T
E
L
K
O
M
NIKA
T
elec
o
mm
un
ica
t
io
n Co
m
pu
t
i
ng
E
lect
ro
nics
a
nd
Co
ntr
o
l
Vo
l.
24
,
No
.
4
,
A
u
g
u
s
t
20
26
,
p
p
.
1
294
~1
3
0
6
I
SS
N:
1
6
9
3
-
6
9
3
0
,
DOI
:
1
0
.
1
2
9
2
8
/
T
E
L
KOM
NI
K
A
.
v
24
i
4
.
27659
1294
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//telko
mn
ika
.
u
a
d
.
a
c.
i
d
Cla
ss
ificatio
n of
P30
0
ev
ent
-
rela
te
d potent
ia
ls usin
g
SNN,
CN
N
a
nd LST
M
deep
l
ea
rning
m
o
d
els
Ahla
a
m
M
.
Sa
ed
1
,
I
btiha
l F
a
w
zi
E
ls
ha
m
i
2
,
Ali I
.
E
lg
a
y
a
r
3
1
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
i
c
a
l
a
n
d
El
e
c
t
r
o
n
i
c
s E
n
g
i
n
e
e
r
i
n
g
,
B
r
i
g
h
t
S
t
a
r
U
n
i
v
e
r
si
t
y
,
B
r
e
g
a
,
L
i
b
y
a
2
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s E
n
g
i
n
e
e
r
i
n
g
,
C
o
l
l
e
g
e
o
f
El
e
c
t
r
i
c
a
l
a
n
d
El
e
c
t
r
o
n
i
c
s T
e
c
h
n
o
l
o
g
y
,
B
e
n
g
h
a
z
i
,
L
i
b
y
a
3
D
e
p
a
r
t
me
n
t
o
f
El
e
c
t
r
o
n
i
c
s E
n
g
i
n
e
e
r
i
n
g
,
U
n
i
v
e
r
si
t
y
o
f
B
e
n
g
h
a
z
i
,
B
e
n
g
h
a
z
i
,
L
i
b
y
a
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
Dec
11
,
2025
R
ev
i
s
ed
A
p
r
3
,
2
0
2
6
A
cc
ep
ted
Ma
y
25
,
2026
A
c
c
u
ra
te
c
las
si
f
ic
a
ti
o
n
o
f
P
3
0
0
e
v
e
n
t
-
re
late
d
p
o
ten
ti
a
ls
re
m
a
in
s
c
h
a
ll
e
n
g
in
g
d
u
e
t
o
th
e
c
o
m
p
lex
,
n
o
n
-
sta
ti
o
n
a
ry
,
a
n
d
lo
w
sig
n
a
l
-
to
-
n
o
ise
c
h
a
r
a
c
teristics
o
f
e
lec
tro
e
n
c
e
p
h
a
lo
g
ra
p
h
y
(
EEG
)
sig
n
a
ls
in
b
ra
in
-
c
o
m
p
u
ter
in
terf
a
c
e
(BCI)
s
y
ste
m
s.
P
3
0
0
-
b
a
se
d
d
e
v
ice
s,
su
c
h
a
s
th
e
P
3
0
0
sp
e
ll
e
r
,
e
n
a
b
l
e
c
o
m
m
u
n
ica
ti
o
n
f
o
r
p
a
ti
e
n
ts
w
it
h
se
v
e
r
e
m
o
to
r
im
p
a
ir
m
e
n
ts,
in
c
l
u
d
in
g
th
o
se
w
it
h
lo
c
k
e
d
-
in
sy
n
d
ro
m
e
;
h
o
w
e
v
e
r,
re
li
a
b
le
b
ra
in
sig
n
a
l
c
las
sif
ica
t
io
n
is
stil
l
a
c
rit
ica
l
li
m
it
a
ti
o
n
.
T
h
is
stu
d
y
p
re
se
n
ts
a
c
o
m
p
a
ra
ti
v
e
e
v
a
lu
a
ti
o
n
o
f
d
e
e
p
lea
rn
in
g
m
o
d
e
ls,
in
c
lu
d
in
g
c
o
n
v
o
lu
ti
o
n
a
l
n
e
u
ra
l
n
e
tw
o
rk
s
(CNN
),
lo
n
g
sh
o
rt
-
term
m
e
m
o
r
y
(L
S
T
M
)
n
e
tw
o
rk
s,
a
n
d
sp
ik
in
g
n
e
u
ra
l
n
e
tw
o
rk
s
(S
NN
),
f
o
r
P
3
0
0
sig
n
a
l
c
las
sif
ica
ti
o
n
.
S
NN
s
re
p
re
se
n
t
a
b
io
l
o
g
ica
ll
y
in
sp
ired
p
a
ra
d
ig
m
th
a
t
m
o
d
e
ls
th
e
d
isc
re
te,
ti
m
e
-
d
e
p
e
n
d
e
n
t
b
e
h
a
v
io
r
o
f
n
e
u
ra
l
sp
ik
in
g
a
c
ti
v
it
y
a
n
d
o
ff
e
r
s
a
d
v
a
n
tag
e
s
in
ter
m
s
o
f
e
n
e
rg
y
e
ff
ici
e
n
c
y
a
n
d
h
a
rd
w
a
re
i
m
p
le
m
e
n
tab
il
it
y
.
Ex
p
e
rime
n
tal
re
su
lt
s
d
e
m
o
n
stra
te
t
h
a
t
CNN
a
c
h
iev
e
d
th
e
h
ig
h
e
st
a
v
e
ra
g
e
c
las
sif
i
c
a
ti
o
n
a
c
c
u
ra
c
y
(8
1
.
0
4
%
),
f
o
ll
o
w
e
d
c
lo
se
ly
b
y
S
NN
(8
0
.
9
4
%
)
a
n
d
L
S
T
M
(8
0
.
6
0
%
).
A
lt
h
o
u
g
h
CN
N
slig
h
tl
y
o
u
t
p
e
rf
o
rm
e
d
th
e
o
th
e
r
m
o
d
e
ls,
S
NN
s
sh
o
w
e
d
c
o
m
p
a
ra
b
le
a
c
c
u
r
a
c
y
w
h
il
e
re
q
u
iri
n
g
f
e
w
e
r
train
in
g
sa
m
p
les
a
n
d
o
f
f
e
rin
g
p
o
ten
t
ial
b
e
n
e
f
it
s
f
o
r
lo
w
-
p
o
w
e
r
a
n
d
re
a
l
-
ti
m
e
BCI
s
y
ste
m
s.
T
h
e
se
f
in
d
in
g
s
h
ig
h
li
g
h
t
th
e
trad
e
-
o
f
f
s
b
e
tw
e
e
n
c
l
a
ss
i
f
ica
ti
o
n
p
e
rf
o
r
m
a
n
c
e
a
n
d
c
o
m
p
u
tatio
n
a
l
e
ff
ici
e
n
c
y
a
n
d
u
n
d
e
rl
in
e
t
h
e
p
ro
m
ise
o
f
S
NN
s
a
s
a
n
e
ff
icie
n
t
a
lt
e
rn
a
ti
v
e
f
o
r
P
3
0
0
-
b
a
se
d
BCI
a
p
p
li
c
a
ti
o
n
s.
K
ey
w
o
r
d
s
:
B
r
ain
-
co
m
p
u
ter
in
ter
f
ac
e
C
o
n
v
o
lu
tio
n
n
e
u
r
al
n
et
w
o
r
k
E
lectr
o
en
ce
p
h
alo
g
r
ap
h
y
L
o
n
g
-
s
h
o
r
t
-
ter
m
-
m
e
m
o
r
y
P
3
0
0
R
ec
ti
f
ied
lin
ea
r
u
n
it
Sp
ik
i
n
g
n
e
u
r
al
n
et
w
o
r
k
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
:
I
b
tih
al
Fa
w
zi
E
ls
h
a
m
i
Dep
ar
t
m
en
t o
f
E
lectr
o
n
ics E
n
g
in
ee
r
i
n
g
,
C
o
lle
g
e
o
f
E
lectr
ica
l a
n
d
E
lectr
o
n
ics T
ec
h
n
o
lo
g
y
B
en
g
h
az
i,
L
ib
y
a
E
m
ail: i
f
el
s
h
a
m
i
@
ce
et.
ed
u
.
l
y
1.
