IAES
I
nt
er
nat
iona
l
Journ
al
of
Ar
tifici
al I
nt
el
li
gence
(I
J
-
AI
)
Vo
l.
8
, No
.
2
,
J
un
e
201
9
, pp.
1
44
~
1
55
IS
S
N:
22
52
-
8938
,
DOI: 10
.11
591/ijai.
v
8
.i
2
.pp
1
44
-
1
55
144
Journ
al h
om
e
page
:
http:
//
ia
es
core
.c
om/o
nline/i
nd
ex
.ph
p
/I
J
A
I
Hybrid
imp
er
iali
stic c
om
petitive
algorith
m in
corp
orated
with
hopfield
neural n
etwork f
or robus
t 3 satisfi
ability l
ogic
program
min
g
Vign
es
hwe
r K
at
hir
vel
1
,
Mohd
.
As
yraf
M
an
s
or
2
,
M
oh
d
Sha
re
duwa
n
Mohd
Kasi
hm
uddin
3
,
Sa
r
at
h
a Sa
tha
siva
m
4
1,
3,
4
School
of
Ma
the
m
at
i
ca
l
Sci
en
ce
s,
Univ
ersiti
S
ai
ns Mal
a
y
s
ia,
1
1800
US
M,
Pula
u
Pinang
,
Mal
a
ysia
2
School
of
Dist
a
nce
Educat
ion
,
Univer
siti
Sains
Malay
s
ia
,
11800
US
M,
Pulau
Pin
ang,
Ma
lay
si
a
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
J
a
n
3
0
, 2
01
9
Re
vised A
p
r
2
8
, 2
019
Accepte
d
M
ay
1
7
, 201
9
Im
per
ia
li
st
Co
m
pet
it
ive
al
gori
thm
(ICA)
is
a
robust
tra
ini
n
g
al
gorit
hm
inspire
d
b
y
the
socio
-
poli
t
ically
m
oti
vat
ed
strateg
y
.
Thi
s
pap
er
foc
uses
on
uti
lizing
a
h
y
b
ridi
z
ed
ICA
w
it
h
Hopfiel
d
Neura
l
Networ
k
on
a
3
-
Sati
sfiability
(3
-
SA
T)
logi
c
prog
ramm
ing.
Eve
nt
ual
l
y
th
e
per
for
m
anc
e
of
th
e
proposed
al
gori
t
hm
will
be
com
par
ed
to
othe
r
2
al
gorit
hm
s,
whi
c
h
are
HN
N
-
3SA
TE
S (E
S) a
nd
HN
N
-
3
SA
TGA (GA).
The
per
form
anc
e
shal
l
be
eva
lu
at
ed
with
the
Root
Mea
n
Square
E
rror
(RMS
E),
Mea
n
Abs
olut
e
Err
or
(MA
E),
Sum
of
Square
s
Err
or
(SS
E),
Schwarz
Ba
y
es
ian
Crit
eri
on
(SBC
),
Globa
l
Minim
a
Rat
io
a
nd
Com
puta
ti
on
Ti
m
e
(CPU
ti
m
e).
The
exp
ec
t
ed
outc
om
e
will
portray
th
at
the
IC
al
gorit
hm
will
outpe
rform
the
othe
r
two
algorithms
in
doing
3
-
SA
T
log
ic
progr
amm
ing.
Ke
yw
or
d
s
:
3
Sati
sfia
bili
ty
Ex
hau
sti
ve
sea
rch
Hopf
ie
l
d neura
l netw
ork
Im
per
ia
li
sti
c c
om
petit
ive
al
gorithm
Lo
gic prog
ram
m
ing
Copyright
©
201
9
Instit
ut
e
o
f Ad
vanc
ed
Engi
n
ee
r
ing
and
S
cienc
e
.
Al
l
rights re
serv
ed
.
Corres
pond
in
g
Aut
h
or
:
Vignes
hw
e
r K
at
hirv
el
,
School
of Mat
hem
atical
Sciences
,
Un
i
ver
sit
i Sai
ns M
al
ay
sia
,
11800 U
SM, P
enang,
Mal
ay
sia
.
Em
a
il
:
vig
nes
hk@
us
m
.stud
e
nt
.
m
y
1.
INTROD
U
CTION
Our
bio
l
og
ic
al
ne
ur
al
net
work’s
st
ru
ct
ur
e
hav
e
ins
pire
d
new
m
od
el
s
of
c
om
pu
ti
ng
to
perform
patte
rn
recog
niti
on
ta
sk
s
.
H
oweve
r,
th
e
ope
rati
on
s
of
t
he
bio
lo
gical
ne
uron
a
nd
the
ne
ural
interco
nnec
ti
on
s
are
no
t
fu
ll
y
unde
rstoo
d
ti
ll
this
ve
ry
day
[
1].
It
a
uto
m
at
i
cal
ly
adap
ts
to
a
ne
w
e
nvir
onm
ent
without
bei
ng
reprog
ram
m
ed,
and
it
is
abl
e
to
deal
with
fu
z
zy
,
pro
ba
bili
sti
c,
no
isy
and
inc
on
sist
ent
inf
or
m
at
io
n
[2
]
.
Ar
ti
fici
al
intel
l
igence
at
tract
e
d
a
pro
fu
se
num
ber
of
r
ese
arch
in
c
om
bin
at
ori
al
op
ti
m
i
zat
ion
pro
ble
m
s
[3
]
.
Ar
ti
fici
al
intel
li
gen
ce
becam
e
popu
la
r
am
ong
resea
rc
he
s
du
e
to
it
s
capab
il
it
y
to
so
lve
c
om
pu
ta
ti
on
al
,
cl
assifi
cat
ion
a
nd
patte
rn
rec
ogniti
on
pr
oble
m
s
us
ing
le
ar
ni
ng
based
al
gorithm
.
Its
arch
it
ect
ur
e
co
ns
ist
s
of
a
two
-
dim
ensional
connecte
r
ne
ur
al
netw
ork
i
n
wh
ic
h
li
nkin
g
the
stre
ng
t
hs
betwee
n
ne
uron
s
are
deci
de
d
base
d
on the c
onstrai
nts a
nd so
l
utio
n basis
of the
optim
iz
at
ion
p
r
oble
m
to
be
s
olve
d [
4].
On
e
of
the
m
os
t
com
m
e
m
or
at
ed
top
ic
s
is
th
e
sat
isfia
bili
ty
pro
blem
(S
AT
).
The
S
AT
prob
le
m
can
be
disti
nguish
e
d
as
the
process
of
obta
inin
g
an
ideal
as
sig
nm
ent
us
in
g
Boo
le
a
n
value
s
to
c
orrob
or
a
te
that
the
f
or
m
ula
is
sat
isfie
d
[
5].
