9
7
3
.
9
--
S
a
j
a
d
M
o
u
s
a
v
i
et
al
.
[2
1
]
S
l
e
e
p
e
d
f
-
2
0
1
3
d
a
t
a
F
p
z
-
Cz
05
20
8
4
.
2
6
7
9
.
6
6
0
.
7
9
S
a
j
a
d
M
o
u
s
a
v
i
et
al
.
[2
1
]
S
l
e
e
p
e
d
f
-
2
0
1
3
d
a
t
a
Pz
-
Oz
05
20
8
2
.
8
3
8
2
.
8
3
0
.
7
7
S
a
j
a
d
M
o
u
sav
i
et
al
.
[2
1
]
S
l
e
e
p
e
d
f
-
2
0
1
8
d
a
t
a
F
p
z
-
Cz
05
10
8
0
.
0
3
8
0
.
0
3
0
.
7
3
S
a
j
a
d
M
o
u
s
a
v
i
et
al
.
[2
1
]
S
l
e
e
p
e
d
f
-
2
0
1
8
d
a
t
a
Pz
-
Oz
05
10
7
7
.
5
6
7
7
.
5
6
6
8
.
9
4
M
i
k
i
t
o
O
g
i
n
o
et
al
.
[2
6
]
M
i
n
d
W
a
v
e
,
N
e
u
r
o
sk
y
F
p
1
-
A1
05
20
7
2
.
7
0
NA
--
C
N
N
-
B
i
L
S
T
M
(
O
u
r
a
l
t
e
r
n
a
t
e
me
t
h
o
d
)
S
l
e
e
p
e
d
f
-
2
0
1
3
d
a
t
a
F
p
z
-
Cz
02
20
9
0
.
2
5
9
3
.
3
2
--
A
l
e
r
t
N
et
(
F
p
z
-
Cz)
S
l
e
e
p
e
d
f
-
2
0
1
3
d
a
t
a
F
p
z
-
Cz
03
20
8
7
.
9
2
7
8
.
0
6
0
.
7
8
A
l
e
r
t
N
et
(Pz
-
O
z
)
S
l
e
e
p
e
d
f
-
2
0
1
3
d
a
t
a
Pz
-
Oz
03
20
8
7
.
7
3
7
8
.
6
0
0
.
7
9
A
l
e
r
t
N
et
-
(
F
p
z
-
C
z
)
S
l
e
e
p
e
d
f
-
2
0
1
8
d
a
t
a
F
p
z
-
Cz
03
10
8
7
.
0
5
7
9
.
6
6
0
.
7
9
A
l
e
r
t
N
et
-
(Pz
-
O
z
)
S
l
e
e
p
e
d
f
-
2
0
1
8
d
a
t
a
Pz
-
Oz
03
10
8
6
.
9
2
7
9
.
4
8
0
.
7
9
5.
CO
NCLU
SI
O
N
T
h
e
p
r
o
p
o
s
ed
d
ee
p
n
eu
r
al
n
et
w
o
r
k
-
b
ased
m
o
d
el
ar
c
h
ite
ctu
r
e
f
o
r
E
E
G
b
ased
d
r
iv
er
aler
tn
es
s
d
etec
tio
n
u
s
es
R
e
s
Net
s
an
d
B
i
-
L
ST
M,
s
eq
u
en
ce
to
t
h
e
s
eq
u
e
n
ce
lear
n
i
n
g
ap
p
r
o
ac
h
.
T
h
e
R
esNets
ar
e
u
s
ed
f
o
r
f
ea
t
u
r
e
ex
tr
ac
tio
n
,
w
it
h
t
h
e
s
k
ip
-
co
n
n
ec
t
io
n
s
h
elp
s
to
r
etai
n
th
e
i
n
f
o
r
m
atio
n
f
r
o
m
t
h
e
al
ter
n
ate
la
y
er
s
.
T
h
is
h
elp
s
i
n
r
etai
n
i
n
g
t
h
e
f
ea
tu
r
e
s
w
it
h
o
u
t
ad
d
in
g
an
y
e
x
tr
a
la
y
er
s
a
n
d
lear
n
s
d
ee
p
er
in
to
t
h
e
n
et
w
o
r
k
w
i
th
n
o
in
cr
ea
s
e
i
n
tr
ain
in
g
er
r
o
r
.
