Int
ern
at
i
onal
Journ
al
of El
e
ctrical
an
d
Co
mput
er
En
gin
eeri
ng
(IJ
E
C
E)
Vo
l.
11
,
No.
1
,
Febr
uar
y
2021
, pp.
71
6
~
727
IS
S
N: 20
88
-
8708
,
DOI: 10
.11
591/
ijece
.
v11
i
1
.
pp
716
-
727
716
Journ
al h
om
e
page
:
http:
//
ij
ece.i
aesc
or
e.c
om
Stoch
astic loc
al sear
ch:
a
state
-
of
-
t
he
-
art r
evie
w
Muhame
t
K
astra
ti,
Ma
re
ng
l
en Bi
ba
Depa
rtment
o
f
C
om
pute
r
Scie
n
ce,
Univer
si
t
y
of
N
ew
York Tirana,
Albania
Art
ic
le
In
f
o
ABSTR
A
CT
Art
ic
le
history:
Re
cei
ved
N
ov
4
, 2
01
9
Re
vised
Ju
l
1
2
,
2020
Accepte
d
J
ul
27, 2
020
The
m
ai
n
objec
ti
ve
of
thi
s
pap
er
is
to
provide
a
stat
e
-
of
-
the
-
a
rt
rev
ie
w
,
ana
l
y
z
e
and
disc
uss
stocha
stic
lo
ca
l
se
arc
h
te
chn
i
ques
used
for
solving
har
d
combinat
ori
al
pr
oble
m
s.
It
begi
n
s
with
a
short
in
troduc
ti
on
,
m
otivati
on
and
som
e
basic
nota
t
ion
on
combin
at
oria
l
probl
ems
,
sea
rch
par
ad
igms
and
othe
r
rel
ev
ant
f
ea
tur
e
s
of
sea
rch
ing
te
chn
ique
s
as
nee
d
ed
for
b
ac
kground
.
In
the
foll
owin
g
a
brie
f
ov
erv
i
ew
of
the
stoch
asti
c
lo
cal
sea
r
c
h
m
et
hods
al
ong
with
an
ana
l
y
sis
of
t
he
stat
e
-
of
-
the
-
art
stocha
st
ic
l
oca
l
se
arc
h
al
gorit
hm
s
is
g
i
ven.
Fin
al
l
y
,
the
la
st
par
t
of
th
e
pape
r
pre
sen
t
a
nd
discuss
som
e
of
the
m
ost
la
t
est
tre
nds
in
applic
at
ion
of
stocha
sti
c
l
oca
l
se
arch
al
gorit
hm
s
in
ma
chi
n
e
le
arn
ing,
dat
a
m
ini
ng
and
som
e
othe
r
are
as
of
scie
nc
e
and
engi
n
ee
r
in
g.
W
e
con
cl
ud
e
with
a
disc
uss
ion
on
ca
pa
bil
ities
and
li
m
it
ations of
sto
cha
sti
c
lo
cal
se
a
rch
a
lgori
thms
.
Ke
yw
or
d
s
:
An
t c
olony
op
t
i
m
iz
ation
Ev
olu
ti
onary a
lgorit
hm
s
Gr
ee
dy r
a
ndom
iz
e adap
ti
ve
Iterate
d
l
ocal s
earch
Stoch
a
sti
c loca
l search
This
is an
open
acc
ess arti
cl
e
un
der
the
CC
B
Y
-
SA
l
ic
ense
.
Corres
pond
in
g
Aut
h
or
:
Muh
am
et
K
ast
rati
,
Dep
a
rtm
ent o
f C
om
pu
te
r
Scie
nce,
Un
i
ver
sit
y o
f New
York
Tira
na,
”Kod
ra e Die
ll
it
”,
Tirana
, Alb
ania.
Em
a
il
:
m
uh
a
m
et
.k
ast
rati
@
gma
il
.co
m
1.
INTROD
U
CTION
Re
s
earche
rs
a
r
e
strug
gling
to
ta
ckle
com
bin
at
ori
al
pro
ble
m
s
as
the
num
ber
of
t
hese
is
pr
es
ent
i
n
sever
al
br
a
nc
he
s
in
sci
ence
and
i
ndus
try
on
w
hich
c
om
pu
ta
ti
on
al
m
et
hods
ha
ve
be
en
wi
dely
ap
plied,
includi
ng
arti
f
ic
ia
l
intel
li
gen
ce,
m
achine
le
arn
i
ng,
data
m
ining
,
operat
ion
s
resea
rch,
bio
-
in
form
at
i
cs
an
d
el
ect
ro
nic
c
omm
erce.
So
m
e
of
the
m
os
t
w
el
l
-
known
c
om
bin
at
or
ia
l
prob
le
m
s
include
find
i
ng
shor
te
st
o
r
cheap
e
st
paths
in
gr
a
phs
,
find
i
ng
s
olu
ti
ons
for
pr
opos
it
ion
al
form
ulae
(S
A
T),
sche
duli
ng,
tim
e
-
ta
blin
g,
plan
ning,
opti
m
iz
ing
resour
c
e
al
locat
ion
an
d
oth
e
r
var
i
ou
s
do
m
ai
ns
[1
]
.
Stoch
ast
ic
local
search
(S
LS
)
ha
s
pro
ved
to
be
su
ccess
fu
ll
y
and
e
xtensi
ve
ly
us
ed
ap
proach
for
s
olvi
ng
c
om
bin
at
or
ia
l
hard
pro
blem
s
.
SLS
al
gorithm
s
us
e
ra
ndom
i
zat
ion
m
et
ho
d
durin
g
the
ge
ner
at
io
n
or
sel
ect
ion
of
ca
ndidate
so
lut
io
ns
[1
]
.
These
al
gorith
m
s
are
com
mo
nly
us
e
d
f
or
so
lvi
ng
hard
c
om
bin
at
or
ia
l
optim
iz
at
ion
and
decisi
on
pr
oble
m
s.
As
presente
d
by
[1
]
so
m
e
of
the
early
and
suc
cessf
ully
app
li
ed
exam
ples
of
SLS
te
c
hniq
ue
s
us
ed
f
or
s
ol
ving
op
ti
m
iz
ation
pro
blem
s
includes
the
L
in
-
K
ern
i
gh
a
n
al
gor
it
h
m
,
wh
ic
h
is
ap
plied
f
or
s
olv
in
g
t
he
tra
velli
ng
sal
es
m
an
prob
l
e
m
(TSP
)
[
2],
evo
l
ution
a
ry
al
gorithm
s
[3
]
and
sim
ulate
d
ann
eal
in
g
[
4]
as
well
as
so
m
e
oth
e
r
su
ccess
fu
l
e
xa
m
ples
of
us
ing
these
al
gorith
m
s
fo
r
s
olv
in
g
NP
-
c
om
plete
decisi
on
pro
bl
e
m
s,
including
gr
a
ph
colo
rin
g
pro
ble
m
(G
CP
)
[5
]
,
the
Sati
sfiabili
ty
pr
oble
m
i
n
pro
posit
ion
al
log
ic
(SAT)
[
6,
7].
Mo
re
rec
ently
,
SLS
al
gorithm
s
hav
e
s
uccess
fu
ll
y
been
a
pp
li
ed
for
pro
ble
m
-
so
lving
a
nd
fo
r
m
od
el
in
g
in
var
i
ous
are
as
of
sci
entifi
c an
d
e
ng
i
neer
i
ng problem
s.
The
w
orkin
g
pr
i
nciple
in
tradit
ion
al
local
search
al
gorit
hm
s
wh
en
de
al
ing
with
an
instance
of
a
com
bin
at
or
ia
l
prob
le
m
rely
on
the
fact
that
searc
h
f
or
s
olu
ti
ons
ha
ppe
n
in
the
s
pace
of
can
did
at
e
s
olut
ion
s.
Additi
on
al
ly
,
the
local
sear
ch
proce
ss
sta
rts
by
m
ov
ing
from
cur
r
ent
so
luti
on
t
o
ot
her
s
olu
ti
on
i
n
a
neighb
orh
ood
sp
ace
of
the
cand
i
date
so
lut
ion
s
,
an
d
decis
ion
on
each
se
arch
ste
p
has
been
m
ade
by
us
in
g
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Sto
c
hasti
c local
sea
rc
h: A st
ate
-
of
-
t
he
-
ar
t rev
ie
w
(
Muham
et
Kastrati
)
717
on
ly
existi
ng
inf
or
m
at
ion
[1
]
.
On
the
ot
her
side,
SLS
al
go
rithm
s,
wh
en
use
d
f
or
so
l
ving
har
d
c
om
bin
at
or
ia
l
pro
blem
s
are
char
act
erize
d
by
ran
dom
iz
ation
,
res
pecti
vely
,
gen
e
rati
on
of
i
niti
al
so
luti
on
s
as
well
as
decisi
on
s
ta
ken
on
eac
h
ste
p
is
ty
pical
ly
base
d
on
ra
ndom
iz
at
ion
.
F
urt
her
m
or
e,
som
e
SLS
m
et
ho
ds
durin
g
t
he
s
ear
c
h
process
us
e
s
om
e
kin
d
of
m
e
m
or
y
te
chn
i
ques
in
orde
r
to
kee
p
a
li
m
i
t
ed
num
ber
of
m
os
t
recently
visit
ed
cand
i
date
so
l
ut
ion
s
[1
]
.
As
m
entioned
ab
ove
se
ver
al
S
L
S
al
gorithm
s
hav
e
bee
n
e
ffec
ti
ve
ap
proac
h
us
ed
m
ai
nly
to
so
lve
hard
com
bina
torial
pro
blem
s,
including
SA
T,
M
AX
-
S
AT,
TS
P,
Sc
he
du
li
ng,
Tim
e
-
Tabli
ng
and
P
r
otein
F
old
in
g.
