T
E
L
KO
M
NIK
A
, V
ol
.
17
,
No.
5,
O
c
tob
er
20
1
9,
p
p.2
6
50
~
26
58
IS
S
N: 1
69
3
-
6
93
0
,
accr
ed
ited
F
irst
Gr
ad
e b
y K
em
en
r
istekdikti,
Decr
ee
No: 2
1/E/
K
P
T
/20
18
DOI:
10.12928/TE
LK
OM
N
IK
A
.v
1
7
i
5
.
11797
◼
26
50
Rec
ei
v
ed
Nov
e
mb
er
12
,
20
1
7
; R
ev
i
s
ed
J
a
nu
ary
2
8
, 2
0
1
9
;
A
c
c
ep
ted
F
eb
r
u
ary
12
,
20
1
9
Pr
op
osi
ng
a
ne
w
met
ho
d
of
i
m
age clas
sifica
ti
on
b
ased
on
t
he
A
d
aB
oo
s
t
dee
p b
eli
ef n
et
w
o
rk h
y
bri
d m
e
th
o
d
Hasa
n
A
s
il
*
1
,
J
amshid
B
a
g
h
er
z
adeh
2
Dep
a
rtm
e
n
t
o
f
Co
m
p
u
t
e
r En
g
i
n
e
e
ri
n
g
,
F
a
c
u
l
ty
o
f
El
e
c
tr
i
c
a
l
a
n
d
Co
m
p
u
te
r
En
g
i
n
e
e
r
i
n
g
,
Urm
i
a
Un
i
v
e
rs
i
ty
,
Ur
m
i
a
,
I
ra
n
*C
o
rre
s
p
o
n
d
i
n
g
a
u
th
o
r,
e
-
m
a
i
l
:
h
.a
s
i
l
@i
a
u
a
z
a
r.a
c
.i
r
1
,
j
.
b
a
g
h
e
r
z
a
d
e
h
@urm
i
a
.a
c
.i
r
2
Ab
strac
t
Im
a
g
e
c
l
a
s
s
i
fi
c
a
t
i
o
n
h
a
s
d
i
ff
e
re
n
t
a
p
p
l
i
c
a
ti
o
n
s
.
U
p
to
n
o
w,
v
a
r
i
o
u
s
a
l
g
o
r
i
th
m
s
h
a
v
e
b
e
e
n
p
re
s
e
n
te
d
fo
r
i
m
a
g
e
c
l
a
s
s
i
fi
c
a
t
i
o
n
.
Ea
c
h
o
f
th
e
s
e
m
e
th
o
d
s
h
a
s
i
t
s
o
wn
wea
k
n
e
s
s
e
s
a
n
d
s
tre
n
g
th
s
.
Re
d
u
c
i
n
g
e
rr
o
r
ra
t
e
i
s
a
n
i
s
s
u
e
wh
i
c
h
m
a
n
y
re
s
e
a
rc
h
e
s
h
a
v
e
b
e
e
n
c
a
rri
e
d
o
u
t
a
b
o
u
t
i
t.
Th
i
s
r
e
s
e
a
rc
h
i
n
t
e
n
d
s
to
o
p
t
i
m
i
z
e
th
e
p
ro
b
l
e
m
wit
h
h
y
b
ri
d
m
e
th
o
d
s
a
n
d
d
e
e
p
l
e
a
rn
i
n
g
.
T
h
e
h
y
b
ri
d
m
e
th
o
d
s
were
d
e
v
e
l
o
p
e
d
to
i
m
p
ro
v
e
th
e
re
s
u
l
ts
o
f
th
e
s
i
n
g
l
e
-
c
o
m
p
o
n
e
n
t
m
e
th
o
d
s
.
O
n
t
h
e
o
th
e
r
h
a
n
d
,
a
d
e
e
p
b
e
l
i
e
f
n
e
tw
o
rk
(DBN
)
i
s
a
g
e
n
e
ra
t
i
v
e
p
ro
b
a
b
i
l
i
s
ti
c
m
o
d
e
l
wit
h
m
u
l
t
i
p
l
e
l
a
y
e
rs
o
f
l
a
te
n
t
v
a
ri
a
b
l
e
s
a
n
d
i
s
u
s
e
d
t
o
s
o
l
v
e
th
e
u
n
l
a
b
e
l
e
d
p
ro
b
l
e
m
s
.
I
n
fa
c
t,
th
i
s
m
e
t
h
o
d
i
s
a
n
u
n
s
u
p
e
rv
i
s
e
d
m
e
t
h
o
d
,
i
n
whi
c
h
a
l
l
l
a
y
e
rs
a
re
o
n
e
-
w
a
y
d
i
re
c
te
d
l
a
y
e
rs
e
x
c
e
p
t
fo
r
th
e
l
a
s
t
l
a
y
e
r.
So
fa
r,
v
a
ri
o
u
s
m
e
th
o
d
s
h
a
v
e
b
e
e
n
p
r
o
p
o
s
e
d
fo
r
i
m
a
g
e
c
l
a
s
s
i
fi
c
a
t
i
o
n
,
a
n
d
th
e
g
o
a
l
o
f
th
i
s
re
s
e
a
r
c
h
p
ro
j
e
c
t
wa
s
to
u
s
e
a
c
o
m
b
i
n
a
ti
o
n
o
f
t
h
e
Ad
a
Bo
o
s
t
m
e
th
o
d
a
n
d
th
e
d
e
e
p
b
e
l
i
e
f
n
e
two
rk
m
e
th
o
d
t
o
c
l
a
s
s
i
fy
i
m
a
g
e
s
.
T
h
e
o
th
e
r
o
b
j
e
c
ti
v
e
w
a
s
t
o
o
b
ta
i
n
b
e
tt
e
r
re
s
u
l
ts
th
a
n
th
e
p
r
e
v
i
o
u
s
re
s
u
l
t
s
.
In
th
i
s
p
ro
j
e
c
t,
a
c
o
m
b
i
n
a
ti
o
n
o
f
th
e
d
e
e
p
b
e
l
i
e
f
n
e
two
r
k
a
n
d
Ad
a
Bo
o
s
t
m
e
th
o
d
was
u
s
e
d
to
b
o
o
s
t
l
e
a
r
n
i
n
g
a
n
d
t
h
e
n
e
two
rk
p
o
te
n
t
i
a
l
wa
s
e
n
h
a
n
c
e
d
b
y
m
a
k
i
n
g
t
h
e
e
n
t
i
re
n
e
two
r
k
r
e
c
u
r
s
i
v
e
.
T
h
i
s
m
e
th
o
d
w
a
s
te
s
te
d
o
n
th
e
M
INIST
d
a
ta
s
e
t
a
n
d
th
e
re
s
u
l
ts
w
e
re
i
n
d
i
c
a
ti
v
e
o
f
a
d
e
c
re
a
s
e
i
n
th
e
e
rr
o
r
ra
t
e
wit
h
t
h
e
p
ro
p
o
s
e
d
m
e
th
o
d
a
s
c
o
m
p
a
re
d
to
t
h
e
A
d
a
Bo
o
s
t
a
n
d
d
e
e
p
b
e
l
i
e
f
n
e
two
r
k
m
e
th
o
d
s
.
Key
w
ords
:
b
o
o
s
t
i
n
g
,
d
e
e
p
b
e
l
i
e
f
n
e
two
rk
,
d
e
e
p
l
e
a
rn
i
n
g
,
h
y
b
ri
d
m
e
th
o
d
s
,
i
m
a
g
e
c
l
a
s
s
i
fi
c
a
t
i
o
n
Copy
righ
t
©
2
0
1
9
Uni
v
e
rsi
t
a
s
Ahm
a
d
D
a
hl
a
n.
All
rig
ht
s
r
e
s
e
rve
d
.
1.
Int
r
o
d
u
ctio
n
In
ord
er
to
i
d
en
t
i
f
y
th
e
c
o
nte
nt
of
an
i
m
ag
e,
i
t
i
s
n
ec
es
s
ar
y
to
ex
tr
ac
t
a
nd
p
r
oc
es
s
i
nf
orm
ati
on
f
r
o
m
i
m
ag
e.
V
a
r
i
ou
s
m
eth
od
s
ha
v
e
be
en
pres
en
te
d
f
or
proc
es
s
i
ng
t
hi
s
i
nf
orm
ati
on
tha
t
ea
c
h
of
the
m
ha
s
i
ts
s
pe
c
i
a
l
f
ea
tures
.
Di
f
f
erent
a
l
go
r
i
thm
s
ha
v
e
be
en
pres
e
nte
d
f
or
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
u
p
to
no
w.
Mo
s
t
of
the
s
e
a
l
g
orit
hm
s
are
b
as
ed
on
arti
f
i
c
i
al
i
nte
l
l
i
ge
nc
e
a
nd
m
ac
hi
ne
l
ea
r
n
i
n
g.
T
he
av
ai
l
a
bl
e
m
eth
od
s
h
av
e
d
i
f
f
erent
po
te
nt
i
al
s
b
as
ed
on
t
he
d
ata
s
ets
an
d
l
ea
r
ni
n
g
m
eth
od
s
.
E
ac
h
c
l
as
s
i
f
i
c
at
i
o
n
pro
bl
em
s
ha
s
err
or
ba
s
e
d
o
n
t
he
ne
t
w
ork
t
y
p
e
an
d
s
tr
uc
ture.
T
hi
s
err
or
c
an
ha
v
e
d
i
f
f
erent
r
e
as
on
s
.
T
hi
s
s
tu
d
y
h
as
ac
te
d
to
r
ed
uc
e
the
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
err
or.