I
n
this
pa
per,
w
e
sh
al
l
stress
on
3
-
Sati
sfia
bili
ty
,
or
c
omm
on
ly
known
as
3
-
SA
T
.
3
-
S
AT
ca
n
be
de
fine
d
as
a
form
ula
in
co
njuncti
ve
nor
m
al
fo
rm
(CN
F)
w
her
e
by
ea
ch
of
the
num
ber
of
neur
on
s
are
li
m
it
ed
to
3
literal
s
or
ne
urons.
Exh
a
us
ti
ve
se
arch
,
or
c
ommon
ly
know
n
as
br
ute
-
f
or
ce
se
a
rch,
is
a
ver
y
c
onve
nt
ion
al
pro
blem
-
so
lvi
ng
m
et
hod
a
nd
al
gorith
m
ic
par
adi
gm
that
is
com
pr
is
ed
of
syst
e
m
atical
ly
identify
in
g
al
l
possible
can
di
dates
f
or
the
so
luti
on
a
nd
t
o
cl
ari
fy
w
het
her
or
not
the
can
did
at
es
sa
ti
sfies
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
IS
S
N:
22
52
-
8938
Hyb
ri
d
i
mp
eri
alist
ic
co
m
petit
iv
e a
lg
or
it
hm i
n
co
rpor
ated
wi
th ho
pfiel
d
neural… (
Vign
e
shw
er Kathirvel
)
145
the
pro
blem
s
ta
tem
ent
resp
e
ct
ively
[6
]
.
H
NN
-
3SA
TES
is
a
popula
r
al
gorithm
-
design
as
it
is
w
idely
app
li
cable
an
d
giv
e
n
c
orrect
ge
ner
at
io
n
a
nd c
heck
i
ng.
In
Ri
dd
le
’s
a
nd
Se
gal’s
pa
pe
r,
t
hey
ha
ve
a
pp
li
ed
in
du
ct
i
ve
cat
egorizat
io
n
m
et
ho
ds
in
a
set
of
data
colle
ct
ion
i
n
a
Boein
g
plant
with
the
ob
j
ect
ive
of
unveili
ng
the
possi
ble
flaws
or
set
ba
c
k
in
a
m
anu
fac
turing
process
.
Tw
o
exp
e
rim
ents
wer
e
ca
rr
ie
d
ou
t
[7
]
w
hi
ch
w
ould
te
s
t
the
pr
e
dicti
on
s
extracte
d
fro
m
the
ass
um
ption
s
that
le
xical
acce
ss
incl
udes
a
searc
h
process
a
re
re
po
rted.
It
is
al
s
o
s
how
n
that
wh
e
n
the
ta
r
get
of
the
sea
rc
h
is
a
no
n
-
e
xistent
entry,
it
i
nvol
ves
a
n
H
NN
-
3SATE
S,
al
though
the
te
st
it
e
m
s
are
words.
From
t
he
res
ults
of
t
he
e
xperim
ent,
it
was
de
duc
ed
t
hat
the
se
arch
m
od
el
ex
plains
t
he
pro
cedure
adequate
ly
,
w
her
e
the
m
os
t
known
m
e
anin
g
of
a
ho
m
ogr
aph
is
acce
sse
d
[
8]
.
The
Im
per
ia
li
st
com
petit
iv
e
al
gorithm
(I
CA)
is
ins
pire
d
by
the
hum
an
so
ci
o
-
poli
ti
cal
evo
l
ution
proc
ess.
In
Luca
s’
and
Na
siri’s
pa
per,
a
novel opti
m
iz
a
ti
on
alg
or
it
hm
base
d
on HNN
-
3SAT
ICA
al
gorithm
is u
sed
for
the d
esi
gn
of
a lo
w
sp
ee
d si
ng
l
e
sided
li
nea
r
in
du
ct
io
n
m
oto
r
(LI
M
)
[
9]
.
Ha
ving
hi
gh
e
ff
ic
ie
ncy
with
hi
gh
po
wer
factor
is
ver
y
vital
in
these
app
li
cat
io
ns
.
T
he
res
ults
sh
ows
that
ICA
is
m
or
e
su
cces
sfu
l
f
or
the
de
sign
of
LIMs
com
par
ed
to
ge
netic
al
gorithm
(
GA
)
an
d
c
o
nve
ntion
al
desig
n
[
10]
.
This
m
et
a
-
heurist
ic
so
luti
on
a
ppr
oac
h
is
al
so
us
e
d
to
optim
iz
e
pro
du
ct
m
ix
-
outso
ur
ci
ng
f
or
m
anu
fact
ur
i
ng
enter
pr
ise
s
.
I
n
Mi
ll
er’
s
pa
per,
they
ha
ve
c
om
bin
ed
two
ad
aptiv
e
proces
ses
,
the
ge
netic
searc
h
th
r
ough
the
netw
ork
arc
hi
te
ct
ur
e
s
pace,
an
d
bac
kpropag
at
io
n
le
arni
ng
i
n
ind
ivi
du
al
net
works.
Cy
cl
es
of
le
a
rn
i
ng
i
n
ind
ivi
du
al
s
a
re
nested
within
cy
cl
es
of
e
vo
l
ution
i
n
popula
ti
on
s,
and
each
of
the
le
ar
ning
cy
cl
e
pr
ese
nts
a
n
in
div
i
du
al
ne
ur
al
netw
ork,
with
the
set
of
i
nput
-
outp
ut
pairs
def
i
ning
the
ta
sk
[
11
]
.
Lam
’s
pap
e
r
on
the
ot
her
ha
nd
pres
ents
the
tu
ning
of
the
str
uctu
r
e
and
par
am
et
e
rs
of
a
neural
netw
or
k
us
i
ng
a
n
im
pro
ved
GA
.
T
he
str
uctur
e
a
nd
par
am
et
ers
of
the
neural
netw
ork
can
be
tun
e
d
us
in
g
t
he
im
pr
ov
e
d
G
A
[12]
.
I
n
this
pa
per,
we
w
il
l
de
velo
p
hybri
d
m
od
e
l
of
IC
A
with
Hopf
ie
l
d
in
3
-
SA
T
(HNN
-
3SATI
CA).
The
co
m
par
ison
will
be
betwee
n
t
he
hybr
i
d
e
xhaustive
sea
rch
(HN
N
-
3SAT
ES)
a
nd
hybri
d
ge
netic
algorit
hm
(H
N
N
-
3SAT
GA).
2.
RESEA
R
CH MET
HO
DS
2
.
1.
3
-
S
at
is
fia
bil
ity
The
3
-
S
AT
pa
rad
i
gm
a
ll
ow
s
bin
ary
val
ue
s
of
each
vari
able,
wh
ic
h
co
ns
ist
of
ei
the
r
1
or
-
1.
The
3
-
S
AT
pro
blem
can
be
seal
ed
as
a
non
-
determ
i
nisti
c
pr
oble
m
[1
3].