S
om
e
oth
e
r
us
e
fu
l
pr
ogress
re
ga
rd
i
ng
to
S
LS
al
gorithm
s
fo
r
S
AT
is
done,
c
ov
e
rin
g
al
gorithm
s
of
t
he
GSAT,
G
W
SA
T,
r
obust
S
LS
al
go
rithm
s
for
M
AX
-
S
AT
(
WalkS
AT
,
I
RoTS,
G
LSS
A
T
an
d
SA
P
S)
a
nd
othe
r hyb
ri
d
S
LS
m
et
ho
ds
[
1].
SLS
al
gorithm
eng
i
neer
i
ng
i
nvolv
e
s
m
any
chall
enges
that
the
researc
h
com
m
un
it
y
mu
st
ad
dr
es
s.
They
are
a
ve
ry
stron
g
s
upporter
of
m
any
e
ff
ect
ive
he
uri
sti
cs
fo
r
s
ol
vin
g
ha
rd
com
pu
ta
ti
on
al
ly
co
m
plex
pro
blem
s.
Gen
erall
y,
eng
inee
rin
g
of
the
se
ty
pes
of
al
go
rithm
s
in
m
anu
al
way
requires
a
ver
y
caref
ul
and
t
o
m
u
c
h
e
f
f
o
r
t
i
n
o
r
d
e
r
t
o
r
e
a
c
h
h
i
g
h
a
n
d
a
c
c
e
p
t
a
b
l
e
p
e
r
f
o
r
m
a
n
c
e
.
O
n
t
h
e
o
t
h
e
r
s
i
d
e
,
a
u
t
o
m
a
t
ic
a
l
g
o
r
i
t
h
m
c
o
n
f
i
g
u
r
a
t
i
o
n
i
s
a
n
i
m
p
r
e
s
s
i
v
e
a
p
p
r
o
a
c
h
i
n
t
r
o
d
u
c
e
d
t
o
i
m
p
r
o
v
e
t
h
e
p
e
r
f
o
r
m
a
n
c
e
o
f
a
l
g
o
r
i
t
h
m
s
f
o
r
c
om
pu
ta
ti
on
al
ly
hard
pro
blem
s.
A
lot
of
w
ork
r
el
at
ed
to
al
gorithm
eng
ineerin
g
has
bee
n
done
by
researc
h
c
omm
un
it
y
wit
h
re
gard
on
autom
at
ic
a
lgorit
hm
con
fig
urat
ion
an
d
pa
ra
m
et
er
tun
ing
t
echn
i
qu
e
s.
E
xam
ples
of
thes
e
include
Pa
ra
m
ILS
al
gorithm
introdu
ce
d
by
H
utter
et
al
.
[8
]
,
aut
om
at
ed
al
go
rit
hm
s
con
fi
gurat
ion
an
d
param
et
er
tu
ning
pr
e
sented
b
y
H
o
o
s
i
n
[
9
]
,
a
u
t
o
m
a
t
i
c
d
e
s
i
g
n
o
f
h
y
b
r
i
d
S
L
S
a
l
g
o
r
i
t
h
m
s
b
y
M
a
r
m
i
o
n
e
t
a
l
.
,
[
1
0
]
,
g
r
a
m
m
a
r
-
b
a
s
e
d
ge
ner
at
i
on
of
SLS
heurist
ic
s
by
Ma
sci
a
et
al
.
[11
]
, p
erm
utati
on
fl
ow
s
hop
p
r
oble
m
s
by
Pagnozzi
a
nd
Stützl
e
[12],
al
go
rithm
c
o
m
p
a
r
i
s
o
n
b
y
a
u
t
o
m
a
t
i
c
a
l
l
y
c
o
n
f
i
g
u
r
a
b
l
e
S
L
S
f
r
a
m
e
w
o
r
k
s
b
y
M
a
s
c
i
a
e
t
a
l
.
,
[
1
3
]
,
r
e
v
i
s
i
t
i
n
g
s
i
m
u
l
a
t
e
d
a
n
n
e
a
l
i
n
g
b
y
F
r
a
n
z
i
n
a
n
d
S
t
ü
t
z
l
e
[
1
4
]
.
O
n
t
h
e
o
t
h
e
r
s
i
d
e
,
F
r
a
n
z
i
n
e
t
a
l
.
,
[
1
5
]
s
t
u
d
i
e
d
t
h
e
e
f
f
e
c
t
o
f
t
r
a
n
s
f
o
r
m
a
t
i
o
n
s
f
o
r
n
u
m
e
r
i
c
a
l
p
a
r
a
m
e
t
e
r
s
,
F
r
a
n
z
i
n
a
n
d
S
t
ü
t
z
l
e
[
1
6
]
p
e
r
f
o
r
m
e
d
c
o
m
p
a
r
i
s
o
n
o
f
a
c
c
e
p
t
a
n
c
e
crit
eri
a
in
ra
ndom
iz
e
d
local
searche
s,
M
u
e
t al
.
,
[
17
]
c
onduct
ed
a
r
ese
arc
h on the im
pact o
f
au
t
om
at
ed
al
gorithm
co
nfi
gurati
on etc.
In
t
he
la
st
ye
ar
s
,
to
m
uch
e
ffo
rt
has
bee
n
m
ade
by
resea
rch
com
m
un
it
y
rel
at
ing
to
the
a
ppli
cat
ion
of
SLS
in
var
i
ous
dom
ai
ns
of
sci
ence
a
nd
eng
i
neer
i
ng.
So
m
e
of
the
m
os
t
su
ccessfu
l
exam
ples
of
S
LS
app
li
cat
io
n
inc
lud
es:
c
om
bin
at
or
ia
l
pro
blem
s
(S
AT
a
nd
TSP),
arti
fici
al
intel
li
gen
ce,
m
a
chine
le
arn
i
ng
a
nd
data
m
ining
,
s
cheduli
ng,
ti
m
e
-
ta
blin
g,
prot
ei
n
fo
l
ding
an
d
to
m
any
oth
er
areas.
I
n
thi
s
pap
e
r
we
ha
ve
bee
n
m
uch
m
or
e
focuse
d
on
the
r
esearch
wor
k
cond
ucted
in
the
la
st
five
ye
ars
in
the
area
of
SLS
.
W
e
will
sta
rt
with
the
work
of
Z
hou
a
n
d
H
u
[
18
]
,
who
presented
t
he
gra
dient
-
base
d
ad
aptive
stoc
hast
ic
search
(
G
A
SS)
for
n
on
-
dif
fer
e
ntia
ble
opti
m
iz
at
i
on
a
nd
la
tt
er
Rosin
[19],
w
ho
presente
d
how
the
un
we
igh
te
d
SLS
c
a
n
b
e
eff
e
ct
ive
f
or
ra
ndom
CSP
be
nch
m
ark
s
,
D
r
ugan
[
20]
,
in
w
hich
was
prese
nted
st
oc
hastic
Paret
o
local
s
earch
for
m
any o
bj
ec
ti
ve
qu
a
drat
ic
assign
m
ent p
r
oble
m
instances
,
Froh
li
ch
et
al
.
,
[
21
]
pr
ese
nte
d
ap
plica
ti
on
of SLS
for
sat
isfia
bili
t
y
m
od
ul
o
the
or
ie
s
.
Anothe
r
si
gn
i
fic
ant
co
ntri
bu
ti
on
to
t
he
fiel
d
include
s
a
pa
pe
r
by
Bo
ughac
i
et
al
.
,
[
22
]
wi
th
re
gard
to
the
app
li
cat
io
n
of
SLS
for
i
m
age
ste
ga
nogra
ph
y,
Wang
et
al
.
,
[
23]
on
est
im
at
ion
of
the
di
stribu
ti
on
al
go
rithm
with
a
SL
S
f
or
un
ce
rtai
n
ca
pa
ci
ta
te
d
arc
r
outi
ng
pro
blem
s
,
f
ollow
e
d
by
Re
zoug
a
nd
B
oughaci
[
24
]
de
al
ing
with
integ
rati
on
of
sel
f
-
a
dap
ti
ve
ha
rm
on
y
search
with
a
SLS
for
ta
ckling
kn
a
ps
ac
k
pro
bl
e
m
.
Fu
rthe
rm
or
e,
in
the
la
st
tw
o
y
ears
the
re
a
re
sever
al
ot
her
s
tud
ie
s
c
onduct
ed
with
reg
a
r
d
to
t
he
SL
S
a
pp
li
cat
io
n,
i
nclud
i
ng
Pu
ti
khin
a
nd
Kasc
heev
[25]
,
they
us
e
d
S
LS
f
or
s
olv
in
g
SA
T
prob
le
m
s
by
exten
ding
con
ti
nu
ou
s
B
oo
le
a
n
f
o
r
m
u
l
a
s
,
C
h
u
e
t
a
l
.
,
[
2
6
]
p
r
e
s
e
n
t
e
d
n
e
i
g
h
b
o
r
i
n
g
v
a
r
i
a
b
l
e
s
b
a
s
e
d
c
o
n
f
i
g
u
r
a
t
i
o
n
c
h
e
c
k
i
n
g
i
n
S
L
S
f
o
r
w
e
i
g
h
t
e
d
p
a
r
t
i
a
l
m
a
x
i
m
u
m
s
a
t
i
s
f
i
a
b
i
l
i
t
y
,
L
u
o
[
2
7
]
,
w
h
o
a
p
p
l
i
e
d
s
t
o
c
h
a
s
t
i
c
i
t
e
r
a
t
i
v
e
e
v
o
l
u
t
i
o
n
C
T
rec
on
st
ru
ct
io
n
al
gorithm
for
lim
it
ed
-
ang
le
sp
ar
se
pro
je
ct
ion
data,
Y
u
et
al
.
,
[
28]
i
ntr
oduce
d
the
Th
om
ps
on
sa
m
pl
ing
f
or
op
t
i
m
izing
SLS.