Deep
l
e
arni
ng
b
as
ed
m
i
x
ed
m
eth
od
s
are us
ed
to
r
e
du
c
e e
r
r
or.
A
r
ti
f
i
c
i
a
l
ne
ura
l
ne
t
wor
k
s
ha
v
e
b
ee
n
em
pl
o
y
e
d
f
or
c
l
as
s
i
f
i
c
ati
on
pu
r
p
os
es
gi
v
e
n
the
p
ote
n
ti
a
l
of
th
e
l
a
y
ers
to
l
e
arn
t
he
ne
w
h
y
bri
d
-
ba
s
ed
f
ea
tur
es
.
T
he
de
ep
be
l
i
ef
m
od
el
s
are
the
ex
te
nd
e
d
v
ers
i
on
s
of
the
arti
f
i
c
i
a
l
n
eu
r
a
l
ne
t
wor
k
m
od
el
s
.
T
he
s
e
m
od
el
s
h
av
e
nu
m
erous
ap
p
l
i
c
at
i
on
s
i
n
c
l
as
s
i
f
y
i
ng
te
x
ts
an
d
i
m
ag
e
s
an
d
proc
es
s
i
ng
s
ate
l
l
i
te
a
nd
m
ed
i
c
al
i
m
ag
es
an
d
the
y
ha
v
e
b
ee
n
wi
de
l
y
us
e
d
r
ec
e
ntl
y
[
1].
D
ee
p
l
ea
r
n
i
ng
h
as
b
ee
n
ba
s
e
d
o
n
th
e
arti
f
i
c
i
a
l
n
eu
r
al
ne
t
w
ork
s
an
d
r
es
ea
r
c
he
r
s
ai
m
to
m
od
el
the
h
i
g
h
-
l
e
v
e
l
ab
s
tr
ac
ti
on
of
the
da
ta.
T
he
s
e
m
od
el
s
e
x
tr
ac
t
m
ul
ti
pl
e
f
ea
tures
f
r
o
m
the
d
ata
an
d
an
a
l
y
z
e
th
em
.
A
da
t
a
c
a
n
be
a
w
or
d,
pi
x
e
l
,
f
r
eq
ue
nc
y
,
etc
.
A
l
tho
ug
h
thi
s
da
t
a
c
an
c
on
v
e
y
an
i
ns
i
gn
i
f
i
c
an
t
m
ea
ni
ng
,
a
c
o
m
bi
na
ti
on
of
thi
s
da
ta
c
a
n
l
ea
d
t
o
be
t
ter
r
es
ul
ts
[2
].
Deep
l
e
arni
ng
i
s
us
ef
ul
f
o
r
c
ertai
n
s
c
en
ario
s
an
d
i
t
i
n
v
o
l
v
es
t
he
us
e
of
th
e
m
ac
hi
n
e
l
ea
r
n
i
n
g
m
od
el
s
an
d
o
the
r
tec
hn
i
qu
es
f
or
the
c
r
ea
ti
o
n
of
m
ea
ni
ng
f
ul
r
es
u
l
ts
[
3].
Deep
m
od
el
s
ha
v
e
s
ev
er
al
l
ate
nt
l
a
y
ers
a
nd
nu
m
erous
pa
r
am
ete
r
s
t
ha
t
ne
ed
to
b
e
tau
g
ht.
T
hi
s
c
o
m
pu
tat
i
o
na
l
c
o
m
pl
ex
i
t
y
an
d
th
e
l
arg
er
p
aram
ete
r
s
pa
c
e
h
av
e
r
e
du
c
ed
t
he
us
e
of
a
l
arge
nu
m
be
r
of
l
a
y
ers
i
n
the
c
om
m
on
ne
ural
n
et
w
or
k
m
eth
od
s
[
4].
T
he
l
arge
n
um
be
r
of
the
l
a
y
ers
i
n
th
es
e
n
et
w
ork
s
no
t
on
l
y
r
ed
uc
es
the
s
p
ee
d
bu
t
al
s
o
r
es
ul
ts
i
n
t
he
l
oc
al
m
i
ni
m
a
an
d
un
s
a
ti
s
f
ac
tor
y
r
e
s
ul
ts
[5]
.
D
ee
p
be
l
i
ef
ne
t
wor
k
s
ha
v
e
pr
ov
i
de
d
t
he
s
o
l
ut
i
o
n
to
th
i
s
pro
bl
em
an
d
t
he
op
p
ortun
i
t
y
t
o
us
e
m
ul
ti
p
l
e
ne
t
w
ork
s
.
In
a
dd
i
ti
on
,
D
ee
p
be
l
i
ef
ne
t
wor
k
s
ha
v
e
a
pp
l
i
c
at
i
on
s
to
f
ea
t
ure
l
e
arni
ng
an
d
c
l
as
s
i
f
i
c
ati
on
[
6].
A
d
ee
p
be
l
i
ef
ne
t
wor
k
i
s
a
ge
ne
r
ati
v
e
proba
bi
l
i
s
ti
c
m
od
el
c
om
po
s
ed
of
m
ul
ti
pl
e
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
P
r
op
os
i
ng
a n
ew m
eth
od
o
f
i
ma
ge
c
l
as
s
i
fi
c
a
ti
o
n
... (
J
a
ms
hi
d
B
a
gh
er
z
a
de
h
)
2651
l
a
y
ers
of
r
an
do
m
hi
dd
e
n
u
ni
ts
,
w
h
i
c
h
are
on
to
p
of
a
l
a
y
er
of
v
i
s
i
b
l
e
v
aria
bl
es
or
a
da
ta
v
ec
tor.
F
i
gu
r
e
1
s
ho
w
s
an
ex
am
pl
e
of
th
es
e n
et
w
ork
s
.
V
ari
ou
s
a
l
go
r
i
thm
s
ha
v
e
be
e
n
prop
o
s
ed
f
or
tr
a
i
ni
n
g
a
de
e
p
be
l
i
ef
ne
t
wor
k
on
the
r
e
l
e
v
an
t
w
e
i
gh
ts
[7]
.
O
ne
of
the
di
s
t
i
nc
ti
v
e
c
ha
r
ac
t
eris
ti
c
s
of
a
de
ep
be
l
i
ef
ne
t
wor
k
i
s
tha
t
a
l
l
of
the
s
tat
es
of
the
hi
dd
en
un
i
ts
of
the
ne
t
w
ork
are
i
de
nti
f
i
ed
o
n
a
f
orw
ar
d
pa
t
h
an
d
r
eg
r
es
s
i
on
i
s
no
t
al
l
o
wed
.
A
l
th
ou
gh
th
i
s
de
du
c
t
i
on
i
s
n
ot
c
om
pl
et
el
y
ac
c
ura
te,
i
t
i
s
r
el
at
i
v
el
y
ac
c
urate.
A
f
ter
tr
ai
n
i
ng
th
e
de
e
p
be
l
i
ef
n
et
wor
k
,
al
l
of
the
pro
ba
b
i
l
i
s
ti
c
m
od
el
s
are
r
u
l
e
d
ou
t
an
d
th
e
r
es
u
l
t
i
ng
wei
g
hts
ar
e
us
e
d
as
a
n
e
w
s
tarti
ng
s
et
f
or
t
he
ne
ural
ne
t
wor
k
w
e
i
g
hts
.
T
hi
s
p
r
oc
es
s
i
s
c
al
l
ed
“
pre
-
tr
ai
n
i
ng
”
.
F
ol
l
o
wi
ng
th
e
pre
-
tr
ai
ni
n
g
p
ha
s
e,
on
e
l
a
y
e
r
i
s
ad
de
d
as
th
e
o
utp
u
t
of
the
n
et
w
or
k
as
aS
of
tm
ax
f
un
c
ti
on
an
d
t
he
en
t
i
r
e
ne
t
wor
k
i
s
tr
ai
ne
d
di
s
c
r
i
m
i
na
ti
v
el
y
.
T
hi
s
ne
t
wor
k
ha
s
be
en
us
ed
i
n
s
ol
v
i
ng
v
ari
ou
s
p
r
ob
l
em
s
[8]
.
T
he
en
s
em
bl
e
c
l
as
s
i
f
i
c
ati
o
ns
be
l
on
g
to
the
f
am
i
l
y
of
the
m
ul
ti
-
c
om
po
ne
nt
c
l
as
s
i
f
i
c
ati
on
ap
pro
ac
he
s
,
whi
c
h
wer
e
pr
op
os
e
d
to
prod
uc
e
be
t
ter
r
es
u
l
ts
tha
n
a
s
i
n
gl
e
-
c
om
po
ne
nt
c
l
as
s
i
f
i
c
ati
on
a
pp
r
o
ac
h
[
9].
In
th
i
s
c
l
as
s
i
f
i
c
ati
o
n,
di
f
f
erent
en
s
em
bl
e
c
l
as
s
i
f
i
c
ati
on
ap
proac
he
s
a
r
e
us
ed
to
ob
t
ai
n
be
t
ter
r
e
s
ul
ts
,
an
d
th
e
h
y
bri
d
ap
pro
ac
he
s
di
f
f
er
i
n
the
i
r
c
l
as
s
i
f
i
c
ati
o
n
m
ec
ha
ni
s
m
s
an
d
ho
w
t
he
y
c
o
m
bi
ne
the
ba
s
e
c
l
as
s
i
f
i
er
i
n
r
e
l
at
i
o
n
to
the
wei
g
hts
[1
0].