The
3
-
S
AT
pro
bl
e
m
in
the con
j
unct
ive
nor
m
al
f
or
m
(
CNF) com
pr
is
es of
four ra
dic
al
f
eat
ures:
1.
The
SAT
form
ula
has
an
ar
ra
y
of
n
var
ia
ble
s,
12
,
,
.
.
.
.
,
n
z
z
z
inside
of
ea
ch
num
ber
of
ne
uro
ns
.
I
n
this
case,
we
s
hall l
i
m
it
i
t t
o
at
m
os
t 3 (
3
n
).
2.
A
set
of m
n
um
ber
o
f ne
uro
ns
i
n
a B
oo
le
a
n form
ula,
12
:
.
.
.
.
m
m
F
c
c
c
.
3.
We
co
ns
i
der
e
d
3
li
te
rals
in
e
ach
num
ber
of
neur
on
s
i
n
3
-
SA
T.
Each
nu
m
ber
of
neurons,
k
c
will
be
com
bin
ed by t
he
lo
gic
op
e
rat
or O
R.
4.
The
li
te
rals ca
n be the
v
a
riab
le
s o
r
the
ne
gat
ion
of the
v
a
ria
bles it
sel
f.
The 3
-
SAT f
or
m
ula can
be
il
lustrate
d as
per
the foll
owin
g:
P
A
B
C
D
E
F
G
H
I
(1)
The
form
ula
can
gen
e
rall
y
be
create
d
in
m
ulti
ple
com
bina
ti
on
s
as
the
num
ber
of
at
om
s
m
ay
var
y.
Re
la
ti
vely
,
the
pro
bab
il
it
ie
s
of
a
nu
m
ber
of
neur
on
s
to
be
sat
isfie
d
will
be
m
axi
m
iz
ed
by
the
great
er
num
ber
of li
te
rals p
e
r n
um
ber
of
neur
on
s
[1
4].
2
.
2.
H
opfie
ld
neural
netw
or
k
Re
current
ne
ural
netw
orks
a
r
e
pr
im
aril
y
dynam
ic
al
sys
tem
s
that
feedb
ack
si
gn
al
s
t
o
them
sel
ves.
Popu
la
rized
by
John
H
opfiel
d,
these
m
od
el
s
hav
e
a
rich
cl
a
ss
of
dy
nam
ic
s
char
act
e
rized
by
the
e
xistenc
e
of
a
few
sta
ble
sta
te
s,
ha
ving
t
heir
res
pecti
ve
ba
sin
of
at
tract
io
n
[
15]
.
S
how
i
n
Fi
gure
1
Hop
fiel
d
Ne
ural
N
et
wor
k.
These
inte
rconn
ect
e
d
unit
s
are
al
so
know
n
as
the
bipo
la
r
thres
hold
un
it
.
Her
e
,
th
e
bin
a
ry
valu
es
are
consi
d
ere
d
to
be
1
or
-
1
[
16]
.
He
nce,
Ni
will
be
the
it
h
act
ivati
on
of
a
ne
uro
n,
hav
i
ng
t
he
f
ollow
in
g
thres
ho
l
d funct
ion
:
1,
1,
ij
j
i
i
if
w
S
N
O
th
e
r
w
is
e
(2)
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8938
IJ
-
AI
V
ol.
8
,
No.
2
,
June
20
1
9
:
1
4
4
–
1
55
1
46
w
he
re
ij
w
is
the
connecti
on
stre
ng
t
h
f
ro
m
un
it
j
to
i
.
j
S
represent
s
the
unit
of
j
a
nd
i
represe
nts
the
thre
shold
of
un
it
i
.
The
r
e
are
no
c
onne
ct
ion
s
withi
n
the
connecti
on
in
the
H
opfi
el
d
net
usual
ly
,
hence
0
ij
w
.
This
sh
ows
that
th
e
connecti
ons
are
bid
irect
io
nal
or
sym
m
e
tric
i
j
j
i
ww
[1
7].
Ne
uro
n
is
in
ge
ner
al
bi
po
la
r,
{
1
,
1
}
i
S
wh
ic
h
fu
lfil
s
the
dynam
ic
s
s
g
n
(
)
ii
Sh
,
w
her
e
the
local
fiel
d
is
re
presen
te
d
by
i
h
.
The
c
om
pu
ta
ti
on
al
m
od
el
will
directl
y
appr
ox
im
at
e
to
a
hi
gh
e
r
order
co
nn
ect
io
n.
T
herefo
re,
t
he
lo
ca
l
fiel
d
m
od
ifie
s to
the
foll
ow
i
ng
:
(
3
)
(
2
)
(
1
)
i
j
k
ij
j
i
ijk
jj
h
w
S
S
w
S
w
(3)
The
e
ne
rg
y
f
unct
ion
by
Pin
ka
s
(
1991)
f
or
t
he
discrete
H
opfiel
d
Ne
ural
Netw
ork
f
or
3
-
Sati
sfiabili
t
y
nu
m
ber
of n
e
uro
ns
ca
n be
w
ritt
en
as foll
ows:
(
3
)
(
2
)
(
1
)
11
32
i
j
k
ij
i
j
i
j
ijk
i
j
k
i
j
i
E
w
S
S
S
w
S
S
w
S
(4)
Hopf
ie
l
d’
s
ene
rg
y
functi
on
is
pa
rtic
ularly
vi
ta
l
as
it
will
de
te
rm
ine
the
de
gr
ee
of
c
onve
r
gen
ce
of
th
e
netw
ork
[
18]
.
The
e
nergy
va
lue
at
ta
ined
from
the
equ
at
i
on
will
be
che
cked
an
d
deter
m
ined
w
hethe
r
it
is
a
global o
r
a
loc
al
m
ini
m
a.
Figure
1. H
opf
ie
ld
ne
ur
al
net
work [
7]
2
.
3.
Ex
ha
u
sti
ve
se
arch
a
l
gori
th
m
The
pur
pose
f
or
ch
oosin
g
th
is
ty
pical
al
go
rithm
is
to
dis
cov
e
r
the
eff
e
ct
iveness
de
gr
ee
of
H
NN
–
3SATES
.
A
pa
r
t
fr
om
that,
the
re
are
th
eo
reti
cal
ly
sat
isfyi
ng
assignm
ents
gi
ven
f
or
a
ny
3
-
SA
T
pro
blem
[
19
]
.
The
sat
is
fied
assignm
ent
f
or
the
ES
al
go
rithm
is
ob
ta
ined
afte
r
c
ondu
ct
in
g
a
brutal
“t
rial
an
d
error”
proce
dure.
T
he
per
f
orm
ance
of
ES
has
a
lready
bee
n
r
econn
oitred
in
the
wo
r
k
of
Nievergelt
(2
00
0).