Oliveira
et
al
.
,
[
29
]
stu
died
a
naly
sis
of
the
AC
O
al
gorithm
fo
r
s
ol
ving
T
SP
,
Pa
quet
e
an
d
Stützl
e
[30]
p
r
e
s
e
n
t
e
d
a
r
e
v
i
e
w
o
f
S
L
S
A
l
g
o
r
i
t
h
m
s
f
o
r
m
u
l
t
i
o
b
j
e
c
t
i
v
e
c
o
m
b
i
n
a
t
o
r
i
a
l
o
p
t
i
m
i
z
a
t
i
o
n
,
N
i
u
e
t
a
l
.
,
[
3
1
]
i
n
t
r
o
d
u
c
e
d
a
n
e
w
S
L
S
a
p
p
r
o
a
c
h
f
o
r
c
o
m
p
u
t
i
n
g
p
r
e
f
e
r
r
e
d
e
x
t
e
n
s
i
o
n
s
o
f
a
b
s
t
r
a
c
t
a
r
g
u
m
e
n
t
a
t
i
o
n
,
S
a
n
t
o
s
e
t
a
l
.
,
[
3
2
]
,
w
h
o
p
e
r
f
o
r
m
e
d
a
n
a
l
y
s
i
s
o
f
S
L
S
m
e
t
h
o
d
s
f
o
r
t
h
e
u
n
r
e
l
a
t
e
d
p
a
r
a
l
l
e
l
m
a
c
h
i
n
e
s
c
h
e
d
u
l
i
n
g
p
r
o
b
l
e
m
a
n
d
Weise
et
al
.
,
[
33
]
,
who
s
howe
d a
n
im
pr
oved
ge
ner
ic
BET
-
AND
-
R
UN strate
gy
w
it
h per
form
ance
pr
e
dicti
on for
SLS.
The
num
ber
of
pa
per
s
pu
blishe
d
in
la
st
t
wen
ty
ye
ars
on
so
m
e
of
the
m
os
t
-
well
known
com
pu
te
r
li
br
aries
(s
uch
as
IEE
E,
Sprin
ger,
Else
vier
a
nd
ACM)
bette
r
s
hows
the
l
at
est
tren
ds
dir
ect
ion
a
nd
sci
e
ntific
sign
ific
a
nce
of
this
fiel
d.
S
LS
an
d
it
s
su
ccessf
ul
app
li
cat
ion
has
a
ls
o
at
tract
ed
the
at
te
ntion
of
sever
a
l
sci
entifi
c
discipli
nes
incl
ud
i
ng
com
pu
te
r
sci
ence
an
d
t
heir
br
a
nc
hes
s
uch
as
arti
fici
al
intel
li
gen
ce,
m
a
chin
e
le
arn
in
g,
data
m
ining
,
ec
on
om
ic
s
and
m
a
nag
em
ent,
ph
ysi
cs,
chem
istr
y
an
d
bi
o
-
i
nfor
m
at
ic
s.
There
has
al
read
y bee
n o
ne
excell
e
nt bo
ok
i
ntr
oduce
d by two pio
nee
r
s o
f
t
he
fiel
d,
Hoos
a
nd
Stütz
le
[
1], which
present
s
an
e
xcell
ent
a
nd
c
om
pr
ehe
ns
ive
ov
e
r
view
on
SL
S
m
et
ho
ds,
te
ch
niques
a
nd
al
gorithm
s
app
li
cat
io
n.
In
2015
,
te
n
ye
ars
la
te
r
, a
gain we
re
Hoos
a
nd St
ützle
[
34]
wh
o pro
vid
e a
ge
ner
al
ov
erv
ie
w on SLS
A
lg
ori
thm
s
.
This
pa
per
se
rv
es
li
k
e
a
c
om
ple
m
entary
on
e
t
o
those
pr
e
viously
pu
blishe
d,
o
n
th
e
sa
m
e
tim
e
pro
vid
es
sta
te
-
of
-
the
-
art
rev
i
ew
on
SL
S
m
et
ho
ds,
al
go
rithm
s
and
m
os
t
rece
nt
de
ve
lop
m
ent
tren
ds
a
nd
app
li
cat
io
n
in
sci
ence
an
d
in
du
st
ry.
First,
it
giv
es
a
sh
ort
ov
e
r
view
on
com
bin
at
or
ia
l
pro
blem
s
and
search
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
71
5
-
72
7
718
par
a
dig
m
s
and
so
m
e
oth
er
use
fu
l
bac
kgr
ound
nota
ti
on
.
T
he
n,
SLS
m
et
ho
ds
are
presente
d
from
three
ki
nd
of
aspects
(classe
s
):
the
first
cl
a
ss
is
known
as
“sim
ple”
SLS
m
et
ho
ds,
as
th
e
te
rm
sh
ows
t
hey
are
m
ai
nly
base
d
on
sim
ple
search
te
ch
ni
ques,
the
sec
ond
cl
ass
com
pr
ise
s
of
hybri
d
SLS
m
et
ho
ds,
w
hi
ch
inte
gr
at
e
se
ver
al
searchi
ng
te
c
hniq
ues
an
d
he
ur
ist
ic
s
an
d
th
e
third
cl
ass
nam
ed
as
“
popu
la
ti
on
-
base
d
”
SLS
m
et
ho
ds
that
procee
ds
with
a
set
or
po
pu
l
at
ion
of
can
di
date
so
luti
on
s
.
These
SL
S
m
et
hods
com
pr
i
se
a
nu
m
ber
of
well
-
known
te
c
hn
i
ques
w
hich
m
ain
ly
hav
e
bee
n
insp
ire
d
by
nat
ur
al
phe
no
m
en
a,
an
d
so
m
e
of
the
m
os
t
su
cc
essfu
l
app
li
cat
io
n
of
these
m
et
ho
ds
in
sci
ence
a
nd
i
ndus
try
ha
ve
bee
n
pr
es
ent
ed.
Finall
y,
besides
t
he
r
ecent
achievem
ents,
so
m
e fu
tu
r
e c
ha
ll
eng
es a
nd li
m
it
a
ti
on
are
br
ie
fly
d
isc
us
se
d.
The
rest
of
thi
s
pa
per
is
orga
nized
as
th
e
f
ol
lowing:
I
n
t
he
seco
nd
sect
io
n
is
giv
e
n
a
bri
ef
ove
rv
ie
w
of
bac
kgr
ound
theo
ry
sta
rtin
g
with
a
ge
ne
r
al
introd
uction
of
c
om
bin
at
ori
al
prob
le
m
s,
search
pa
ra
digm
s
a
nd
SLS
te
c
hn
i
qu
e
s.
T
he
t
hird
s
ect
ion
giv
es
c
om
pr
ehe
ns
ive
rev
ie
w
of
the
sta
te
-
of
-
the
-
art
in
SLS
al
gori
thm
s,
m
et
ho
ds
a
nd
t
echn
i
qu
e
s
fro
m
it
s
early
stag
e
of
de
velo
pm
ent
to
the
present
days.
T
he
four
t
h
sect
io
n
prese
nts
the
m
os
t
recent
app
li
cat
ion
of
SLS
al
gori
th
m
s
in
m
achine
le
arn
in
g,
da
ta
m
ining
an
d
ot
her
ar
eas
in
sc
ie
nce
and in
dustry. T
he
la
st sect
io
n pro
vid
es
bri
efly
so
m
e con
cl
udin
g rem
ark
s
.
2.
BACKG
ROU
ND AN
D NO
TATION
In
the
f
ollo
wing
is
gi
ven
a
sh
ort
intr
oduct
ion
in
bac
kgr
ound
the
or
y
an
d
nota
ti
on
as
it
has
been
c
o
n
s
i
d
e
r
e
d
a
s
m
o
r
e
t
h
a
n
n
e
c
e
s
s
a
r
y
i
n
o
r
d
e
r
t
o
f
a
c
i
l
i
t
a
t
e
u
n
d
e
r
s
t
a
n
d
i
n
g
o
f
t
h
e
S
L
S
m
e
t
h
o
d
s
,
t
e
c
h
n
i
q
u
e
s
a
n
d
a
l
g
o
r
i
t
h
m
s
.
2.1
.
Co
m
bina
t
oria
l
prob
le
m
s
Com
bin
at
or
ia
l
pro
blem
s
can
be
cat
e
gorized
in
tw
o
m
ajor
gro
up
s:
c
om
bin
at
ori
al
decisi
on
pr
ob
le
m
s
and
com
bin
at
ori
al
op
ti
m
isa
t
i
on
pro
blem
s.
The
tw
o
prot
ot
ypic
al
com
bin
at
or
ia
l
pro
ble
m
s
widely
known
a
re
SA
T
an
d
TSP
.
The
nu
m
ber
of
these
pro
blem
s
is
la
rg
e,
an
d
the
y
are
pr
es
ent
in
sever
al
br
a
nc
hes
of
c
om
pu
te
r
sci
ences,
eco
nom
ic
s
and
m
anag
em
ent,
physi
cs,
c
hem
ist
ry,
bio
-
inf
orm
atic
s
an
d
oth
e
r
res
earch
areas
in
science
and
in
dustry
in
w
hich
com
pu
ta
ti
onal
m
eth
ods
fi
nd
ap
pl
ic
at
ion
s.
So
m
e
oth
er
well
known
c
om
bi
nato
rial
pro
blem
s
includes
fin
ding
s
hortest
r
ound
tr
ips
(TSP),
s
ol
ving
pro
posit
ion
al
f
or
m
ulae
(S
A
T),
plan
ni
ng
a
nd
sche
du
li
ng,
ti
m
e
-
ta
bling
,
r
es
ource all
ocati
on, p
red
ic
ti
on
of pro
te
in
str
uctu
res,
etc
.
2.
2
.
