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.
8
,
No
.
1
,
Feb
r
u
ar
y
201
8
,
p
p
.
3
2
6
~
3
3
2
I
SS
N:
2088
-
8708
,
DOI
: 1
0
.
1
1
5
9
1
/
i
j
ec
e
.
v
8
i
1
.
p
p
3
2
6
-
3
3
2
326
J
o
ur
na
l ho
m
ep
a
g
e
:
h
ttp
:
//ia
e
s
co
r
e
.
co
m/
jo
u
r
n
a
ls
/in
d
ex
.
p
h
p
/
I
JE
C
E
Sy
m
bo
lic
-
Co
nnec
tionis
t
Re
presen
t
a
tiona
l Mo
del f
o
r
O
p
ti
m
i
z
ing
Decisio
n Ma
k
ing
Beha
v
io
r in
Intell
ig
ent
Sy
ste
m
s
R.
J
o
hn
M
a
rt
in
1
,
Su
j
a
t
ha
2
1
S
c
h
o
o
l
o
f
Co
m
p
u
ter S
c
ien
c
e
a
n
d
En
g
in
e
e
rin
g
,
B
h
a
ra
th
iar Un
iv
e
rsi
ty
,
Co
i
m
b
a
to
re
,
In
d
ia
2
De
p
a
rtme
n
t
o
f
Co
m
p
u
ter S
c
ien
c
e
,
Bh
a
ra
th
i
W
o
m
e
n
‟s Co
ll
e
g
e
(Au
to
n
o
m
o
u
s),
T
a
m
il
Na
d
u
6
0
0
1
0
8
,
In
d
ia
Art
icle
I
nfo
AB
ST
RAC
T
A
r
ticle
his
to
r
y:
R
ec
eiv
ed
A
p
r
2
,
2
0
1
7
R
ev
i
s
ed
J
u
l 3
0
,
2
0
1
7
A
cc
ep
ted
A
u
g
1
4
,
2
0
1
7
M
o
d
e
li
n
g
h
ig
h
e
r
o
r
d
e
r
c
o
g
n
it
iv
e
p
ro
c
e
ss
e
s li
k
e
h
u
m
a
n
d
e
c
isio
n
m
a
k
in
g
c
o
m
e
in
t
h
re
e
re
p
re
se
n
tatio
n
a
l
a
p
p
ro
a
c
h
e
s
n
a
m
e
l
y
s
y
m
b
o
li
c
,
c
o
n
n
e
c
t
io
n
ist
a
n
d
s
y
m
b
o
li
c
-
c
o
n
n
e
c
ti
o
n
ist.
M
a
n
y
c
o
n
n
e
c
ti
o
n
ist
n
e
u
ra
l
n
e
tw
o
rk
m
o
d
e
ls
a
r
e
e
v
o
lv
e
d
o
v
e
r
th
e
d
e
c
a
d
e
s
f
o
r
o
p
ti
m
izin
g
d
e
c
isio
n
m
a
k
in
g
b
e
h
a
v
io
rs
a
n
d
t
h
e
ir
a
g
e
n
ts
a
re
a
lso
in
p
lac
e
.
T
h
e
re
h
a
d
b
e
e
n
a
tt
e
m
p
ts
to
im
p
lem
e
n
t
sy
m
b
o
li
c
stru
c
tu
re
s
w
it
h
in
c
o
n
n
e
c
ti
o
n
ist
a
rc
h
it
e
c
tu
re
s
w
it
h
d
istri
b
u
ted
re
p
re
se
n
tatio
n
s.
Ou
r
w
o
rk
w
a
s
a
i
m
e
d
a
t
p
ro
p
o
si
n
g
a
n
e
n
h
a
n
c
e
d
c
o
n
n
e
c
ti
o
n
ist
a
p
p
r
o
a
c
h
o
f
o
p
ti
m
iz
in
g
th
e
d
e
c
isio
n
s
w
it
h
in
th
e
f
ra
m
e
w
o
rk
o
f
a
s
y
m
b
o
li
c
c
o
g
n
it
iv
e
m
o
d
e
l.
T
h
e
a
c
ti
o
n
se
lec
ti
o
n
m
o
d
u
le
o
f
th
is
f
ra
m
e
w
o
rk
is
f
o
re
f
ro
n
t
in
e
v
o
lv
in
g
in
telli
g
e
n
t
a
g
e
n
ts
th
r
o
u
g
h
a
v
a
riet
y
o
f
so
f
t
c
o
m
p
u
ti
n
g
m
o
d
e
ls.
A
s
a
c
o
n
ti
n
o
u
s
e
ff
o
rt,
a
Co
n
n
e
c
ti
o
n
ist
Co
g
n
it
iv
e
M
o
d
e
l
(CCN)
h
a
d
b
e
e
n
e
v
o
lv
e
d
b
y
b
rin
g
in
g
a
trad
it
io
n
a
l
s
y
m
b
o
li
c
c
o
g
n
it
iv
e
p
ro
c
e
ss
m
o
d
e
l
p
ro
p
o
se
d
b
y
L
ID
A
a
s
a
n
in
sp
iratio
n
to
a
f
e
e
d
f
o
r
w
a
rd
n
e
u
ra
l
n
e
t
w
o
rk
m
o
d
e
l
f
o
r
o
p
ti
m
izin
g
d
e
c
io
n
m
a
k
in
g
b
e
h
a
v
io
u
rs
in
in
telli
g
e
n
t
a
g
e
n
ts.
S
ig
n
if
ica
n
c
t
p
ro
g
re
ss
w
a
s o
b
se
rv
e
d
w
h
il
e
c
o
m
p
a
rin
g
it
s
p
e
rf
o
r
m
a
n
c
e
w
it
h
o
th
e
r
v
a
rien
ts.
K
ey
w
o
r
d
:
C
o
g
n
iti
v
e
p
r
o
ce
s
s
C
o
n
n
ec
tio
n
i
s
t
m
o
d
el
Dec
is
io
n
m
a
k
in
g
Ma
ch
i
n
e
lear
n
i
n
g
Neu
r
al
n
et
w
o
r
k
Co
p
y
rig
h
t
©
2
0
1
8
In
stit
u
te o
f
A
d
v
a
n
c
e
d
E
n
g
i
n
e
e
rin
g
a
n
d
S
c
ien
c
e
.
Al
l
rig
h
ts
re
se
rv
e
d
.
C
o
r
r
e
s
p
o
nd
ing
A
uth
o
r
:
R
.
J
o
h
n
Ma
r
tin
,
Sch
o
o
l o
f
C
o
m
p
u
ter
Scie
n
ce
,
B
h
ar
ath
iar
Un
iv
er
s
it
y
,
C
o
i
m
b
ato
r
e,
I
n
d
ia
.
E
m
ail:
j
m
ar
ti
n
.
in
@
g
m
ail.
co
m
1.
I
NT
RO
D
UCT
I
O
N
Op
ti
m
izi
n
g
th
e
a
g
en
t
b
ased
d
ec
is
io
n
m
a
k
i
n
g
s
y
s
te
m
s
i
s
a
ce
n
tr
e
o
f
in
te
lli
g
en
t
s
y
s
te
m
s
r
esear
ch
.
Ag
e
n
t
b
ased
d
ec
is
io
n
m
a
k
in
g
is
f
u
n
d
a
m
e
n
tall
y
d
if
f
er
e
n
t
f
r
o
m
h
u
m
an
d
ec
is
io
n
m
ak
in
g
a
n
d
m
a
k
e
d
ec
is
io
n
s
ac
co
r
d
in
g
to
p
r
o
g
r
a
m
m
ed
p
r
o
ce
d
u
r
es.
I
n
co
g
n
iti
v
e
m
o
d
el
lin
g
,
d
ec
is
io
n
m
a
k
i
n
g
is
v
ie
w
ed
as
a
h
i
g
h
le
v
el
m
en
tal
p
r
o
ce
s
s
w
h
ic
h
in
v
o
l
v
e
s
j
u
d
g
in
g
m
u
lt
ip
le
o
p
tio
n
s
in
o
r
d
er
t
o
ch
o
o
s
e
o
n
e,
s
o
as
to
f
u
l
f
il
th
e
o
b
j
ec
tiv
e
o
f
th
e
d
ec
is
io
n
m
a
k
i
n
g
a
g
en
t
[
1
]
,
[
2
]
.
T
h
is
p
ap
er
d
escr
ib
es
t
h
e
ef
f
o
r
t
o
f
m
o
u
ld
i
n
g
a
co
n
n
ec
tio
n
is
t
m
o
d
el
f
o
r
o
p
tim
izatio
n
b
y
i
n
co
r
p
o
r
atin
g
th
e
p
r
in
c
ip
les
o
f
L
I
D
A
(
L
ea
r
n
in
g
I
n
tel
lig
e
n
t
Di
s
tr
ib
u
t
i
o
n
Ag
e
n
t)
co
g
n
iti
v
e
ar
ch
itect
u
r
e
[
3
]
.