I
NT
RO
D
UCT
I
O
N
B
r
ain
-
co
m
p
u
ter
in
ter
f
ac
e
s
(
B
C
I
s
)
en
ab
le
d
ir
ec
t
co
m
m
u
n
icatio
n
b
et
w
ee
n
t
h
e
h
u
m
a
n
b
r
ain
an
d
ex
ter
n
al
d
ev
ices,
o
f
f
er
in
g
a
n
ess
e
n
tial
co
m
m
u
n
icatio
n
c
h
an
n
el
f
o
r
p
atien
t
s
w
it
h
s
e
v
er
e
m
o
to
r
i
m
p
air
m
e
n
ts
,
p
ar
ticu
lar
l
y
t
h
o
s
e
a
f
f
ec
ted
b
y
lo
ck
ed
-
i
n
s
y
n
d
r
o
m
e.
B
y
tr
a
n
s
lat
in
g
n
eu
r
al
ac
ti
v
it
y
i
n
to
co
n
tr
o
l
co
m
m
a
n
d
s
,
B
C
I
s
b
r
id
g
e
th
e
g
ap
b
et
w
ee
n
co
g
n
iti
o
n
an
d
ac
tio
n
an
d
allo
w
u
s
er
s
to
in
ter
ac
t
w
it
h
ass
i
s
t
iv
e
tech
n
o
lo
g
ies,
as
illu
s
tr
ated
in
Fi
g
u
r
e
1
.
Dep
en
d
in
g
o
n
t
h
e
s
ig
n
al
ac
q
u
is
itio
n
ap
p
r
o
ac
h
,
B
C
I
s
ca
n
b
e
cla
s
s
if
ied
i
n
to
i
n
v
a
s
i
v
e
s
y
s
te
m
s
,
w
h
ich
r
eq
u
ir
e
i
m
p
la
n
ted
elec
tr
o
d
es,
an
d
n
o
n
-
i
n
v
a
s
iv
e
s
y
s
te
m
s
t
h
at
r
el
y
o
n
elec
tr
o
en
ce
p
h
alo
g
r
ap
h
y
(
E
E
G)
s
ig
n
als
r
ec
o
r
d
ed
f
r
o
m
t
h
e
s
ca
lp
.
D
u
e
to
t
h
eir
s
af
et
y
,
a
cc
ess
ib
ilit
y
,
a
n
d
ea
s
e
o
f
u
s
e,
n
o
n
-
i
n
v
asi
v
e
E
E
G
-
b
ased
B
C
I
s
ar
e
t
h
e
m
o
s
t
w
id
el
y
ad
o
p
ted
d
esp
ite
th
e
in
h
er
en
t
co
m
p
le
x
it
y
a
n
d
n
o
n
-
li
n
ea
r
it
y
o
f
E
E
G
s
ig
n
als,
w
h
ic
h
m
a
k
e
s
ac
c
u
r
ate
b
r
ain
s
i
g
n
al
cla
s
s
i
f
ica
tio
n
c
h
allen
g
i
n
g
[
1
]
,
[
2
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
la
s
s
if
ica
tio
n
o
f P
3
0
0
ev
en
t
-
r
ela
ted
p
o
ten
tia
ls
u
s
in
g
S
N
N
,
C
N
N
a
n
d
LS
TM
…
(
A
h
la
a
m
M.
S
a
ed
)
1295
Fig
u
r
e
1
.
W
o
r
k
in
g
o
f
a
g
en
er
i
c
B
C
I
[
2
]
T
h
e
P3
0
0
-
b
ased
B
C
I
,
a
w
id
ely
u
s
ed
E
E
G
p
ar
ad
ig
m
,
elici
ts
a
p
o
s
itiv
e
ev
e
n
t
-
r
elate
d
p
o
ten
tial
(
E
R
P
)
ar
o
u
n
d
3
0
0
m
s
a
f
ter
task
-
r
ele
v
an
t
s
ti
m
u
li,
en
ab
li
n
g
u
s
er
s
to
s
elec
t
s
y
m
b
o
ls
o
r
co
m
m
a
n
d
s
,
an
d
is
p
ar
ticu
lar
l
y
u
s
e
f
u
l
f
o
r
in
d
iv
id
u
als
w
it
h
s
e
v
er
e
d
is
ab
ilit
ies
[
3
]
,
[
4
]
.
C
lass
if
icatio
n
is
ch
alle
n
g
ed
b
y
lo
w
s
i
g
n
al
-
to
-
n
o
i
s
e
r
atio
(
SNR
)
,
n
o
n
-
s
ta
tio
n
ar
it
y
,
an
d
in
ter
-
s
u
b
j
ec
t
v
ar
iab
ilit
y
.
R
ec
u
r
r
en
t
n
e
u
r
al
n
et
w
o
r
k
s
(
R
N
Ns
)
,
esp
ec
iall
y
lo
n
g
s
h
o
r
t
-
ter
m
m
e
m
o
r
y
(
L
ST
M
)
n
et
w
o
r
k
s
,
ca
p
t
u
r
e
lo
n
g
-
ter
m
te
m
p
o
r
al
d
ep
en
d
en
cie
s
i
n
E
E
G,
i
m
p
r
o
v
i
n
g
in
ter
p
r
etatio
n
o
f
u
s
er
in
ten
t
an
d
o
v
er
all
B
C
I
p
er
f
o
r
m
a
n
c
e
[
5
]
.
Sp
ik
in
g
n
e
u
r
al
n
et
w
o
r
k
s
(
S
NNs)
e
n
co
d
e
in
f
o
r
m
atio
n
v
ia
d
is
cr
ete
s
p
i
k
e
s
,
o
f
f
er
i
n
g
te
m
p
o
r
al
p
r
ec
is
io
n
,
en
er
g
y
e
f
f
icie
n
c
y
,
a
n
d
s
u
itab
i
lit
y
f
o
r
lo
w
-
p
o
w
er
,
r
ea
l
-
ti
m
e
B
C
I
s
.
T
r
a
i
n
i
n
g
S
NNs
i
s
c
o
m
p
l
i
c
at
e
d
b
y
n
o
n
-
d
if
f
e
r
en
t
ia
b
l
e
s
p
ik
es
,
b
u
t
s
u
r
r
o
g
a
te
g
r
a
d
i
e
n
t
m
et
h
o
d
s
a
n
d
a
r
t
if
i
c
ia
l
n
eu
r
a
l
n
etw
o
r
k
(
A
NN
)
-
to
-
S
N
N
c
o
n
v
e
r
s
i
o
n
te
ch
n
iq
u
e
s
c
o
m
b
in
e
A
N
N
t
r
ain
in
g
e
f
f
i
ci
en
cy
w
ith
SN
N
i
n
f
e
r
en
c
e
a
d
v
an
t
ag
e
s
[
6
]
,
[
7
]
.
Mo
tiv
ated
b
y
t
h
is
g
ap
,
th
e
o
b
j
ec
tiv
e
o
f
t
h
is
s
tu
d
y
is
to
co
n
d
u
ct
a
co
m
p
r
eh
e
n
s
i
v
e
c
o
m
p
ar
ati
v
e
ev
alu
a
tio
n
o
f
co
n
v
o
lu
tio
n
al
n
eu
r
al
n
et
w
o
r
k
(
C
N
N
)
,
L
ST
M,
an
d
SNN
ar
ch
itectu
r
es
f
o
r
P
3
0
0
s
ig
n
a
l
class
i
f
icatio
n
u
s
in
g
a
s
tan
d
ar
d
ized
P
3
0
0
s
p
eller
d
ataset.
T
h
e
m
ai
n
co
n
tr
ib
u
t
io
n
s
o
f
t
h
i
s
w
o
r
k
ar
e
th
r
ee
f
o
ld
:
(i
)
a
f
air
co
m
p
ar
is
o
n
o
f
s
p
at
ial,
te
m
p
o
r
al,
an
d
s
p
ik
in
g
n
e
u
r
al
m
o
d
els
u
s
i
n
g
t
h
e
s
a
m
e
p
r
ep
r
o
ce
s
s
in
g
an
d
ev
alu
a
tio
n
p
r
o
to
co
l;
(
ii
)
an
a
n
al
y
s
i
s
o
f
clas
s
i
f
icatio
n
ac
cu
r
ac
y
ac
r
o
s
s
s
u
b
j
ec
ts
to
as
s
es
s
m
o
d
el
r
o
b
u
s
tn
es
s
;
a
n
d
(
iii
)
an
in
v
est
ig
atio
n
o
f
t
h
e
p
o
ten
tial o
f
SNNs
a
s
an
e
n
er
g
y
-
e
f
f
icien
t a
lter
n
ati
v
e
f
o
r
P
3
0
0
-
b
ased
B
C
I
s
y
s
te
m
s
.