The o
bj
ect
ive
fun
ct
io
n
is
d
em
on
st
rated a
s th
e f
ollow
i
ng
:
m
a
x
{
}
ES
f
(5)
Her
e
with a
re t
he
ste
ps
of im
p
lem
enting
ES:
Stage
1: Init
ia
li
zat
ion
The
ca
ndidate
bit stri
ng is i
niti
at
ed
an
d ge
nerat
ed.
Stage
2: Fitnes
s Ev
al
uatio
n
The
ca
ndidate
bit stri
ng is test
ed
a
nd h
e
nce,
t
he fit
ness wil
l be c
om
pu
te
d b
y uti
li
zi
ng
the
foll
ow
i
ng
:
1
2
3
(
)
(
)
(
)
.
.
.
(
)
E
S
t
o
t
a
l
N
C
f
c
x
c
x
c
x
c
x
(6)
Stage
3:
Ev
al
ua
ti
on
As
an
ou
tc
ome
,
return
the
a
ssign
m
ent
with
the
m
axi
m
u
m
fitness.
Else
oth
er
wise,
identify
a
ne
w
cand
i
date
bit stri
ng.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
IS
S
N:
22
52
-
8938
Hyb
ri
d
i
mp
eri
alist
ic
co
m
petit
iv
e a
lg
or
it
hm i
n
co
rpor
ated
wi
th ho
pfiel
d
neural… (
Vign
e
shw
er Kathirvel
)
147
2
.
4.
Imperi
alist
c
omp
e
titive
a
lg
orit
hm
HNN
-
3SAT
IC
A
al
gorithm
is
a
new
m
et
a
-
heu
risti
c
op
ti
m
iz
ation
de
velo
pe
d
base
d
on
a
so
ci
o
poli
ti
cal
l
y
m
otivate
d
str
at
egy an
d
c
on
t
ai
ns
tw
o
m
ai
n
ste
ps
[20]:
The
m
ov
em
ent of the
co
l
on
ie
s
The
im
per
ia
li
st
ic
co
m
petit
ion
It
is
a
br
and
ne
w
ev
olu
ti
onary
te
chn
iq
ue
gal
van
iz
e
d
by
the
i
m
per
ia
li
st
ic
c
om
petit
ion
a
m
ong
natio
ns
.
The
i
niti
al
p
op
ulati
on
of s
olu
t
ion
is
r
e
pr
ese
nt
ed
as
cou
ntry.
Her
e
are t
he 6
m
ai
n
sta
ges of
ICA:
Stage
1: Init
ia
li
zat
ion
The
par
am
et
e
rs
are
dete
rm
ined
a
nd
the
co
un
try
’s
popula
ti
on
are
ini
ti
al
iz
ed.
Fo
r
insta
nce
,
Country =
100.
1
2
3
(
,
,
,
.
.
.
.
)
n
c
o
u
n
t
r
y
x
x
x
x
(7)
Stage
2: Fitnes
s of the
Co
untr
ie
s
The fit
ness o
f
t
he
c
ountries
a
r
e cal
culat
ed.
1
2
3
(
)
(
)
(
)
.
.
.
(
)
c
o
u
n
t
r
y
t
o
t
a
l
N
C
f
c
x
c
x
c
x
c
x
(8)
Stage
3: I
m
peri
al
ist
Select
ion
Gen
e
rate
the
e
m
pire.
The
im
per
ia
li
st
will
be
sel
ect
ed
fro
m
the
country
with
the
highe
st
fitness
an
d
the o
t
her
s
r
em
ai
n
as c
olonies.
Stage
4: Assi
m
il
at
ion
Sele
ct
N=5
fro
m
the
m
os
t
powerfu
l
c
ount
ries
to
form
e
m
pires.
T
he
rem
ai
nin
g
co
untrie
s
or
c
olonies
sh
al
l be
d
esi
gnat
ed
to t
hese e
m
pires
resp
ect
i
vely
.
Stage
5: Rev
ol
ution
This
is
wh
ere
t
her
e
will
be
an
exch
a
ng
e
in
posit
ion
bet
wee
n
a
colon
y
an
d
the
I
m
per
ia
li
st.
A
colon
y
that
posses
s
a
bette
r
posit
ion
will
hav
e
the
l
ikeli
hood
t
o
ta
ke
cha
r
ge
of
the
em
pire
by
r
eplaci
ng
t
he
e
xisti
ng
Im
per
ia
li
st.
Stage
6: I
m
peri
al
ist
ic
Co
m
pet
it
ion
This
would
be
the
cr
ucial
sta
ge
,
w
her
e
by
al
l
i
m
per
ia
li
st
will
com
pete
to
ta
ke
c
on
tr
ol
of
e
ach
oth
e
r’s
colo
nies. T
he
total
pow
e
r of t
he
em
pire
will
be
c
om
pu
te
d
a
s foll
ow
s:
{}
n
i
m
p
e
r
i
a
l
i
s
t
c
o
l
o
n
i
e
s
o
f
e
m
p
i
r
e
p
o
w
e
r
f
m
e
a
n
f
(9)
wh
e
re
f
im
perialist
ind
ic
at
es
the
f
it
ness
of
eac
h
i
m
per
ia
li
st,
f
colo
nies
of
em
pire
rep
re
sents
the
fitnes
s
of
the
c
oloni
es
of
the
em
pire
a
nd
ε
=
0.05
ref
e
rs
to
a
I
C
fr
ee
pa
ram
et
er.
Em
pires
wh
ic
h
a
re
powe
rless
will
fall
in
the
i
m
per
ia
li
st
ic
com
petition
and
the
im
per
ia
li
st
with
the
hig
he
st
pow
er
will
be
ch
ose
n
as
the
sur
viv
in
g
i
m
per
ia
li
st.
Rep
eat
ste
p
5
if
the
highest
po
wer
c
on
sist
s
of
{}
i
m
p
e
r
i
a
l
i
s
t
c
o
l
o
n
i
e
s
o
f
e
m
p
i
r
e
f
m
e
a
n
f
.
The
so
l
ution
will
be
the
n st
or
e
d
i
n HNN.
2
.
5.
Gene
tic
a
l
go
ri
th
m
The
H
N
N
-
3SATG
A
(
G
A)
are
al
gorithm
s
for
optim
iz
at
i
on
a
nd
le
arn
i
ng
base
d
e
ntirel
y
on
a
few
bio
lo
gical
e
vo
l
ution feat
ur
es
[21
]
.
Th
is
alg
ori
th
m
r
eq
uires m
ai
nly 5
com
po
ne
nts:
A
m
et
ho
d o
f
e
ncodin
g sol
ution
s
to
the
pro
bl
e
m
o
n
c
hrom
os
om
es
An ev
al
uatio
n functi
on that
re
tur
ns
a
rati
ng for ea
ch
chrom
os
om
e g
ive
n
t
o i
t
A way
of
de
velop
i
ng the
popu
la
ti
on
of c
hro
m
os
o
m
es
Op
e
rato
rs
t
hat
m
ay
b
e ap
plied
to pare
nts
wh
e
n
they
re
produ
ce to m
od
ify
th
ei
r
ge
netic
com
po
sit
ion
.