In
f
ac
t,
the
r
e
are
t
wo
po
s
s
i
b
l
e
f
r
a
m
ew
ork
s
f
or
the
en
s
em
bl
es
:
de
pe
nd
en
t
(
s
eq
ue
n
ti
al
)
an
d
i
nd
ep
e
nd
e
nt
(
pa
r
a
l
l
el
)
[1
1].
In
a
de
p
en
d
en
t
f
r
am
ew
or
k
,
the
ou
tpu
t
of
on
e
c
l
as
s
i
f
i
er
i
s
us
ed
to
c
r
ea
te
t
he
s
ub
s
eq
ue
nt
c
l
as
s
i
f
i
er.
He
nc
e,
i
t
i
s
po
s
s
i
b
l
e
to
us
e
t
he
k
no
wl
ed
ge
ge
ne
r
ate
d
thro
ug
h
the
pr
ev
i
ou
s
c
y
c
l
es
t
o
d
i
r
e
c
t
the
l
ea
r
n
i
n
g
proc
es
s
i
n
t
he
s
ub
s
e
qu
e
nt
c
y
c
l
es
[1
2].
B
o
os
ti
n
g
i
s
an
ex
am
pl
e
of
the
ap
pl
i
c
ati
on
of
thi
s
ap
pro
ac
h.
In
t
he
s
ec
on
d
f
r
am
ew
ork
,
i
.e.
t
h
e
i
nd
e
pe
n
de
n
t
f
r
a
m
ew
ork
,
ea
c
h
c
l
as
s
i
f
i
er
i
s
bu
i
l
t
i
nd
i
v
i
d
ua
l
l
y
an
d
the
o
utp
uts
of
the
c
l
as
s
i
f
i
ers
are
c
o
m
bi
ne
d
w
i
th
the
po
l
l
i
n
g
m
eth
od
s
[13
].
T
he
pres
en
t
r
es
e
arc
h
go
a
l
was
to
us
e
the
de
e
p
be
l
i
ef
ne
t
w
ork
h
y
brid
m
eth
od
s
to
op
t
i
m
i
z
e t
h
e re
s
ul
ts
of
i
m
ag
e c
l
as
s
i
f
i
c
ati
o
n
.
Natu
r
a
l
l
y
,
the
m
ai
n
go
a
l
of
th
i
s
pro
bl
em
an
d
t
he
ot
he
r
l
e
arni
ng
prob
l
em
s
i
s
to
p
r
ov
i
de
be
tte
r
r
es
ul
ts
. T
he
h
y
br
i
d
m
eth
o
ds
w
ere
us
ed
i
n
th
i
s
r
e
s
ea
r
c
h t
o
i
m
prov
e t
h
e res
u
l
ts
. A
s
h
um
an
s
us
e
th
e
pre
v
i
o
us
r
es
u
l
ts
a
n
d
r
es
ea
r
c
h
f
i
nd
i
ng
s
to
m
a
k
e
d
ec
i
s
i
o
ns
a
nd
ob
t
ai
n
be
tt
er
r
es
ul
ts
,
th
i
s
m
eth
od
ha
s
be
en
al
s
o
em
p
l
o
y
ed
i
n
v
ar
i
ou
s
s
tu
di
es
on
l
ea
r
n
i
n
g.
O
ne
of
the
r
e
l
at
ed
tec
hn
i
qu
e
s
i
s
the
m
e
m
or
y
r
ev
i
v
al
tec
hn
i
q
ue
.
V
ario
us
s
tud
i
es
ha
v
e
b
ee
n
c
on
du
c
te
d
on
t
hi
s
to
pi
c
.
T
he
pres
en
t
r
es
ea
r
c
h
go
a
l
w
as
to
c
om
bi
ne
the
A
d
aB
o
os
t,
de
ep
b
el
i
ef
ne
twork
,
an
d
ne
ura
l
ne
t
w
ork
s
m
eth
od
s
.
In
f
ac
t,
th
e
go
a
l
was
to
c
o
m
bi
ne
A
d
aB
oo
s
t
wi
th
d
ee
p
be
l
i
ef
ne
t
wor
k
s
to
c
r
ea
te
a
m
e
m
ory
-
ba
s
ed
ne
t
w
ork
,
us
e
the
pre
v
i
ou
s
l
e
arni
n
g
r
es
u
l
ts
i
n
th
e
s
u
bs
eq
ue
nt
i
t
erat
i
on
s
,
an
d
produc
e
be
t
ter
r
es
u
l
ts
.
W
i
th
the
pro
po
s
ed
h
y
bri
d
m
eth
od
,
the
r
es
ul
ts
f
r
om
t
he
l
arge
-
s
c
al
e
s
tud
i
es
are
ex
pe
c
ted
to
be
c
l
as
s
i
f
i
ed
m
ore s
pe
c
i
f
i
c
al
l
y
an
d t
he
c
om
pu
tat
i
on
al
di
m
en
s
i
o
n i
s
ex
p
ec
ted
to
d
ec
r
e
as
e.
2.
L
it
er
atu
r
e R
ev
iew
Rec
en
t
l
y
t
he
de
ep
b
el
i
ef
ne
t
w
ork
s
ha
v
e
be
en
us
e
d
t
o
s
ol
v
e
v
ario
us
t
y
p
es
of
probl
em
s
s
uc
h
as
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
,
o
bj
ec
t
r
ec
og
n
i
ti
on
,
an
d
f
ea
ture
ex
tr
ac
ti
on
[1
4].
T
he
s
e
ne
t
w
ork
s
em
p
l
o
y
di
f
f
erent
t
ec
hn
i
qu
e
s
to
op
t
i
m
i
z
e
the
s
e
m
eth
od
s
.
T
he
no
ti
o
n
of
the
d
ee
p
be
l
i
ef
n
et
w
ork
s
was
pro
po
s
ed
b
y
Hi
n
ton
.
T
he
s
e
n
et
w
ork
s
of
f
er
pl
en
t
y
of
ad
v
a
nta
g
es
a
nd
ha
v
e
b
ee
n
us
ed
to
s
ol
v
e
di
f
f
erent
t
y
p
es
of
prob
l
em
s
[15
].
O
n
e
of
the
s
e
prob
l
em
s
w
as
th
e
c
l
as
s
i
f
i
c
ati
on
of
the
s
oc
i
a
l
ne
t
w
ork
s
us
i
ng
a
ne
ural
D
B
N
ne
t
wor
k
an
d
the
g
en
e
t
i
c
al
g
orit
hm
(
G
A
)
[16
].
In
an
ot
he
r
s
tud
y
,
a
ne
w
m
eth
od
of
ne
ural
ne
twork
pre
-
tr
ai
ni
n
g
w
as
i
ntro
du
c
ed
b
as
ed
o
n
B
ol
t
z
m
an
n
m
ac
hi
ne
s
to
ac
c
el
erat
e
t
he
tr
ai
ni
ng
pro
c
es
s
an
d
en
ha
nc
e
t
he
ph
o
ne
m
e
r
ec
og
ni
ti
o
n.
O
the
r
r
e
s
ea
r
c
he
r
s
al
s
o
us
ed
th
e
d
ee
p
be
l
i
ef
ne
t
wor
k
s
to
r
ec
og
n
i
z
e
th
e
F
ars
i
nu
m
be
r
s
[17
].
I
n
an
ot
h
er
s
tud
y
on
de
e
p
l
ea
r
n
i
n
g,
the
c
o
nte
x
t
-
b
as
e
d
wor
d
r
ec
og
ni
t
i
o
n
ab
i
l
i
t
y
was
an
a
l
y
z
e
d
[18
].
F
r
i
t
z
et
al
.
pro
po
s
ed
a
deep
-
l
ea
r
n
i
n
g
m
od
el
f
or
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
.
T
he
i
r
m
od
el
c
om
bi
ne
s
a
de
ep
ne
t
wor
k
w
i
th
the
P
i
x
el
RNN
a
nd
DCG
A
N
m
od
el
s
an
d
i
t
i
s
us
ed
f
or
i
m
ag
e
r
ec
og
ni
t
i
o
n.
T
he
s
e
m
od
el
s
w
ere
de
v
el
op
e
d
wi
th
r
eg
ard
to
P
i
x
e
l
RNN
an
d
DCG
A
N
f
or
ha
n
d
w
r
i
t
ten
da
t
a
[
19
].
O
th
er
r
es
ea
r
c
he
r
s
c
on
du
c
te
d
r
es
ea
r
c
h
o
n
th
e
A
DG
M
proj
ec
t
c
l
as
s
i
f
i
c
a
ti
on
[20
]
.
T
he
A
r
ti
f
i
c
i
a
l
D
e
ep
G
en
erati
v
e
Neura
l
N
et
w
ork
s
Mo
de
l
(
A
DG
M)
w
as
de
v
e
l
op
ed
u
s
i
n
g
a
d
i
v
ers
e
s
et
of
en
c
o
de
r
s
an
d
d
ec
od
ers
.
T
hi
s
proj
ec
t
w
as
tes
te
d
on
t
he
MIN
IS
T
da
tas
et
[21
]
.
T
he
G
A
N
proj
ec
t
was
al
s
o
a
no
the
r
proj
ec
t
on
i
m
ag
e
c
l
as
s
i
f
i
c
ati
o
n.
T
he
r
e
s
ul
ts
f
r
o
m
thi
s
proj
ec
t
c
a
n
b
e
us
ed
f
or
K
-
c
l
as
s
c
l
as
s
i
f
i
c
a
ti
on
.
M
oreo
v
er,
on
e
of
the
m
a
j
or
de
ep
-
n
et
wor
k
ap
proac
he
s
i
s
ba
s
e
d
on
the
c
r
ea
t
i
on
of
di
f
f
erent
l
a
y
ers
f
or
f
ea
ture
l
ea
r
n
i
n
g [
2
2
].