T
h
e
s
eq
u
en
ce
to
s
eq
u
en
ce
m
o
d
el
h
elp
s
i
n
lear
n
i
n
g
t
h
e
co
m
p
lex
d
ep
en
d
en
cies
p
r
esen
t
in
t
h
e
E
E
G
s
i
g
n
al.
T
h
e
m
o
d
el
’
s
p
er
f
o
r
m
a
n
ce
f
o
r
t
h
e
N1
s
leep
s
tag
e
i
s
b
etter
c
o
m
p
a
r
e
d
t
o
e
x
i
s
t
i
n
g
m
o
d
e
l
s
.
H
e
n
c
e
,
t
h
e
m
o
d
e
l
c
a
n
b
e
u
s
e
d
f
o
r
f
u
t
u
r
e
u
s
a
g
e
o
f
a
u
t
o
m
a
t
i
c
c
l
a
s
s
i
f
i
c
a
t
i
o
n
o
f
s
l
e
e
p
u
s
i
n
g
r
a
w
E
E
G
s
ig
n
al
s
.
RE
F
E
R
E
NC
E
S
[1
]
O.
Rich
,
T
o
n
y
a
L
,
a
n
d
Be
rn
a
d
e
tt
e
T
G
il
li
c
k
,
“
El
e
c
tro
d
e
P
lac
e
m
e
n
t
in
T
ra
n
sc
ra
n
ial
Dir
e
c
t
Cu
rre
n
t
S
ti
m
u
latio
n
-
Ho
w
Re
li
a
b
le Is t
h
e
De
term
in
a
ti
o
n
o
f
C3
/C4
,
”
Br
a
in
sc
ien
c
e
s
,
v
o
l.
9
,
n
o
.
3
,
p
p
.
6
9
,
2
0
1
9
.
[2
]
Ra
jee
v
Ag
a
r
wa
l
a
n
d
Je
a
n
G
o
t
m
a
n
,
“
Co
m
p
u
ter
-
a
ss
isted
sle
e
p
s
tag
in
g
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Bi
o
me
d
ica
l
En
g
i
n
e
e
rin
g
,
v
o
l.
4
8
,
n
o
.
1
2
,
p
p
.
1
4
1
2
-
1
4
2
3
,
2
0
0
1
.
[3
]
S
lee
p
F
o
u
n
d
a
ti
o
n
,
2
0
2
0
,
[
On
l
in
e
].
A
v
a
i
la
b
le:
h
tt
p
s://
w
ww
.
sl
e
e
p
f
o
u
n
d
a
ti
o
n
.
o
rg
/h
o
w
-
sle
e
p
-
w
o
rk
s/sta
g
e
s
-
of
-
sle
e
p
.
[4
]
Re
c
h
tsc
h
a
ff
e
n
A
,
“
A
m
a
n
u
a
l
f
o
r
sta
n
d
a
rd
ize
d
term
in
o
lo
g
y
,
t
e
c
h
n
iq
u
e
s
a
n
d
sc
o
rin
g
s
y
ste
m
f
o
r
s
lee
p
sta
g
e
s
in
h
u
m
a
n
su
b
jec
ts,”
Bra
i
n
i
n
fo
rm
a
ti
o
n
se
rv
ice
,
1
9
6
8
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
n
t J
E
lec
&
C
o
m
p
E
n
g
I
SS
N:
2
0
8
8
-
8708
A
lert
N
et
:
Dee
p
co
n
vo
lu
tio
n
a
l
-
r
ec
u
r
r
en
t n
eu
r
a
l n
etw
o
r
k
mo
d
el
fo
r
d
r
ivin
g
…
(
P
.
C
.
N
i
s
s
ima
g
o
u
d
a
r
)
3537
[5
]
R.
B.
Be
rry
,
R.
Bro
o
k
s,
C.
E.
G
a
m
a
ld
o
,
C.
E.