L
I
DA
f
ac
ilit
ates
th
e
co
g
n
iti
v
e
p
r
o
ce
s
s
o
f
d
ec
is
io
n
m
ak
in
g
th
r
o
u
g
h
its
„
ac
tio
n
s
elec
tio
n
‟
m
o
d
u
le
[
4
]
,
[
5
]
.
T
h
e
L
I
D
A
ac
t
io
n
s
elec
tio
n
m
o
d
u
le
is
b
ei
n
g
ad
o
p
ted
to
ev
o
lv
e
co
n
n
ec
tio
n
is
t
n
eu
r
al
ar
c
h
i
tectu
r
e
to
en
h
a
n
ce
its
co
m
p
u
tatio
n
al
ab
ilit
y
.
T
h
e
co
g
n
iti
v
el
y
m
o
ti
v
ated
co
m
p
u
tatio
n
al
co
n
n
ec
tio
n
is
t
m
o
d
el
w
il
l
b
e
u
s
ed
f
o
r
v
ar
io
u
s
o
p
ti
m
izatio
n
p
r
o
b
le
m
s
w
it
h
co
n
s
id
er
ab
l
y
g
o
o
d
im
p
ac
t
i
n
it
s
d
ec
is
io
n
m
a
k
i
n
g
ag
en
t
s
.
E
v
en
th
o
u
g
h
co
g
n
iti
v
e
m
o
d
elli
n
g
(
s
y
m
b
o
li
c)
an
d
n
eu
r
al
m
o
d
elli
n
g
(
c
o
n
n
ec
tio
n
is
t)
u
s
e
t
w
o
d
i
f
f
er
en
t
ap
p
r
o
ac
h
es
in
b
u
ild
in
g
i
n
telli
g
e
n
t
ag
e
n
t
s
,
th
er
e
ar
e
r
ec
en
t
d
ev
elo
p
m
e
n
ts
i
n
co
m
b
i
n
i
n
g
b
o
th
ap
p
r
o
ac
h
es
(
s
y
m
b
o
lic
-
co
n
n
ec
tio
n
is
t)
to
r
ein
f
o
r
ce
th
e
f
ield
o
f
co
g
n
iti
v
e
ag
e
n
t b
u
ild
i
n
g
[
6
-
8
]
.
T
h
e
f
o
llo
w
i
n
g
s
ec
t
io
n
s
o
f
t
h
is
p
ap
er
d
escr
ib
e
th
e
co
n
ce
p
ts
an
d
p
r
in
cip
les
b
e
h
in
d
t
h
e
co
m
p
u
tat
io
n
al
m
et
h
o
d
s
to
b
r
in
g
in
g
o
u
t
an
e
n
h
a
n
ce
d
C
o
n
n
ec
tio
n
i
s
t
C
o
g
n
i
tiv
e
Net
w
o
r
k
(
C
C
N)
m
o
d
el
f
o
r
o
p
tim
izi
n
g
r
ea
l
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:
2088
-
8708
S
ymb
o
lic
-
C
o
n
n
ec
tio
n
is
t R
ep
r
esen
ta
tio
n
a
l Mo
d
el
fo
r
Op
timiz
in
g
Dec
is
io
n
Ma
kin
g
…
(
R
.
Jo
h
n
Ma
r
tin
)
327
ti
m
e
d
ec
is
io
n
m
a
k
i
n
g
p
r
o
b
lem
s
.
T
h
i
s
w
il
l
b
e
a
m
o
ti
v
atio
n
f
o
r
th
e
p
r
o
s
p
ec
tiv
e
r
esear
c
h
er
s
to
ap
p
r
o
ac
h
t
h
e
n
e
w
d
i
m
e
n
tio
n
o
f
co
g
n
iti
v
e
m
o
d
el
b
u
ild
in
g
.
2.
ACTI
O
N
SE
L
E
CT
I
O
N
M
O
DUL
E
O
F
L
I
DA
T
h
e
L
I
D
A
co
g
n
iti
v
e
ar
ch
itec
t
u
r
e
ex
te
n
s
i
v
el
y
d
ef
in
e
s
th
e
h
i
g
h
er
o
r
d
er
co
g
n
iti
v
e
p
r
o
ce
s
s
o
f
d
ec
is
io
n
m
ak
in
g
w
it
h
it
s
ac
tio
n
s
elec
t
io
n
m
o
d
u
le.
T
h
e
ac
tio
n
s
elec
tio
n
p
h
ase
is
b
asical
l
y
a
lear
n
in
g
p
h
ase
w
h
er
e
s
ev
er
al
p
r
o
ce
s
s
es
o
p
er
ate
i
n
p
ar
allel
in
t
h
e
co
m
p
lete
co
g
n
itiv
e
c
y
cle
o
f
L
I
D
A
(
Fig
u
r
e
1
)
f
o
r
th
e
d
ec
is
io
n
w
h
ic
h
w
il
l
lead
to
t
h
e
n
ex
t
c
y
cle.
Var
io
u
s
as
s
o
ciatio
n
s
f
o
r
m
ed
w
i
th
i
n
t
h
e
e
n
ti
ties
o
f
its
ar
c
h
itect
u
r
e
b
y
r
ein
f
o
r
ci
n
g
o
ld
s
elec
tio
n
s
o
cc
u
r
as
t
h
e
co
n
s
cio
u
s
b
r
o
ad
ca
s
t
r
ea
ch
es
P
er
ce
p
tu
al
A
s
s
o
ciati
v
e
Me
m
o
r
y
.
Ne
w
ev
en
t
s
f
r
o
m
t
h
e
co
n
s
cio
u
s
b
r
o
ad
ca
s
t
ar
e
en
co
d
ed
as
k
n
o
w
led
g
e
p
atter
n
s
i
n
T
r
an
s
ie
n
t
E
p
is
o
d
ic
Me
m
o
r
y
.
P
o
ten
tial
ac
tio
n
p
atter
n
s
,
to
g
eth
er
w
it
h
t
h
eir
co
n
te
x
ts
a
n
d
ex
p
ec
ted
r
esu
lts
,
ar
e
lear
n
ed
in
to
P
r
o
ce
d
u
r
al
Me
m
o
r
y
f
r
o
m
t
h
e
co
n
s
cio
u
s
b
r
o
a
d
ca
s
t.
T
h
is
is
m
o
r
e
s
i
m
ilar
to
th
e
tr
ain
in
g
p
atter
n
o
f
th
e
f
ee
d
f
o
r
w
ar
d
n
et
w
o
r
k
s
o
f
lear
n
i
n
g
.
A
c
t
i
o
n
S
e
l
e
c
t
i
o
n
(
B
e
h
a
v
i
o
r
N
e
t
)
S
e
n
s
o
r
-
M
o
t
o
r
M
e
m
o
r
y
P
r
o
c
e
d
u
r
a
l
M
e
m
o
r
y
(
S
c
h
e
m
a
N
e
t
)
A
c
t
i
o
n
S
e
l
e
c
t
e
d
P
e
r
c
e
p
t
u
a
l
A
s
s
o
c
i
a
t
i
v
e
M
e
m
o
r
y
(
S
l
i
p
N
e
t
)
I
n
s
t
a
n
t
i
a
t
e
S
c
h
e
m
a
s
Fig
u
r
e
1
.
A
ctio
n
-
s
elec
tio
n
m
o
d
u
le
o
f
L
I
D
A
co
g
n
iti
v
e
ar
ch
it
ec
tu
r
e
A
lo
n
g
s
id
e
o
f
t
h
is
lear
n
i
n
g
,
u
s
in
g
th
e
co
n
s
cio
u
s
co
n
te
n
t
s
,
p
o
s
s
ib
le
s
c
h
e
m
a
s
f
o
r
t
h
e
ac
tio
n
b
eh
a
v
io
r
ar
e
ev
o
lv
ed
f
r
o
m
t
h
e
P
r
o
ce
d
u
r
al
Me
m
o
r
y
.
Sa
m
e
ac
tio
n
p
atte
r
n
is
s
e
n
t
to
A
ctio
n
Selectio
n
,
w
h
er
e
it
c
o
m
p
ete
s
to
b
e
th
e
b
eh
a
v
io
r
s
elec
ted
f
o
r
th
i
s
co
g
n
iti
v
e
c
y
cle.
T
h
e
s
ele
cted
b
eh
av
io
r
tr
ig
g
er
s
Sen
s
o
r
y
-
Mo
to
r
Me
m
o
r
y
to
p
r
o
d
u
ce
a
s
u
itab
le
m
o
to
r
p
lan
f
o
r
th
e
b
e
h
av
io
r
p
atter
n
to
b
e
ca
r
r
ied
o
u
t
[
9
]
.
T
h
is
p
ar
t
o
f
t
h
e
co
g
n
it
iv
e
c
y
cle
is
th
e
m
o
ti
v
atio
n
f
o
r
th
e
p
r
o
p
o
s
ed
co
n
n
ec
tio
n
i
s
t a
p
p
r
o
ac
h
o
f
d
ec
is
io
n
m
ak
in
g
.