T
h
e
r
em
ai
n
d
er
o
f
th
is
p
ap
er
is
o
r
g
an
ized
as
f
o
llo
w
s
.
Sectio
n
2
p
r
esen
ts
th
e
th
eo
r
etica
l b
ac
k
g
r
o
u
n
d
o
f
n
eu
r
al
n
et
w
o
r
k
m
o
d
els
a
n
d
t
h
e
P
3
0
0
p
o
ten
tial.
Sectio
n
3
d
escr
ib
es
th
e
d
ataset,
p
r
ep
r
o
ce
s
s
i
n
g
p
r
o
ce
d
u
r
es,
ex
p
er
i
m
e
n
tal
s
et
u
p
,
an
d
r
es
u
l
ts
.
Fi
n
all
y
,
s
ec
t
io
n
4
co
n
c
lu
d
es
t
h
e
p
ap
er
an
d
o
u
t
li
n
e
s
d
ir
ec
tio
n
s
f
o
r
f
u
t
u
r
e
r
esear
ch
.
2.
M
E
T
H
O
D
Neu
r
al
n
et
w
o
r
k
s
ar
e
a
n
et
wo
r
k
o
f
in
ter
co
n
n
ec
ted
p
r
o
ce
s
s
in
g
u
n
it
s
d
esig
n
ed
to
m
i
m
i
c
d
esire
d
b
eh
av
io
r
.
B
ec
au
s
e
o
f
t
h
e
s
i
m
i
lar
ities
to
b
io
lo
g
ical
n
e
u
r
al
n
e
t
w
o
r
k
s
,
t
h
ese
u
n
its
ar
e
co
m
m
o
n
l
y
r
e
f
er
r
ed
to
as
n
eu
r
o
n
s
a
n
d
th
e
ir
co
n
n
ec
tio
n
s
as
s
y
n
ap
s
e
s
.
A
n
an
a
lo
g
n
et
w
o
r
k
’
s
n
e
u
r
o
n
i
s
a
t
y
p
e
o
f
c
ell
th
at
co
n
v
er
ts
a
ce
r
tain
in
p
u
t si
g
n
al
in
to
an
eq
u
iv
a
len
t o
u
tp
u
t
s
ig
n
al
[
2
]
,
[
5
]
.
T
h
e
ac
tiv
atio
n
f
u
n
ctio
n
o
f
t
h
e
n
eu
r
o
n
is
u
s
ed
to
ac
co
m
p
lis
h
t
h
ese
m
o
d
if
icat
io
n
s
a
s
s
h
o
w
in
Fig
u
r
e
2
.
T
h
e
f
u
n
ctio
n
s
u
m
s
t
h
e
s
i
g
n
a
ls
,
o
r
in
f
o
r
m
atio
n
,
it
g
ets
f
r
o
m
s
y
n
ap
s
e
s
(
u
s
u
a
ll
y
r
ea
l
n
u
m
b
er
s
)
,
an
d
th
en
o
u
tp
u
t
s
th
e
r
es
u
lt.
S
y
n
ap
s
e
s
ar
e
i
n
v
o
lv
ed
in
b
o
th
s
i
g
n
al
tr
an
s
m
is
s
io
n
a
n
d
s
tr
e
n
g
t
h
r
e
g
u
latio
n
.
T
h
e
ac
tiv
at
io
n
“
f
ir
es,
”
o
r
g
en
er
ates
a
s
tr
o
n
g
o
u
tp
u
t,
i
f
th
e
to
tal
in
p
u
t
is
s
tr
o
n
g
en
o
u
g
h
[
3
]
,
[
6
]
.
A
lth
o
u
g
h
th
er
e
ar
e
n
u
m
er
o
u
s
v
ar
ieties
o
f
n
eu
r
al
n
et
w
o
r
k
s
,
w
e
w
il
l
talk
ab
o
u
t
th
r
ee
h
er
e:
r
ec
u
r
r
en
t,
s
p
ik
in
g
,
an
d
co
n
v
o
lu
tio
n
al
C
NN
s
ar
e
ex
a
m
p
le
s
o
f
n
e
u
r
al
n
et
w
o
r
k
s
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
4
,
A
u
g
u
s
t
20
26
:
1
294
-
1
3
0
6
1296
Fig
u
r
e
2
.
Mo
d
el
o
f
a
n
eu
r
o
n
n
et
w
o
r
k
[
5
]
2
.
1
.
C
NN
I
n
co
m
p
u
ter
v
i
s
io
n
,
w
h
ich
d
e
als
w
ith
cr
ea
ti
n
g
s
y
s
te
m
s
t
h
a
t
ev
alu
ate
v
is
u
al
d
ata
an
d
r
ec
o
g
n
ize
o
r
s
eg
m
e
n
t
i
m
p
o
r
tan
t
f
ea
t
u
r
es
o
r
o
b
j
ec
ts
,
C
NNs
ar
e
s
o
p
h
i
s
ti
ca
ted
d
ee
p
lear
n
in
g
m
o
d
el
s
t
h
at
ar
e
f
r
eq
u
e
n
tl
y
u
tili
ze
d
.
Am
o
n
g
th
e
f
ir
s
t
d
e
ep
lear
n
in
g
m
o
d
els
to
ac
co
m
p
lis
h
o
b
j
ec
t
d
etec
tio
n
at
a
lev
el
co
m
p
ar
ab
le
to
h
u
m
a
n
s
w
er
e
C
NNs.
C
NNs
h
av
e
b
ee
n
u
til
ized
in
th
e
r
e
ce
n
t
p
ast
f
o
r
a
v
ar
iet
y
o
f
b
r
ain
s
i
g
n
a
l
s
t
u
d
ies,
in
cl
u
d
in
g
t
h
e
d
etec
tio
n
o
f
P
3
0
0
,
m
e
n
tal
tas
k
r
ec
o
g
n
itio
n
,
e
m
o
tio
n
r
ec
o
g
n
it
io
n
,
a
n
d
s
ei
zu
r
e
d
etec
tio
n
.
Fo
r
in
s
ta
n
ce
,
t
h
e
f
o
u
r
s
ep
ar
ate
B
C
I
p
ar
ad
ig
m
s
o
f
v
is
u
al
-
ev
o
k
ed
p
o
ten
tials
(
VE
P
)
,
er
r
o
r
-
r
elate
d
n
eg
ati
v
it
y
(
E
R
N)
re
s
p
o
n
s
es,
t
h
e
m
o
d
el
’
s
ab
ilit
y
to
class
i
f
y
d
ata
w
as e
v
al
u
ated
as sh
o
w
i
n
Fi
g
u
r
e
3
[
8
]
.
(
)
=
(
∗
[
−
,
…
.
;
+
]
+
)
(
1
)
Fig
u
r
e
3
.
C
NN
[
8
]
2
.
2
.
RNN
R
NNs
ca
n
d
escr
ib
e
s
eq
u
en
t
ial
d
ata
an
d
cr
ea
te
co
n
tex
t
-
a
w
ar
e
p
r
e
d
ictio
n
s
b
y
f
ee
d
in
g
p
r
ev
io
u
s
o
u
tp
u
t
s
in
to
cu
r
r
en
t
i
n
p
u
t
s
.
T
h
ey
p
er
f
o
r
m
w
ell
f
o
r
ti
m
e
-
s
er
ie
s
ap
p
licatio
n
s
an
d
ca
n
b
e
en
h
an
ce
d
w
it
h
co
n
v
o
lu
t
io
n
al
la
y
er
s
.
B
u
t
tr
ain
i
n
g
is
ch
a
llen
g
in
g
d
u
e
to
v
an
i
s
h
in
g
an
d
ex
p
a
n
d
in
g
g
r
ad
ien
t
s
,
w
h
ic
h
led
to
t
h
e
d
ev
elo
p
m
en
t
o
f
L
ST
M
n
et
w
o
r
k
s
f
o
r
i
m
p
r
o
v
ed
p
er
f
o
r
m
a
n
ce
o
n
lo
n
g
-
ter
m
d
e
p
en
d
en
cies.