,
Ha
rd
i
n
g
,
S
.
M
.
,
M
a
rc
u
s,
C.
,
a
n
d
V
a
u
g
h
n
,
B.
V.,
“
T
h
e
AA
S
M
M
a
n
u
a
l
f
o
r
th
e
sc
o
ri
n
g
o
f
sle
e
p
a
n
d
a
ss
o
c
iate
d
e
v
e
n
ts,
Ru
les
,
T
e
r
m
in
o
lo
g
y
,
a
n
d
T
e
c
h
n
ica
l
S
p
e
c
if
ica
ti
o
n
s,”
Ame
ric
a
n
Aca
d
e
my
o
f
S
lee
p
M
e
d
i
c
in
e
,
v
o
l.
1
7
6
,
p
p
.
1
-
7
.
2
0
1
2
.
[6
]
M
o
se
r,
D.,
A
n
d
e
re
r,
P
.
,
G
ru
b
e
r,
G
.
,
P
a
ra
p
a
ti
c
s,
S
.
,
L
o
re
tz,
E
.
,
B
o
e
c
k
,
M
e
t
a
l.
,
“
S
lee
p
c
las
sif
ica
ti
o
n
a
c
c
o
r
d
in
g
t
o
AA
S
M
a
n
d
Re
c
h
tsc
h
a
ff
e
n
&
Ka
l
e
s: effe
c
ts o
n
sle
e
p
sc
o
rin
g
p
a
ra
m
e
ters
,
”
S
lee
p
,
v
o
l.
3
2
,
n
o
.
2
,
p
p
.
1
3
9
-
1
4
9
,
2
0
0
9
.
[7
]
L
ib
o
u
re
l,
P
.
A
.
,
C
o
rn
e
y
ll
ie,
A
.
,
L
u
p
p
i,
P
.
H.,
C
h
o
u
v
e
t,
G
.
,
a
n
d
G
e
r
v
a
so
n
i,
D
.
,
“
Un
s
u
p
e
rv
ise
d
o
n
li
n
e
c
las
sif
ier
in
sle
e
p
sc
o
rin
g
f
o
r
sle
e
p
d
e
p
riv
a
ti
o
n
stu
d
ies
,
”
S
lee
p
,
v
o
l.
3
8
,
n
o
.
5
,
p
p
.
8
1
5
-
8
2
8
,
2
0
1
5
.
[8
]
A
.
R.
Ha
ss
a
n
a
n
d
A
.
S
u
b
a
si,
“
A
d
e
c
isio
n
su
p
p
o
rt
sy
ste
m
f
o
r
a
u
to
m
a
ted
id
e
n
ti
f
ica
ti
o
n
o
f
sle
e
p
sta
g
e
s
f
ro
m
sin
g
le
-
c
h
a
n
n
e
l
EE
G
sig
n
a
ls,”
Kn
o
wle
d
g
e
-
Ba
se
d
S
y
ste
ms
,
v
o
l.
1
2
8
,
p
p
.
1
1
5
-
1
2
4
,
2
0
1
7
.
[9
]
E.
A
li
c
k
o
v
i
c
a
n
d
A
.
S
u
b
a
si,
"
En
se
m
b
le
S
V
M
M
e
th
o
d
f
o
r
A
u
to
m
a
ti
c
S
lee
p
S
tag
e
Clas
si
fica
ti
o
n
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
In
str
u
me
n
ta
t
io
n
a
n
d
M
e
a
su
re
me
n
t,
v
o
l.
6
7
,
n
o
.
6
,
p
p
.
1
2
5
8
-
1
2
6
5
,
2
0
1
8
.
[1
0
]
S
u
n
a
g
a
w
a
,
G
.
A
.
,
S
é
i,
H.,
S
h
im
b
a
,
S
.
,
Ura
d
e
,
Y.,
a
n
d
Ue
d
a
,
H.
R
,
“
F
a
ste
r:
a
n
u
n
su
p
e
rv
ise
d
f
u
ll
y
a
u
to
m
a
ted
sle
e
p
sta
g
in
g
m
e
th
o
d
f
o
r
m
ice
,
”
Ge
n
e
s
to
Ce
ll
s,
v
o
l
.