Param
et
er s
et
tin
gs f
or the
alg
or
it
hm
, oper
at
or
s
and
oth
e
rs.
Ther
e
are
princ
ipall
y 5 s
ta
ge
s
in the p
ro
ces
s
of GA:
Stage
1: Init
ia
li
zat
ion
100
popula
ti
ons
of
c
hrom
os
om
es
rando
m
iz
ed
as
the
i
nterpr
et
at
ion
s
(
bit
str
ing
s
).
H
e
nce,
t
he
pro
bab
le
interp
retat
ion o
f
E
HNN
-
3SAT
w
il
l be
de
picte
d by the c
hrom
os
om
es.
Stage
2: Fitnes
s Ev
al
uatio
n
The
fitness
of
the
chrom
os
om
es
are
co
m
pu
te
d
bas
ed
on
the
nu
m
ber
of
sat
isfie
d
num
ber
of
neurons
in each
of t
he
e
xpos
it
io
n.
T
he t
rainin
g proces
s’
e
ff
ect
ive
nes
s w
il
l be
d
et
e
r
m
ined
by t
he m
axi
m
u
m
f
it
ness.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8938
IJ
-
AI
V
ol.
8
,
No.
2
,
June
20
1
9
:
1
4
4
–
1
55
148
Stage
3: Select
ion
Stage
10
pros
pect
ch
ro
m
os
om
es
out
of
the
10
0
th
at
are
no
te
d
to
hav
e
m
axi
m
um
fitness
will
adv
a
nce
to
the
f
ollow
i
ng
gen
e
rati
on
an
d
sta
ge
of
G
A.
Lat
er
on,
th
ese
chosen
c
hrom
os
om
es
will
carry
out
the
c
rosso
ve
r
proce
dure to a
m
pl
ify
the f
it
ne
ss and
va
riabi
li
ty
.
Stage
4: Cr
os
s
ov
e
r
The
c
ro
ss
over
op
e
rato
r
incl
udes
the
m
ai
n
trans
f
or
m
at
ion
ste
ps
in
t
he
H
NN
-
3SA
TG
A.
Inform
at
ion
exch
a
nge
will
occu
r
durin
g
this
ph
ase
be
tween
tw
o
su
b
-
str
uctu
re
of
the
chrom
os
om
es
(b
it
str
ing
s
)
.
The
c
rosso
ve
r
point
c
hrom
os
om
es
is
de
f
ined
m
utably
to
s
us
ta
in
t
he
ch
ro
m
os
ome
s
ge
netic
di
ver
sit
y.
Cros
s
over
nor
m
al
l
y i
ncr
ease
s the
num
ber
of sati
sfied
num
ber o
f neu
r
on
s
of the
ne
w
c
hrom
os
om
e p
ai
rs.
Stage
5: Mutat
ion
The
no
n
-
im
pro
vi
ng
i
nter
pr
et
at
ion
s
that
sti
ll
exists
will
be
enh
a
nce
d
by
m
uta
ti
on
.
T
he
m
uta
ti
on
in
GA
e
ntail
s
the
flipp
i
ng
of
t
he
sta
te
of
the
bit
string,
ei
th
er
f
ro
m
1
to
-
1
or
vice
vers
a.
This
will
resu
lt
in
a
bette
r
ch
r
om
os
om
e
being
generate
d
a
fter
m
utati
on
.
Since
th
e
m
utati
on
ha
s
occ
urred,
th
e
fitness
value
will
hav
e
to
be
re
-
eval
uated
for
the
new
ly
f
orm
ed
chrom
os
om
es.
The
sta
ges
s
hall
re
pe
at
from
sta
ge
1
if
the m
axi
m
u
m
f
it
ness
val
ue
is
no
t at
ta
ine
d.
3.
PERFO
R
MANC
E E
V
ALU
ATIO
N MAT
RI
X
In
this
pa
per
,
we
sh
al
l
util
iz
e
C+
+
to
run
th
e
3
-
S
AT
pro
ble
m
us
ing
the
a
bove
m
entione
d
al
gorithm
s.
In
orde
r
to
ve
rify
w
hich
al
gorithm
sh
al
l
ou
t
perform
the
rest,
a
serie
s
of
perform
a
nce
e
valuati
on
sh
al
l
determ
ine w
hi
ch would
b
e
th
e b
et
te
r
al
gorithm
to
be use
d.
3
.
1.
R
oot
mea
n sq
u
are
erro
r
Roo
t
Me
an
S
quare
E
rror,
or
com
m
on
ly
kn
own
as
RM
SE,
is
widely
us
ed
to
m
easur
e
the
diff
e
ren
ces
betwee
n
pr
e
di
ct
ed
values
of
a
m
od
el
an
d
t
he
act
ual
val
ue
s.
T
hey
s
ugge
ste
d
that
RM
SE
is
m
or
e
s
ui
ta
ble
to
represe
nt
m
od
el
per
f
orm
ance
wh
e
n
the
di
stribu
ti
on
of
the
e
rro
r
is
a
ntici
pated
to
be
Ga
us
sia
n.
RM
SE
is def
i
ned [
22]
as foll
ows:
2
m
a
x
1
1
()
n
i
i
R
M
S
E
f
f
n
(10)
3
.
2.
Me
an
abs
olut
e
error
The
m
easur
e
of
dif
fer
e
nces
be
tween
tw
o
co
ntinuo
us
va
ria
ble
is
known
a
s
the
Me
an
A
bsolute
Er
r
or
(MAE
).
Un
li
ke
RM
SE,
the
cal
culat
ion
of
MAE
is
relat
ively
si
m
ple,
w
her
e
by
the
m
agn
it
udes
a
re
s
umm
ed
(ab
s
olu
te
val
ue
s)
of
the
er
r
ors
to
at
ta
in
the
cum
ulativ
e
error
[
23]
.
Th
e
MAE
fo
rm
ula
is
descr
ibed
as
the
fo
ll
owin
g:
m
a
x
1
1
n
i
i
M
A
E
f
f
n
(11)
Si
m
il
arly
,
the
i
f
represe
nts
the
fitnes
s
va
lue
obser
ve
d,
wh
e
reas
the
m
a
x
f
will
be
t
he
m
axi
m
u
m
f
it
ness.
3
.
3.
Sum
of
s
qua
res
err
or
This
e
rror
is
to
m
easur
e
how
f
ar
t
he
data
are
from
the
assum
ed
value
s
of
the
m
od
el
.