T
h
ese
n
et
w
o
r
k
s
s
o
lv
e
is
s
u
es
l
ik
e
lo
n
g
-
ter
m
d
ep
en
d
en
c
ies
a
n
d
th
e
v
a
n
i
s
h
in
g
g
r
ad
ie
n
t
p
r
o
b
le
m
th
at
co
m
e
w
it
h
co
n
v
en
tio
n
al
r
ec
u
r
r
en
t
lo
o
p
s
b
y
u
tili
zi
n
g
s
p
ec
ialized
L
ST
M
ce
lls
.
T
h
e
im
p
le
m
en
tatio
n
s
o
f
L
ST
M
ce
lls
d
if
f
er
,
b
u
t
a
s
i
m
p
le
m
o
d
el
co
n
s
is
ts
o
f
a
r
ec
u
r
r
e
n
t
n
e
u
r
o
n
eq
u
ip
p
ed
w
it
h
a
f
o
r
g
e
t
g
at
e,
w
h
ich
i
s
a
v
ec
to
r
th
at
d
ec
id
es
w
h
a
t in
f
o
r
m
atio
n
in
th
e
ce
l
l s
h
o
u
ld
b
e
k
ep
t o
r
d
elete
d
[
9
]
,
[
1
0
]
.
T
h
e
L
ST
M
ar
ch
itectu
r
e
i
n
tr
o
d
u
ce
s
g
a
te
s
to
co
n
tr
o
l
d
ata
f
l
o
w
,
allo
w
i
n
g
i
t
to
o
v
er
co
m
e
ch
alle
n
g
e
s
ass
o
ciate
d
w
it
h
v
a
n
i
s
h
i
n
g
an
d
ex
p
lo
d
in
g
g
r
ad
ien
ts
i
n
R
N
Ns.
E
ac
h
L
ST
M
u
n
it
h
as
t
wo
o
u
tp
u
ts
:
th
e
m
a
in
o
u
tp
u
t
a
n
d
th
e
u
n
it
’
s
m
e
m
o
r
y
.
T
h
e
ar
ch
itectu
r
e
u
s
e
s
g
ate
s
t
o
r
eg
u
late
d
ata
f
lo
w
,
f
ac
ilit
at
i
n
g
t
h
e
lear
n
i
n
g
o
f
lo
n
g
-
ter
m
d
ep
en
d
en
cies.
A
co
m
p
a
r
is
o
n
b
etw
e
en
a
c
l
as
s
i
c
al
R
NN
an
d
an
L
S
T
M
is
d
e
p
i
c
t
e
d
in
F
ig
u
r
e
4
,
em
p
h
as
i
z
in
g
th
e
c
o
m
p
le
x
i
ty
an
d
ef
f
e
c
ti
v
e
n
e
s
s
o
f
L
S
T
M
w
i
th
i
ts
f
o
u
r
l
ay
e
r
s
in
th
e
r
e
p
e
a
te
d
m
o
d
u
l
e
[
1
1
]
.
I
n
s
i
d
e
t
h
e
L
S
T
M
,
t
h
e
c
o
m
p
u
t
a
t
i
o
n
s
a
r
e
d
e
f
in
e
d
b
y
d
o
o
r
s
th
at
a
l
l
o
w
o
r
n
o
t
th
e
t
r
a
n
s
m
is
s
i
o
n
o
f
i
n
f
o
r
m
a
ti
o
n
[
4
]
,
[
1
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
la
s
s
if
ica
tio
n
o
f P
3
0
0
ev
en
t
-
r
ela
ted
p
o
ten
tia
ls
u
s
in
g
S
N
N
,
C
N
N
a
n
d
LS
TM
…
(
A
h
la
a
m
M.
S
a
ed
)
1297
Fig
u
r
e
4
.
Diag
r
a
m
o
f
an
L
ST
M
[
1
2
]
(
)
=
(
ℎ
(
−
1
)
+
(
)
+
)
(
2
)
(
)
=
(
ℎ
(
−
1
)
+
(
)
+
)
(
3
)
̂
(
)
=
ℎ
(
ℎ
(
−
1
)
+
(
)
+
)
(
4
)
(
)
=
(
)
⊙
(
−
1
)
+
(
)
⊙
̂
(
)
(
5
)
=
(
ℎ
(
−
1
)
+
(
)
+
)
(
6
)
ℎ
(
)
=
(
)
⊙
ℎ
(
)
(
7
)
(
)
=
(
ℎ
(
)
+
)
(
8
)
T
h
e
p
r
e
d
icted
ch
ar
ac
ter
̂
in
a
s
in
g
le
tr
ial
is
co
m
p
u
ted
b
y
f
i
n
d
in
g
t
h
e
ch
ar
ac
ter
w
it
h
m
a
x
i
m
al
v
a
lu
e
o
f
(
)
a
m
o
n
g
t
h
e
g
r
o
u
p
o
f
all
t
h
e
letter
s
:
̂
=
{
(
(
)
)
}
∈
(
9
)
T
h
e
f
o
r
m
u
la
f
o
r
p
r
ed
ictin
g
t
h
e
tar
g
et
s
ti
m
u
li
is
:
̂
=
{
1
∑
(
(
,
)
)
∈
}
∈
(
1
0
)
I
n
n
eu
r
a
l
n
et
w
o
r
k
,
th
e
w
ei
g
h
ts
ar
e
u
p
d
ated
b
y
d
er
iv
i
n
g
th
e
lo
s
s
f
u
n
ctio
n
w
i
th
r
esp
ec
ted
to
th
e
n
et
w
o
r
k
w
ei
g
h
ts
.
=
(
(
(
,
)
)
,
,
)
(
1
1
)
Her
e
,
r
ef
er
to
th
e
tr
u
e
lab
el
o
f
s
a
m
p
le
(
,
)
,
th
e
b
in
ar
y
lo
g
lo
s
s
f
u
n
ctio
n
:
(
(
(
,
)
)
,
,
)
=
−
(
,
(
(
,
)
)
−
(
1
−
,
)
(
1
−
(
(
,
)
)
)
)
(
1
2
)
T
h
e
er
r
o
r
in
a
n
eu
r
al
n
e
t
w
o
r
k
i
s
t
y
p
icall
y
ca
lc
u
lated
o
n
m
u
lti
p
le
s
a
m
p
les,
as
f
o
llo
w
s
:
=
1
∑
(
(
,
)
)
(
1
3
)
Her
e
in
d
icate
s
th
e
s
ize
o
f
t
h
e
b
atch
o
f
s
a
m
p
les
[
1
0
]
,
[
1
1
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
4
,
A
u
g
u
s
t
20
26
:
1
294
-
1
3
0
6
1298
2
.
3
.
SNN
A
n
e
w
b
r
ee
d
o
f
co
m
p
u
ter
-
s
i
m
u
lated
n
e
u
r
al
n
e
t
w
o
r
k
s
i
s
r
ep
r
esen
ted
b
y
S
NNs.
I
n
ad
d
itio
n
to
tr
y
i
n
g
to
ad
d
r
ess
s
o
m
e
o
f
th
e
d
r
a
w
b
ac
k
s
o
f
an
a
lo
g
n
et
s
,
s
u
c
h
as
th
eir
s
l
u
g
g
i
s
h
r
esp
o
n
s
e
ti
m
es,
h
ig
h
p
o
w
er
co
n
s
u
m
p
tio
n
,
o
r
lack
o
f
a
s
y
n
c
h
r
o
n
o
u
s
p
r
o
ce
s
s
i
n
g
,
SNNs
ar
e
in
te
n
d
ed
to
m
o
r
e
ac
c
u
r
atel
y
s
i
m
u
late
b
io
lo
g
ica
l
b
r
ain
n
et
w
o
r
k
s
.
Sp
ik
in
g
n
et
wo
r
k
s
r
eq
u
ir
ed
a
m
o
d
if
icat
io
n
t
o
th
eir
s
tr
u
ct
u
r
e
d
u
e
to
t
h
e
ex
p
ec
tatio
n
s
t
h
at
w
er
e
cr
ea
ted
.
A
s
s
ee
n
i
n
Fi
g
u
r
e
5
,
s
p
ik
in
g
n
e
u
r
o
n
s
,
w
h
ic
h
ar
e
e
s
s
e
n
tiall
y
d
is
tin
c
t
f
r
o
m
t
h
e
n
e
u
r
o
n
s
f
o
u
n
d
i
n
A
NN
s
,
m
ak
e
u
p
a
s
p
i
k
in
g
n
et
w
o
r
k
.