1
8
,
n
o
.
6
,
p
p
.
5
0
2
-
5
1
8
,
2
0
1
3
.
[1
1
]
I.
G
a
th
a
n
d
A
.
B.
G
e
v
a
,
"
Un
su
p
e
rv
ise
d
o
p
ti
m
a
l
f
u
z
z
y
c
lu
ste
rin
g
,
"
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
rn
An
a
lys
is
a
n
d
M
a
c
h
in
e
In
telli
g
e
n
c
e
,
v
o
l
.
1
1
,
n
o
.
7
,
p
p
.
7
7
3
-
7
8
0
,
1
9
8
9
.
[1
2
]
M
ik
o
lo
v
,
T
.
,
Ka
ra
f
iát,
M
.
,
Bu
rg
e
t
,
L
.
,
Če
rn
o
c
k
ý
,
J.,
a
n
d
Kh
u
d
a
n
p
u
r,
S
.
,
“
Re
c
u
rre
n
t
n
e
u
ra
l
n
e
tw
o
rk
-
b
a
se
d
lan
g
u
a
g
e
m
o
d
e
l,
”
El
e
v
e
n
th
a
n
n
u
a
l
c
o
n
fer
e
n
c
e
o
f
t
h
e
in
ter
n
a
ti
o
n
a
l
s
p
e
e
c
h
c
o
mm
u
n
ica
ti
o
n
a
ss
o
c
i
a
ti
o
n
,
2
0
1
0
,
p
p
.
1
0
4
5
-
1
0
4
8
.
[1
3
]
A
.
G
ra
v
e
s,
A
.
M
o
h
a
m
e
d
a
n
d
G
.
Hin
to
n
,
"
S
p
e
e
c
h
re
c
o
g
n
it
io
n
w
it
h
d
e
e
p
re
c
u
rre
n
t
n
e
u
ra
l
n
e
tw
o
rk
s,
"
IEE
E
In
ter
n
a
t
io
n
a
l
C
o
n
fer
e
n
c
e
o
n
Ac
o
u
stics
,
S
p
e
e
c
h
a
n
d
S
ig
n
a
l
Pro
c
e
ss
in
g
,
2
0
1
3
,
p
p
.
6
6
4
5
-
6
6
4
9
.
[1
4
]
A
n
d
re
j
Ka
rp
a
th
y
a
n
d
L
i
F
e
i
-
F
e
i
,
“
De
e
p
v
isu
a
l
-
se
m
a
n
ti
c
a
li
g
n
m
e
n
ts
f
o
r
g
e
n
e
ra
ti
n
g
i
m
a
g
e
d
e
sc
r
ip
ti
o
n
s,”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Pa
tt
e
rn
A
n
a
lys
is
a
n
d
M
a
c
h
i
n
e
In
telli
g
e
n
c
e
,
v
o
l.
3
9
,
n
o
.
4
,
p
p
.
6
6
4
-
6
7
6
,
2
0
1
7
.
[1
5
]
P
.
R.
Da
v
id
so
n
,
R.
D
.
Jo
n
e
s
,
a
n
d
M
.
T
.
R
.
P
e
iri
s,
"
De
tec
ti
n
g
Be
h
a
v
io
ra
l
M
icro
sle
e
p
s
u
sin
g
E
EG
a
n
d
L
S
T
M
Re
c
u
rre
n
t
Ne
u
ra
l
Ne
t
w
o
rk
s,"
2
0
0
5
IEE
E
E
n
g
i
n
e
e
rin
g
i
n
M
e
d
icin
e
a
n
d
Bi
o
l
o
g
y
2
7
th
A
n
n
u
a
l
Co
n
fer
e
n
ce
,
2
0
0
5
,
p
p
.
5
7
5
4
-
5
7
5
7
.
[1
6
]
A
.
S
u
p
ra
tak
,
H.
Do
n
g
,
C.