Search
p
aradigms
In
ge
ne
ral
almost
al
l
co
m
pu
ta
ti
on
al
appr
oa
ch
es
us
e
d
to
de
al
with
hard
com
bin
at
or
ia
l
pro
blem
s
can
be
re
pr
ese
nted
as
search
al
gorithm
s.
T
he
work
i
ng
pri
nci
ple
em
plo
ye
d
in
case
of
t
he
search
a
ppr
oac
h
is
to
gen
e
rate
a
nd
evaluate
ca
nd
i
dat
e
so
l
utio
ns
in
it
erati
ve
way.
On
one
hand,
w
he
n
c
om
bin
at
or
ia
l
de
ci
si
on
pro
blem
s
app
r
oach
is
us
e
d,
the
process
of
eval
uating
a
cand
i
date
s
olut
ion
is
relat
ed
with
the
proc
ess
of
decisi
on
w
hether
it
represe
nts
an
act
ual
so
luti
on.
O
n
t
he
oth
e
r
ha
nd
,
wh
e
n
deali
ng
with
opti
m
i
zat
ion
pro
blem
s,
it
g
ener
al
ly
im
plies
d
et
erm
ining
t
he
res
pecti
ve va
lue of t
he obje
ct
ive fun
ct
io
n
.
a.
So
l
ution
met
hods
for
co
m
bina
torial
pro
blem
s
-
t
he
re
ha
s
bee
n
a g
reat
inte
re
st
and
m
any
ef
forts
ha
ve
been
m
ade
to
the
de
sign
of
c
om
bin
at
or
ia
l
(optim
i
zat
ion
)
pro
blem
s,
and
a
nu
m
ber
al
go
rithm
f
or
s
olv
i
ng
the
s
e
kinds
o
f
ha
rd
and
com
plex
pro
blem
s
hav
e
bee
n
de
sig
ne
d.
In
ge
ner
al
,
these
al
gorith
m
s
are
t
y
p
i
c
a
l
l
y
c
a
t
e
g
o
r
i
z
e
d
a
s
e
i
t
h
e
r
e
x
a
c
t
o
r
a
p
p
r
o
x
i
m
a
t
e
a
l
g
o
r
i
t
h
m
s
.
T
h
e
f
i
r
s
t
c
l
a
s
s
i
s
k
n
o
w
n
a
s
e
x
a
c
t
a
l
g
o
r
i
t
h
m
s
a
n
d
c
o
m
p
r
i
s
e
s
a
l
g
o
r
i
t
h
m
s
t
h
a
t
g
u
a
r
a
n
t
e
e
t
o
s
o
l
v
e
e
v
e
r
y
f
i
n
i
t
e
s
i
z
e
i
n
s
t
a
n
c
e
o
f
a
com
bin
at
ori
al
op
ti
m
iz
at
ion
pro
blem
.
Wh
il
e
the
oth
e
r
cl
a
ss
com
pr
is
es
a
lgorit
hm
s
wh
ic
h
m
ake
a
tra
de
-
off
bet
ween
t
he
guara
ntee
of
fin
ding
op
ti
m
a
l solutio
ns
a
nd
getti
ng
good
s
olu
ti
ons i
n po
l
ynom
ia
l
-
tim
e [3
5].
b.
Con
str
uctiv
e
al
go
rit
hms
-
ge
ne
rate
init
ia
l
so
lu
ti
on
s
from
scratch
an
d
la
te
r
s
olu
ti
on
c
om
po
nen
ts
are
a
d
d
e
d
t
o
t
h
e
i
n
i
t
i
a
l
s
o
l
u
t
i
o
n
s
a
c
c
o
r
d
i
n
g
t
o
s
o
m
e
r
u
l
e
s
u
n
t
i
l
a
s
o
l
u
t
i
o
n
i
s
c
o
m
p
l
e
t
e
.
T
h
i
s
t
y
p
e
o
f
al
go
rit
hm
s
i
s
consi
der
e
d
to
be
the f
ast
est
appr
oxim
a
te
m
e
t
hods
, h
ow
e
ve
r
the so
luti
ons
pro
vid
e
d
by the
m
q
uite of
te
n
i
s
with lo
we
r qu
a
li
ty
co
m
par
ed wit
h on
e
pr
ov
i
ded b
y l
ocal se
a
rch al
gorithm
s.
c.
Lo
c
al
Sea
rc
h
-
the
w
orkin
g
pr
i
nciple
of
local
search
al
gorithm
s
is
as
fo
ll
ow
in
g,
they
sta
rt
by
gen
erati
ng
an
init
ia
l
so
luti
on
an
d
the
n
iterate
fr
om
current
so
l
ution
t
o
oth
e
r
can
dida
te
so
luti
on
in
a
neighborho
od
sp
ace
of the c
a
nd
i
date s
olu
ti
o
ns
, by
rep
la
ci
ng actual
so
l
utio
n by bette
r
one
foun
d
[
35]
.
d.
Pert
urba
ti
ve
(
local)
searc
h
-
wh
e
n
deali
ng
with
c
om
bin
at
or
ia
l
prob
le
m
s
can
did
at
e
s
ol
utions
c
on
sist
of
so
luti
on
c
om
po
ne
nts
a
nd
ty
pi
cal
case
is
with
assi
gn
i
ng
tru
th
val
ues
t
o
in
div
id
ual
pro
posit
ion
al
v
a
riabl
es
wh
e
n
deali
ng
with
S
AT
pro
blem
s.
On
the
oth
e
r
hand,
c
and
i
date
s
olu
ti
on
s
can
easi
ly
be
c
ha
ng
e
d
in
to
new
ca
ndidate
so
luti
ons
by
app
ly
in
g
one
or
m
or
e
m
od
ific
at
ion
on
res
pecti
ve
s
olu
ti
on
com
ponen
ts
.
This
proce
ss
of
cha
ng
i
ng
is
known
as
t
he
per
t
urbin
g
of
a
giv
e
n
can
dida
te
so
luti
on.
A
ccordin
g
to
[1]
search
al
gorith
m
s
wh
ic
h
a
re
base
d
on
this
m
echan
ism
(te
chn
i
qu
e
)
in
order
to
ge
ner
at
e
the
can
did
at
e
so
luti
ons a
re c
al
le
d
pe
rtu
rb
at
i
ve
sea
rch m
et
ho
ds.
e.
Con
str
uctiv
e
(
local)
search
-
t
he
ge
ner
at
i
on
of
the
ca
ndida
te
so
luti
ons
in
com
bin
at
or
ia
l
prob
le
m
s
are
m
ade
by
re
pea
te
dly
exten
ding
par
ti
al
s
olu
ti
on
s
that
c
an
be
de
fine
d
as
a
searc
h
pro
ble
m
and
the
m
ai
n
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Sto
c
hasti
c local
sea
rc
h: A st
ate
-
of
-
t
he
-
ar
t rev
ie
w
(
Muham
et
Kastrati
)
719
obj
ect
ive
he
re
is
to
get
a
‘
good’
can
did
a
te
so
luti
on,
w
her
e
use
d
f
or
opti
m
iz
at
ion
pro
blem
s,
and
the
“goo
dn
e
ss
”
cor
relat
e
to
the
obj
ect
ive
f
un
ct
io
n.
Algor
it
h
m
s
based
on
this
m
echan
is
m
are
cal
le
d
const
ru
ct
ive
se
arch m
et
ho
ds
(
al
so
known
as
const
ru
ct
ive
he
ur
ist
ic
m
et
ho
ds).
2.3
.
St
oc
hastic
l
oc
al se
arch
The
wor
king
pr
i
nciple
in
t
r
aditi
on
al
local
searc
h
al
gorit
hm
s
wh
en
ap
pl
ie
d
to
s
olv
e
c
om
bin
at
or
ia
l
pro
blem
is
cha
racteri
zed
wit
h
fact
that
search
for
so
l
utio
ns
happen
in
the
sp
ace
siz
e
of
cand
i
date
so
lut
ion
s
.
In
a
ddit
ion
,
th
e
local
searc
h
procee
ds
by
first
ta
king
a
n
ge
ner
at
ed
init
ia
l
cand
idate
s
olu
ti
on,
a
nd
then
is
it
erated
f
ro
m
on
e
can
dida
te
so
l
ution
to
a
no
t
her
can
did
a
te
so
luti
on
wi
thin
pr
e
def
i
ne
d
neig
hborh
oo
d,
a
nd
the
decisi
on
on
each
ste
p
is
ta
ken
base
d
on
l
y
on
e
xisti
ng
l
ocal
inf
orm
at
ion
[1
]
.
H
ow
e
ve
r,
local
search
ty
pical
ly
is
char
act
erized
wit
h
tw
o
ki
nd
of
pro
blem
s
su
ch
as:
(i
)
getti
ng
stuck
in
local
op
ti
m
a
and
(ii)
bei
ng
m
isgu
ide
d by e
valuati
on
(ob
j
e
ct
ive)
f
unct
io
n.
3.
SLS M
ET
HO
DS
In
t
his
pa
rt,
a
re
s
hortly
pr
e
sented
so
m
e
of
t
he
m
os
t
prom
inent
SLS
m
e
tho
ds,
inc
lud
in
g
t
hei
r
app
li
cat
io
n
to
s
olv
e
ha
rd com
bin
at
ori
al
pr
oblem
s.
3.1
.
Simpl
e SLS
m
eth
od
s
I
n
orde
r
to
en
su
re
t
he
searc
h
process
to
esc
ape
f
ro
m
a
local
op
ti
m
a
(
m
ini
m
u
m
),
a
nu
m
ber
of
S
LS
m
et
ho
ds
acce
pt
wo
rse
ning
m
ov
e
s.