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 1
69
3
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
265
0
-
26
58
2652
3.
T
h
e P
r
o
p
o
se
d
M
eth
o
d
T
he
de
ep
be
l
i
ef
ne
t
w
ork
s
are
RB
M
-
ba
s
e
d
n
et
w
ork
s
[23
,
2
4].
T
he
r
e
are
a
l
s
o
t
wo
oth
er
ne
t
w
ork
t
y
pe
s
,
na
m
el
y
t
he
Deep
B
ol
t
z
m
an
Ma
c
hi
ne
s
(
DB
Ms
)
an
d
D
ee
p
E
ne
r
g
y
M
od
e
l
s
(
DE
Ms
)
,
whi
c
h
are
d
ev
el
op
e
d
ba
s
e
d
on
the
s
e
n
et
w
ork
s
[25
].
F
i
g
ure
2
pres
en
ts
th
e
i
nt
erc
on
ne
c
ti
o
ns
wi
th
i
n
the
s
e
n
et
w
ork
s
.
A
s
s
ee
n
i
n
F
i
gu
r
e
2,
t
he
D
B
Ns
ha
v
e
un
d
i
r
ec
ted
c
o
nn
ec
t
i
on
s
i
n
the
u
pp
er
t
wo
l
a
y
ers
,
w
hi
c
h
f
orm
an
RB
M.
In
a
dd
i
ti
o
n,
th
e
di
r
ec
t
ed
c
on
ne
c
ti
on
s
are
i
n
the
l
o
wer
l
a
y
ers
[26
].
T
he
m
ai
n
c
ha
r
ac
ter
i
s
ti
c
of
thi
s
n
et
wor
k
i
s
tha
t
th
e
tr
ai
n
i
ng
i
s
u
ns
up
er
v
i
s
e
d,
wh
i
c
h
e
l
i
m
i
na
tes
the
n
ee
d
f
or
the
l
a
be
l
ed
da
t
a
f
or
tr
ai
ni
ng
.
A
d
ee
p
be
l
i
ef
ne
t
w
ork
i
s
a
ge
ne
r
at
i
v
e
pro
ba
b
i
l
i
s
ti
c
m
od
el
,
i
n
whi
c
h
t
he
j
o
i
nt
pro
ba
b
i
l
i
t
y
di
s
tr
i
bu
t
i
o
n o
f
t
he
da
t
a
i
s
v
i
s
i
bl
e
a
nd
pro
v
i
de
s
t
he
l
a
be
l
s
. A
DB
N
f
i
r
s
t
us
es
an
o
pti
m
u
m
l
a
y
er
ed
g
r
ee
d
y
s
tr
a
teg
y
f
or
the
i
n
i
ti
al
i
z
ati
on
(
of
the
de
e
p
ne
t
w
or
k
pa
r
a
m
ete
r
s
)
,
an
d
th
en
s
ets
a
l
l
of
t
he
w
e
i
gh
ts
j
oi
n
tl
y
i
n
r
el
ati
on
to
th
e
ex
p
ec
ted
o
utp
u
ts
.
T
he
greed
y
l
e
arni
ng
proc
ed
ure
ha
s
t
w
o
a
dv
an
tag
es
[2
7]:
f
i
r
s
tl
y
,
i
t
pro
v
i
de
s
f
or
proper
ne
t
wor
k
i
ni
ti
al
i
z
ati
on
a
nd
the
r
ef
ore
i
t
r
e
d
uc
es
t
he
di
f
f
i
c
ul
t
y
of
s
el
ec
ti
n
g
t
he
pa
r
a
m
ete
r
s
(
w
hi
c
h
m
a
y
l
e
ad
t
o
the
s
e
l
ec
t
i
on
of
l
oc
al
op
t
i
m
a);
s
ec
on
dl
y
,
t
h
e
l
e
arni
ng
proc
es
s
i
s
u
ns
u
pe
r
v
i
s
ed
,
wi
t
ho
ut
a
ne
e
d
f
or
a
c
l
as
s
l
ab
e
l
.
Henc
e,
the
tr
a
i
n
i
ng
ne
ed
f
or
the
l
ab
e
l
e
d
d
ata
i
s
av
oi
de
d
[
28
].
Ho
wev
er,
t
he
de
v
e
l
op
m
en
t
of
a
DB
N
m
od
el
i
c
urs
he
av
y
c
om
pu
ta
ti
on
al
c
os
ts
be
c
au
s
e
i
t
i
s
no
t
k
no
wn
ho
w
to
a
pp
r
ox
i
m
ate
the
m
ax
i
m
u
m
tr
ai
n
i
ng
l
i
k
el
i
ho
od
f
or the
m
od
e
l
op
ti
m
i
z
at
i
on
[2
9
].
A
s
s
tat
ed
i
n
s
ec
ti
o
n
1,
th
e
e
ns
em
bl
e
c
l
as
s
i
f
i
c
ati
on
s
w
er
e
propos
e
d
to
i
m
prov
e
t
he
r
es
ul
t
s
.
T
he
s
e
m
ul
ti
-
c
o
m
po
ne
nt
c
l
a
s
s
i
f
i
c
ati
on
ap
proac
he
s
r
es
ul
t
i
n
be
tt
er
r
es
ul
ts
.
T
he
prop
os
ed
al
go
r
i
t
hm
was
al
s
o
de
s
i
gn
ed
to
us
e
t
he
h
y
bri
d
m
eth
od
s
an
d
th
e
de
ep
b
el
i
ef
m
eth
od
to
i
m
pr
ov
e
t
he
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
r
es
ul
ts
.
In
thi
s
r
es
ea
r
c
h,
a
bo
os
ti
n
g
h
y
br
i
d
a
l
go
r
i
thm
w
as
us
ed
t
o
c
o
m
bi
ne
the
da
t
a,
an
d t
he
propos
e
d d
es
i
gn
i
s
i
l
l
us
tr
ate
d
i
n Fi
gu
r
e
3
.
F
i
gu
r
e
1.
A
de
ep
be
l
i
ef
ne
t
wor
k
F
i
gu
r
e
2.
A
c
om
pa
r
i
s
on
be
t
ween
the
Dee
p B
ol
t
z
m
an
n
et
w
ork
s
F
i
gu
r
e
3
.
T
he
pro
po
s
ed
al
g
orit
hm
s
tr
uc
ture
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
P
r
op
os
i
ng
a n
ew m
eth
od
o
f
i
ma
ge
c
l
as
s
i
fi
c
a
ti
o
n
... (
J
a
ms
hi
d
B
a
gh
er
z
a
de
h
)
2653
4.
A
s
se
s
sment
of
t
h
e P
r
o
p
o
se
d
A
lgo
r
it
h
m
In
order
to
as
s
es
s
the
propos
e
d
al
go
r
i
t
hm
,
i
t
was
i
m
pl
e
m
en
ted
i
n
M
A
T
LA
B
o
n
the
MINI
S
T
da
t
as
et.
T
hi
s
d
ata
s
et
c
on
s
i
s
ts
of
ap
pr
ox
i
m
ate
l
y
60
00
0
r
ec
ords
of
ha
nd
w
r
i
tte
n
E
ng
l
i
s
h
nu
m
be
r
s
f
or
l
ea
r
ni
n
g
a
nd
10
00
0
d
ata
r
ec
ords
f
or
tes
ti
ng
.
In
order
to
c
o
m
pa
r
e
the
r
es
u
l
ts
,
t
he
de
ep
be
l
i
ef
ne
t
w
ork
,
A
d
aB
o
os
t,
an
d
th
e
pro
po
s
e
d
D
B
N
-
ba
s
e
d
A
da
B
oo
s
t
i
ng
m
eth
od
s
w
ere
t
es
ted
.
O
n
t
he
ot
he
r
ha
nd
,
f
or
a
m
ore
prec
i
s
e
an
a
l
y
s
i
s
,
the
r
es
ul
ts
f
r
o
m
ea
c
h
tes
t
wer
e
c
om
pa
r
ed
c
y
c
l
e
-
wi
s
e
as
pres
en
te
d
i
n t
he
f
ol
l
o
wi
n
g.
4
.
1
.
A
d
aBoo
st
F
i
gu
r
e
4
d
ep
i
c
ts
t
he
l
ea
r
n
i
ng
proc
es
s
i
n
th
e
A
da
B
oo
s
ti
ng
al
g
orit
hm
throug
h
d
i
f
f
erent
c
y
c
l
es
.
A
s
s
ee
n,
wi
th
an
i
n
c
r
ea
s
e
i
n
t
he
c
y
c
l
es
the
err
or
r
ate
d
ec
r
ea
s
es
.
T
o
s
ol
v
e
thi
s
pr
ob
l
em
,
the
pro
bl
em
s
pa
c
e
was
di
v
i
de
d
i
nt
o
10
s
e
gm
en
ts
an
d
the
l
e
arni
ng
pr
oc
es
s
w
as
c
o
m
pl
ete
d
f
o
r
the
s
e
c
l
as
s
es
.
T
he
A
d
aB
oo
s
ti
n
g
proc
es
s
a
l
s
o
t
oo
k
pl
ac
e.
Nat
ural
l
y
,
wi
th
a
n
i
nc
r
e
as
e
i
n
the
nu
m
be
r
of
th
e
c
y
c
l
es
, t
he
l
ea
r
ni
n
g
ti
m
e e
s
c
al
at
ed
.