S
p
ik
in
g
n
e
u
r
o
n
s
lack
a
n
ac
tiv
at
io
n
f
u
n
ctio
n
,
ev
e
n
i
f
th
e
y
co
m
m
u
n
icate
th
r
o
u
g
h
s
y
n
ap
tic
co
n
n
ec
tio
n
s
as
w
ell.
R
ath
er
,
ea
c
h
n
e
u
r
o
n
t
h
at
f
ir
es
h
as
a
m
e
m
b
r
an
e
p
o
ten
tia
l.
Sp
i
k
e
tr
ain
s
,
o
r
s
er
ies
o
f
ac
tio
n
p
o
te
n
tial
s
t
h
at
c
h
an
g
e
th
e
m
e
m
b
r
an
e
p
o
ten
t
ial,
ar
e
h
o
w
n
eu
r
o
n
s
e
x
ch
a
n
g
e
a
n
d
r
ec
eiv
e
i
n
f
o
r
m
atio
n
.
T
h
e
n
eu
r
o
n
g
en
er
ate
s
a
s
p
ik
e,
o
r
s
ti
m
u
latio
n
,
w
h
en
e
v
er
a
s
p
ec
if
ic
th
r
es
h
o
ld
v
o
ltag
e
is
cr
o
s
s
ed
,
an
d
th
e
m
e
m
b
r
a
n
e
p
o
ten
tial
i
s
th
e
n
r
eset
to
a
p
r
ed
eter
m
in
ed
b
aseli
n
e.
An
o
th
er
p
o
p
u
lar
ter
m
f
o
r
t
h
is
o
cc
u
r
r
e
n
ce
is
a
n
eu
r
o
n
f
ir
i
n
g
[
1
3
]
–
[
1
9
]
.
Fig
u
r
e
5
.
Vis
u
al
izatio
n
o
f
a
s
p
ik
i
n
g
n
eu
r
o
n
[
1
4
]
A
leak
y
i
n
te
g
r
ate
-
a
n
d
-
f
ir
e
(
L
I
F
)
n
e
u
r
o
n
es
s
en
tiall
y
ac
c
u
m
u
l
ates
(
in
teg
r
ates)
a
ll
o
f
th
e
in
p
u
ts
th
at
it
r
ec
eiv
es
o
v
er
ti
m
e,
d
is
ch
ar
g
e
s
(
leak
s
)
ex
p
o
n
e
n
tiall
y
co
n
s
ta
n
tl
y
,
a
n
d
f
ir
es
to
r
elea
s
e
an
a
ctio
n
p
o
ten
tial.
T
h
e
d
ev
elo
p
m
en
t
o
f
th
e
s
p
i
k
i
n
g
n
eu
r
o
n
m
o
d
el
is
ess
e
n
tia
l
to
th
e
r
ea
liza
tio
n
o
f
S
NN.
Neu
r
o
n
s
d
i
f
f
er
i
n
th
ei
r
elec
tr
ical
an
d
g
eo
m
etr
ic
p
r
o
p
e
r
ties
.
T
h
e
s
p
ik
in
g
n
eu
r
o
n
m
o
d
el,
as
s
ee
n
in
Fi
g
u
r
e
5
,
is
in
teg
r
ate
-
an
d
-
f
ir
e
(
I
A
F)
m
o
d
el
as
d
ep
icted
in
Fi
g
u
r
e
6
[
1
5
]
w
ill
b
e
co
m
p
u
ted
u
s
i
n
g
t
h
e
SNN
a
n
d
L
I
F
m
eth
o
d
.
T
h
e
te
m
p
o
r
al
d
y
n
a
m
ics
ar
e
f
o
r
m
u
la
ted
:
=
−
+
∗
(
−
)
(
1
4
)
w
h
er
e:
=
m
e
m
b
r
an
e
p
o
ten
tia
l
an
d
=
ti
m
e
co
n
s
ta
n
t
.
=
g
e
ts
m
o
d
u
l
a
te
d
b
y
th
e
s
y
n
a
p
t
i
c
w
ei
g
h
t
an
d
(
−
)
=
in
c
o
m
in
g
s
p
ik
e
,
a
s
p
ik
e
t
r
ai
n
c
an
b
e
d
e
s
c
r
i
b
e
d
as
:
(
)
=
∑
(
−
)
(
1
5
)
w
h
er
e
=
1
,
2
,
…
T
h
e
s
o
f
t
m
a
x
f
o
r
m
u
la,
s
o
m
eti
m
es c
alled
t
h
e
n
o
r
m
a
lized
ex
p
o
n
en
tial
[
9
]
is
g
iv
e
n
:
=
(
)
∑
(
)
(
1
6)
L
I
F
m
o
d
el
ar
e
d
escr
ib
ed
b
y
a
s
in
g
le
f
ir
s
t
-
o
r
d
er
lin
ea
r
d
if
f
er
e
n
tial e
q
u
at
io
n
:
=
−
(
−
)
+
(
1
7
)
=
−
1
(
−
)
+
+
(
)
(
−
)
(
1
8
)
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
la
s
s
if
ica
tio
n
o
f P
3
0
0
ev
en
t
-
r
ela
ted
p
o
ten
tia
ls
u
s
in
g
S
N
N
,
C
N
N
a
n
d
LS
TM
…
(
A
h
la
a
m
M.
S
a
ed
)
1299
T
h
e
m
e
m
b
r
an
e
ti
m
e
co
n
s
tan
t
s
atis
f
ies
[
1
4
]
.
=
(
1
9
)
T
h
e
lo
s
s
f
u
n
ctio
n
d
u
r
in
g
tr
ai
n
i
n
g
,
w
h
ic
h
ca
n
b
e
ca
lcu
la
te
d
as
:
(
(
(
)
)
,
(
)
)
=
−
1
(
(
(
)
)
−
(
1
−
)
l
og
(
1
−
(
(
)
)
)
)
(
2
0
)
T
h
e
E
E
G
c
an
o
b
t
a
in
th
e
b
es
t
r
e
c
o
v
e
r
y
f
r
o
m
th
e
en
c
o
d
e
d
s
p
ik
e
t
r
a
in
s
b
y
m
in
im
iz
in
g
th
e
e
r
r
o
r
w
h
i
ch
i
s
d
ef
in
ed
as
(
2
1
)
.
=
∑
|
−
|
∑
(
2
1
)
Sp
ik
e
tr
ai
n
s
o
f
t
h
e
o
u
tp
u
t
la
y
er
ar
e
f
ilter
ed
b
y
a
d
o
u
b
le
e
x
p
o
n
en
tial
s
y
n
ap
tic
f
ilter
f
o
r
ea
ch
n
e
u
r
o
n
j
as
s
h
o
w
n
in
(
2
2
)
,
(
2
3
)
.
̇
=
−
+
ℎ
(
2
2
)
ℎ
̇
=
−
ℎ
+
1
∑
(
−
)
(
2
3
)
As s
h
o
w
n
i
n
(
2
4
)
,
f
o
r
a
s
in
g
le
ch
an
n
el
c
,
w
h
er
e
*
in
d
icate
s
d
o
t p
r
o
d
u
ct
m
u
ltip
licatio
n
.
=
∑
∗
(
2
4
)
Fig
u
r
e
6
.
E
q
u
iv
ale
n
t c
ir
cu
it o
f
co
n
s
tan
t
-
lea
k
I
A
F
m
o
d
el
[
1
9
]
2
.
4
.
P
3
0
0
po
t
ent
ia
l in B
CI
A
llo
w
s
u
s
er
s
to
u
tili
ze
t
h
eir
b
r
ain
s
ig
n
als
to
ch
o
o
s
e
o
b
j
ec
ts
o
n
a
s
cr
ee
n
o
r
s
p
ell
w
o
r
d
s
.
T
h
e
ter
m
“
P
3
0
0
”
d
esig
n
a
tes
a
p
ar
ticu
lar
k
in
d
o
f
E
R
P
th
at
h
ap
p
e
n
s
ab
o
u
t
3
0
0
m
s
af
ter
an
i
n
d
iv
id
u
al
n
o
tice
a
n
o
te
w
o
r
t
h
y
s
ti
m
u
l
u
s
,
l
ik
e
a
t
ar
g
et
letter
o
r
s
y
m
b
o
l
[
1
5
]
,
[
1
6
]
.