This
part
is
dev
oted
t
o
these
m
et
ho
ds
wh
ic
h
are
know
n
as
sim
pl
e
SLS
m
et
ho
ds,
as t
he
se m
et
ho
ds
e
m
plo
y on
ly
on
e ty
pe
of sea
rc
h
te
ch
niques
, o
n
li
m
it
ed
neig
hbor
hood sets
[
34
]
.
a.
Ra
ndomize
d
it
erativ
e
im
pr
ov
emen
t
(
R
II
)
-
is
sim
ply
based
on
it
erati
ve
i
m
pr
ovem
ent
exten
de
d
with
rand
om
iz
a
ti
on
.
Mor
e
pr
eci
sel
y, in each
ste
p i
te
rati
on
is p
e
rfor
m
ed
base
d o
n
pr
ob
a
bili
ty
, t
he
sel
ect
ion
of
ne
xt
sea
rc
hi
ng
posit
ion
′
is
do
ne
us
in
g
un
i
form
l
y
at
r
andom
within
current
nei
ghbor
hood
(
)
,
known
as
uni
nfor
m
ed
rand
om
walk
ste
p,
or
in
oth
er
ci
r
cum
sta
nces
1
−
i
s
use
d
to
perf
or
m
an
i
m
pr
ovem
ent
ste
p.
He
re,
t
he
is
cal
le
d
as
w
al
k
pro
ba
bili
ty
or
so
m
et
i
m
es
it
is
al
so
known
a
s
noise
par
am
et
er.
O
ne
of the
m
ai
n
adv
a
ntage
s
of
R
II
is
that
they
a
re easy to
b
e
im
ple
m
ented
[
1]
.
The
s
uccess
fu
l
ap
plica
ti
on
of
RII
al
gorithm
s
ha
ve
be
en
pro
ve
n
in
se
ve
r
al
op
ti
m
iz
a
ti
on
s
a
nd
decisi
ons
pro
bl
e
m
s.
As
pr
es
e
nted
by
[34]
in
the
1990s
ar
e
i
m
ple
m
ented
so
m
e
ver
sion
of
RI
I
by
us
in
g
m
ino
r
va
riat
ion
s,
in
s
uch
cas
es
the
rando
m
it
erati
on
is
defi
ned
with
r
ega
rd
to
the
am
ou
nt
of
co
ns
trai
nt
vio
la
ti
ons
in
pl
ace
of
c
hoos
in
g
un
if
orm
l
y
at
rand
om
,
chang
es
those
that
e
nab
le
R
II
al
gor
it
h
m
s
to
be
on
e
of the c
urre
nt s
ta
te
-
of
-
the
-
art
al
gorithm
s u
sed
to
s
olv
e
SA
T
problem
s an
d C
SP
s [3
4].
b.
Probabil
ist
ic
i
te
ra
ti
ve
impro
vement
(
PI
I)
-
un
li
ke
so
m
e
oth
er
m
et
ho
ds
t
hat
acce
pts
s
om
e
wo
rse
ning
search
ste
ps
,
he
re
this
m
et
ho
d
is
base
d
on
t
he
idea
of
us
in
g
so
m
e
acce
ptance
crit
eria
w
hich
is
based
on
pro
bab
il
ist
ic
evaluati
on
f
unct
ion
.
In
c
ontras
t
t
o
RII,
eac
h
ste
p
of
PII
re
quires
tw
o
ste
ps:
the
first
ste
p
deals
with
sel
e
ct
ion
of
nei
ghbori
ng
can
did
at
e
so
l
ution
′
in
c
urren
t
nei
ghbo
rho
od
(
)
,
w
hich
t
ypic
al
ly
is
done
by
usi
ng
un
if
orm
l
y
at
rand
om
app
r
oa
ch;
the
sec
ond
ste
p
is
use
d
t
o
determ
ine
wh
e
ther
to
acce
pt
or
no
t
′
as
the
new
ca
nd
i
dat
e
so
luti
on.
I
n
add
it
io
n,
f
or
c
la
ss
of
prob
le
m
s
wh
ere
m
ini
m
iz
ation
is
require
d,
t
he M
et
ro
poli
s c
onditi
on
s
i
s used
as accepta
nce
pro
bab
il
it
y [34
]
.
(
;
;
′
)
:
=
(
1
if
(
′
)
<
(
)
(
(
′
)
−
(
)
)
ot
h
er
wi
se
,
wh
e
re
(
;
;
′
)
repres
ents
pr
ob
a
bili
ty
of
acce
ptanc
e
an
d
denote
s
the
e
valuati
on
f
unct
ion
w
hi
ch
sh
oul
d
be
re
duced.
T
he
para
m
et
er
nam
ed
as
re
pr
ese
nts
the
im
pact
of
t
he
pr
ob
a
bili
ty
w
hethe
r
to
acce
pt
or
not
w
orseni
ng
search
ste
ps
a
nd
is
eq
uiv
al
e
nt
to
c
onsta
nt
-
te
m
per
at
ur
e
de
fine
d
to
sim
ulat
ed
ann
eal
in
g.
It
i
s
worth
m
entio
ni
ng
t
hat
for
value
of
=
0
,
the
n,
P
II
will
ef
fe
ct
ively
tur
ns
into
a
n
it
erati
ve
i
m
pr
ovem
en
t pr
oce
dure,
and
f
or v
al
ue of T=
∞, th
e
n
acc
ompli
sh
es
unif
orm
r
and
om
w
al
k [34].
c.
Simulated
anne
aling
(
SA
)
-
is
si
m
ple
SLS
m
et
hod,
wh
ic
h
s
h
ares
sim
il
ar
featur
es
with
P
II
,
a
par
t
f
r
om
tem
per
at
ur
e
pa
ram
et
er
that
is
m
od
ifie
d
at
r
un
ti
m
e.
This
m
et
ho
d
is
ins
pi
red
by
natu
ral
ph
e
nom
eno
n
analo
gy,
res
pe
ct
ively
the
slow
co
oling
of
s
olid
m
a
te
rial
s.
The
par
am
et
e
r
T
(te
m
per
at
ure)
init
ia
ll
y
ha
s
the
hi
gh
val
ue
,
w
hich
the
n
i
s
pro
gressi
vel
y
decr
ease
d
unti
l
the
lo
west
tem
per
at
ure
va
lue
is
reac
he
d.
High
te
m
per
at
ur
e v
al
ue
s
m
ea
ns
that pro
ba
bili
ty
of
acce
ptin
g
worse
ning
ca
nd
i
date
so
l
utio
ns
w
il
l
be
h
i
g
h
.
T
h
u
s
,
a
s
t
e
m
p
e
r
a
t
u
r
e
d
e
c
r
e
a
s
e
s
,
t
h
e
p
r
o
b
a
b
i
l
i
t
y
o
f
a
c
c
e
p
t
i
n
g
w
o
r
s
e
n
i
n
g
c
a
n
d
i
d
a
t
e
s
o
l
u
t
i
o
n
s
dec
reases
as
well
,
wh
ic
h
m
eans
that
searc
h
proces
s
pro
gressi
vely
be
gin
s
to
be
gr
ee
dy
.
Fu
rt
her
m
or
e
,
for
ve
ry
low
tem
per
at
ur
es
(
sm
a
ll
values
of
T
),
al
m
os
t
on
ly
nei
ghbors
that
ha
ve
bett
er
or
at
le
ast
equ
al
val
ue
of
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
71
5
-
72
7
720
evaluati
on
t
o
t
he
c
urre
nt
can
did
at
e
s
olu
ti
on
can
be
acce
pt
ed
[34].
SA
ha
s
bee
n
su
cces
s
fu
ll
y
an
d
wide
ly
u
s
e
d
t
o
s
o
l
v
e
h
a
r
d
c
o
m
p
u
t
a
t
i
o
n
a
l
p
r
o
b
l
e
m
s
,
b
y
i
n
v
o
l
v
i
n
g
v
a
r
i
o
u
s
f
o
r
m
s
o
f
a
n
n
e
a
l
i
n
g
a
n
d
a
c
c
e
p
t
a
n
c
e
t
e
s
t
s
.
d.
Tabu
sea
rch
(
t
s)
-
is
ano
ther
si
m
ple
SLS
m
eth
od,
w
hich
is
c
har
a
ct
erize
d
by
exp
li
ci
t
m
e
mo
ry
dep
e
nde
ncy
durin
g
the
sea
rch
process
[36].
In
it
s
sim
ple
fo
rm
,
known
as
sim
ple
tab
u
sea
rc
h,
it
pe
rfor
m
s
it
erativ
e
i
m
pr
ovem
ent
fr
om
one
po
te
nt
ia
l
so
luti
on
t
o
a
n
im
pr
ov
e
d
so
l
utio
n
′
in
the
neig
hbor
hood
of
unti
l
so
m
e
stop
pi
ng
crit
erion
has
be
en
sat
isfie
d.
I
n
or
der
t
o
av
oi
d
getti
ng
st
uck
on
local
opti
m
a
an
d
to
e
nsur
e
that
al
l
reg
ions
in
their
nei
ghbo
rho
od
hav
e
been
e
xp
l
or
e
d,
so
m
e
kin
d
of
sh
ort
-
te
rm
m
e
m
or
y
has
been
us
e
d.
I
n
order
to
sk
ip
m
e
m
or
iz
ing
w
ho
le
se
t
of
cand
i
date
so
luti
ons
an
d
e
xp
li
ci
tl
y
fo
rb
i
dden
these
,
he
r
e
is
assigne
d
a
ta
bu
sta
tus
t
o
each
com
ponen
t
and
is
kee
pt
track
of
each
s
olu
ti
on
c
om
po
ne
nt
wh
e
n
it
wa
s
la
st
m
od
ifie
d.
TS
al
gorithm
s
even
sim
ple
on
e
has
pr
ov
e
d
to
perform
qu
it
e
go
od
in
sev
eral
nu
m
be
r
of
pro
blem
s.