W
u
,
a
n
d
Yik
e
G
u
o
,
“
De
e
p
sle
e
p
n
e
t:
A
m
o
d
e
l
f
o
r
a
u
to
m
a
ti
c
sle
e
p
sta
g
e
s
c
o
rin
g
b
a
se
d
o
n
ra
w
sin
g
le
-
c
h
a
n
n
e
l
EE
G
,
”
IEE
E
T
ra
n
sa
c
ti
o
n
s
o
n
Ne
u
ra
l
S
y
ste
ms
a
n
d
Reh
a
b
il
i
ta
ti
o
n
En
g
in
e
e
rin
g
,
v
o
l.
2
5
,
n
o
.
1
1
,
p
p
.
1
9
9
8
-
20
0
8
,
2
0
1
7
.
[1
7
]
A
S
o
rs,
S
.
Bo
n
n
e
t,
S
e
b
a
stien
M
i
re
k
,
L
.
V
e
rc
u
e
il
,
a
n
d
J
.
-
F
.
o
is
P
a
y
e
n
,
“
A
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
t
w
o
rk
f
o
r
sle
e
p
sta
g
e
sc
o
rin
g
f
ro
m
ra
w
sin
g
le
-
c
h
a
n
n
e
l
EE
G
,
”
Bi
o
me
d
ica
l
S
ig
n
a
l
Pro
c
e
ss
in
g
a
n
d
C
o
n
tro
l,
v
o
l.
4
2
,
p
p
.
1
0
7
-
1
1
4
,
2
0
1
8
.
[1
8
]
S
.
W
u
,
S
.
Zh
o
n
g
,
a
n
d
Y
.
L
iu
,
“
D
e
e
p
re
sid
u
a
l
lea
rn
i
n
g
f
o
r
im
a
g
e
st
e
g
a
n
a
l
y
sis,”
M
u
lt
ime
d
ia
t
o
o
ls
a
n
d
a
p
p
l
ica
ti
o
n
s,
v
o
l.
7
7
,
n
o
.
9
,
p
p
.
1
0
4
3
7
-
1
0
4
5
3
,
2
0
1
8
.
[1
9
]
A
.
S
h
e
rstin
sk
y
,
“
F
u
n
d
a
m
e
n
tals
o
f
th
e
re
c
u
rre
n
t
n
e
u
ra
l
n
e
tw
o
rk
(rn
n
)
a
n
d
l
o
n
g
s
h
o
rt
-
term
m
e
m
o
r
y
(
L
S
T
M
)
n
e
tw
o
rk
,
”
P
h
y
sic
a
D:
No
n
li
n
e
a
r
P
h
e
n
o
m
e
n
a
,
v
o
l.
4
0
4
,
2
0
2
0
,
A
rt
.
n
o
.
1
3
2
3
0
6
.
[2
0
]
I
.
S
u
tsk
e
v
e
r,
O
.
V
i
n
y
a
ls,
a
n
d
Q
.
V
.
L
e
,
“
S
e
q
u
e
n
c
e
to
se
q
u
e
n
c
e
lea
rn
in
g
w
it
h
n
e
u
ra
l
n
e
tw
o
rk
s,”
NIPS
'1
4
:
Pro
c
e
e
d
in
g
s
o
f
t
h
e
2
7
th
I
n
te
rn
a
ti
o
n
a
l
Co
n
fer
e
n
c
e
o
n
Ne
u
ra
l
In
f
o
rm
a
ti
o
n
Pro
c
e
ss
in
g
S
y
ste
ms
,
v
o
l.
2
,
p
p
.
3
1
0
4
-
3
1
1
2
,
2
0
1
4
.
[2
1
]
M
o
u
sa
v
i,
S
.
,
Af
g
h
a
h
,
F
.
,
a
n
d
A
c
h
a
r
y
a
,
U.
R
.
,
“
S
lee
p
EE
G
Ne
t:
A
u
to
m
a
ted
sle
e
p
sta
g
e
sc
o
rin
g
w
it
h
se
q
u
e
n
c
e
-
to
-
se
q
u
e
n
c
e
d
e
e
p
lea
rn
i
n
g
a
p
p
r
o
a
c
h
,
”
Pl
o
S
o
n
e
,
v
o
l.
1
4
,
n
o
.