T
h
e
g
en
er
al
o
p
er
atio
n
o
f
a
P
3
0
0
-
b
ased
B
C
I
s
p
eller
task
i
s
as
f
o
llo
w
s
:
t
h
e
P
3
0
0
s
p
eller
s
y
s
te
m
b
eg
in
s
w
ith
E
E
G
s
e
tu
p
an
d
ca
lib
r
atio
n
,
w
h
er
e
s
ca
l
p
elec
tr
o
d
es
r
ec
o
r
d
b
r
ain
ac
tiv
i
t
y
.
A
6
×6
m
atr
i
x
(
3
6
ch
ar
ac
ter
s
)
is
d
is
p
la
y
ed
,
a
n
d
th
e
u
s
er
f
o
cu
s
es
o
n
th
e
d
esire
d
ch
ar
ac
ter
[
1
9
]
.
W
h
en
th
e
tar
g
et
f
las
h
es,
a
P
3
0
0
r
esp
o
n
s
e
i
s
g
e
n
er
ated
an
d
d
etec
ted
.
A
cla
s
s
i
f
icatio
n
alg
o
r
it
h
m
id
en
ti
f
ies
t
h
e
s
elec
ted
ch
ar
ac
ter
b
ased
o
n
P
3
0
0
r
esp
o
n
s
es,
p
r
o
v
id
in
g
r
ea
l
-
ti
m
e
f
ee
d
b
ac
k
a
n
d
er
r
o
r
co
r
r
e
ctio
n
i
f
n
ee
d
ed
[
4
]
,
[
1
3
]
,
[
1
5
]
,
[
1
6
]
.
E
lectr
o
d
es
ar
e
ty
p
icall
y
p
lace
d
ac
co
r
d
in
g
to
th
e
I
n
ter
n
atio
n
al
1
0
–
2
0
s
y
s
te
m
,
a
n
d
s
p
ec
if
ic
ch
a
n
n
els
(
F3
,
F4
,
C
3
,
C
4
,
P
3
,
P
4
,
O1
,
O2
)
h
av
e
b
ee
n
s
h
o
w
n
to
i
m
p
r
o
v
e
ef
f
icie
n
c
y
an
d
r
ed
u
ce
co
s
t
[
1
0
]
,
[
1
1
]
,
[
1
3
]
–
[
1
9
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
4
,
A
u
g
u
s
t
20
26
:
1
294
-
1
3
0
6
1300
3.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
N
3
.
1
.
Cla
s
s
if
ica
t
io
n
m
o
de
ls
Ou
r
s
u
g
g
e
s
ted
ar
ch
itect
u
r
e
c
o
m
b
i
n
es
v
ar
io
u
s
A
N
N
la
y
er
t
y
p
es,
as
f
o
llo
w
s
:
f
u
ll
y
co
n
n
ec
ted
la
y
er
(
FC
)
,
a
la
y
er
is
co
m
m
o
n
l
y
r
e
f
er
r
ed
to
as
s
u
ch
w
h
e
n
e
v
er
y
n
eu
r
o
n
in
t
h
e
i
n
p
u
t
i
s
co
u
p
led
to
ev
er
y
n
e
u
r
o
n
i
n
th
e
o
u
tp
u
t
[
1
6
]
.
T
h
e
f
o
llo
w
i
n
g
(
2
5
)
y
ield
s
th
e
o
u
tp
u
t i
n
an
F
C
la
y
er
:
=
(
+
)
(2
5
)
(
·
)
d
en
o
tes
a
n
ele
m
e
n
t
-
w
i
s
e
n
o
n
-
li
n
ea
r
it
y
,
s
u
ch
as
h
y
p
er
b
o
lic
tan
g
e
n
t
(
T
an
H)
,
s
i
g
o
m
id
,
o
r
r
ec
tif
ied
li
n
ea
r
u
n
i
t (
R
e
L
U)
as s
h
o
w
in
Fig
u
r
e
s
7
(
a
)
an
d
(
b
)
.
(
a)
(
b
)
Fig
u
r
e
7
.
A
cti
v
atio
n
f
u
n
c
tio
n
:
(
a
)
T
an
H
-
s
i
g
m
o
id
ac
tiv
at
io
n
f
u
n
ct
io
n
an
d
(
b
)
R
eL
U
f
u
n
ctio
n
3
.
2
.
Reso
urce
s
f
o
r
s
i
m
ula
t
in
g
T
h
e
s
ec
o
n
d
s
ec
tio
n
r
ev
ie
w
s
t
o
o
ls
f
o
r
m
o
d
elin
g
an
d
s
i
m
u
l
atin
g
v
ar
io
u
s
n
e
u
r
al
n
et
w
o
r
k
s
.
Sev
er
al
p
o
w
er
f
u
l
f
r
a
m
e
w
o
r
k
s
ar
e
av
ai
lab
le,
m
o
s
tl
y
b
ased
o
n
P
y
t
h
o
n
d
u
e
to
its
s
i
m
p
le
s
y
n
ta
x
an
d
co
m
p
atib
il
it
y
w
i
th
C
/C
++
.
E
x
a
m
p
les
i
n
cl
u
d
e
T
e
n
s
o
r
Flo
w
,
Ker
as,
NE
UR
ON,
Nen
g
o
,
P
y
t
h
o
n
n
e
u
r
al
n
et
w
o
r
k
s
(
P
y
NN
)
,
B
r
ian
,
B
in
d
in
g
s
f
o
r
n
eu
r
al
n
et
w
o
r
k
s
(
B
in
d
s
NE
T
)
,
an
d
n
eu
r
al
s
i
m
u
latio
n
to
o
l
(
NE
ST
)
.
T
h
ese
lib
r
ar
ies
ar
e
also
u
s
ed
f
o
r
er
r
o
r
ev
alu
atio
n
,
g
r
ap
h
e
x
t
r
ac
tio
n
,
an
d
n
e
u
r
al
n
et
w
o
r
k
s
i
m
u
latio
n
[
2
0
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
la
s
s
if
ica
tio
n
o
f P
3
0
0
ev
en
t
-
r
ela
ted
p
o
ten
tia
ls
u
s
in
g
S
N
N
,
C
N
N
a
n
d
LS
TM
…
(
A
h
la
a
m
M.
S
a
ed
)
1301
3
.
3
.
E
v
a
lua
t
ed
m
o
del
s
3
.
3
.
1
.
CNN
m
o
del
s
A
C
NN
la
y
er
ca
p
t
u
r
es
lo
ca
l
c
o
r
r
elatio
n
s
b
et
w
ee
n
ad
j
ac
en
t
i
n
p
u
t
ce
l
ls
,
p
r
o
d
u
ci
n
g
m
u
lti
-
d
i
m
en
s
io
n
a
l
f
ea
t
u
r
e
m
ap
s
u
s
i
n
g
tr
ain
ab
le
k
er
n
el
s
.
A
o
n
e
-
d
i
m
e
n
s
io
n
al
(
1
D)
C
NN
ex
a
m
p
le
is
s
h
o
w
n
i
n
Fi
g
u
r
e
8
[
9
]
,
[
1
1
]
,
[
1
4
]
.
T
h
e
f
ir
s
t
la
y
er
h
as
1
0
s
p
atial
f
ilter
s
(
5
5
×1
)
,
th
e
s
ec
o
n
d
la
y
er
h
a
s
1
3
te
m
p
o
r
al
f
ilte
r
s
(
1
×5
)
p
r
o
ce
s
s
in
g
f
i
v
e
co
n
s
ec
u
tiv
e
ti
m
e
p
o
in
ts
w
it
h
o
u
t
o
v
er
lap
.
T
h
e
th
ir
d
an
d
f
o
u
r
th
la
y
er
s
ar
e
f
u
ll
y
co
n
n
ec
ted
,
en
d
in
g
w
ith
a
s
in
g
le
s
i
g
m
o
id
-
ac
t
iv
ated
ce
ll t
h
at
o
u
tp
u
ts
a
s
ca
lar
.
3
.
3
.
2
.
L
ST
M
mo
del
L
S
T
M
m
a
d
e
u
p
o
f
1
0
0
an
d
3
0
h
i
d
d
en
ce
l
ls
,
r
es
p
e
c
ti
v
e
ly
,
in
e
ac
h
o
f
i
ts
s
in
g
l
e
L
S
T
M
l
ay
e
r
s
.