Ho
wev
e
r,
it
s
pe
rfor
m
ance
is
str
ongly
relat
ed
on
t
he
ta
bu
te
nure
set
ti
ngs.
I
n
or
der
to
esca
pe
the
prob
le
m
of
fi
nd
i
ng
fi
xe
d
set
ti
ng
s
,
wh
ic
h
a
re
s
uitable
for
a
gi
ven
s
pe
ci
fic
pro
blem
,
reacti
ve
ta
bu
search
m
echani
s
m
h
as b
ee
n
i
ntr
oduce
d
in
or
der
to m
od
ify
t
he
ta
bu te
nure
set
ti
ng
s at
run
-
tim
e [3
4].
e.
Dyna
mic
local
s
earch
(
DLS)
-
in
the
previ
ous
sect
ion
we
hav
e
prese
nted
seve
ral
te
chn
iq
ues
us
e
d
f
or
escapin
g
f
ro
m
local
op
ti
m
a
by
us
ing
m
ini
m
al
ly
wo
rsen
i
ng
ste
ps
in
local
search
.
U
nlike
oth
er
‘sim
ple’
SLS
m
e
tho
ds p
resen
te
d
ab
ov
e
, D
LS [1] do
es
n’
t pe
rm
i
t wo
r
senin
g
searc
h
s
te
ps
, but in
cont
rast it
u
pd
at
es
the ev
al
uatio
n functi
on
durin
g
the
local sea
r
ch
in
or
der
t
o
a
vo
i
d gett
ing st
uck on l
ocal
optim
a [3
4].
Accor
ding
t
o
[
34
]
t
he
update
d
e
valuati
on
f
unct
ion
′
is
cal
cu
la
te
d
as
t
he
sum
′
(
)
an
d
(
)
,
that i
s
:
′
(
)
:
=
(
)
+
∑
∈
(
)
(
)
(1)
′
(
)
represe
nts
th
e
or
i
gin
al
eva
luati
on
functi
on
val
ue,
(
)
de
note
s
pe
nalti
es
for
eve
ry
so
l
ut
ion
com
po
ne
nt
,
(
)
represe
nts
a
s
et
of
s
olu
ti
on
com
ponen
ts
of
.
At
t
he
be
ginnin
g,
al
l
pe
nalti
es
ha
ve
the
va
lue
e
qua
l
to
zer
o.
Depend
i
ng
on
the
pen
al
ty
updating
m
echan
ism
an
d
c
hoic
e
of
so
l
ution
com
pone
nts
that
are
us
e
d
t
o
a
dju
st
t
he
pe
nalti
es,
the
re
e
xist
dif
fere
nt
va
ria
nts
of
D
LS
.
I
n
t
he
beg
i
nn
i
ng,
(
)
is
cal
culat
ed
for
e
a
c
h
a
c
c
o
r
d
i
n
g
t
o
e
q
u
a
t
i
o
n
:
(
)
:
=
(
)
/
(
1
+
(
)
)
w
h
e
r
e
(
)
i
s
u
s
e
d
t
o
m
e
a
s
u
r
e
t
h
e
e
f
f
e
c
t
o
f
o
n
t
h
e
e
v
a
l
u
a
t
i
o
n
f
u
n
c
t
i
o
n
a
n
d
t
h
e
n
p
e
n
a
l
t
i
e
s
a
r
e
i
n
c
r
e
a
s
e
d
f
o
r
s
o
l
u
t
i
o
n
c
o
m
p
o
n
e
n
t
s
t
h
a
t
h
a
v
e
m
a
x
i
m
a
l
u
t
i
l
i
t
y
[
3
4
]
.
3.2
.
Hy
brid
SLS
meth
od
s
Ther
e
e
xist
sev
eral
m
or
e
com
plex
SLS
m
et
ho
ds
known
as
hybri
d
m
et
ho
ds,
wh
ic
h
integ
r
at
e
diff
ere
nt
ty
pes
of
sea
rc
h t
echn
iq
ues (ex
.,
co
ns
tr
uctio
n sea
rch
with
pert
urbati
ve
local
search
)
or in
vo
lv
ing
oth
e
r
ty
pe
s o
f
com
plex
m
od
i
ficat
ion
s
of
c
urre
nt
can
did
at
e
so
luti
ons,
i
n
orde
r
to
gen
e
rat
e
eff
ect
ive
i
niti
al
sta
rting
poi
nt
f
or
fo
ll
owin
g
it
era
ti
ve
im
pr
ov
em
ent
sea
rch
[
34]
.
The
f
ollow
i
ng
sect
io
n
pr
e
sents
a
sho
rt
over
view
of
t
he
m
os
t
prom
inent h
yb
rid
m
et
ho
d
s.
a.
G
r
e
e
d
y
r
a
n
d
o
m
i
z
e
d
a
d
a
p
t
i
v
e
s
e
a
r
c
h
p
r
o
c
e
d
u
r
e
s
(
G
R
A
S
P
)
-
t
h
e
w
o
r
k
i
n
g
p
r
i
n
c
i
p
l
e
u
s
e
d
b
y
G
R
A
S
P
[
3
7
]
i
s
i
n
t
e
g
r
a
t
i
o
n
o
f
r
a
n
d
o
m
i
z
e
d
g
r
e
e
d
y
c
o
n
s
t
r
u
c
t
i
v
e
s
e
a
r
c
h
w
i
t
h
a
s
u
c
c
e
s
s
i
v
e
p
e
r
t
u
r
b
a
t
i
v
e
l
o
c
a
l
sear
ch
T
he
process
o
f
g
e
n
e
r
a
t
i
n
g
i
m
p
r
o
v
e
d
s
o
l
u
t
i
o
n
s
b
y
c
o
n
s
t
r
u
c
t
i
o
n
h
e
u
r
i
s
t
i
c
s
(
g
r
e
e
d
y
r
a
n
d
o
m
i
z
e
d
p
r
o
c
e
d
u
r
e
)
i
s
co
ntin
ually
rep
eat
e
d
unti
l
a
te
r
m
inati
on
crit
erion
is
m
e
t.
The
pr
e
senc
e
of
the
w
ord
adapti
ve
sp
eci
fied
in
GR
ASP
nam
e, is to
denote that hy
br
i
d sea
rch p
ro
ce
du
re invo
l
ves
a
n adaptive
constr
uction he
ur
ist
i
c. In
t
h
e
s
i
m
i
l
a
r
w
a
y
,
t
h
e
t
e
r
m
r
a
n
d
o
m
i
z
e
d
d
e
n
o
t
e
s
t
h
a
t
r
a
n
d
o
m
i
z
a
t
i
o
n
i
s
u
s
e
d
,
a
n
d
i
t
i
s
r
e
a
l
i
z
e
d
b
y
u
s
i
n
g
t
h
e
s
o
-
c
a
l
l
e
d
r
e
s
t
r
i
c
t
e
d
c
a
n
d
i
d
a
t
e
l
i
s
t
t
h
a
t
k
e
e
p
s
t
h
e
b
e
s
t
-
s
c
o
r
i
n
g
c
o
m
p
o
n
e
n
t
s
d
e
p
e
n
d
i
n
g
o
n
a
h
e
u
r
i
s
t
i
c
s
f
u
n
c
t
i
o
n
[
3
4
]
.
b.
Iterated
gr
eed
y
(
IG
)
-
is
an
othe
r
al
g
ori
thm
that
belo
ng
to
hybr
i
d
S
LS
m
et
ho
ds.
The
w
orki
ng
pr
inci
ple
of
IG
is
that
it
i
te
rati
vely
per
f
or
m
s
gr
eedy
c
on
st
ru
ct
io
n
he
ur
ist
ic
s
in
ord
er
to
pro
du
ce
a
sequ
e
nce
of
cand
i
date
so
lu
ti
on
s
that
ha
ve
hig
h
-
qu
al
it
y.
In
this
al
gorithm
,
the
key
pr
inciple
is
to
sw
it
ch
betwe
e
n
so
luti
on
c
onstr
uction
an
d
des
tructi
on
ph
as
es
,
w
hile
this
e
nsures
bette
r
pe
rfor
m
ance
th
rough
i
nteg
rati
on
of tw
o
ty
pes o
f
searc
h
te
ch
ni
ques
dif
fer
e
nt to
each othe
r [34
]
.
The
basic
pr
i
nc
iple
ap
plied
t
o
I
G
al
gorithm
,
was
re
disco
ve
red
a
nu
m
ber
of
t
im
es
and
pr
esented
with
diff
e
re
nt
nam
es,
su
ch
as
:
ru
in
-
a
nd
rec
r
eat
e,
it
erati
ve
f
la
tt
ening
a
nd
it
erati
ve
c
on
st
ruct
ion
heurist
ic
.
IG
al
gorithm
s
hav
e
bee
n
s
ucc
essfu
ll
y
use
d
f
or
so
l
ving
a
num
ber
of
pro
blem
s
and
obta
ine
d
re
su
lt
s
s
how
that,
especial
ly
wh
e
n
inte
gr
at
ed
with
per
t
urbati
ve
local
search
,
they
a
chieve
d
im
pr
essive
res
ults
on
m
any o
ptim
iz
a
ti
on
pro
blem
s [
38, 39].
c.
Iterated
loc
al
searc
h
(
ILS)
-
is
an
oth
er
suc
ce
ssfu
l
hybri
d
S
LS
m
et
ho
d,
w
hich
is
m
ai
nly
known
as
hybri
d
betwee
n
pe
rtu
r
bation
m
echani
s
m
and
local
se
arch
al
gorith
m
.