A
su
m
of
sq
ua
res
(
SSE
)
m
ini
m
u
m
can
f
re
qu
e
ntly
be
searche
d
ve
ry
eff
ic
ie
ntly
via
app
li
cat
io
n
of
a
gen
e
rali
zat
ion
of
the
le
ast
sq
ua
re m
et
hod [24].
T
he
ge
ner
at
e
d
the
for
m
ula o
f
SSE as
belo
w:
2
m
a
x
1
()
n
i
i
S
S
E
f
f
(12)
The
i
f
re
pr
ese
nts
the
fitness
val
ue obse
r
ved,
w
her
eas
the
m
a
x
f
will
be
the
m
axi
m
um
f
it
ness
3
.
4.
Schw
arx
bayesia
n crite
ri
on
Schwarz
Ba
ye
sia
n
Crit
eri
on
(S
BC
)
,
or
al
so
cal
le
d
Ba
ye
sia
n
I
nfo
rm
ation
C
rite
rio
n
(BIC)
,
is
a
crit
erion
for
se
le
ct
ing
m
od
el
s
fr
om
a
finite
s
et
of
m
od
el
s
[25].
Wh
en
m
odel
s
are
fitt
ed,
it
is
l
ikely
to
increase
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
IS
S
N:
22
52
-
8938
Hyb
ri
d
i
mp
eri
alist
ic
co
m
petit
iv
e a
lg
or
it
hm i
n
co
rpor
ated
wi
th ho
pfiel
d
neural… (
Vign
e
shw
er Kathirvel
)
149
wh
e
n
the
pa
ra
m
et
ers
are
added.
T
hey
hav
e
ta
ken
the
m
od
el
com
par
iso
n
ap
proac
h
to
analy
sing
sta
ti
sti
cal
ly
[26]
. Ass
um
in
g
that t
he dist
urba
nces
or m
od
el
er
rors
a
re i
nd
e
p
e
nd
e
nt a
nd
distrib
uted
e
qu
al
ly
in
acc
or
dan
ce
to
a
norm
al
distri
bu
ti
on
a
nd
that
the
co
ndit
ion
of
t
he
bo
unda
r
y
that
the
log’s
der
i
vative
wit
h
res
pect
to
t
he
true
var
ia
nce is ze
r
o,
t
he fo
rm
ula g
ene
rated
w
il
l
be:
.
l
n
(
)
.
l
n
(
)
S
B
C
n
M
S
E
p
a
n
(13)
The
n
represe
nts
the
num
ber
of
s
olu
ti
ons
,
pa
w
ould
be
the
nu
m
ber
of
par
am
et
ers
and
M
SE
is
the
m
ean
sq
ua
re e
rro
r.
3
.
5.
C
omp
u
t
ati
on
t
im
e
Com
pu
ta
ti
on
ti
m
e
or
widely
known
as
CP
U
tim
e
(also
cal
l
ed
run
ning
tim
e)
is
the
m
eas
ur
e
of
tim
e
require
d
to
pe
r
form
a
co
m
pu
ta
ti
on
al
pr
ocess
[27]
.
It
is
pro
portio
nal
to
th
e
nu
m
ber
of
r
ule
ap
plica
ti
ons
and
the co
m
pu
ta
ti
on ti
m
e is
m
eas
ur
e
d usin
g sec
onds
.
Com
pu
ta
ti
on
ti
m
e is defi
ned
by the
foll
ow
i
ng equati
on.
_
(
)
_
(
)
_
(
)
C
P
U
T
i
m
e
s
T
r
a
i
n
i
n
g
T
i
m
e
s
R
e
t
r
i
e
v
a
l
T
i
m
e
s
(14)
3
.
6.
Glob
al
mi
nima r
at
i
o
The
gl
ob
al
m
i
nim
a
rat
io
is
def
ine
d
by
the
rati
o
of
the
cu
m
ula
ti
ve
glo
ba
l
m
ini
m
u
m
ener
gy
to
the
nu
m
ber
of
tria
ls
or
nu
m
ber
of
neur
on
s
[
28]
.
It
will
be
fe
asi
ble
to
in
dicat
e
the
si
m
ulatio
n
by
chec
kin
g
t
he
global
m
ini
ma
rati
o
si
nce
the
pr
ogra
m
is
seekin
g
f
or
e
ve
ry
ne
uron’s
st
at
e
global
m
ini
m
u
m
energy i
n
3
-
S
A
T
[29]
.
T
he
e
qu
at
ion
of g
l
ob
al
m
ini
m
a rati
o
is give
n by the
foll
ow
i
ng.
m
in
1
1
(
)
(
)
n
E
i
G
lo
b
a
l
M
in
im
a
R
a
tio
N
N
Tr
C
O
M
B
M
A
X
(15)
NTr
ind
ic
at
es
the
nu
m
ber
of
tria
ls,
m
i
n
E
N
is
the
glo
bal
m
ini
m
u
m
so
luti
ons
an
d
COMBMAX
is
the
neuron’s
m
axi
m
u
m
co
m
bin
at
io
n.
4.
METHO
DOL
OGY
/I
MPLE
MENT
ATIO
N
HNN
–
3SAT
ES,
H
N
N
–
3SAT
GA
a
nd
HNN
–
3SAT
IC
A
will
be
pro
po
se
d
to
e
valuate
t
he
perform
ance. The im
ple
m
ent
at
ion
of these
m
od
el
s to
the
3
-
SAT
prob
le
m
w
il
l be acc
ordi
ng to
t
he
flo
w char
t
:
5.
RESU
LT
S
A
ND AN
ALYSIS
The
sim
ulati
on
s
of
the
3
-
S
AT
pr
ob
le
m
us
ing
var
i
ous
al
gorithm
s
was
cond
ucted
with
an
I
ntel®
Core
™
i5
-
72
00U
@
2.50GH
z
,
4.0
0G
B
R
AM
ASUS
A
556U
Se
ries,
via
DEV
C+
+.
F
r
om
the
trai
ning
ph
ase
il
lustrate
d
in
Figure
2
,
ob
se
rv
e
d
that
after
12
num
ber
of
neur
on
s
,
the
HN
N
-
3SATE
S
al
go
rithm
pr
od
uce
the
hi
ghest
num
ber
of
er
rors.
It
s
howe
d
a
cl
os
e
pro
portio
na
te
relat
ion
s
hi
p
betwee
n
t
he
nu
m
ber
of
num
ber
of
neur
on
s
a
nd
er
rors
obta
ine
d.
This
sim
i
la
r
pa
tt
ern
was
al
s
o
portrayed
at
t
he
te
sti
ng
phase
for
ES
i
n
Fig
ur
e
3
,
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8938
IJ
-
AI
V
ol.
8
,
No.
2
,
June
20
1
9
:
1
4
4
–
1
55
150
bu
t
t
her
e
we
re
so
m
e
m
il
d
fluctuati
ons
i
n
t
he
e
rror
rea
dings.