F
or i
ns
t
an
c
e,
10
0,
50
0,
an
d
20
00
l
e
arni
ng
proc
es
s
es
were c
om
pl
ete
d
f
or 10,
50
,
an
d
20
0 c
y
c
l
es
,
r
es
pe
c
ti
v
el
y
.
(
a)
(
b)
(
c
)
(
d)
F
i
gu
r
e
4.
T
he
proc
es
s
of
l
e
arni
n
g b
as
ed
o
n t
h
e n
um
be
r
of
c
y
c
l
es
us
i
ng
the
A
da
B
o
os
ti
n
g l
ea
r
n
i
ng
m
eth
od
:
(
a)
10
c
y
c
l
es
, (b)
5
0 c
y
c
l
es
, (c
)
20
0 c
y
c
l
es
, (d)
50
0 c
y
c
l
es
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 1
69
3
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
265
0
-
26
58
2654
4.2
. Con
v
o
lut
ion
In
t
he
s
u
bs
eq
u
en
t
ex
p
erim
en
ts
,
th
e
c
o
nv
ol
u
ti
o
na
l
ne
t
wo
r
k
err
or
r
ate
s
at
di
f
f
erent
i
ter
ati
o
ns
are
ex
am
i
ne
d
an
d
c
om
pa
r
ed
to
the
propos
e
d
al
go
r
i
thm
.
T
he
c
on
v
ol
uti
on
a
l
a
l
go
r
i
th
m
i
s
c
o
m
po
s
ed
of
three
l
a
y
ers
,
na
m
el
y
t
he
f
ul
l
y
-
c
on
ne
c
te
d,
p
oo
l
i
n
g
,
a
nd
c
on
v
o
l
ut
i
on
al
l
a
y
ers
.
T
hi
s
m
eth
od
was
tes
ted
on
th
e
MINI
S
T
da
t
a
s
et
an
d
the
r
es
u
l
ts
are
pr
e
s
en
ted
i
n
F
i
gu
r
e
5.
T
hi
s
f
i
gu
r
e
i
l
l
us
tr
at
es
the
c
om
pl
eti
on
of
the
c
on
v
ol
ut
i
o
na
l
l
ea
r
ni
ng
al
go
r
i
t
hm
proc
es
s
f
or
10
,
50
,
an
d
2
00
c
y
c
l
es
.
A
s
s
ee
n,
t
he
A
da
B
o
os
ti
n
g
m
eth
od
ou
tp
ac
ed
t
hi
s
m
eth
od
.
(
a)
(
b)
(
c
)
(
d)
F
i
gu
r
e
5.
T
he
proc
es
s
of
l
e
arni
n
g b
as
ed
o
n t
h
e n
um
be
r
of
c
y
c
l
es
us
i
ng
the
c
on
v
o
l
ut
i
o
na
l
l
ea
r
ni
ng
m
eth
od
:
(
a)
10
c
y
c
l
es
, (b)
5
0 c
y
c
l
es
, (c
)
20
0 c
y
c
l
es
, (d)
50
0 c
y
c
l
es
4.3
. D
ee
p
B
eli
ef N
etw
o
r
k
In
or
de
r
t
o
c
om
pa
r
e
t
he
p
r
op
os
ed
m
eth
od
wi
t
h
t
he
de
ep
b
el
i
ef
n
et
w
ork
,
the
MINI
S
T
da
tas
et
was
c
on
v
erted
an
d
i
m
pl
e
m
en
ted
us
i
n
g
a
t
y
p
i
c
al
ne
ura
l
ne
t
wor
k
.
T
he
r
es
ul
ts
are
pres
en
t
ed
i
n t
he
Fi
gu
r
e
6.
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
P
r
op
os
i
ng
a n
ew m
eth
od
o
f
i
ma
ge
c
l
as
s
i
fi
c
a
ti
o
n
... (
J
a
ms
hi
d
B
a
gh
er
z
a
de
h
)
2655
(
a)
(
b)
(
c
)
(
d)
F
i
gu
r
e
6
.
T
he
proc
es
s
of
l
e
arni
n
g b
as
ed
o
n t
h
e n
um
be
r
of
c
y
c
l
es
us
i
ng
the
d
ee
p
be
l
i
ef
ne
t
w
ork
m
eth
od
:
(
a)
10
c
y
c
l
es
,
(
b)
50
c
y
c
l
es
,
(
c
)
20
0 c
y
c
l
es
,
(
d)
50
0 c
y
c
l
es
4.4
. D
BN
-
Bas
ed
A
d
aBoo
s
t
ing
T
he
propos
e
d
m
eth
od
i
s
a
c
om
bi
na
t
i
on
of
t
he
A
da
B
oo
s
t
i
ng
a
nd
de
ep
be
l
i
ef
ne
t
w
ork
m
eth
od
s
.
In
order
t
o
s
i
m
ul
at
e
t
hi
s
m
eth
od
,
th
e
s
a
m
pl
e
s
pa
c
e
was
c
om
bi
ne
d
i
n
t
he
s
pa
c
e
wi
th
the
de
ep
be
l
i
ef
m
eth
od
.
F
i
g
ure
7
al
s
o
pres
en
ts
the
r
es
ul
ts
of
the
l
e
arni
ng
proc
e
s
s
i
n
d
i
f
f
eren
t
c
y
c
l
es
.
5.
Co
mp
a
r
ison
of
t
h
e
A
s
s
es
sm
ent
Res
u
lt
s
T
hi
s
proj
ec
t
ai
m
ed
to
op
ti
m
i
z
e
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
an
d
as
a
r
es
ul
t,
r
ed
uc
i
ng
err
or
r
ate
i
n
i
m
ag
e
c
l
as
s
i
f
i
c
ati
o
n.
F
o
u
r
al
g
orit
hm
s
w
ere
e
v
a
l
u
ate
d
i
n
th
i
s
r
es
ea
r
c
h.
A
c
c
ordi
n
g
to
the
e
v
a
l
ua
ti
o
n,
l
ea
r
n
i
ng
t
i
m
e
i
n
t
he
pro
po
s
ed
m
eth
od
was
hi
g
he
r
tha
n
ot
he
r
m
eth
od
s
.
T
hi
s
i
s
w
h
i
l
e
the
prec
i
s
i
o
n o
f
s
y
s
tem
i
nc
r
ea
s
ed
an
d
as
a
r
es
u
l
t,
err
or r
ate
d
ec
r
ea
s
ed
.
A
c
c
ordi
n
g
to
th
e
r
es
ul
ts
o
f
the
c
om
pa
r
i
s
on
be
t
w
ee
n
thi
s
m
eth
od
an
d
th
e
pro
po
s
ed
m
eth
od
s
,
the
pro
po
s
ed
m
eth
od
w
as
s
l
o
wer
th
an
th
e
a
f
ores
ai
d
m
eth
od
s
,
bu
t
wi
th
an
i
nc
r
ea
s
e
i
n
the
nu
m
be
r
of
c
y
c
l
es
,
t
he
c
on
v
erge
nc
e
i
nc
r
ea
s
e
d.
O
n
the
oth
er
ha
n
d,
t
he
r
es
ul
ts
of
the
c
y
c
l
e
-
wi
s
e
as
s
es
s
m
en
t
of
al
l
t
hree
m
eth
od
s
are
l
i
s
ted
i
n
T
ab
l
e
1.
A
s
s
e
en
,
w
i
t
h
a
n
i
nc
r
ea
s
e
i
n
th
e
nu
m
be
r
of
c
y
c
l
es
t
he
err
or
r
ate
de
c
l
i
ne
s
an
d
thi
s
m
eth
od
ou
tp
erf
or
m
s
the
m
en
ti
on
ed
m
eth
od
s
i
n t
erm
s
of
th
e e
r
r
or r
ate
.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 1
69
3
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
265
0
-
26
58
2656
T
he
r
es
ul
ts
f
or
three
err
or
r
ate
s
are
pres
en
t
ed
i
n
T
ab
l
e
1.
T
ab
l
e
1
i
n
di
c
at
es
err
or
f
o
r
three
r
ate
s
.
T
he
err
or
r
ate
i
n
A
da
B
oo
s
t
i
ng
f
or
50
0
i
tera
ti
on
s
was
0.0
0
8,
0.
19
7
f
or
c
on
v
o
l
ut
i
on
,
0.0
02
4
f
or dee
p b
e
l
i
ef
, a
nd
0.
00
0
8
f
or Ad
aB
o
os
tba
s
e
d o
n
de
e
p b
e
l
i
ef
th
at
err
or r
ate
h
as
s
i
gn
i
f
i
c
an
t e
r
r
or
r
ate
.
T
he
di
ag
r
am
of
the
c
om
pa
r
i
s
on
of
th
e
pr
op
os
e
d
a
l
go
r
i
thm
wi
th
th
e
prev
i
o
us
m
eth
od
s
i
s
de
p
i
c
ted
i
n F
i
gu
r
e
8.
(
a)
(
b)
(
c
)
(
d)
F
i
gu
r
e
7.
A
c
y
c
l
e
-
w
i
s
e c
om
pa
r
i
s
on
be
t
wee
n t
h
e e
r
r
or r
ate
s
us
i
ng
the
d
ee
p
-
be
l
i
ef
A
d
aB
oo
s
ti
n
g m
eth
od
:
(
a)
10
c
y
c
l
es
, (b)
50
c
y
c
l
es
, (c
)
20
0 c
y
c
l
es
, (
d)
50
0 c
y
c
l
es
T
ab
l
e
1
.