3.
SYM
B
O
L
I
C
RE
P
RE
S
E
N
T
AT
I
O
NA
L
M
O
DE
L
O
F
DE
CIS
I
O
N
M
AK
I
NG
Mo
d
elin
g
o
f
h
u
m
a
n
d
ec
is
io
n
-
m
ak
in
g
is
q
u
ite
ch
al
len
g
i
n
g
e
s
p
ec
iall
y
u
n
d
er
a
co
m
p
lex
a
n
d
u
n
ce
r
tai
n
en
v
ir
o
n
m
e
n
t.
E
th
ical
d
ec
is
io
n
s
ar
e
a
m
o
n
g
t
h
e
m
o
r
e
co
m
p
lex
d
ec
is
io
n
s
t
h
at
ag
e
n
t
s
f
a
ce
.
E
th
ical
d
ec
is
io
n
m
ak
in
g
ca
n
b
e
u
n
d
er
s
to
o
d
as a
ctio
n
s
elec
tio
n
u
n
d
er
co
n
d
iti
o
n
s
w
h
e
r
e
co
n
s
tr
ai
n
ts
,
p
r
in
cip
l
es a
n
d
v
al
u
es p
la
y
a
ce
n
tr
al
r
o
le
in
d
eter
m
i
n
i
n
g
w
h
ich
b
eh
a
v
io
r
al
attitu
d
e
s
an
d
r
esp
o
n
s
es a
r
e
ac
ce
p
tab
le
[
1
0
]
.
Ma
n
y
d
ec
is
io
n
s
r
eq
u
ir
e
h
a
v
in
g
to
s
elec
t
an
ac
tio
n
w
h
e
n
in
f
o
r
m
atio
n
is
u
n
clea
r
,
in
co
m
p
lete,
co
n
f
u
s
i
n
g
,
an
d
ev
e
n
f
a
ls
e,
wh
er
e
th
e
p
o
s
s
ib
le
r
es
u
lt
s
o
f
a
n
ac
tio
n
ca
n
n
o
t
b
e
p
r
ed
icted
w
it
h
a
n
y
s
ig
n
i
f
ica
n
t
d
eg
r
ee
o
f
ce
r
tain
t
y
,
an
d
w
h
er
e
co
n
f
licti
n
g
v
al
u
es
ca
n
i
n
f
o
r
m
th
e
d
ec
is
io
n
-
m
a
k
i
n
g
p
r
o
ce
s
s
.
I
n
o
r
d
er
to
m
ak
e
th
e
p
r
o
ce
s
s
o
f
ac
tio
n
s
e
lectio
n
to
b
e
u
n
f
u
s
s
y
b
y
an
ag
e
n
t,
a
b
o
tto
m
-
up
ap
p
r
o
ac
h
h
a
s
b
ee
n
d
er
iv
ed
f
r
o
m
ac
tio
n
s
elec
tio
n
p
h
a
s
e
o
f
L
I
D
A
co
g
n
itiv
e
ar
ch
i
tectu
r
e
[
1
1
]
.
T
h
e
Fig
u
r
e
2
is
an
e
x
p
an
s
io
n
o
f
ac
ti
o
n
s
elec
tio
n
m
o
d
u
l
e
d
escr
ib
ed
in
th
e
p
r
ev
io
u
s
s
ec
ti
o
n
an
d
is
v
ie
w
ed
as a
s
y
m
b
o
li
c
r
ep
r
esen
tatio
n
al
m
o
d
el.
T
h
is
s
y
m
b
o
lic
r
ep
r
ese
n
tatio
n
al
m
o
d
el
h
as
th
r
ee
la
y
er
s
f
o
r
o
p
tim
izin
g
t
h
e
d
ec
is
io
n
w
h
i
ch
w
i
ll
b
e
d
er
iv
ed
as
a
n
e
w
s
c
h
e
m
a.
T
h
e
f
ir
s
t
la
y
er
is
s
ch
ema
n
et
th
at
r
ec
eiv
es
i
n
p
u
t
f
r
o
m
t
h
e
p
r
o
ce
d
u
r
al
m
e
m
o
r
y
,
w
h
ic
h
is
v
er
y
s
i
m
ilar
to
th
e
in
p
u
t
la
y
er
o
f
th
e
f
ee
d
f
o
r
w
ar
d
co
n
n
ec
tio
n
is
t
n
et
w
o
r
k
ar
ch
i
tectu
r
e.
T
h
e
h
id
d
en
la
y
er
o
f
th
e
f
ee
d
f
o
r
w
ar
d
n
et
is
as
s
o
ci
ated
w
it
h
s
lip
n
et
in
t
h
e
co
g
n
it
iv
e
m
o
d
el
w
h
ich
is
u
s
ed
to
ev
alu
ate
th
e
ad
eq
u
ac
y
o
f
alter
n
at
iv
e
s
f
o
r
t
h
e
ac
tio
n
s
elec
tio
n
p
r
o
ce
s
s
.
T
h
e
o
u
tp
u
t
l
a
y
er
o
f
t
h
e
n
eu
r
al
n
et
i
s
m
o
ld
ed
w
it
h
t
h
ir
d
p
a
r
t
o
f
th
e
co
g
n
i
tiv
e
m
o
d
el
i.e
,
b
eh
a
vio
r
n
et
.
T
h
e
n
e
w
s
elec
tio
n
(
n
e
w
s
c
h
e
m
a)
i
s
f
ed
b
ac
k
to
th
e
s
e
n
s
o
r
m
o
to
r
m
e
m
o
r
y
f
o
r
lear
n
i
n
g
.
T
h
i
s
c
o
n
ce
p
tu
al
f
r
a
m
e
w
o
r
k
o
f
co
g
n
iti
v
e
m
o
d
el
is
t
h
e
m
o
ti
v
ati
o
n
al
f
ac
to
r
to
th
e
e
m
er
g
e
n
ce
o
f
en
h
a
n
ce
d
s
y
m
b
o
lic
-
co
n
n
ec
t
io
n
is
t
n
et
w
o
r
k
m
o
d
el
f
o
r
o
p
ti
m
izi
n
g
d
ec
is
io
n
s
in
i
n
telli
g
en
t
s
y
s
te
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.
8
,
No
.
1
,
Feb
r
u
ar
y
201
8
:
3
2
6
–
3
3
2
328
4.
B
ACK
P
RO
P
AG
AT
I
O
N
NE
T
WO
RK
AS A
CO
NN
E
C
T
I
O
NIS
T
M
O
DE
L
I
n
co
n
n
ec
tio
n
is
t
n
eu
r
al
n
et
w
o
r
k
,
th
e
m
o
s
t
p
o
p
u
lar
m
et
h
o
d
o
f
lear
n
in
g
is
ca
lled
B
ac
k
p
r
o
p
ag
atio
n
(
B
P
N)
[
1
2
]
.
L
ea
r
n
in
g
i
n
f
ee
d
-
f
o
r
w
ar
d
n
e
t
w
o
r
k
s
u
s
u
a
l
l
y
s
u
p
er
v
is
ed
lear
n
i
n
g
,
m
ea
n
i
n
g
t
h
at
t
h
e
n
et
w
o
r
k
is
tr
ain
ed
b
y
p
r
o
v
id
in
g
it
w
it
h
in
p
u
t
a
n
d
m
atc
h
i
n
g
o
u
tp
u
t
p
atter
n
s
.
T
h
e
co
m
p
u
tatio
n
al
ab
ilit
y
o
f
B
P
N
in
o
p
tim
izin
g
d
iv
er
s
i
f
ied
d
ec
i
s
io
n
m
a
k
in
g
s
ce
n
ar
io
s
h
as
a
ls
o
p
av
ed
w
a
y
f
o
r
i
m
p
le
m
e
n
ti
n
g
c
o
g
n
iti
v
e
m
o
d
els
o
f
d
ec
is
io
n
m
a
k
i
n
g
[
1
3
-
1
5
]
.
S
c
h
e
m
a
(
C
o
n
t
e
x
t
,
A
c
t
i
o
n
&
R
e
s
u
l
t
s
)
I
d
e
n
t
i
f
y
D
e
c
i
s
i
o
n
G
o
a
l
S
e
a
r
c
h
A
l
t
e
r
n
a
t
i
v
e
s
A
S
e
a
r
c
h
C
r
i
t
e
r
i
a
C
Q
u
a
n
t
i
f
y
A
Q
u
a
n
t
i
f
y
C
E
v
a
l
u
a
t
e
A
d
e
q
u
a
c
y
o
f
A
E
v
a
l
u
a
t
e
A
d
e
q
u
a
c
y
o
f
C
S
e
l
e
c
t
D
e
c
i
s
i
o
n
d
E
v
a
l
u
a
t
e
S
a
t
i
s
f
a
c
t
i
o
n
o
f
d
N
e
w
S
c
h
e
m
a
S
e
n
s
o
r
y
M
o
t
o
r
M
e
m
o
r
y
S
c
h
e
m
a
N
e
t
(
P
r
o
c
e
d
u
r
a
l
M
e
m
o
r
y
)
S
l
i
p
N
e
t
(
P
e
r
c
e
p
t
u
a
l
A
s
s
o
c
i
a
t
i
v
e
M
e
m
o
r
y
)
B
e
h
a
v
i
o
u
r
N
e
t
(
A
c
t
i
o
n
S
e
l
e
c
t
i
o
n
)
E
n
v
i
r
o
n
m
e
n
t
N
N
Y
N
Y
Fig
u
r
e
2
.