H
NN
-
3SA
TGA,
on
the
ot
her
ha
nd,
s
howed
a
m
or
e con
sist
e
nt
incr
ease
as th
e num
ber
of
num
ber
o
f ne
uro
ns
go
t
h
i
gh
e
r.
Howe
ver,
this
patte
rn
is
not
the
sam
e
in
th
e
te
sti
ng
ph
as
e.
As
obser
ve
d,
there
is
qu
it
e
a
sign
i
ficant
var
ia
ti
on
of
t
he
errors
f
or
the
first
5
to
6
nu
m
ber
of
ne
uro
ns
,
a
nd
e
ven
t
ua
ll
y
beco
m
es
con
sist
e
nt
an
d
c
losin
g
to
the
perf
orm
ance
of
the
ES
al
go
rit
hm
.
HN
N
-
3SA
TICA
al
gorit
hm
see
m
ed
a
li
ttle
m
or
e
com
p
elling
com
par
ed
t
o
t
he
oth
e
r
2
al
gorithm
s.
In
the
trai
ning
phase,
it
s
howe
d
a
n
a
bsolute
0
er
ror
th
r
ough
ou
t
the
nu
m
ber
of
neur
on
s
,
an
d
a
ver
y
m
ini
m
al
error
tre
nd
on
the
te
sti
ng
pha
se
in
com
par
ison
with
ES
a
nd
G
A.
On
t
he
trai
ning
phase
,
it
is
ob
s
er
ved
that
HNN
-
3SATE
S
al
gorithm
had
a
huge
valu
e
unti
l
the
val
ue
of
NN
=
108.
The
po
ssi
bili
ty
of
ob
ta
inin
g
the
cor
rect
inter
pret
at
ion
s
will
reduce
drast
ic
al
ly
wh
en
ther
e
is
an
increase
in
the
nu
m
ber
of
ne
uro
ns
.
This
is
du
e
to
it
s
natu
re
that
kicks
off
the
“ge
ne
rate
and
te
st”
and
“t
rial
and er
ror” tec
hniq
ue
in
or
der
t
o
at
ta
in t
he
c
orrect sol
utio
ns
i
n
a
sp
eci
fic
sea
rch sp
ace
.
Figure
2
.
RM
S
E ev
al
uation o
f
HNN
-
3SA
TE
S, HNN
-
3SAT
GA &
H
NN
-
3SAT
ICA
t
raini
ng phase
Figure
3
.
RM
S
E ev
al
uation o
f
HNN
-
3SA
TE
S, HNN
-
3SAT
GA &
H
NN
-
3SAT
ICA
-
te
sti
ng
ph
a
se
S
how
n
in
Fig
ure
4
,
t
he
E
S
al
gorithm
again
portrayed
the
highest
num
ber
of
er
r
or
s
,
f
ollow
e
d
by
the
GA
al
go
rithm
.
The
IC
al
gorit
hm
disp
la
ye
d
abso
l
utely
no
er
rors
thr
ou
ghou
t
the
run.
H
ow
ever,
this
is
dif
fer
e
nt
in
the
te
sti
ng
phase
as
s
how
n
in
Fig
ur
e
5
,
w
her
e
by
al
l
3
al
gorithm
s
exh
ib
it
a
ver
y
si
m
il
a
r
tren
d
am
on
gst
each
oth
e
r.
IC
al
go
rithm
sh
ow
s
a
dr
ast
ic
undul
at
ion
on
an
i
ncr
easi
ng
m
ann
er
th
r
oughou
t
the
12
num
ber
of
neur
on
s
,
but
m
os
tl
y
hav
i
ng
a
n
er
ror
lo
wer
to
that
of
GA
a
n
d
ES,
w
her
e
t
his
2
s
how
s
a
m
or
e
ste
ady
increase
.
As
plo
tt
ed
,
the
ES
again
sho
ws
the
highest
nu
m
ber
of
er
r
or
s
produce
d,
reachi
ng
sli
ghtl
y
m
or
e
than
12
00
in
the trainin
g phase
and abo
ut 3
60
0000 in
the
test
ing
phase
, a
s por
tray
e
d
in
Figure
6
a
nd F
igure
7
. r
es
pect
iv
el
y
.
GA
is
sig
nifica
ntly
lower
in
t
he
trai
ni
ng
pha
se,
al
m
os
t
3
ti
m
es
lower
t
ha
n
ES
,
bu
t
s
ho
ws
a
ver
y
hom
og
e
nous
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
IS
S
N:
22
52
-
8938
Hyb
ri
d
i
mp
eri
alist
ic
co
m
petit
iv
e a
lg
or
it
hm i
n
co
rpor
ated
wi
th ho
pfiel
d
neural… (
Vign
e
shw
er Kathirvel
)
151
resu
lt
to
ES.
IC,
oth
e
rw
is
e,
pro
ves
to
be
a
m
or
e
st
able
run
w
he
n
it
pr
od
uces
0
error
s
a
ga
in
in
the
trai
ni
ng
ph
ase,
an
d
al
m
os
t
100
ti
m
es
le
s
ser
c
om
par
ed
t
o
ES
an
d
GA
in
the
te
sti
ng
phase.
From
this,
it
can
be
de
duced
t
ha
t
the
error
s
a
ccum
ulate
d
wh
en
t
he
3
-
S
A
T
was
trai
ne
d
and
te
ste
d
wit
h
G
A
an
d
E
S.
This
is
diff
e
re
nt
f
or
I
C
wh
e
re
it
wa
s
su
sta
ini
ng
a
ve
ry
low
val
ue
of
e
rro
rs.
In
ot
her
w
ords,
the
com
p
il
at
ion
of
errors
was
le
sse
r
with
the
inc
rease
of
N
N.
Com
par
ed
to
ES
a
nd
GA,
the
IC
al
gorithm
entai
le
d
le
sse
r
it
erati
on
s
t
o
com
pu
te
the
desire
d
so
l
ution
s
.
ES
in
t
hi
s
case
need
e
d
m
or
e
it
eratio
ns
t
o
at
ta
in
a
global
so
lut
ion
,
a
s
the
inter
pr
et
at
ion
will
disin
te
gr
at
e
if
it
do
e
s
not
su
ccee
d
t
o
achieve
the
m
axim
u
m
fitness.
T
his
will
the
n
le
ad
to the h
un
ti
ng
of n
e
w
i
nterpre
ta
ti
on
w
it
h a
ne
w fit
ness valu
e.
Figure
4
.
MAE
ev
al
uati
on of
HNN
-
3S
ATE
S
, HN
N
-
3SAT
G
A
&
HN
N
-
3SATIC
A
-
t
rainin
g ph
a
se
Figure
5
.
MAE
ev
al
ua
ti
on of
HNN
-
3S
ATE
S
, HN
N
-
3SAT
G
A
&
HN
N
-
3SATIC
A
-
te
sti
ng
phase
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2252
-
8938
IJ
-
AI
V
ol.