T
he
C
om
pa
r
i
s
on
o
f
th
e
E
r
r
or Rat
es
B
as
e
d
o
n
A
l
g
orit
hm
and the
Num
be
r
of
C
y
c
l
es
N
o
.
C
y
c
les
A
d
a
B
o
o
s
t
ing
C
o
n
v
o
lut
ion
D
e
e
p
B
e
li
e
f
D
e
e
p
-
B
e
li
e
f
A
d
a
B
o
o
s
t
ing
1
10
0
.
0
1
3
5
.
0
4
6
8
0
0
.
0
1
9
8
0
.
0
0
6
1
2
50
0
.
0
0
9
4
0
.
5
0
0
0
0
.
0
0
4
8
0
.
0
0
5
7
3
200
0
.
0
0
9
0
.
0
1
9
7
0
.
0
0
2
4
0
.
0
0
2
8
4
500
0
.
0
0
8
0
.
0
0
8
7
0
.
0
0
1
5
0
.
0
0
0
8
Evaluation Warning : The document was created with Spire.PDF for Python.
T
E
L
KO
M
NIK
A
IS
S
N: 1
69
3
-
6
93
0
◼
P
r
op
os
i
ng
a n
ew m
eth
od
o
f
i
ma
ge
c
l
as
s
i
fi
c
a
ti
o
n
... (
J
a
ms
hi
d
B
a
gh
er
z
a
de
h
)
2657
F
i
gu
r
e
8
.
A
c
y
c
l
e
-
w
i
s
e
an
a
l
y
s
i
s
of
th
e
al
go
r
i
t
hm
err
or r
ate
6
.
Dis
cussion
T
he
pres
en
t
r
es
ea
r
c
h
go
al
w
as
to
c
om
bi
ne
th
e
A
da
B
o
os
t,
de
e
p
be
l
i
ef
ne
t
w
ork
,
an
d
ne
ura
l
n
et
w
ork
m
eth
o
ds
to
ob
tai
n
b
ett
er
r
es
u
l
ts
f
r
o
m
i
m
ag
e
c
l
as
s
i
f
i
c
ati
on
.
O
ne
go
al
o
f
m
a
c
hi
ne
l
e
arni
ng
i
s
t
o
de
v
el
o
p
p
att
er
ns
b
as
ed
on
th
e
pre
v
i
ou
s
d
ata
to
i
m
pl
em
en
t
th
e
m
eth
od
s
.
S
tud
i
es
ha
v
e
al
s
o
be
en
c
on
du
c
ted
o
n
m
e
m
orie
s
an
d
the
i
r
ef
f
ec
ts
on
l
ea
r
ni
ng
.
G
i
v
e
n
the
po
te
nti
al
s
of
the
A
da
B
o
os
t
m
eth
od
f
or
the
c
l
as
s
i
f
i
c
a
ti
on
of
the
l
e
arni
ng
da
t
a
a
nd
r
ei
nf
orc
em
en
t
o
f
l
ea
r
n
i
n
g
a
nd
gi
v
e
n
t
he
c
ap
ac
i
t
y
of
the
de
ep
l
e
arn
i
ng
m
eth
od
f
or
l
arge
-
s
c
al
e
op
erati
on
s
,
a
c
o
m
bi
na
t
i
on
of
the
s
e
m
eth
od
s
y
i
e
l
ds
be
tte
r
r
es
u
l
ts
.
B
as
ed
o
n
th
e
i
nv
es
ti
g
ati
on
r
es
ul
ts
,
t
he
err
or
r
ate
d
ec
r
ea
s
ed
ap
pro
x
i
m
ate
l
y
b
y
0.0
0
07
%
an
d
0.0
08
%
as
c
o
m
pa
r
ed
to
the
de
ep
b
el
i
ef
ne
t
w
ork
an
d
c
on
v
ol
u
ti
o
na
l
ap
pr
oa
c
h,
r
es
pe
c
ti
v
el
y
.
T
he
m
ai
n
ad
v
an
ta
ge
of
thi
s
m
eth
od
i
s
th
e
r
e
i
nf
orc
em
en
t
of
l
ea
r
n
i
n
g
thro
ug
h
s
ev
era
l
i
tera
ti
o
ns
,
w
h
i
c
h
i
s
i
n
he
r
i
t
ed
f
r
o
m
A
da
B
o
os
t,
an
d
i
ts
c
ap
ac
i
t
y
f
or
l
arge
-
s
c
al
e
op
era
ti
o
ns
.
T
he
i
m
pl
em
en
tat
i
on
of
thi
s
m
eth
od
wi
th
m
an
y
i
t
erati
on
s
y
i
e
l
de
d
be
tte
r
r
es
ul
ts
.
Ho
wev
er,
the
l
ea
r
ni
ng
t
i
m
e
w
as
i
nc
r
ea
s
e
d
be
c
au
s
e
of
the
c
l
as
s
i
f
i
c
ati
o
n
a
nd
r
ep
e
ate
d
l
e
arni
ng
proc
es
s
es
.
E
v
i
de
nt
l
y
,
t
he
ac
hi
ev
em
en
t
of
be
tte
r
r
es
u
l
ts
ha
s
a
h
i
g
he
r
pri
orit
y
t
ha
n
l
e
arni
n
g.
7
.
Co
n
clus
ion
D
e
ep
b
el
i
e
f
n
etw
o
r
k
s
a
r
e
te
c
h
ni
qu
e
s
w
i
t
h
a
pp
l
i
c
at
i
o
n
s
t
o
fe
a
t
u
r
e
l
ea
r
ni
ng
an
d
c
l
a
s
s
i
fi
c
a
ti
on
.
A
de
ep
be
l
i
e
f
n
e
tw
o
r
k
i
s
a
g
en
e
r
at
i
v
e
p
r
o
ba
bi
l
i
s
ti
c
m
o
d
el
c
o
n
s
i
s
ti
ng
o
f
m
u
l
t
i
p
l
e
l
ay
er
s
o
f
r
a
n
do
m
h
i
d
de
n
u
ni
t
s
tha
t
a
r
e
pl
a
c
e
d
o
n
to
p
o
f
a
l
a
y
e
r
o
f
v
i
s
i
b
l
e
d
a
ta
o
r
a
da
ta
v
e
c
to
r
.
T
h
e
s
e
ne
tw
o
r
k
s
a
r
e
u
s
e
d
to
i
n
c
r
e
a
s
e
t
he
n
u
m
b
e
r
o
f
l
a
y
er
s
a
nd
c
o
nd
u
c
t
m
o
r
e
p
r
e
c
i
s
e
i
nv
e
s
ti
ga
ti
on
s
.
T
h
e
p
r
e
s
e
nt
r
e
s
e
a
r
c
h
w
a
s
al
s
o c
on
du
c
t
ed
u
s
i
ng
t
he
h
y
br
i
d
m
e
t
ho
d
s
a
nd
th
e d
e
ep
be
l
i
e
f n
e
tw
o
r
k
s
t
o
p
r
op
o
s
e
a
n
ew
m
e
t
h
od
fo
r
t
he
c
l
a
s
s
i
fi
c
a
ti
on
o
f
h
an
dw
r
i
t
te
n
da
t
a.
T
he
hy
br
i
d
m
e
t
ho
d
s
c
l
a
s
s
i
f
y
l
ea
r
ni
ng
a
nd
tu
r
n
p
oo
r
l
e
a
r
n
i
n
g
i
nt
o
s
t
r
o
ng
l
e
ar
n
i
n
g
.
T
he
d
ee
p
b
el
i
e
f
ne
tw
or
k
s
a
r
e
u
n
l
a
be
l
e
d,
bu
t
th
e
y
i
n
t
en
d
t
o
s
el
e
c
t
t
h
e
p
r
op
e
r
l
e
a
r
ni
ng
p
a
r
a
m
e
t
e
r
s
u
s
i
n
g
th
e
g
r
e
ed
y
ap
p
r
oa
c
h
e
s
.
T
h
e
ov
e
r
a
r
c
h
i
n
g
g
oa
l
o
f
t
hi
s
pa
pe
r
w
a
s
to
c
o
m
b
i
n
e
t
he
tw
o
m
e
t
ho
d
s
t
o
i
m
p
r
ov
e
t
he
r
e
s
u
l
t
s
a
nd
r
e
du
c
e
e
r
r
o
r
s
.
T
hi
s
c
o
m
b
i
n
a
ti
o
n
i
s
b
a
s
ed
on
r
eg
r
e
s
s
i
o
n
a
nd
ea
c
h
c
l
a
s
s
u
s
e
s
t
he
p
r
ev
i
ou
s
c
l
a
s
s
pa
r
a
m
e
t
e
r
s
fo
r
a
s
s
e
s
s
m
e
n
t
.
H
e
n
c
e
,
i
t
c
a
n
pe
r
fo
r
m
b
e
t
te
r
w
i
t
h
th
e
s
e
m
i
-
s
u
p
e
r
v
i
s
e
d
m
e
t
ho
d
s
.
In
fa
c
t
,
thi
s
m
e
t
ho
d
c
r
ea
t
e
s
a
m
e
m
o
r
y
-
ba
s
ed
hy
br
i
d
n
etw
o
r
k
.
T
h
e
r
e
s
ul
t
s
a
l
s
o
r
e
fl
e
c
te
d
t
he
i
m
p
r
o
v
e
d
c
l
a
s
s
i
fi
c
a
ti
on
s
an
d
th
e
de
c
r
e
a
s
ed
l
ev
e
l
o
f
e
r
r
o
r
r
a
t
e.