5
,
2
0
1
9
,
A
rt
.
n
o
.
e
0
2
1
6
4
5
6
.
[2
2
]
S
.
W
a
n
g
,
W
e
i
L
iu
,
Jia
W
u
,
L
.
C
a
o
,
Q.
M
e
n
g
,
a
n
d
P
.
J
Ke
n
n
e
d
y
,
“
T
ra
in
in
g
d
e
e
p
n
e
u
ra
l
n
e
tw
o
rk
s
o
n
im
b
a
lan
c
e
d
d
a
ta
se
ts,”
2
0
1
6
I
n
ter
n
a
t
io
n
a
l
J
o
in
t
C
o
n
fer
e
n
c
e
o
n
Ne
u
r
a
l
Ne
t
wo
rk
s
(
I
J
CN
N),
V
a
n
c
o
u
v
e
r,
BC,
Ca
n
a
d
a
,
2
0
1
6
,
p
p
.
4
3
6
8
-
4
3
7
4
.
[2
3
]
S
.
M
.
V
ieira
,
U
.
Ka
y
m
a
k
,
a
n
d
J
.
M
c
S
o
u
sa
,
“
Co
h
e
n
’s
k
a
p
p
a
c
o
e
ff
ici
e
n
t
a
s
a
p
e
r
f
o
r
m
a
n
c
e
m
e
a
s
u
re
f
o
r
fe
a
tu
re
se
lec
ti
o
n
,
”
In
ter
n
a
ti
o
n
a
l
C
o
n
fer
e
n
c
e
o
n
Fu
zz
y
S
y
ste
ms
,
Ba
rc
e
lo
n
a
,
S
p
a
in
,
2
0
1
0
,
p
p
.
1
-
8.
[2
4
]
H.
Je
o
n
g
a
n
d
W
.
P
a
rk
,
"
De
v
e
lo
p
in
g
a
n
d
Ev
a
lu
a
ti
n
g
a
M
ix
e
d
S
e
n
so
r
S
m
a
rt
Ch
a
ir
S
y
ste
m
f
o
r
Re
a
l
-
ti
m
e
P
o
stu
r
e
Clas
sif
ic
a
ti
o
n
:
Co
m
b
in
in
g
P
re
ss
u
re
a
n
d
Dista
n
c
e
se
n
so
rs,"
IEE
E
J
o
u
rn
a
l
o
f
Bi
o
me
d
ica
l
a
n
d
He
a
lt
h
In
f
o
rm
a
ti
c
s,
2
0
2
0
.
[2
5
]
T
sin
a
li
s,
O.,
M
a
tt
h
e
w
s,
P
.
M
.
,
Gu
o
,
Y.
,
a
n
d
Zaf
e
iri
o
u
,
S
.
,
“
A
u
to
m
a
ti
c
sle
e
p
sta
g
e
sc
o
rin
g
w
it
h
sin
g
le
-
c
h
a
n
n
e
l
EE
G
u
sin
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s,”
a
rXiv p
re
p
rin
t
a
rXiv:1
6
1
0
.
0
1
6
8
3
,
2
0
1
6
.
[2
6
]
M
ik
it
o
Og
in
o
,
Ya
su
e
M
it
su
k
u
ra
,
"
P
o
rtab
le
Dro
w
sin
e
ss
De
tec
ti
o
n
th
ro
u
g
h
Us
e
o
f
a
P
re
f
ro
n
tal
S
in
g
le
-
Ch
a
n
n
e
l
El
e
c
tro
e
n
c
e
p
h
a
lo
g
ra
m
,
"
S
e
n
so
rs
,
v
o
l.
1
8
,
n
o
.
1
2
,
2
0
1
8
,
A
rt
.
n
o
.
4
4
7
7
.
B
I
O
G
RAP
H
I
E
S
O
F
AUTH
O
RS
Pra
b
h
a
v
a
th
i
C.
Niss
i
m
a
g
o
u
d
a
r
is
c
u
rre
n
tl
y
p
u
rsu
in
g
P
h
D
a
n
d
w
o
rk
in
g
a
s
As
so
c
iate
P
ro
f
e
ss
o
r,
in
S
c
h
o
o
l
o
f
ECE
,
KL
ET
U,
Hu
b
b
a
ll
i.