B
o
th
m
o
d
e
ls
cu
lm
in
a
t
e
in
a
s
in
g
l
e
ce
l
l
t
h
a
t
p
r
o
d
u
c
e
s
a
s
c
al
a
r
an
d
h
a
s
a
s
ig
m
o
i
d
a
ct
iv
a
ti
o
n
l
ay
e
r
,
as
s
h
o
w
in
Fi
g
u
r
e
9
[
9
]
–
[
1
1
]
:
(
i
)
e
ac
h
s
u
b
j
ec
t
’
s
d
ata
is
u
s
ed
s
ep
ar
atel
y
f
o
r
tr
ain
in
g
an
d
te
s
ti
n
g
i
n
o
r
d
er
to
lo
o
k
at
i
n
tr
a
-
s
u
b
j
ec
t
d
if
f
er
e
n
ce
s
;
(
i
i
)
e
x
a
m
i
n
i
n
g
th
e
ef
f
ec
t
s
o
f
ad
d
itio
n
al
d
ata
b
y
co
m
b
i
n
i
n
g
tr
ai
n
i
n
g
a
n
d
test
i
n
g
d
ata
f
r
o
m
d
if
f
er
en
t
s
u
b
j
ec
ts
;
an
d
(
i
i
i
)
t
r
an
s
f
er
le
ar
n
in
g
:
teac
h
in
g
all
b
u
t
o
n
e
s
u
b
j
ec
t.
T
h
is
ex
p
er
i
m
en
t
aim
s
to
ex
p
lo
r
e
th
e
ad
v
an
ta
g
es
o
f
ad
ap
tin
g
a
m
o
d
el
th
at
w
as
tr
ai
n
ed
o
f
f
li
n
e
o
n
s
ev
er
al
s
u
b
j
ec
ts
to
a
n
e
w
s
u
b
j
ec
t,
eith
er
w
i
th
o
r
w
it
h
o
u
t f
u
r
th
er
ca
lib
r
atio
n
[
1
8
]
,
[
2
1
]
.
3
.
3
.
3
.
SNN
m
o
del
A
t
t
h
e
n
e
u
r
o
n
al
lev
el,
b
r
ain
d
y
n
a
m
ics
ar
e
m
o
d
eled
m
ath
e
m
atica
ll
y
,
b
u
t
s
i
m
u
lati
n
g
a
n
d
tr
ain
in
g
SNNs
w
it
h
m
an
y
s
y
n
ap
s
e
s
is
ch
alle
n
g
i
n
g
d
u
e
to
h
ig
h
e
v
e
n
t
r
ates
[
1
9
]
,
[
2
0
]
.
A
f
ee
d
f
o
r
w
ar
d
SNN
w
it
h
L
I
F
n
eu
r
o
n
s
w
as
d
es
ig
n
ed
to
g
en
e
r
ate
E
E
G
s
ig
n
als.
Si
n
ce
t
h
e
tr
u
e
s
p
ik
e
tr
ain
f
o
r
P
3
0
0
is
u
n
a
v
ailab
le,
a
P
o
is
s
o
n
s
p
ik
e
tr
ain
at
1
0
Hz
s
er
v
es
as
in
p
u
t,
f
o
llo
w
ed
b
y
o
n
e
h
id
d
en
an
d
o
n
e
o
u
tp
u
t
la
y
er
.
L
a
y
er
s
izes
v
ar
y
p
er
P
3
0
0
.
Ou
tp
u
t
s
p
i
k
es
ar
e
co
n
v
er
ted
t
o
r
ate
s
ig
n
als
v
ia
a
d
o
u
b
le
e
x
p
o
n
en
tial
s
y
n
ap
tic
f
ilter
a
n
d
w
ei
g
h
ted
to
p
r
o
d
u
ce
E
E
G
[
2
0
]
.
A
ll
w
ei
g
h
t
m
atr
ic
es,
ex
ce
p
t
p
er
tu
r
b
atio
n
la
y
er
s
,
w
er
e
tr
ain
ed
,
w
h
ile
th
e
p
e
r
tu
r
b
atio
n
la
y
er
o
f
P
o
is
s
o
n
n
e
u
r
o
n
s
m
o
d
el
s
b
ac
k
g
r
o
u
n
d
n
o
is
e,
as s
h
o
w
in
Fi
g
u
r
e
10
[
1
9
]
–
[
2
3
]
.
Fig
u
r
e
8
.
C
NN
m
ode
l
Fig
u
r
e
9
.
L
ST
M
m
o
d
el
Fig
u
r
e
1
0
.
SNN
m
o
d
el
3
.
4
.
De
s
cr
iptio
n o
f
ex
peri
m
e
nta
l da
t
a
D
e
e
p
l
e
a
r
n
in
g
m
o
d
e
ls
,
in
cl
u
d
in
g
SN
N
an
d
L
S
T
M
,
w
e
r
e
u
s
e
d
t
o
c
l
as
s
i
f
y
P3
0
0
d
a
t
a
o
b
t
ai
n
e
d
f
r
o
m
K
a
g
g
le
.
T
h
e
d
a
t
as
e
t
w
as
g
en
e
r
a
t
e
d
u
s
in
g
th
e
c
o
n
v
en
t
i
o
n
a
l 6
×6
D
o
n
ch
i
n
an
d
F
a
r
ew
el
l P3
0
0
s
p
e
l
l
e
r
m
a
t
r
ix
.
E
a
ch
o
f
th
e
s
ev
en
w
o
r
d
s
c
o
n
t
a
in
e
d
f
i
v
e
l
et
t
e
r
s
,
a
n
d
th
e
m
a
t
r
i
x
w
as
s
t
im
u
la
t
e
d
1
2
0
t
im
es
p
e
r
l
e
t
te
r
(
1
0
t
im
e
s
p
e
r
r
o
w
a
n
d
c
o
lu
m
n
)
.
T
h
e
g
o
a
l
w
as
t
o
d
e
c
o
d
e
f
o
u
r
w
o
r
d
s
f
r
o
m
th
e
r
em
ai
n
in
g
tw
en
ty
l
e
tt
e
r
s
.
A
0
.
1
–
3
0
H
z
b
a
n
d
-
p
a
s
s
f
il
t
e
r
an
d
a
5
0
H
z
n
o
t
ch
f
il
t
e
r
w
e
r
e
a
p
p
l
i
e
d
,
an
d
d
a
t
a
w
e
r
e
c
o
l
l
e
c
t
e
d
f
r
o
m
e
ig
h
t
s
u
b
je
c
t
(
T
a
b
l
e
1
)
[
1
9
]
,
[
2
3
]
.
Evaluation Warning : The document was created with Spire.PDF for Python.
I
SS
N
:
1
6
9
3
-
6930
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
,
Vo
l.
24
,
No
.
4
,
A
u
g
u
s
t
20
26
:
1
294
-
1
3
0
6
1302
T
ab
le
1
.
B
asic in
f
o
r
m
atio
n
o
f
s
u
b
j
ec
ts
S
u
b
j
e
c
t
A
g
e
G
e
n
d
e
r
1
56
M
a
l
e
2
21
M
a
l
e
3
23
F
e
mal
e
4
29
M
a
l
e
5
20
F
e
mal
e
6
24
M
a
l
e
7
29
M
a
l
e
8
30
M
a
l
e
3
.
5
.
Da
t
a
p
re
pro
ce
s
s
ing
T
h
e
co
d
e
w
a
s
i
m
p
le
m
en
ted
u
s
in
g
Ker
as.
T
h
e
d
ataset
w
a
s
d
iv
id
ed
in
to
t
w
o
clas
s
es:
tar
g
et
P
3
0
0
an
d
n
o
n
-
tar
g
et.
E
ac
h
s
a
m
p
le
co
n
ta
in
s
5
5
ch
an
n
el
s
w
ith
1
7
5
s
am
p
lin
g
p
o
in
ts
p
er
ch
an
n
el.
Af
te
r
p
r
e
p
r
o
ce
s
s
in
g
,
th
e
tr
ain
i
n
g
s
et
i
n
cl
u
d
ed
2
8
1
0
n
o
n
-
tar
g
et
s
a
m
p
les
a
n
d
5
4
8
ta
r
g
et
P
3
0
0
s
am
p
les.
T
o
ev
alu
ate
th
e
m
o
d
el,
f
i
v
e
f
o
ld
cr
o
s
s
-
v
alid
atio
n
w
a
s
ap
p
lied
.