I
n
t
hi
s
n
etw
o
r
k
,
e
r
r
o
r
r
a
t
e
ha
s
s
i
gn
i
fi
c
a
nt
r
ed
u
c
ti
on
r
e
l
a
ti
v
e
to
A
da
B
oo
s
t
de
ep
l
ea
r
ni
n
g
a
nd
c
o
nv
ol
u
ti
on
.
I
t
i
s
,
ho
w
e
v
e
r
,
po
s
s
i
b
l
e
t
o
u
s
e
ot
he
r
h
y
b
r
i
d
al
go
r
i
t
h
m
s
o
r
t
he
m
e
m
o
r
y
-
ba
s
ed
m
e
t
h
o
d
s
to
i
m
p
r
o
v
e
the
r
e
s
u
l
t
s
.
Ref
er
en
ce
[1
]
Sc
h
m
i
d
h
u
b
e
r
J
.
Dee
p
l
e
a
rn
i
n
g
i
n
n
e
u
ra
l
n
e
tw
o
rk
s
:
An
o
v
e
rv
i
e
w
.
Neu
ra
l
Ne
two
r
k
s
.
2
0
1
5
;
6
1
:
8
5
-
1
1
7
.
[2
]
Ba
g
h
e
rz
a
d
e
h
J
,
A
s
i
l
H.
A
re
v
i
e
w
o
f
v
a
ri
o
u
s
s
e
m
i
-
s
u
p
e
rv
i
s
e
d
l
e
a
rn
i
n
g
m
o
d
e
l
s
w
i
th
a
d
e
e
p
l
e
a
rn
i
n
g
a
n
d
m
e
m
o
ry
a
p
p
ro
a
c
h
.
Ira
n
J
o
u
rn
a
l
o
f
C
o
m
p
u
te
r S
c
i
e
n
c
e
.
2
0
1
8
:
1
-
6.
[3
]
Su
d
i
a
t
m
i
k
a
IBK
,
Ra
h
m
a
n
F
,
T
r
i
s
n
o
T
,
Su
y
o
to
S.
Im
a
g
e
fo
r
g
e
r
y
d
e
te
c
ti
o
n
u
s
i
n
g
e
rro
r
l
e
v
e
l
a
n
a
l
y
s
i
s
a
n
d
d
e
e
p
l
e
a
r
n
i
n
g
.
TEL
KO
M
NIKA
Te
l
e
c
o
m
m
u
n
i
c
a
ti
o
n
Com
p
u
t
i
n
g
El
e
c
tro
n
i
c
s
a
n
d
Con
tr
o
l
.
2
0
1
9
;
1
7
(
2
)
:
653
-
65
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.
◼
IS
S
N: 1
69
3
-
6
93
0
T
E
L
KO
M
NIK
A
V
ol
.
17
,
No
.
5,
O
c
tob
er 20
19
:
265
0
-
26
58
2658
[4
]
Be
n
g
i
o
Y
,
Cou
rv
i
l
l
e
A,
V
i
n
c
e
n
t
P.
Rep
re
s
e
n
ta
t
i
o
n
L
e
a
rn
i
n
g
:
A
Rev
i
e
w
a
n
d
Ne
w
Pe
rs
p
e
c
t
i
v
e
s
.
IEE
E
Tra
n
s
a
c
t
i
o
n
s
o
n
Pa
tt
e
rn
An
a
l
y
s
i
s
a
n
d
M
a
c
h
i
n
e
In
t
e
l
l
i
g
e
n
c
e
.
2
0
1
3
;
3
5
(
8
):
1
7
9
8
–
1
8
2
8
.
a
rX
i
v
:1
2
0
6
.5
5
3
8
Fre
e
l
y
a
c
c
e
s
s
i
b
l
e
.
d
o
i
:1
0
.1
1
0
9
/t
p
a
m
i
.2
0
1
3
.
5
0
.
[5
]
Su
m
a
T
,Ku
m
a
ra
Sw
a
m
y
Y
S
.
Pl
a
n
n
i
n
g
a
n
d
Coo
rd
i
n
a
t
i
o
n
i
n
H
i
e
ra
rc
h
i
e
s
o
f
In
te
l
l
i
g
e
n
t
Dy
n
a
m
i
c
Sy
s
te
m
s
.
TEL
KO
M
NIKA T
e
l
e
c
o
m
m
u
n
i
c
a
ti
o
n
Co
m
p
u
ti
n
g
El
e
c
tr
o
n
i
c
s
a
n
d
Co
n
tr
o
l
.
2016
;
1
4
(4
):
1
4
0
8
-
1416
.
[6
]
L
i
u
Y
,
Zh
o
u
S,
Che
n
Q
.
Di
s
c
ri
m
i
n
a
ti
v
e
d
e
e
p
b
e
l
i
e
f
n
e
tw
o
rk
s
fo
r
v
i
s
u
a
l
d
a
ta
c
l
a
s
s
i
fi
c
a
ti
o
n
.
Pa
tt
e
r
n
Rec
o
g
n
i
ti
o
n
.
2
0
1
1
;
4
4
(
1
0
-
1
1
)
:
2287
–
2
2
9
6
.
[7
]
Dah
l
G
E,
Y
u
D,
De
n
g
L
,
Ac
e
ro
A
.
Con
t
e
x
t
-
d
e
p
e
n
d
e
n
t
p
re
-
tra
i
n
e
d
d
e
e
p
n
e
u
r
a
l
n
e
tw
o
rk
s
fo
r
l
a
rg
e
-
v
o
c
a
b
u
l
a
ry
s
p
e
e
c
h
r
e
c
o
g
n
i
ti
o
n
.
IEEE
Tr
a
n
s
a
c
t
i
o
n
s
o
n
a
u
d
i
o
,
s
p
e
e
c
h
,
a
n
d
l
a
n
g
u
a
g
e
p
ro
c
e
s
s
i
n
g
.
2
0
1
1
;
2
0
(1
)
:
30
–
42.
[8
]
Hs
u
C
W
,
L
i
n
C
J
.
A
c
o
m
p
a
ri
s
o
n
o
f
m
e
th
o
d
s
fo
r
m
u
l
ti
c
l
a
s
s
s
u
p
p
o
rt
v
e
c
to
r
m
a
c
h
i
n
e
s
.
IEEE
tr
a
n
s
a
c
ti
o
n
s
o
n
Ne
u
ra
l
Ne
two
r
k
s
.
2
0
0
2
;
1
3
(
2
):
415
–
4
2
5
.
[9
]
T
anha
J.
E
n
s
e
m
b
l
e
a
p
p
ro
a
c
h
e
s
to
s
e
m
i
-
s
u
p
e
rv
i
s
e
d
l
e
a
r
n
i
n
g
.
Doc
to
ra
l
T
h
e
s
i
s
.
Un
i
v
e
rs
i
ty
o
f
Am
s
te
r
d
a
m
.
2013
[1
0
]
Ras
m
u
s
A
,
V
a
l
p
o
l
a
H
,
Ho
n
k
a
l
a
M
,
Be
rg
l
u
n
d
M
,
Rai
k
o
T
.
Se
m
i
-
S
u
p
e
rv
i
s
e
d
L
e
a
rn
i
n
g
w
i
th
L
a
d
d
e
r
Net
w
o
rk
s
.
Neu
ra
l
a
n
d
Ev
o
l
u
t
i
o
n
a
ry
Co
m
p
u
ti
n
g
.
2
0
1
5
[1
1
]
J
a
fa
rT
a
n
h
a
,
M
a
a
rte
n
v
a
n
So
m
e
re
n
,
Ham
i
d
e
h
Afs
a
rm
a
n
e
s
h
.
An
Ad
a
Bo
o
s
t
Al
g
o
r
i
th
m
fo
r
M
u
l
ti
c
l
a
s
s
Se
m
i
-
s
u
p
e
rv
i
s
e
d
L
e
a
r
n
i
n
g
.
IE
EE 1
2
th
I
n
te
rn
a
ti
o
n
a
l
C
o
n
fe
r
e
n
c
e
o
n
Da
ta
M
i
n
i
n
g
(I
CD
M
)
.
2012
.
[1
2
]
Y
i
n
S,
Liu
J
,
T
e
n
g
L
.
A
New
Se
m
i
-
s
u
p
e
rv
i
s
e
d
Clu
s
te
ri
n
g
A
l
g
o
ri
th
m
Ba
s
e
d
o
n
V
a
ri
a
t
i
o
n
a
l
B
a
y
e
s
i
a
n
a
n
d
It
s
Ap
p
l
i
c
a
t
i
o
n
.
TEL
KO
M
NIKA
Te
l
e
c
o
m
m
u
n
i
c
a
ti
o
n
C
o
m
p
u
ti
n
g
El
e
c
tr
o
n
i
c
s
a
n
d
Con
tr
o
l
.
2
0
1
6
;
14
(
3
)
:
1150
-
1
1
5
6
.
[1
3
]
Be
n
g
i
o
Y
.
L
e
a
rn
i
n
g
d
e
e
p
a
r
c
h
i
te
c
t
u
re
s
fo
r
AI
.
Fo
u
n
d
a
t
i
o
n
s
a
n
d
tre
n
d
s
R
i
n
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
.
2
0
0
9
;
2
(1
):
1
–
1
2
7
.
[1
4
]
W
i
b
o
w
o
H
,
F
i
rd
a
u
s
i
F
,
Su
h
a
r
s
o
W
,
Ku
s
u
m
a
WA
,
H
a
rm
a
n
t
o
D
.
Fa
c
i
a
l
e
x
p
re
s
s
i
o
n
re
c
o
g
n
i
ti
o
n
o
f
3
D
i
m
a
g
e
u
s
i
n
g
f
a
c
i
a
l
a
c
t
i
o
n
c
o
d
i
n
g
s
y
s
t
e
m
(
FACS).