8
,
No.
2
,
June
20
1
9
:
1
4
4
–
1
55
152
Figure
6
.
SS
E
evaluati
on
of
HNN
-
3S
ATE
S
, HN
N
-
3SAT
G
A
&
HN
N
-
3SATIC
A
-
trai
nin
g p
hase
Figure
7
.
SS
E
evaluati
on
of
HNN
-
3SATE
S
, HN
N
-
3SAT
G
A
&
HN
N
-
3SATIC
A
-
te
sti
ng
ph
a
se
Fr
om
Fig
ur
e
8
,
w
he
n
t
he
num
ber
of
ne
uro
ns
is
1,
ES
a
nd
IC
wer
e
abl
e
to
at
ta
in
a
glo
bal
m
ini
m
a
rati
o
of
1,
e
xce
pt
f
or
GA.
As
the
nu
m
ber
of
nu
m
ber
of
neurons
i
ncr
ea
ses
ti
ll
12
,
it
ca
n
be
seen
that
t
he
rati
o
gr
a
dual
ly
decre
ases
f
or
E
S
a
nd
G
A.
On
th
e
con
t
rar
y,
e
ve
n
with
a
re
duc
ing
ra
ti
o,
th
e
I
C
al
go
rit
hm
ten
ds
to
m
ake
an
at
te
m
pt
to
at
ta
in
a
r
at
io
of
1,
or
at
le
ast
cl
os
e
to
it
.
This
ca
n
be
s
een
al
on
g
the
r
un
that
the
rea
dings
are
act
ually
var
yi
ng.
At
the
12th
num
ber
of
n
eu
r
on
s
,
ES,
G
A
an
d
IC
are
appr
ox
im
at
ely
a
t
the
sa
m
e
value,
bu
t
IC
ha
ving
a
sli
gh
tl
y
hi
gh
e
r
ra
ti
o
in
c
om
par
ison
with
the
ot
her
2
al
gorith
m
s.
In
IC
al
gorithm
,
it
has
a
highe
r
eff
ic
ie
ncy
to
neur
o
-
sea
rch
i
ng
m
e
tho
d
generate
d
in
com
par
is
on
with
ES
and
GA.
The
net
work
becam
e
sign
ific
a
ntly
beca
m
e
m
or
e
c
om
plica
te
d
as
the
nu
m
ber
o
f
neu
r
ons
incre
ased,
as
the
siz
e
of
the
restri
ct
ions
becam
e
la
rg
er in
de
finite
ly
.
The
syst
e
m
was
al
so
able
to
cat
egorize
feasibl
e
so
luti
ons
co
nst
ru
ct
ively
and
adap
t
with m
or
e c
onstrai
nts.
From
t
he gra
ph
disp
la
ye
d
in
Fig
ur
e
9
, th
e
ES al
gorithm
sh
ows the
low
e
st
SBC v
al
ue
a
t
the
1s
t
num
ber
of
ne
uro
n,
w
hi
ch
is
at
the
ne
gative
value,
w
her
eas
IC
is
sli
gh
tl
y
bel
ow
0,
wh
il
e
GA
is
al
ready
at
20000.
T
he
HNN
-
3SAT
IC
A
is
m
or
e
ef
fe
ct
ive
in
see
king
f
or
the
c
orre
ct
interp
retat
io
n
with
out
the
us
e
of
m
or
e
it
erati
on
s.
HNN
-
3SAT
E
S
on
the o
the
r
hand
pe
naliz
ed
the
SBC
values
du
e
to
the
c
ollec
ti
on
of
MSE
a
nd
retrieval
pha
se.
T
his
ex
plains
t
he
higher
val
ues
o
f
SBC
i
n
HNN
-
3SA
TES
c
om
par
ed
to
the o
t
her tw
o
a
lgorit
hm
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
IJ
-
AI
IS
S
N:
22
52
-
8938
Hyb
ri
d
i
mp
eri
alist
ic
co
m
petit
iv
e a
lg
or
it
hm i
n
co
rpor
ated
wi
th ho
pfiel
d
neural… (
Vign
e
shw
er Kathirvel
)
153
Figure
8. Gl
obal
Mi
ni
m
a Rat
i
o of H
NN
-
3SA
TES, H
NN
-
3S
ATGA
&
HNN
-
3SA
TIC
A
Figure
9. SBC
of HNN
-
3SAT
ES, HN
N
-
3SA
TGA &
HNN
-
3SATIC
A
The
sp
ee
d
of
t
he
E
S
a
nd
GA
al
gorithm
are
nota
bly
sim
ilar
in
the
12
di
ff
e
ren
t
num
ber
of
neur
ons
sh
ow
n
in
Fi
gure
10
.
It
show
s
that
wh
e
n
the
se
al
gorithm
s
are
util
iz
ed
f
or
m
or
e
num
ber
of
neur
on
s
,
the
tim
e
ta
ken
will
gra
du
al
ly
inc
reas
e
as
w
hat
is
sh
ow
n
in
the
gr
a
ph.
T
he
ti
m
e
ta
ken
for
the
IC
al
gorithm
is
i
m
pr
essivel
y
low,
eve
n
at
the
12
th
nu
m
ber
of
ne
uro
ns
,
m
ark
in
g
at
alm
os
t
on
ly
24
seco
nd
s
.
The
powe
r
of
the
al
gorithm
i
s
basical
ly
dem
on
strat
ed
by
the
eff
ect
ive
ne
ss
of
the
com
pu
ta
ti
on
pr
ocess
in
whole.
The
refor
e
,
par
ti
cula
r
acce
le
rati
ng
syst
e
m
is
req
uire
d
for
the
trai
ning
proc
ess.
T
he
tim
e
of
com
pu
ta
ti
on
was
hi
gh
e
r
f
or
HNN
-
3SAT
IC
A
du
e
to
t
he
e
f
fecti
ven
es
s
of
t
he
im
per
ia
li
st
t
o
im
pr
ove
to
w
ard
s
the
prefe
r
red
so
l
utions.
GA
is
com
par
at
ively
si
m
i
la
r,
ye
t
i
n
real
is
sti
ll
lowe
r
than
E
S
,
as
the
cr
os
s
ov
e
r
an
d
m
utati
on
proce
sse
s
hav
e
the
abili
ty
to
i
m
pr
ove
the
inter
pr
et
at
io
ns
.
The
ES
brut
e
f
or
ce
nee
de
d
m
uch
m
or
e
com
p
utati
on
tim
e
as
the “g
e
ne
rate a
nd test
” syst
em
r
e
qu
ire
d
m
or
e
tim
e to att
ai
n
the c
orrect i
nter
pr
et
at
io
ns
.
Evaluation Warning : The document was created with Spire.PDF for Python.