S
y
m
b
o
lic
r
ep
r
esen
ta
tio
n
al
m
o
d
el
o
f
ac
tio
n
-
s
elec
tio
n
T
h
e
s
tan
d
ar
d
f
o
r
m
o
f
b
ac
k
p
r
o
p
ag
atio
n
al
g
o
r
ith
m
is
ill
u
s
tr
a
ted
in
Fi
g
u
r
e
3
.
I
ts
co
m
p
u
ta
tio
n
al
p
h
a
s
e
s
ar
e
b
ein
g
p
r
esen
ted
h
er
e
f
o
r
m
o
r
e
clar
ity
.
Y
1
Z
1
X
1
Y
k
Z
j
X
i
Y
m
Z
p
X
n
I
n
p
u
t
H
i
d
d
e
n
O
u
t
p
u
t
V
1
1
V
1
j
V
1
p
V
i
1
V
i
j
V
i
p
V
n
1
V
n
j
V
n
p
W
1
k
W
1
m
W
j
1
W
j
k
W
j
m
W
p
1
W
p
k
W
p
m
W
1
1
1
1
V
o
1
V
o
j
V
o
p
W
o
1
W
o
k
W
o
m
Fig
u
r
e
3
.
Mu
ltil
a
y
er
f
ee
d
f
o
r
war
d
n
eu
r
al
Net
w
o
r
k
m
o
d
el
T
h
e
to
p
o
lo
g
y
co
n
s
is
t
s
o
f
th
r
ee
la
y
er
s
o
f
u
n
its
n
a
m
el
y
in
p
u
t,
h
id
d
en
a
n
d
o
u
tp
u
t
la
y
er
s
.
T
h
e
co
n
n
ec
tio
n
s
b
et
w
ee
n
th
e
la
y
e
r
s
f
ee
d
in
g
t
h
eir
in
p
u
ts
to
t
h
e
o
th
er
la
y
er
w
it
h
w
ei
g
h
ted
c
o
n
n
ec
tio
n
s
.
A
ll
t
h
e
w
ei
g
h
ts
a
r
e
ch
a
n
g
ea
b
le
in
s
u
c
h
a
w
a
y
to
lear
n
th
e
n
et
w
o
r
k
t
o
o
b
tain
th
e
d
esire
d
g
o
al.
T
h
e
n
et
w
o
r
k
lear
n
s
b
y
ad
j
u
s
tin
g
its
w
ei
g
h
ts
u
n
til to
f
i
n
d
a
s
et
o
f
w
ei
g
h
t
s
t
h
at
p
r
o
d
u
ce
th
e
co
r
r
ec
t o
u
tp
u
t f
o
r
ev
er
y
f
u
r
th
er
i
n
p
u
t.
On
co
n
s
id
er
in
g
a
f
ee
d
f
o
r
w
ar
d
n
et
w
o
r
k
w
i
th
n
in
p
u
t
(
i)
u
n
its
,
p
h
id
d
en
(
j
)
u
n
it
s
a
n
d
m
o
u
tp
u
t
(
k
)
u
n
i
ts
,
t
h
e
b
ac
k
p
r
o
p
ag
atio
n
tr
ain
i
n
g
w
i
ll b
e
p
er
f
o
r
m
ed
in
t
w
o
p
h
ases
n
a
m
el
y
f
o
r
w
ar
d
p
ass
a
n
d
b
ac
k
w
ar
d
p
ass
.
4
.
1
.
F
o
rwa
rd
P
a
s
s
On
e
o
f
t
h
e
s
et
o
f
n
tr
ain
in
g
i
n
p
u
t p
atter
n
s
i
s
ap
p
lied
to
th
e
in
p
u
t la
y
er
.
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:
2088
-
8708
S
ymb
o
lic
-
C
o
n
n
ec
tio
n
is
t R
ep
r
esen
ta
tio
n
a
l Mo
d
el
fo
r
Op
timiz
in
g
Dec
is
io
n
Ma
kin
g
…
(
R
.
Jo
h
n
Ma
r
tin
)
329
x
p
=(
x
p1
,
x
p2,
... x
pn
)
w
h
ic
h
m
a
y
b
e
a
b
in
ar
y
o
r
r
ea
l
-
v
ec
to
r
.
T
h
e
ac
tiv
atio
n
s
o
f
th
e
n
eu
r
o
n
s
in
th
e
h
id
d
en
la
y
er
ar
e
ca
lcu
l
ated
b
y
u
s
i
n
g
th
e
s
u
m
o
f
t
h
e
ac
tiv
atio
n
s
o
f
th
e
n
u
m
b
er
ed
in
p
u
t
la
y
er
u
n
its
t
h
e
y
ar
e
co
n
n
ec
ted
to
m
u
ltip
l
y
b
y
th
eir
r
esp
ec
tiv
e
co
n
n
ec
tio
n
w
ei
g
h
ts
a
n
d
p
ass
in
g
it t
h
r
o
u
g
h
a
tr
an
s
f
er
f
u
n
ctio
n
.
Net
in
p
u
t to
h
id
d
en
la
y
er
u
n
it
j
is
ca
lcu
lated
as
x
w
=
n
e
t
i
ji
n
1
=
i
j
T
h
e
h
id
d
en
la
y
er
o
u
tp
u
t,
i.e
.
ta
k
e
n
et
i
n
p
u
t o
f
u
n
it j
an
d
p
ass
it
th
r
o
u
g
h
a
tr
a
n
s
f
er
f
u
n
c
tio
n
)
n
e
t
(
=
oh
j
j
T
h
e
ac
tiv
atio
n
s
o
f
th
e
h
id
d
en
la
y
er
u
n
i
ts
ca
lc
u
lated
ar
e
t
h
e
n
u
s
ed
i
n
u
p
d
ati
n
g
th
e
ac
ti
v
at
io
n
o
f
t
h
e
o
u
tp
u
t u
n
it
s
.
Net
i
n
p
u
t to
o
u
tp
u
t u
n
it
k
is
ca
lc
u
lated
as
oh
w
=
n
e
t
j
kj
L
1
=
j
k
T
h
e
tr
an
s
f
er
f
u
n
ctio
n
f
o
r
th
e
o
u
tp
u
t o
f
o
u
tp
u
t
u
n
it k
)
n
e
t
(
=
oo
k
k
4
.
2
.
B
a
ck
w
a
rd
P
a
s
s
T
h
e
er
r
o
r
s
ig
n
al
f
o
r
ea
ch
in
p
u
t
is
ca
lcu
lated
b
y
u
s
i
n
g
th
e
d
if
f
er
en
ce
b
et
w
ee
n
t
h
e
ac
tu
al
ac
t
iv
atio
n
o
f
ea
ch
o
u
tp
u
t
u
n
it (
o
o
k)
an
d
t
h
e
d
esire
d
tar
g
et
ac
tiv
atio
n
(
d
k
)
f
o
r
th
at
u
n
it.
)
oo
-
d
(
k
k
Delta
(
)
f
o
r
ea
ch
o
u
tp
u
t
u
n
i
t is
th
en
ca
lc
u
lated
.
)
oo
-
(1
oo
)
oo
-
d
(
=
o
k
k
k
k
k
T
h
e
er
r
o
r
s
ig
n
als
f
o
r
th
e
h
id
d
en
la
y
er
u
n
it
s
ar
e
th
e
n
ca
lcu
lat
ed
b
y
tak
i
n
g
th
e
s
u
m
o
f
t
h
e
d
eltas
o
f
th
e
o
u
tp
u
t
u
n
its
a
p
ar
ticu
lar
h
id
d
en
u
n
it
co
n
n
ec
ts
to
m
u
ltip
l
y
b
y
t
h
e
w
ei
g
h
t
t
h
at
co
n
n
ec
t
s
th
e
h
id
d
en
a
n
d
o
u
tp
u
t
u
n
i
t.