He
r
re
se
a
rc
h
is
in
t
h
e
a
re
a
s
o
f
a
u
to
m
o
ti
v
e
e
m
b
e
d
d
e
d
s
y
ste
m
s,
A
d
v
a
n
c
e
Driv
e
r
A
ss
ist
a
n
c
e
S
y
ste
m
s (
A
D
A
S
)
a
n
d
In
telli
g
e
n
t
Bio
m
e
d
ica
l
s
y
ste
m
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
2
0
8
8
-
8708
I
n
t J
E
lec
&
C
o
m
p
E
n
g
,
Vo
l.
11
,
No
.
4
,
A
u
g
u
s
t 2
0
2
1
:
3
5
2
9
-
3538
3538
An
il
k
u
m
a
r
V
.
N
a
n
d
i
is
c
u
rre
n
tl
y
w
o
rk
in
g
a
s
P
ro
f
e
ss
o
r,
in
De
p
a
rtm
e
n
t
o
f
ECE
,
B.
V
.
C.
E.
T
Hu
b
b
a
ll
i,
In
d
ia.
He
p
o
st
g
ra
d
u
a
ted
f
ro
m
II
T
Kh
a
rg
h
p
u
r,
In
d
ia
in
t
h
e
a
re
a
o
f
M
EM
S
a
n
d
re
c
e
iv
e
d
P
h
D
f
ro
m
V
T
U,
Be
la
g
a
v
i,
In
d
ia.
His
re
se
a
rc
h
is
in
th
e
a
re
a
o
f
V
L
S
I
d
e
sig
n
,
E
m
b
e
d
d
e
d
S
y
ste
m
s,
M
EM
S
a
n
d
I
n
telli
g
e
n
t
Bi
o
m
e
d
ica
l
S
y
ste
m
s.
Aa
k
a
n
k
sha
K
.
Pa
til
is
w
o
rk
in
g
a
s
a
n
u
n
d
e
rg
ra
d
u
a
te
re
se
a
rc
h
stu
d
e
n
t
a
t
t
h
e
KL
ET
U,
In
d
ia.
He
r
re
se
a
rc
h
in
tere
sts
in
c
lu
d
e
d
e
v
e
lo
p
m
e
n
t
o
f
d
e
e
p
lea
rn
in
g
a
lg
o
rit
h
m
s
f
o
r
a
u
to
m
o
ti
v
e
in
telli
g
e
n
t
s
y
ste
m
s.
S
h
e
w
o
rk
e
d
a
t
Co
n
ti
n
e
n
tal
a
u
to
m
o
ti
v
e
a
s
a
n
in
tern
sh
ip
train
e
e
d
u
rin
g
h
e
r
f
in
a
l
y
e
a
r
o
f
g
ra
d
u
a
ti
o
n
.
G
ire
e
s
h
a
H
.
M
.
is
w
o
rk
in
g
a
s
As
sista
n
t
P
ro
f
e
ss
o
r
in
th
e
S
c
h
o
o
l
o
f
ECE
a
t
KL
E
T
e
c
h
n
o
l
o
g
ica
l
Un
iv
e
rsit
y
,
V
id
y
a
n
a
g
a
r,
Hu
b
b
a
ll
i.
He
p
o
st
g
ra
d
u
a
ted
i
n
B
io
m
e
d
ica
l
S
ig
n
a
l
P
ro
c
e
ss
in
g
a
n
d
In
stru
-
m
e
n
tatio
n
f
ro
m
S
JCE
M
y
so
re
.
His
re
se
a
rc
h
is
in
th
e
a
re
a
o
f
so
f
t
c
o
m
p
u
ti
n
g
,
sig
n
a
l
a
n
d
im
a
g
e
p
ro
c
e
ss
in
g
,
a
n
d
a
d
v
a
n
c
e
d
r
iv
e
r
a
ss
istan
c
e
s
y
ste
m
s.
Evaluation Warning : The document was created with Spire.PDF for Python.