As
pr
ese
nt
ed
in
[
34
]
an
ILS
al
gorithm
com
pr
ise
s
by
four
basic
ty
pes
of
m
echan
ism
s
:
(i)
m
echan
ism
us
ed
to
gen
e
rate
an
init
ia
l
so
luti
ons
(ex.,
co
ns
tr
uction
he
ur
ist
ic
s),
(ii)
a
subsidiar
y
(p
ert
urbati
ve
)
local
searc
h
proce
dure,
(iii
)
a
per
tu
rb
at
i
on
proce
dure
that
perform
s
m
o
dificat
ion
t
o
c
and
i
date
so
l
ution
a
nd
(iv
)
an
acce
pta
nce
crit
erion.
IL
S
al
gorithm
s,
esp
eci
al
ly
it
s
basic
ve
rsion,
is
ch
aracte
rized
as fa
st
an
d
easi
ly
im
ple
m
ented.
F
ur
t
her
m
or
e, b
y
app
ly
in
g
s
om
e
ty
pes
of
m
od
ific
at
ion
,
IL
S
r
epr
e
se
nts
sta
te
-
of
-
t
he
-
a
rt
m
eth
od
us
e
d
t
o
s
ol
ve
a
wide
ra
nge
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Sto
c
hasti
c local
sea
rc
h: A st
ate
-
of
-
t
he
-
ar
t rev
ie
w
(
Muham
et
Kastrati
)
721
of
c
om
bin
at
or
i
al
prob
le
m
s
[34].
D
uri
ng
the
la
st
ye
ars
sev
eral
ve
rsions
of
IL
S
hav
e
be
en
re
disc
ov
e
re
d
and
f
ound
unde
r
di
ff
e
ren
t
na
m
es,
li
ke
la
rge
-
ste
p
Ma
rko
v
chains
[
40
]
a
nd
chai
ned
l
oc
al
op
ti
m
iz
a
ti
on
[41],
but
al
so
there
e
xist
so
m
e
con
ce
ptu
al
c
onnecti
on
of
the
un
der
ly
ing
al
gorithm
s
wit
h
dif
fer
e
nt
ty
pe
s
of
var
ia
ble
nei
ghbor
hood
sea
rch,
f
ur
th
erm
or
e
V
NS
a
nd
S
VNS
m
et
ho
ds
are
co
ns
ide
re
d
as
ver
si
ons
of
ILS, w
hich
b
e
ne
fit by
us
in
g
th
e ad
van
ta
ges o
f pertu
r
bat
ion
at
r
un ti
m
e [3
4]
.
3.3
.
Pop
ul
at
i
on
-
b
ase
d S
LS
met
h
od
s
These
m
et
ho
ds
are
c
onside
r
ed
m
or
e
c
om
plex
com
par
e
t
o
oth
e
r
SL
S
m
et
ho
ds
,
beca
us
e
t
hey
are
est
ablished
ba
s
ed
on princi
ple of
us
in
g
a
set
or po
pu
la
ti
on
of ca
nd
i
date so
l
utions.
a.
Ant
c
olony
opti
miza
ti
on
(
ACO)
-
is
par
t
of
popula
ti
on
-
bas
ed
S
LS
m
et
hods
t
hat
has
be
en
widely
a
nd
su
ccess
fu
ll
y
e
m
plo
ye
d
to
s
olv
e
m
any
com
plex
com
bin
at
or
ia
l
prob
le
m
s
[42
-
44]
.
T
his
al
gorithm
reli
es
on
bio
lo
gical
phe
no
m
ena
(a
nts),
an
d
he
re
te
r
m
arti
fici
al
a
nts
re
pr
ese
nts
m
ulti
-
agen
t
m
e
tho
ds
whos
e
insp
irat
io
n
c
om
es
fr
om
collecti
ve
beh
a
viors
of
real
ant
colo
nies.
Ty
pi
cal
ly
,
co
m
m
un
ic
at
ion
base
d
on
ph
e
r
om
on
es
of
the
bi
ologica
l
ant
is
the
par
a
dig
m
us
ed.
Co
m
bin
at
ion
of
a
rtific
ia
l
ants
w
it
h
local
searc
h
al
gorithm
s
ha
ve
bee
n
su
cce
ssfu
ll
y
us
e
d
i
n
seve
ral
op
ti
m
iz
at
ion
ta
sk
s.
Fu
rt
her
de
ta
il
s
abo
ut
thes
e
m
oth
od
s ca
n b
e f
ound in
[29,
44
]
.
b.
Evolutio
nary
al
go
rit
hms
(
EAs)
-
is
on
e
of
the
m
os
t
widely
us
ed
an
d
su
c
cesfu
ll
popula
ti
on
-
base
d
S
L
S
m
et
ho
d
us
e
d
f
or
so
l
ving
c
om
plex
com
pu
ta
ti
on
al
prob
le
m
s
in
recent
y
e
ars.
EA
s
have
bee
n
ge
ner
al
ly
insp
ire
d
f
ro
m
bio
lo
gical
evaluati
on
w
hi
ch
cha
racteri
z
ed
by
three
m
os
t
well
-
known
ev
olu
ti
on
m
echan
ism
su
ch
as
m
utati
on
,
rec
om
bin
at
ion
an
d
sel
ect
io
n.
Anothe
r
si
gn
i
ficant
m
echan
ism
is
evaluati
on
functi
on, which
is k
now
n
as fi
tness
.
T
hese
a
lgorit
hm
s
are
al
so
c
har
act
eriz
ed
with r
an
do
m
iz
at
ion
proce
ss
wh
ic
h
is
use
d
to
ge
ner
at
e
t
he
init
ia
l
set
of
c
and
i
date
so
l
ution
s
,
a
nd
t
hen
gr
ee
dy
co
ns
tr
uc
ti
on
heurist
ic
s
that
is
m
a
inly
e
m
plo
ye
d
to
s
eed
the
popula
ti
on
.
Lat
er
,
this
under
ly
in
g
popula
ti
on
is
subj
ect
of
th
ree
m
os
t
well
known
ge
netic
-
base
d
m
echani
s
m
known
a
s
m
utati
on
,
r
ecom
bin
at
ion
an
d
s
el
ect
ion
.
In
ge
ner
al
,
E
A
s
perform
ance
is
strongly
rela
te
d
with
the
ri
gh
t
us
e
of
the
evo
l
ution
a
ry
m
echan
ism
,
due
to
this
fact,
to
o
m
uch
resear
ch
w
ork
has
be
e
n
do
ne
with
reg
a
rd
t
o
desi
gn
an
d
ef
fecti
ve
us
e
of
m
utati
on
and
rec
om
bin
at
ion
m
echan
is
m
[3
4].
EAs
ha
ve
been
exte
ns
sively
us
ed
to
s
olv
e
dif
fer
e
nt
ki
nd
s
of
rea
l
world
prob
le
m
s
and
re
su
lt
s
ob
ta
ine
d
s
how
that
EAs
achi
eved
sta
te
-
of
-
t
he
-
a
rt
resu
lt
w
hen
a
pp
li
e
d
to
so
lve
c
o
m
b
i
n
a
t
o
r
i
a
l
r
e
l
a
t
e
d
p
r
o
b
l
e
m
s
i
n
c
l
u
d
i
n
g
f
i
n
d
i
n
g
s
h
o
r
t
a
n
d
i
m
p
l
e
m
e
n
t
a
t
i
o
n
-
f
r
i
e
n
d
l
y
a
d
d
i
t
i
o
n
c
h
a
i
n
s
[
4
5
]
a
n
d
f
i
n
d
i
n
g
n
a
s
h
e
q
u
i
l
i
b
r
i
u
m
i
n
e
l
e
c
t
r
i
c
i
t
y
m
a
r
k
e
t
s
[
4
6
,
4
7
]
.
4.
STATE
-
OF
-
T
HE
-
A
RT I
N S
LS
In
this
sect
io
n
,
we
hav
e
br
ie
fl
y
pr
esente
d
a
sta
te
-
of
-
the
-
art
rev
ie
w
in
SLS
m
et
ho
ds
a
nd
it
s
su
ccessf
ul
app
li
cat
io
n
in
s
ci
ence a
nd eng
ineerin
g
.
4.1.
Int
e
gratio
n
of sy
s
tem
at
ic
se
arch
and SLS
SLS
a
nd
syst
em
at
ic
search
usuall
y hav
e
b
e
en
co
ns
i
der
e
d as t
wo
sepa
rate
ap
pr
oach
e
s appli
ed
to s
ol
ve
com
plex
com
bin
at
or
ia
l
opti
m
iz
at
ion
an
d
de
ci
sion
pro
blem
s.
D
ue
to
t
heir
sp
eci
fic
c
har
a
ct
erist
ic
s
fo
r
a
wh
il
e
these
ap
proac
hes
ha
ve
been
con
si
der
e
d
m
or
e
c
on
c
urre
nt
then
c
om
ple
m
entary.
H
owever,
duri
ng
t
he
la
st
decad
e
ha
s
be
en
ide
ntifie
d
an
inc
reased
at
te
ntion
to
th
e
e
xp
l
or
i
ng,
de
sign
a
nd
de
ve
lop
m
ent
of
hy
br
i
d
m
et
ho
ds,
w
hic
h
in
vo
l
ve
inte
gr
at
io
n
of
sys
tem
at
ic
search
and
S
LS
a
ppr
oac
hes
with
purpose
e
nha
ncin
g
the
al
go
rit
hm
s
.
These
hybri
d
m
e
tho
ds
ca
n
be
cl
assifi
ed
in
two
m
ajo
r
cl
asses
de
pe
nd
i
ng
on
the
ro
le
of
com
bin
ed
te
ch
ni
ques
a
nd
th
e
ro
le
t
hey
pl
ay
.
Acc
ordin
g
to
this
cat
e
gorizat
io
n
the
f
irst
cl
ass
com
pr
ise
s
appr
oach
es
where
the
syst
e
m
at
ic
search
te
chn
i
qu
e
s
ser
ves
as
m
ast
er
pr
oc
ess
wh
il
e
othe
r
proce
dure
(
SLS)
is
eng
a
ge
d
to
ta
c
kle
any
issue
t
hat
can
a
rise
thr
ough
ou
t
the
syst
e
m
atic
search
ste
ps
.