E
r
r
o
r
s
ig
n
al
f
o
r
ea
ch
h
id
d
en
u
n
it j
w
o
kj
k
W
1
=
k
Delta
ter
m
f
o
r
ea
ch
h
id
d
en
u
n
it
j
i
s
eq
u
a
l
to
i
ts
er
r
o
r
s
ig
n
a
l
m
u
lt
ip
lied
b
y
its
o
u
tp
u
t,
m
u
ltip
lied
b
y
(
1
-
its
o
u
tp
u
t)
.
w
o
)
oh
-
)(1
oh
(
=
h
kj
k
W
1
=
k
j
j
j
T
h
e
w
ei
g
h
t
er
r
o
r
d
er
iv
ativ
es
f
o
r
ea
ch
w
e
ig
h
t
b
et
w
ee
n
t
h
e
h
i
d
d
en
an
d
o
u
tp
u
t
u
n
it
s
ar
e
ca
lc
u
lated
b
y
tak
i
n
g
t
h
e
d
elta
o
f
ea
ch
o
u
tp
u
t
u
n
it
a
n
d
m
u
ltip
l
y
i
n
g
it
b
y
th
e
ac
tiv
a
tio
n
o
f
t
h
e
h
id
d
en
u
n
i
t
it
co
n
n
ec
ts
to
.
T
h
ese
w
ei
g
h
t e
r
r
o
r
d
er
iv
ativ
es
ar
e
th
en
u
s
ed
to
ch
an
g
e
t
h
e
weig
h
ts
b
et
w
ee
n
th
e
h
id
d
en
an
d
o
u
tp
u
t la
y
er
s
.
)
oh
(
o
=
w
e
d
j
k
jk
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.
8
,
No
.
1
,
Feb
r
u
ar
y
201
8
:
3
2
6
–
3
3
2
330
T
h
e
w
ei
g
h
t
er
r
o
r
d
er
iv
ativ
es
f
o
r
ea
ch
w
e
ig
h
t b
et
w
ee
n
t
h
e
i
n
p
u
t u
n
it i
an
d
h
id
d
en
u
n
it
j
ar
e
ca
lcu
lated
b
y
ta
k
i
n
g
t
h
e
d
elta
o
f
ea
ch
h
id
d
en
u
n
i
t
an
d
m
u
ltip
l
y
in
g
it
b
y
th
e
ac
tiv
a
tio
n
o
f
t
h
e
in
p
u
t
u
n
it
it
co
n
n
ec
ts
to
(
i.e
.
th
at
i
n
p
u
t
p
atter
n
x
i
)
.
T
h
ese
w
ei
g
h
t
er
r
o
r
d
er
iv
ati
v
es
ar
e
t
h
en
u
s
ed
to
c
h
an
g
e
t
h
e
w
eig
h
ts
b
et
w
ee
n
t
h
e
i
n
p
u
t
an
d
h
id
d
en
la
y
er
s
.
)
x
(
h
=
w
e
d
i
j
ij
T
o
ch
an
g
e
th
e
ac
t
u
al
w
e
ig
h
t
s
t
h
e
m
s
el
v
es,
a
lear
n
in
g
r
at
e
co
ef
f
icie
n
t
y
eta
(
)
is
u
s
ed
,
w
h
ic
h
co
n
tr
o
ls
th
e
a
m
o
u
n
t
t
h
e
w
e
i
g
h
ts
ar
e
u
p
d
ated
d
u
r
in
g
ea
ch
b
ac
k
p
r
o
p
ag
atio
n
c
y
cle.
T
h
e
w
eig
h
ts
at
a
ti
m
e
(
t
+
1
)
b
et
w
ee
n
t
h
e
h
id
d
en
a
n
d
o
u
tp
u
t
la
y
er
s
ar
e
s
et
u
s
in
g
t
h
e
weig
h
ts
at
a
ti
m
e
an
d
t
h
e
w
ei
g
h
t
er
r
o
r
d
er
iv
ativ
e
s
b
et
w
ee
n
th
e
h
id
d
en
an
d
o
u
tp
u
t
la
y
er
s
u
s
i
n
g
t
h
e
f
o
llo
w
i
n
g
eq
u
atio
n
.
)
w
e
d
(
+
(
t
)
w
=
1)
+
(t
w
jk
jk
jk
I
n
a
s
i
m
ilar
w
a
y
t
h
e
w
ei
g
h
ts
a
r
e
ch
an
g
ed
b
et
w
ee
n
th
e
i
n
p
u
t a
n
d
h
id
d
en
u
n
it
s
)
w
e
d
(
+
(
t
)
w
=
1)
+
(t
w
ij
ij
ij
T
h
e
s
i
m
i
lar
ities
b
et
w
ee
n
th
e
a
ctio
n
s
elec
tio
n
m
o
d
el
o
f
L
I
DA
co
g
n
iti
v
e
c
y
c
le
a
n
d
th
e
B
P
N
alg
o
r
it
h
m
is
th
e
m
o
ti
v
ati
n
g
f
ac
to
r
f
o
r
m
o
d
ellin
g
o
f
th
e
e
n
h
an
ce
d
C
o
n
n
e
ctio
n
is
t
C
o
g
n
iti
v
e
Net
w
o
r
k
(
C
C
N)
.
I
n
B
P
N,
ea
ch
u
n
it
in
t
h
e
n
et
w
o
r
k
r
ec
eiv
e
s
a
n
er
r
o
r
s
i
g
n
al
th
at
d
e
s
cr
ib
es
it
s
r
elati
v
e
co
n
t
r
ib
u
tio
n
to
th
e
to
tal
er
r
o
r
b
et
w
ee
n
t
h
e
ac
t
u
al
o
u
tp
u
t
a
n
d
th
e
tar
g
et
o
u
tp
u
t.
B
ased
o
n
t
h
e
er
r
o
r
s
i
g
n
a
l
r
ec
eiv
ed
,
th
e
w
eig
h
t
s
co
n
n
ec
ti
n
g
t
h
e
u
n
its
in
d
if
f
er
en
t
la
y
er
s
ar
e
u
p
d
ated
.
T
h
e
tw
o
p
as
s
es
ar
e
r
ep
ea
ted
s
e
v
er
al
ti
m
es
f
o
r
d
is
ti
n
ct
in
p
u
t p
atter
n
s
a
n
d
t
h
eir
tar
g
et
s
,
u
n
t
il t
h
e
er
r
o
r
b
et
w
ee
n
th
e
ac
tu
al
o
u
tp
u
t o
f
t
h
e
n
et
w
o
r
k
a
n
d
its
tar
g
et
o
u
tp
u
t i
s
co
n
v
i
n
c
i
n
g
l
y
s
m
al
l f
o
r
all
th
e
u
n
i
ts
o
f
t
h
e
s
et
o
f
tr
ain
i
n
g
i
n
p
u
ts
.
Feed
f
o
r
w
ar
d
co
n
n
ec
tio
n
i
s
t n
et
w
o
r
k
s
li
k
el
y
to
d
ev
elo
p
in
ter
n
al
r
elatio
n
s
h
ip
s
b
et
w
ee
n
u
n
it
s
s
o
as
to
o
r
g
an
ize
th
e
tr
ain
in
g
d
ata,
as
a
r
es
u
lt
th
e
y
d
e
v
elo
p
an
in
ter
n
a
l
r
ep
r
esen
tatio
n
t
h
at
e
n
ab
les
th
e
m
to
g
e
n
er
ate
t
h
e
d
esire
d
o
u
tp
u
t
s
w
h
en
g
i
v
e
n
t
h
e
tr
ain
i
n
g
in
p
u
ts
.
T
h
e
s
a
m
e
in
ter
n
al
r
ep
r
ese
n
tatio
n
c
an
b
e
ap
p
lied
to
in
p
u
ts
th
a
t
wer
e
n
o
t
u
s
ed
d
u
r
i
n
g
tr
ain
in
g
;
t
h
e
n
e
w
i
n
p
u
t
s
ar
e
class
i
f
ied
b
y
t
h
e
n
et
w
o
r
k
ac
co
r
d
in
g
to
th
e
f
ea
t
u
r
es th
e
y
s
h
ar
e
w
it
h
t
h
e
tr
ai
n
i
n
g
in
p
u
ts
.
5.
CO
NNEC
T
I
O
N
I
S
T
CO
G
N
I
T
I
V
E
NE
T
WO
RK
(
CCN)
M
O
DE
L
C
o
n
n
ec
tio
n
i
s
t
C
o
g
n
i
tiv
e
Ne
t
w
o
r
k
(
C
C
N)
is
a
co
n
ce
p
t
u
al
f
r
a
m
e
w
o
r
k
w
h
ic
h
co
m
b
in
es
t
h
e
f
u
n
ctio
n
alitie
s
o
f
L
I
D
A
b
as
ed
ac
tio
n
s
elec
tio
n
m
o
d
el
an
d
f
ee
d
f
o
r
w
ar
d
n
e
u
r
al
n
e
t
w
o
r
k
m
o
d
el.
T
h
e
ar
ch
itect
u
r
al
s
i
m
i
lar
itie
s
o
f
b
o
th
m
o
d
el
s
ar
e
illu
s
tr
ated
in
t
h
e
C
C
N
ar
c
h
itect
u
r
e
(
Fig
u
r
e
4
)
.
Si
m
ilar
k
in
d
s
o
f
f
r
a
m
e
w
o
r
k
s
h
a
v
e
d
o
n
e
w
i
th
t
h
e
co
g
n
iti
v
e
m
o
d
el
o
f
ac
tio
n
s
elec
tio
n
p
r
o
p
o
s
ed
b
y
L
I
D
A
.