Fiv
e
s
u
b
s
et
s
o
f
th
e
tr
ai
n
in
g
d
ata
w
er
e
cr
ea
ted
.
Fo
u
r
s
u
b
s
et
s
w
er
e
u
tili
ze
d
f
o
r
tr
ain
i
n
g
an
d
o
n
e
w
as
u
s
ed
f
o
r
v
alid
atio
n
in
ea
c
h
iter
atio
n
.
T
h
e
f
in
al
ev
al
u
atio
n
m
etr
ic
was
ca
lcu
lated
as
th
e
av
er
ag
e
ac
c
u
r
ac
y
ac
r
o
s
s
t
h
e
f
iv
e
f
o
ld
s
[
1
9
]
,
[
2
2
]
,
[
2
4
]
.
T
h
e
P
3
0
0
p
o
ten
tial
is
id
e
n
ti
f
ied
b
y
a
p
o
s
iti
v
e
p
ea
k
o
cc
u
r
r
in
g
ar
o
u
n
d
3
0
0
m
s
a
f
t
er
s
ti
m
u
l
u
s
o
n
s
et,
as
s
h
o
w
n
in
Fi
g
u
r
es
11
(
a
)
an
d
(
b
)
.
I
t
i
s
elicited
w
h
en
a
n
in
d
iv
id
u
al
d
etec
ts
a
n
in
f
r
eq
u
e
n
t
b
u
t
e
x
p
ec
ted
s
ti
m
u
l
u
s
a
m
o
n
g
f
r
eq
u
e
n
t
o
n
es.
I
n
t
w
o
-
cla
s
s
class
i
f
icatio
n
tas
k
s
,
r
ar
e
s
ti
m
u
li
g
e
n
er
ate
a
P
3
0
0
r
esp
o
n
s
e
in
t
h
e
E
E
G
s
i
g
n
a
l.
T
h
is
f
r
a
m
e
w
o
r
k
is
k
n
o
w
n
a
s
th
e
o
d
d
b
all
p
ar
ad
ig
m
an
d
h
as b
ee
n
e
x
te
n
s
i
v
el
y
s
t
u
d
i
ed
[
1
5
]
,
[
1
9
]
,
[
2
5
]
.
(
a)
(
b
)
Fig
u
r
e
1
1
.
W
av
e
ch
ar
ac
ter
izatio
n
:
(
a
)
P
3
0
0
p
o
ten
tial d
iag
r
a
m
an
d
(
b
)
p
o
s
tiv
e
p
ea
k
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KOM
NI
K
A
T
elec
o
m
m
u
n
C
o
m
p
u
t E
l
C
o
n
tr
o
l
C
la
s
s
if
ica
tio
n
o
f P
3
0
0
ev
en
t
-
r
ela
ted
p
o
ten
tia
ls
u
s
in
g
S
N
N
,
C
N
N
a
n
d
LS
TM
…
(
A
h
la
a
m
M.
S
a
ed
)
1303
3
.
6
.
Resul
t
s
T
h
e
m
o
d
els
’
p
r
ed
icted
ac
tio
n
s
ar
e
f
ir
s
t
d
is
p
la
y
ed
in
t
h
ese
Fi
g
u
r
es
12
an
d
1
3
.
T
h
e
m
o
d
els
’
ac
cu
r
ac
y
in
id
en
ti
f
y
i
n
g
t
h
e
n
ee
d
ed
lette
r
is
th
e
n
d
is
p
la
y
ed
i
n
t
h
e
T
ab
l
es
2
to
4
th
at
f
o
llo
w
,
alo
n
g
w
i
th
a
lo
s
s
p
er
ce
n
ta
g
e
ca
lcu
latio
n
f
o
r
ea
ch
m
o
d
el.
T
h
e
m
e
m
b
r
an
e
d
y
n
a
m
ics
an
d
d
ef
o
r
m
atio
n
o
f
t
h
e
co
n
s
ta
n
t
-
l
ea
k
I
AF
m
o
d
el
ar
e
s
h
o
w
n
i
n
Fi
g
u
r
e
1
4
.
T
ab
les
2
to
4
p
r
esen
t
th
e
s
u
b
j
ec
t
-
w
is
e
class
if
icatio
n
p
er
f
o
r
m
a
n
ce
o
f
th
e
C
NN,
L
ST
M,
an
d
SNN
m
o
d
els
in
ter
m
s
o
f
l
o
s
s
an
d
ac
cu
r
ac
y
.
T
h
e
r
esu
lts
d
em
o
n
s
tr
ate
v
ar
iab
ilit
y
ac
r
o
s
s
s
u
b
j
ec
ts
,
w
h
ich
i
s
ex
p
ec
ted
d
u
e
to
in
ter
-
s
u
b
j
ec
t d
if
f
er
en
ce
s
i
n
E
E
G
p
atter
n
s
an
d
th
e
n
o
n
-
s
tatio
n
ar
y
n
a
tu
r
e
o
f
b
r
ain
s
i
g
n
al
s
.
T
h
e
C
NN
m
o
d
el
ac
h
iev
ed
c
o
n
s
is
ten
tl
y
h
i
g
h
cla
s
s
i
f
ica
tio
n
ac
cu
r
ac
y
ac
r
o
s
s
m
o
s
t
s
u
b
j
ec
ts
,
w
it
h
s
u
b
j
ec
ts
1
an
d
8
o
b
tain
in
g
t
h
e
h
i
g
h
est
p
er
f
o
r
m
a
n
ce
an
d
lo
w
lo
s
s
v
al
u
es,
in
d
icati
n
g
s
tab
le
tr
ain
in
g
an
d
ef
f
ec
tiv
e
s
p
atial
f
ea
t
u
r
e
ex
tr
ac
tio
n
.
T
ab
le
3
s
h
o
w
s
th
at
th
e
L
ST
M
m
o
d
el
also
ac
h
ie
v
ed
co
m
p
e
titi
v
e
r
es
u
lts
b
y
ca
p
tu
r
in
g
te
m
p
o
r
al
d
ep
en
d
en
c
ies in
E
E
G
s
i
g
n
als.
T
h
e
SNN
m
o
d
el
ac
h
ie
v
ed
s
tr
o
n
g
s
u
b
j
ec
t
-
w
i
s
e
p
er
f
o
r
m
a
n
ce
,
w
it
h
ac
c
u
r
ac
y
co
m
p
ar
ab
le
t
o
C
NN
f
o
r
s
ev
er
al
s
u
b
j
ec
ts
a
n
d
r
elati
v
e
l
y
lo
w
lo
s
s
v
al
u
es,
co
n
f
ir
m
in
g
t
h
e
e
f
f
ec
tiv
e
n
es
s
o
f
s
p
i
k
e
-
b
a
s
ed
te
m
p
o
r
al
en
co
d
in
g
f
o
r
P
3
0
0
s
ig
n
als.
Am
o
n
g
t
h
e
m
o
d
els,
C
NN
ac
h
iev
ed
t
h
e
h
ig
h
es
t
av
er
a
g
e
ac
cu
r
ac
y
(
8
1
.
0
4
%),
f
o
llo
w
ed
b
y
SNN
(
8
0
.
9
4
%)
a
n
d
L
ST
M
(
8
0
.
6
0
%).
A
lth
o
u
g
h
C
NN
s
li
g
h
tl
y
o
u
tp
er
f
o
r
m
ed
th
e
o
th
er
s
,
th
e
g
ap
b
et
w
ee
n
C
NN
an
d
SNN
is
m
i
n
i
m
al
th
ese
r
es
u
lts
s
h
o
w
t
h
at
SNNs
b
alan
ce
ac
cu
r
ac
y
an
d
ef
f
icien
c
y
,
p
r
o
v
id
in
g
co
m
p
eti
tiv
e
p
er
f
o
r
m
an
ce
f
o
r
r
ea
l
-
ti
m
e,
lo
w
-
p
o
w
er
B
C
I
ap
p
licatio
n
s
.
Fig
u
r
e
1
2
.
C
NN
p
r
ed
icatio
n
Fig
u
r
e
1
3
.
L
ST
M
p
r
ed
ictio
n
Fig
u
r
e
1
4.
C
o
n
s
ta
n
t
-
leak
I
A
F
m
e
m
b
r
a
n
e
d
y
n
a
m
ics
Evaluation Warning : The document was created with Spire.PDF for Python.