TEL
K
O
M
NIKA
Te
l
e
c
o
m
m
u
n
i
c
a
t
i
o
n
Com
p
u
ti
n
g
El
e
c
tr
o
n
i
c
s
a
n
d
Co
n
tro
l
.
2
0
1
9
;
1
7
(2
)
:
628
-
636
.
[1
5
]
O
d
e
n
a
A.
Se
m
i
-
Su
p
e
rv
i
s
e
d
L
e
a
rn
i
n
g
w
i
th
G
e
n
e
ra
ti
v
e
Ad
v
e
rs
a
ri
a
l
Net
w
o
rk
s
.
Dat
a
Eff
i
c
i
e
n
t
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
wo
r
k
s
h
o
p
ICM
L
2
0
1
6
.
2016
.
a
rX
i
v
:1
6
0
6
.
0
1
5
8
3
v
2
.
[1
6
]
Za
b
i
h
i
F.
U
s
e
t
h
e
Be
l
i
e
f
n
e
two
rk
T
o
r
e
c
o
g
n
i
z
e
F
a
rs
i
n
u
m
b
e
r
s
.
Ei
g
h
t
h
c
o
n
fe
r
e
n
c
e
Ira
n
M
i
n
i
n
g
.
2015
.
[1
7
]
Va
l
p
o
l
a
H.
Fro
m
n
e
u
ra
l
PCA
t
o
d
e
e
p
u
n
s
u
p
e
rv
i
s
e
d
l
e
a
rn
i
n
g
.
Ad
v
.
i
n
In
d
e
p
e
n
d
e
n
t
Com
p
o
n
e
n
t
An
a
l
y
s
i
s
a
n
d
L
e
a
rn
i
n
g
M
a
c
h
i
n
e
s
.
2
0
1
5
:
143
–
171.
a
rX
i
v
:1
4
1
1
.7
7
8
3
.
[1
8
]
O
o
rd
AV,
Ka
l
c
h
b
r
e
n
n
e
r
N,
Ka
v
u
k
c
u
o
g
l
u
K.
Pi
x
e
l
Rec
u
rre
n
t
N
e
u
ra
l
Net
w
o
rk
s
.
I
n
te
rn
a
ti
o
n
a
l
Con
fe
re
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
.
New
Y
o
rk
.
2
0
1
6
.
a
rX
i
v
:1
6
0
1
.0
6
7
5
9
v
3
.
[1
9
]
O
d
e
n
a
A.
Se
m
i
-
Su
p
e
rv
i
s
e
d
L
e
a
rn
i
n
g
w
i
th
G
e
n
e
ra
ti
v
e
Ad
v
e
rs
a
ri
a
l
Net
w
o
rk
s
.
Dat
a
Eff
i
c
i
e
n
t
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
wo
r
k
s
h
o
p
ICM
L
2
0
1
6
.
2016
.
a
rX
i
v
:1
6
0
6
.
0
1
5
8
3
v
2
.
[2
0
]
Si
m
o
n
y
a
n
K,
Zi
s
s
e
r
m
a
n
A.
Ve
r
y
d
e
e
p
c
o
n
v
o
l
u
ti
o
n
a
l
n
e
tw
o
rk
s
fo
r
l
a
rg
e
s
c
a
l
e
i
m
a
g
e
r
e
c
o
g
n
i
t
i
o
n
.
ArX
i
v
p
re
p
ri
n
t
a
rXi
v
:
1
4
0
9
.1
5
5
6
.
2
0
1
4
.
[2
1
]
Sz
e
g
e
d
y
C,
L
i
u
W
,
J
i
a
Y
,
Se
rm
a
n
e
t
P,
Ree
d
S,
An
g
u
e
l
o
v
D,
Erh
a
n
D,
Va
n
h
o
u
c
k
e
V,
Rab
i
n
o
v
i
c
h
A.
G
o
i
n
g
d
e
e
p
e
r w
i
th
c
o
n
v
o
l
u
t
i
o
n
s
.
ArX
i
v
p
r
e
p
ri
n
t
a
rX
i
v
:
1
4
0
9
.
4
8
4
2
.
2
0
1
4
.
[2
2
]
Ngi
a
m
J
,
Che
n
Z,
K
o
h
P
W
,
Ng
AY
.
L
e
a
rn
i
n
g
d
e
e
p
e
n
e
rg
y
m
o
d
e
l
s
.
Pro
c
e
e
d
i
n
g
s
o
f
th
e
2
8
th
I
n
te
rn
a
t
i
o
n
a
l
Con
fe
re
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
rn
i
n
g
(I
CM
L
-
11)
.
2
0
1
1
:
1
1
0
5
-
1
1
1
2
.
[2
3
]
Are
l
I,
Ro
s
e
DC
,
Ka
r
n
o
w
s
k
i
T
P
.
Dee
p
m
a
c
h
i
n
e
l
e
a
r
n
i
n
g
-
a
n
e
w
fro
n
ti
e
r
i
n
a
rti
f
i
c
i
a
l
i
n
te
l
l
i
g
e
n
c
e
re
s
e
a
rc
h
[re
s
e
a
rc
h
f
ro
n
t
i
e
r].
Com
p
u
ta
t
i
o
n
a
l
I
n
te
l
l
i
g
e
n
c
e
M
a
g
a
z
i
n
e
.
2
0
1
0
;
5
(4
):
13
-
1
8
[2
4
]
Be
n
g
i
o
Y
,
Cou
rv
i
l
l
e
A,
Vi
n
c
e
n
t
P.
Rep
re
s
e
n
ta
ti
o
n
l
e
a
rn
i
n
g
:
A
re
v
i
e
w
a
n
d
n
e
w
p
e
rs
p
e
c
t
i
v
e
s
.
IEE
E
tra
n
s
a
c
t
i
o
n
s
o
n
p
a
tt
e
rn
a
n
a
l
y
s
i
s
a
n
d
m
a
c
h
i
n
e
i
n
te
l
l
i
g
e
n
c
e
.
2
0
1
3
;
3
5
(
8
):
1
7
9
8
-
828.
[2
5
]
Ba
g
h
e
ri
A,
Sa
ra
e
e
M
,
d
e
J
o
n
g
F.
S
e
n
ti
m
e
n
t
Cl
a
s
s
i
f
i
c
a
ti
o
n
i
n
Pe
r
s
i
a
n
:
I
n
tro
d
u
c
i
n
g
a
M
u
tu
a
l
In
fo
rm
a
ti
o
n
-
b
a
s
e
d
M
e
th
o
d
f
o
r
Fe
a
tu
re
S
e
l
e
c
ti
o
n
.
21
st
Ira
n
i
a
n
Con
fe
re
n
c
e
o
n
El
e
c
tr
i
c
a
l
E
n
g
i
n
e
e
ri
n
g
(ICEE
)
.
2013
:
1
-
6
.
[2
6
]
Al
e
x
a
n
d
e
r
Y
a
k
o
v
l
e
v
i
c
h
Fr
i
d
m
a
n
.
Pl
a
n
n
i
n
g
a
n
d
Coo
r
d
i
n
a
ti
o
n
i
n
Hi
e
ra
rc
h
i
e
s
o
f
In
te
l
l
i
g
e
n
t
Dy
n
a
m
i
c
Sy
s
te
m
s
.
TEL
KO
M
NIKA
Te
l
e
c
o
m
m
u
n
i
c
a
ti
o
n
Com
p
u
ti
n
g
El
e
c
tr
o
n
i
c
s
a
n
d
Con
tr
o
l
.
2
0
1
6
;
14
(
4
):
1408
-
1
4
1
6
.
[2
7
]
M
o
g
h
i
m
i
M
,
Be
l
o
n
g
i
e
SJ
,
Sa
b
e
ri
a
n
M
J
,
Y
a
n
g
J
,
Va
s
c
o
n
c
e
l
o
s
N,
L
i
L
J
.
Bo
o
s
t
e
d
Con
v
o
l
u
ti
o
n
a
l
Neu
ra
l
Net
work
s
.
2
7
t
h
Bri
t
i
s
h
M
a
c
h
i
n
e
Vi
s
i
o
n
Co
n
fe
re
n
c
e
.
2
0
1
6
:2
4
-
1.
ISBN 1
-
9
0
1
7
2
5
-
59
-
6
.
[2
8
]
As
i
l
H.
In
n
o
v
a
ti
o
n
Stra
t
e
g
i
e
s
:
Cha
l
l
e
n
g
e
s
o
r
O
p
p
o
rt
u
n
i
t
i
e
s
i
n
So
ft
w
a
re
Dev
e
l
o
p
m
e
n
t
T
e
a
m
s
.
J
o
u
rn
a
l
o
f
So
ft
ware
En
g
i
n
e
e
ri
n
g
.
2
0
1
3
;
7
(4
):
1
5
1
-
15
5.
[2
9
]
Ram
a
d
a
n
ZM
.
O
p
ti
m
u
m
I
m
a
g
e
Fi
l
te
rs
fo
r
V
a
ri
o
u
s
T
y
p
e
s
o
f
Noi
s
e
.
TEL
KO
M
NIKA
Te
l
e
c
o
m
m
u
n
i
c
a
ti
o
n
Com
p
u
ti
n
g
E
l
e
c
tro
n
i
c
s
a
n
d
Co
n
tro
l
.
2
0
1
8
;
1
6
(
5
):
2
4
5
8
-
24
64.
Evaluation Warning : The document was created with Spire.PDF for Python.