T
h
er
e
ar
e
r
esear
ch
o
u
tco
m
es
t
h
at
s
h
o
w
t
h
e
i
m
p
o
r
tan
ce
o
f
b
r
in
g
i
n
g
co
g
n
iti
v
e
a
n
d
co
n
n
ec
tio
n
is
t
ap
p
r
o
ac
h
es
t
o
g
eth
er
to
p
r
o
d
u
ce
o
u
t p
er
f
o
r
m
ed
in
tel
lig
e
n
t a
g
en
ts
[
1
6
]
.
T
h
e
au
th
o
r
s
h
o
p
e
t
h
at
t
h
i
s
f
r
am
e
w
o
r
k
w
ill
e
li
m
in
ate
t
h
e
li
m
itatio
n
s
o
f
a
B
P
N
th
r
o
u
g
h
t
h
e
ad
o
p
tio
n
o
f
co
g
n
it
iv
e
m
o
d
el
b
y
r
ed
u
c
in
g
n
u
m
b
er
o
f
lear
n
i
n
g
s
tep
s
,
s
o
th
at
th
e
co
m
p
u
tin
g
ti
m
e
w
ill
al
s
o
b
e
less
co
m
p
ar
ed
w
it
h
p
u
r
e
B
P
N.
T
h
e
m
aj
o
r
im
p
r
o
v
e
m
e
n
t
i
s
ac
h
iev
ed
b
y
e
n
h
a
n
ci
n
g
th
e
m
o
m
en
tu
m
f
ac
to
r
o
n
u
p
d
atin
g
w
eig
h
t
s
.
Du
r
in
g
ac
tio
n
s
elec
tio
n
p
h
ase
o
f
L
I
D
A
co
g
n
iti
v
e
c
y
cle,
ea
c
h
co
g
n
it
iv
e
c
y
cle
w
ill
p
r
o
d
u
ce
th
e
n
e
w
s
ch
e
m
a
w
h
ich
w
o
u
ld
b
e
b
ased
o
n
th
e
p
r
ev
io
u
s
s
c
h
e
m
a
s
g
e
n
er
ated
in
p
r
ev
io
u
s
c
y
c
les.
T
h
e
p
r
ev
io
u
s
s
c
h
e
m
a
s
ar
e
b
ein
g
r
ef
er
r
ed
f
r
o
m
p
er
ce
p
tu
al
ass
o
ciati
v
e
m
e
m
o
r
y
d
u
r
i
n
g
t
h
e
cu
r
r
e
n
t
c
y
cle
a
s
illu
s
tr
a
ted
in
Fig
u
r
e
1
an
d
s
u
b
s
eq
u
en
t
l
y
i
n
Fi
g
u
r
e
2
.
T
h
e
w
e
ig
h
t
f
r
o
m
o
n
e
o
f
m
o
r
e
p
r
ev
io
u
s
tr
ain
i
n
g
p
atter
n
s
m
u
s
t
b
e
s
to
r
ed
in
o
r
d
e
r
to
u
s
e
m
o
m
e
n
t
u
m
.
Her
e,
th
e
n
e
w
w
e
ig
h
t
s
f
o
r
tr
ain
in
g
s
tep
i+2
i
s
b
ased
o
n
i
a
n
d
i+1
.
T
h
is
m
a
k
es
th
e
cu
r
r
e
n
t
w
e
ig
h
t
ad
j
u
s
t
m
e
n
t
w
it
h
a
f
r
ac
tio
n
o
f
t
h
e
r
ec
en
t
w
ei
g
h
t
ad
j
u
s
t
m
e
n
t.
T
h
e
w
ei
g
h
t
u
p
d
atin
g
f
o
r
m
u
la
w
h
ic
h
d
i
f
f
er
en
tiate
s
n
o
r
m
al
f
ee
d
f
o
r
w
ar
d
n
et
w
o
r
k
is
t
w
t
w
)
w
e
d
(
+
(
t
)
w
=
1)
+
(t
w
jk
jk
jk
jk
jk
)]
1
(
)
(
[
A
s
li
g
h
t
e
n
h
an
ce
m
e
n
t
in
t
h
is
alg
o
r
it
h
m
w
h
ic
h
w
i
ll
g
u
ar
an
tee
th
e
co
n
v
er
g
e
n
ce
is
b
y
li
m
iti
n
g
t
h
e
m
o
m
e
n
t
u
m
f
ac
to
r
w
it
h
th
e
r
a
n
g
e
0
<
<0
.
5
in
s
tead
o
f
0
<
<1
.
T
h
is
i
s
v
er
y
s
i
m
i
lar
to
s
ce
n
ar
io
o
f
f
i
n
e
t
u
n
i
n
g
t
h
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:
2088
-
8708
S
ymb
o
lic
-
C
o
n
n
ec
tio
n
is
t R
ep
r
esen
ta
tio
n
a
l Mo
d
el
fo
r
Op
timiz
in
g
Dec
is
io
n
Ma
kin
g
…
(
R
.
Jo
h
n
Ma
r
tin
)
331
d
ec
is
io
n
alter
n
ati
v
es
th
r
o
u
g
h
e
v
al
u
ati
n
g
th
e
s
ea
r
c
h
cr
it
er
ia
in
th
e
p
er
ce
p
tu
al
a
s
s
o
ciativ
e
m
e
m
o
r
y
as
illu
s
tar
ated
in
Fi
g
u
r
e
2.
Y
1
Z
1
X
1
Y
k
Z
j
X
i
Y
m
Z
p
X
n
P
r
o
c
e
d
u
r
a
l
M
e
m
o
r
y
A
c
t
i
o
n
S
e
l
e
c
t
i
o
n
V
1
1
V
1
j
V
1
p
V
i
1
V
i
j
V
i
p
V
n
1
V
n
j
V
n
p
W
1
k
W
1
m
W
j
1
W
j
k
W
j
m
W
p
1
W
p
k
W
p
m
W
1
1
1
1
V
o
1
V
o
j
V
o
p
W
o
1
W
o
k
W
o
m
P
e
r
c
e
p
t
u
a
l
A
s
s
o
c
i
a
t
i
v
e
M
e
m
o
r
y
C
o
n
t
e
n
t
A
c
t
i
o
n
R
e
s
u
l
t
s
C
o
n
v
e
r
g
e
n
c
e
w
i
t
h
I
m
p
r
o
v
e
d
M
o
m
e
n
t
u
m
F
a
c
t
o
r
D
e
c
i
s
i
o
n
G
o
a
l
Fig
u
r
e
4
.
C
o
n
n
ec
t
io
n
i
s
t c
o
g
n
it
iv
e
n
et
w
o
r
k
m
o
d
el
6.
RE
SU
L
T
S AN
D
D
I
SCU
SS
I
O
NS
T
h
e
p
er
f
o
r
m
a
n
ce
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
is
v
er
i
f
ied
b
y
s
i
m
u
lati
n
g
a
n
a
g
e
n
t
f
o
r
f
ig
u
r
e
p
r
i
n
t
id
en
ti
f
icat
io
n
a
s
a
b
en
ch
m
ar
k
p
r
o
b
lem
.
T
h
e
co
n
v
er
g
en
ce
b
eh
av
io
r
o
f
th
e
p
r
o
p
o
s
ed
a
lg
o
r
ith
m
(
C
C
N)
i
s
co
m
p
ar
ed
w
it
h
w
ell
k
n
o
w
n
B
P
N
tr
ain
in
g
alg
o
r
it
h
m
s
s
u
ch
as
g
r
ad
ien
t
d
esce
n
t
(
B
P
N
-
GD
)
,
g
r
ad
ien
t
d
e
s
ce
n
t
w
it
h
ad
ap
tiv
e
lear
n
i
n
g
r
ate
(
B
P
N
-
GDA
)
[
1
7
]
.
E
p
o
ch
n
u
m
b
er
s
,
m
ea
n
s
q
u
ar
ed
er
r
o
r
s
(
MSE
)
an
d
e
x
ec
u
tio
n
ti
m
e
s
w
ill
b
e
co
n
s
id
er
ed
f
o
r
ev
alu
a
tin
g
t
h
e
co
n
v
er
g
e
n
c
e
p
er
f
o
r
m
an
ce
o
f
t
h
e
alg
o
r
it
h
m
s
.
T
o
h
av
e
ef
f
ec
tiv
e
co
m
p
ar
is
o
n
i
n
ea
c
h
al
g
o
r
ith
m
,
th
e
p
ar
a
m
e
ter
v
al
u
es
an
d
ter
m
i
n
atio
n
co
n
d
itio
n
ar
e
co
n
s
id
er
ed
as
co
n
s
ta
n
t
s
.
T
o
g
et
b
est
r
esu
lts
t
h
e
in
i
tial
w
ei
g
h
ts
ar
e
s
et
to
r
an
d
o
m
an
d
u
n
if
o
r
m
l
y
g
en
er
ated
b
etw
ee
n
[
-
4
,
+4
]
.
T
h
e
lear
n
in
g
r
ate
co
ef
f
icie
n
t
is
a
s
s
i
g
n
ed
t
h
e
v
al
u
e
0
.
0
1
.
T
h
e
f
ig
u
r
e
p
r
in
t
id
en
ti
f
icat
io
n
is
a
w
e
ll
k
n
o
w
n
d
atab
ase
i
n
t
h
e
p
atter
n
r
ec
o
g
n
itio
n
.
T
h
r
ee
class
es
o
f
d
ata
s
et
ea
ch
h
as
3
0
in
s
tan
c
es,
to
tall
y
9
0
p
atter
n
s
ar
e
u
s
ed
.
Ou
t
o
f
9
0
p
atter
n
s
,
7
0
p
atter
n
s
ar
e
u
s
ed
f
o
r
tr
ain
i
n
g
a
n
d
th
e
r
e
m
ai
n
i
n
g
2
0
u
s
ed
f
o
r
test
i
n
g
.
T
h
e
in
p
u
t
v
al
u
es a
r
e
n
o
r
m
alize
d
b
ef
o
r
e
ap
p
lied
to
th
e
n
et
w
o
r
k
.
T
ab
le
1
s
h
o
w
s
t
h
e
r
es
u
lt
s
o
f
n
et
w
o
r
k
co
n
v
er
g
en
ce
w
h
ile
r
u
n
n
in
g
t
h
e
r
esp
ec
tiv
e
al
g
o
r
ith
m
s
.
T
ab
le
1
.
Netw
o
r
k
C
o
n
v
er
g
e
n
c
e
A
l
g
o
r
i
t
h
m
T
r
a
i
n
i
n
g
P
a
r
a
me
t
e
r
s
Ep
o
c
h
s
M
S
E
@
T
r
a
i
n
i
n
g
M
S
E
@
T
e
st
i
n
g
T
i
me
(
mse
c
s)
B
P
N
-
GD
=
0
.
0
1
3
9
5
0
.
0
0
0
3
0
0
.
0
0
9
5
1
1
4
3
B
P
N
-
G
D
A
=
0
.
0
1
3
5
8
0
.
0
0
0
2
9
0
.
0
0
7
1
9
1
2
9
P
r
o
p
o
se
d
C
C
N
=
0
.
0
1
,
=
0
.
0
9
1
3
9
0
.
0
0
0
2
8
0
.
0
0
6
9
2
67
Fig
u
r
e
5
.
Net
w
o
r
k
lear
n
i
n
g
c
u
r
v
e
o
f
C
C
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.
8
,
No
.
1
,
Feb
r
u
ar
y
201
8
:
3
2
6
–
3
3
2
332
T
h
e
p
r
o
p
o
s
ed
C
C
N
al
g
o
r
ith
m
co
n
v
er
g
es
at
1
3
9
th
iter
atio
n
w
h
ile
o
th
er
t
w
o
B
P
N
alg
o
r
it
h
m
s
a
t
t
h
e
ep
o
ch
s
o
f
3
9
5
an
d
3
5
8
r
esp
ec
tiv
el
y
.
I
t
a
ls
o
ta
k
es
m
in
i
m
u
m
ti
m
e
o
f
6
7
m
s
ec
s
to
co
n
v
er
g
e.
B
u
t
B
P
N
-
G
D
a
n
d
B
P
N
-
GDA
h
a
v
e
tak
e
n
1
4
3
m
s
ec
s
a
n
d
1
2
9
m
s
ec
s
r
esp
e
ctiv
el
y
to
r
ea
c
h
t
h
e
ter
m
i
n
at
io
n
w
it
h
co
n
d
itio
n
MSE
=
0
.
0
0
0
3
.
T
h
e
n
o
tab
le
v
ar
iatio
n
f
o
u
n
d
i
n
t
h
e
p
r
o
p
o
s
ed
C
C
N
i
s
its
test
i
n
g
MSE
i.e
0
.
6
9
2
,
w
h
ic
h
i
s
m
i
n
i
m
u
m
a
m
o
n
g
al
l
al
g
o
r
ith
m
s
.
T
h
e
lear
n
in
g
c
u
r
v
e
d
r
aw
n
a
g
ai
n
s
t
ep
o
ch
s
a
n
d
MS
E
f
o
r
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
is
s
h
o
w
n
i
n
Fi
g
u
r
e
5
.
7.
CO
NCLU
SI
O
N
T
h
e
s
tr
ateg
y
ad
o
p
ted
to
m
a
k
e
a
u
n
iq
u
e
C
o
n
n
ec
tio
n
i
s
t
C
o
g
n
iti
v
e
Ne
t
w
o
r
k
(
C
C
N)
m
o
d
el
b
y
ass
o
ciati
n
g
th
e
B
P
N
an
d
L
I
DA
C
o
g
n
iti
v
e
m
o
d
el
o
f
ac
ti
o
n
s
elec
tio
n
is
co
m
p
u
tat
io
n
a
ll
y
p
r
o
v
ed
w
it
h
it
s
en
h
a
n
ce
d
n
et
w
o
r
k
p
er
f
o
r
m
an
ce
co
m
p
ar
ed
w
it
h
o
th
er
t
w
o
v
ar
iatio
n
s
o
f
B
P
N.
T
h
e
co
n
v
er
g
en
ce
s
p
ee
d
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
i
n
ter
m
s
o
f
ti
m
e
an
d
ep
o
ch
s
i
s
s
h
o
w
n
b
y
s
i
m
u
lati
n
g
th
e
b
en
c
h
m
ar
k
i
n
g
p
r
o
b
lem
o
f
f
i
g
u
r
e
p
r
in
t
v
er
if
ica
tio
n
.
T
h
e
h
ig
h
li
g
h
t
o
f
th
e
p
r
o
p
o
s
ed
alg
o
r
ith
m
i
s
th
e
m
i
n
i
m
u
m
MSE
o
b
s
er
v
ed
w
h
ile
test
i
n
g
.
T
h
is
ca
n
b
e
i
m
p
r
o
v
ed
b
y
tr
ai
n
i
n
g
w
it
h
m
o
r
e
i
n
p
u
t
p
atter
n
s
.
T
h
e
p
r
o
p
o
s
ed
C
C
N
m
o
d
el
s
h
o
u
ld
also
b
e
co
m
p
ar
ed
w
it
h
o
th
er
lead
i
n
g
ar
c
h
itect
u
r
es lik
e
S
VM
to
k
n
o
w
i
ts
li
m
itat
io
n
s
.
I
t
is
ev
id
en
t
t
h
at
t
h
is
i
n
itiati
v
e
w
ill
eli
m
i
n
ate
th
e
l
i
m
i
tat
io
n
s
o
f
b
o
th
tr
ad
itio
n
al
s
y
m
b
o
lic
an
d
co
n
n
ec
tio
n
is
t a
p
p
r
o
ac
h
es a
n
d
w
il
l le
ad
u
s
to
a
n
e
w
d
i
m
e
n
tio
n
o
f
h
ig
h
er
o
r
d
er
co
g
n
itiv
e
a
g
en
t b
u
ild
i
n
g
.
RE
F
E
R
E
NC
E
S
[1
]
Yo
u
n
g
-
Ju
n
S
o
n
,
e
t
a
l.
,
“
An
e
x
ten
d
e
d
BDI
m
o
d
e
l
fo
r
h
u
m
a
n
b
e
h
a
v
i
o
rs
:
d
e
c
isio
n
-
ma
k
in
g
,
lea
rn
i
n
g
,
i
n
ter
a
c
ti
o
n
s,
a
n
d
a
p
p
li
c
a
ti
o
n
s
,”
In
P
r
o
c
e
e
d
in
g
s
o
f
th
e
2
0
1
3
W
in
ter
S
im
u
latio
n
Co
n
f
e
re
n
c
e
:
S
i
m
u
latio
n
:
M
a
k
in
g
De
c
isio
n
s
in
a
Co
m
p
lex
W
o
rld
(W
S
C
'
1
3
),
IEE
E
P
re
ss
.
2
0
1
3
;
4
0
1
-
4
1
1
.
[2
]
Yin
g
x
u
W
,
G
u
e
n
th
e
r
R,
“
T
h
e
Co
g
n
it
iv
e
P
r
o
c
e
ss
o
f
De
c
isio
n
M
a
k
in
g
,”
In
ter
n
a
ti
o
n
a
l
J
o
u
r
n
a
l
o
f
C
o
g
n
it
ive
In
fo
rm
a
t
ics
a
n
d
N
a
tu
r
a
l
I
n
telli
g
e
n
c
e
.
2
0
0
7
;
1
(
2
):
7
3
-
85.
[3
]
S
tan
F
ra
n
k
li
n
,
e
t
a
l.
,
“
L
ID
A
:
A
S
y
ste
m
s
-
le
v
e
l
A
rc
h
it
e
c
tu
re
fo
r
Co
g
n
it
i
o
n
,
Em
o
ti
o
n
,
a
n
d
L
e
a
rn
in
g
,
”
IEE
E
Tr
a
n
sa
c
ti
o
n
s
o
n
Au
to
n
o
mo
u
s M
e
n
ta
l
De
v
e
lo
p
me
n
t
.
2
0
1
4
;
6
(
1
):1
9
-
41.
[4
]
W
a
ll
a
c
h
,
W
,
e
t
a
l.
,
“
Co
n
c
e
p
tu
a
l
a
n
d
Co
m
p
u
tati
o
n
a
l
M
o
d
e
l
o
f
M
o
ra
l
De
c
isio
n
M
a
k
in
g
in
Hu
m
a
n
a
n
d
A
rti
f
icia
l
Ag
e
n
ts
,
”
T
o
p
ics
i
n
Co
g
n
i
ti
v
e
S
c
ie
n
c
e
.
2
0
1
0
;
2
(
3
):
4
5
4
–
4
8
5
.
[5
]
Jo
h
n
M
a
rti
n
R,
S
u
jat
h
a
S
,
“
V
o
l
it
io
n
a
l
De
c
isio
n
M
a
k
in
g
o
n
In
te
ra
c
ti
v
it
y
a
s
a
Re
su
lt
o
f
M
u
lt
i
-
c
y
c
li
c
c
o
g
n
it
iv
e
P
r
o
c
e
ss
e
s an
d
Em
o
ti
o
n
s,”
J
o
u
rn
a
l
o
f
C
o
mp
u
ti
n
g
T
e
c
h
n
o
lo
g
ies
.
2
0
1
5
;
4
(
1
0
)
:2
2
7
8
–
3
8
1
4
.
[6
]
T
h
o
m
a
s,
e
t
a
l.
,
“
C
o
n
n
e
c
ti
o
n
ist
m
o
d
e
ls
o
f
c
o
g
n
it
io
n
.
In
R
o
n
S
u
n
(e
d
.
),
”
T
h
e
C
a
mb
ri
d
g
e
Ha
n
d
b
o
o
k
o
f
Co
mp
u
t
a
ti
o
n
a
l
Psy
c
h
o
lo
g
y
,
”
Ca
m
b
rid
g
e
Un
iv
e
rsit
y
P
re
ss
.
2
0
1
1
;
2
3
-
58.
[7
]
Ili
n
,
e
t
a
l
.
,
“
Co
g
n
i
ti
v
e
ly
In
sp
ired
Ne
u
ra
l
Ne
tw
o
r
k
f
o
r
Re
c
o
g
n
it
io
n
o
f
S
it
u
a
ti
o
n
s,”
IJ
NCR
.
2
0
1
0
;
1
:
3
6
-
5
5
.
[8
]
F
risto
n
K.,
P
a
rk
H,
“
S
tru
c
t
u
ra
l
a
n
d
f
u
n
c
ti
o
n
a
l
b
ra
in
n
e
tw
o
rk
s
:
fro
m
c
o
n
n
e
c
ti
o
n
s
to
c
o
g
n
it
i
o
n
,
”
S
c
ien
c
e
.
2
0
1
3
;
3
4
2
(6
1
5
8
).
[9
]
Do
n
g
D,
F
ra
n
k
li
n
S
,
“
S
e
n
so
ry
M
o
to
r
S
y
ste
m:
M
o
d
e
li
n
g
t
h
e
Pro
c
e
ss
o
f
Actio
n
Exe
c
u
ti
o
n
,”
P
r
o
c
e
e
d
in
g
s
o
f
th
e
3
6
t
h
A
n
n
u
a
l
Co
n
f
e
re
n
c
e
o
f
th
e
Co
g
n
it
i
v
e
S
c
ien
c
e
S
o
c
iety
.
2
0
1
4
;
2
1
4
5
-
2
1
5
0
.
[1
0
]
L
e
b
iere
C,
A
n
d
e
rso
n
JR,
“
Co
g
n
it
iv
e
c
o
n
stra
in
ts
o
n
d
e
c
isio
n
m
a
k
in
g
u
n
d
e
r
u
n
c
e
rtain
ty
,
”
Fro
n
t.
Psy
c
h
o
lo
g
y
.
2
0
1
1
;
2
(3
0
5
)
.
[1
1
]
Jo
h
n
M
a
rti
n
R,
S
u
jat
h
a
S
,
“
Bo
tt
o
m
-
u
p
A
p
p
ro
a
c
h
o
f
M
o
d
e
li
n
g
Hu
m
a
n
De
c
isio
n
M
a
k
in
g
f
o
r
Bu
il
d
in
g
In
telli
g
e
n
t
Ag
e
n
ts,
”
In
d
ia
n
J
o
u
rn
a
l
o
f
S
c
ie
n
c
e
a
n
d
T
e
c
h
n
o
l
o
g
y
.
2
0
1
6
;
9
(
4
).
[1
2
]
Da
v
id
E.
Ru
m
e
lh
a
rt,
Ja
m
e
s
L
.
M
c
Clellan
d
,
“
Pa
ra
l
lel
Distrib
u
ted
Pro
c
e
ss
in
g
:
Exp
lo
r
a
ti
o
n
s in
th
e
M
icr
o
stru
c
tu
re
o
f
Co
g
n
it
io
n
,
”
Psy
c
h
o
l
o
g
ic
a
l
a
n
d
Bi
o
lo
g
ica
l
M
o
d
e
ls,
M
IT
Pre
ss
,
Ca
m
b
rid
g
e
.
1
9
8
6
;
2.
[1
3
]
M
a
n
a
s
S
in
g
h
a
l,
M
a
it
re
y
e
e
Du
tt
a
,
M
a
n
ish
T
rik
h
a
,
“
S
ig
n
a
tu
re
V
e
r
if
ica
ti
o
n
u
sin
g
No
rm
a
li
z
e
d
S
tatic
F
e
a
tu
re
s
a
n
d
Ne
u
ra
l
Ne
t
w
o
rk
Clas
si
f
ica
ti
o
n
,
”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
E
lec
trica
l
a
n
d
C
o
mp
u
ter
En
g
in
e
e
rin
g
(
IJ
ECE
)
.
De
c
e
m
b
e
r
2
0
1
6
;
6
(6
)
:
2
6
6
5
-
2
6
7
3
.
[1
4
]
A
rp
a
d
Ke
l
e
m
e
n
,
Yu
lan
L
ian
g
,
S
tan
F
ra
n
k
li
,
“
Lea
rn
in
g
Hig
h
Qu
a
li
ty
De
c
isi
o
n
s
w
it
h
Ne
u
ra
l
Ne
t
w
o
r
k
s
in
Co
n
sc
io
u
s
S
o
f
tw
a
r
e
A
g
e
n
ts,
”
W
S
EA
S
T
ra
n
s
a
c
ti
o
n
s o
n
S
y
ste
ms
.
2
0
0
5
;
9
(4
):
1
4
8
2
-
1
4
9
2
.
[1
5
]
Na
ra
sin
g
a
Ra
o
M
R,
e
t
a
l.
,
“
P
r
e
d
ictiv
e
M
o
d
e
l
f
o
r
M
in
i
n
g
Op
in
io
n
s
o
f
a
n
Ed
u
c
a
ti
o
n
a
l
Da
tab
a
se
Us
in
g
Ne
u
ra
l
Ne
tw
o
rk
s,”
In
ter
n
a
ti
o
n
a
l
J
o
u
rn
a
l
o
f
El
e
c
tr
ica
l
a
n
d
C
o
mp
u
ter
En
g
i
n
e
e
rin
g
(
IJ
ECE
)
.
2
0
1
5
;5
(
5
):1
1
5
8
-
1
1
6
3
.
[1
6
]
F
o
d
o
r,
Je
rry
A
,
e
t
a
l.
,
“
Co
n
n
e
c
ti
o
n
ism
a
n
d
c
o
g
n
it
iv
e
a
r
c
h
it
e
c
tu
re
:
A
Crit
ica
l
A
n
a
l
y
sis,”
Co
g
n
it
i
o
n
.
1
9
9
8
;
2
8
:
3
-
7
1
.
[1
7
]
G
.
Á
.
Wern
e
r
a
n
d
L
.
Ha
n
k
a
,
“
T
u
n
i
n
g
a
n
a
rtif
icia
l
n
e
u
r
a
l
n
e
t
wo
rk
to
in
c
re
a
se
th
e
e
ff
icie
n
c
y
o
f
a
fi
n
g
e
rp
rin
t
ma
tch
in
g
a
lg
o
rit
h
m
,”
I
EE
E
1
4
t
h
In
tern
a
ti
o
n
a
l
S
y
m
p
o
siu
m
o
n
A
p
p
li
e
d
M
a
c
h
i
n
e
In
tell
ig
e
n
c
e
a
n
d
I
n
f
o
rm
a
ti
c
s
(S
A
M
I)
,
He
rlan
y
.
2
0
1
6
;
1
0
5
-
1
0
9
.
Evaluation Warning : The document was created with Spire.PDF for Python.