In
t
he
seco
nd
cl
ass
,
these
ro
le
s
hav
e
bee
n
cha
nged
,
here
SLS
al
go
rith
m
serv
es
as
t
he
m
ast
er
proces
s,
w
hile
syst
em
at
ic
search
te
chn
i
qu
e
is use
d wh
e
n d
eal
ing
with s
pe
ci
fic ta
sk
s t
hat
arise d
uri
ng ru
nn
i
ng tim
e o
f SLS al
gorithm
[34].
4.2
.
SLS a
l
gori
thm
en
gineeri
ng
The
def
i
niti
on
of
ge
ner
al
-
pu
rpose
S
LS
m
et
hods
is
no
t
s
i
m
ple
as
that
of
fu
ll
y
de
fin
ed
reci
pes:
m
eaning
that
there
are
s
om
e
op
en
c
ho
ic
es
durin
g
the
al
gorithm
s
desig
n,
f
urt
he
rm
or
e
on
ly
pro
per
com
bin
at
ion
s
of
t
hese
ch
oice
s
can
he
lp
on
desig
ning
of
e
f
fecti
ve
al
gorit
hm
s,
wh
ic
h
ca
n
be
us
e
d
f
or
s
olv
in
g
sp
eci
fic
dom
ain
pro
blem
s.
As
sug
gested
by
[
34
]
m
or
e
a
nd
m
or
e
m
et
ho
do
l
og
ic
al
rese
arch
is
need
ed
to
be
unde
rtake
n
to
wards
a
n
im
pr
ov
e
d
desi
gn
a
nd
im
pl
e
m
enta
ti
on
of
SL
S
al
gorithm
s.
Alth
ough
S
LS
al
gorithm
s
eng
i
neer
i
ng
fol
low
s
the
m
ot
ivati
on
of
al
gorithm
s
eng
in
eerin
g;
they
are
m
a
inly
us
ed
to
so
l
ve
N
P
-
ha
r
d
pro
blem
s
that
are
c
harac
te
rized
with
c
omplex
a
nd
unpr
edict
able
beh
a
vior,
a
ddit
ion
a
ll
y,
the
prese
nc
e
of
stochastic
it
y m
akes a
naly
sis
of these
alg
or
it
hm
s
m
or
e h
ar
d and com
plex.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
71
5
-
72
7
722
4.3
.
Au
t
om
at
ic
configu
ra
tion o
f
SLS
algorith
ms
As
m
any
oth
e
r
al
gorithm
s,
th
e
pe
rfor
m
ance
of
SLS
al
gorit
hm
s
is
strongl
y
relat
ed
t
o
th
e
num
ber
of
desig
n
c
ho
ic
es
and
par
am
et
er
set
ti
ng
s
.
Fin
ding
ri
gh
t
de
sign
a
nd
co
nf
i
gurati
on
of
SL
S
al
gorithm
s
i
nvolv
e
s
set
ti
ng
a
la
rg
e
num
ber
of
ca
te
gorical
,
or
din
al
a
nd
num
erical
par
am
et
ers.
T
he
m
ai
n
goal
in
desi
gning
of
autom
at
ic
con
fig
ur
at
io
n
of
SLS
al
go
rithm
s
is
to
fi
nd
th
e
pro
per
set
ti
ng
s
of
t
hese
pa
ram
et
ers
in
order
t
o
achieve
t
he
op
tim
a
l
perform
a
nce
[34].
Acc
ordi
ng
to
[17]
t
he
a
uto
m
at
ic
c
onfig
ur
at
io
n
of
SLS
al
gorith
m
s
i
s
a
powerf
ul
an
d
wi
dely
us
ed
appr
oach
t
hat
has
sig
nifica
nt
ro
le
in
th
e
de
sign
a
nd
de
vel
op
m
ent
of
al
gorithm
s
that p
rovide
b
e
tt
er
perf
or
m
ance
in
pro
blem
s
olv
in
g
.
As
w
e
m
entioned
a
bove
the
perform
ance
of
SLS
al
gorithm
s
us
ually
de
pends
on
the
nu
m
ber
of
desig
n
ch
oices
and
pa
ram
et
e
r
tun
in
g.
I
n
doin
g
so
,
m
any
research
e
r
s
ha
ve
bee
n
for
a
wh
il
e
unhapp
y
with
m
anu
al
al
gorithm
con
fig
ur
at
ion
,
a
nd
in
a
nu
m
ber
of
fi
el
ds
they
ha
ve
introd
uced
t
heir
a
ppro
ac
he
s
for
autom
at
ic
par
a
m
et
er
tun
in
g.
A
ppro
ac
hes
for
set
ti
ng
pa
ram
et
ers
hav
e
bee
n
pr
ese
nt
ed
a
nd
desc
ri
bed
by
a
nu
m
ber
of
a
uthors,
but
we
will
descr
i
be
so
m
e
of
the
m
os
t
prom
inent
on
e
s
that
hav
e
bee
n
pu
blishe
d
their
researc
h
w
ork
in
the
la
st
deca
de
.
H
utter
et
al
.
,
[8
]
i
n
thei
r
st
ud
y
present
ed
m
et
ho
ds
that
s
et
pro
per
pa
ra
m
et
ers
autom
at
ic
ally
i
n
order
to
en
ha
nce
t
he
perform
ance
on
a
gi
ven
cl
ass
of
pro
blem
instances
.
T
wo
ye
ars
la
te
r,
ano
t
her
e
xtensi
ve
resear
ch
wa
s
condu
ct
e
d
by
Hoos
[9
]
on
autom
at
ed
al
go
r
it
h
m
s
con
fig
ur
at
ion
an
d
par
a
m
et
er
tun
in
g.
In
thi
s
stud
y,
they
pr
ese
nted
a
n
exten
de
d
int
rod
uction
to
t
he
sig
nificant
ro
le
that
aut
om
at
e
d
al
gorithm
s
config
ur
at
io
n
a
nd
par
am
et
er
tun
i
ng
te
c
hn
i
ques
play
in
the
im
ple
m
entat
ion
of
al
gorithm
s
that
hav
e
bette
r
perf
or
m
ance.
T
his
study
al
so
giv
es
a
br
ie
f
s
urvey
on
the
area
of
al
gorithm
con
fig
urat
ion
a
nd
par
a
m
et
er
tun
in
g
te
c
hniq
ues.
A
n
o
t
h
e
r
s
t
u
d
y
o
n
a
u
t
o
m
a
t
i
c
d
e
s
i
g
n
o
f
h
y
b
r
i
d
S
L
S
a
l
g
o
r
i
t
h
m
s
h
a
s
b
e
e
n
i
n
t
r
o
d
u
c
e
d
b
y
M
a
r
m
i
o
n
e
t
a
l
.
,
[
1
0
]
w
h
o
p
r
o
p
o
s
e
d
a
p
r
a
c
t
i
c
a
l
,
u
n
i
f
i
e
d
s
t
r
u
c
t
u
r
e
t
h
a
t
i
n
t
e
g
r
a
t
e
s
s
e
v
e
r
a
l
S
L
S
m
e
t
h
o
d
s
.
T
h
i
s
a
p
p
r
o
a
c
h
i
s
u
n
i
f
i
e
d
du
e
t
o
the
fact
that
inv
ol
ves
these
m
et
aheurist
ic
s
into
a
sing
le
str
uc
ture
w
hich
ca
n
be
se
par
at
el
y
instanti
at
ed
an
d
ca
n
be
us
e
d
t
o
ge
ner
at
e
c
om
plex
c
om
bin
at
ions
an
d
va
riants
as
well
.
Aro
und
t
he
sam
e
tim
e,
gr
am
m
ar
-
base
d
gen
e
rati
on
of
SLS
he
ur
ist
i
cs
thr
ough
au
tom
a
ti
c
al
go
rithm
con
fig
ur
a
ti
on
to
ols
has
been
i
ntrod
uc
e
d
by
Ma
sci
a
et
al
.
,
[11
]
.
A
uthor
s
pro
pose
d
he
re
a
ne
w
a
ppro
ac
h
t
hat
is
base
d
on
us
i
ng
s
om
e
sequ
e
nce
of
cat
eg
ori
cal
,
int
eger,
a
nd
real
-
value
d
par
am
et
ers
an
d
a
uto
m
at
ic
al
go
rithm
config
ur
at
io
n
i
n
order
to
fin
d
out
the
al
g
or
it
hm
that
perf
or
m
s
bette
r
f
or
t
he
sp
eci
fic
prob
le
m
at
han
d.
T
he
re
is
an
oth
e
r
stud
y
c
onduct
ed
by
Ma
sci
a
et
al
.
,
[
13
]
,
c
oncer
ning
to
th
e
al
gorit
h
m
co
m
par
iso
n
by
a
uto
m
at
icall
y
con
fig
urab
le
SLS
f
ram
ewo
r
ks
,
wh
ic
h
is
a
case
stud
y
us
i
ng
flo
w
-
s
hop
sc
he
du
li
ng
pr
ob
le
m
(F
SP)
.
Th
e
ob
ta
ine
d
res
ult
s
sh
owe
d
that
hybri
d
al
gorithm
s that are i
ns
ta
ntiat
ed were
ab
le
t
o m
at
ch
and im
pr
ove
over t
he b
est
conv
e
ntio
na
l SL
S m
et
ho
d.
Anothe
r
resea
r
ch
with
re
gard
to
the
i
m
pact
of
aut
om
at
ed
al
go
rithm
con
fi
gurati
on
on
the
scal
in
g
beh
a
vior
of
sta
te
-
of
-
the
-
art
in
exact
TSP
so
l
ve
rs
was
pre
se
nt
ed
by
Mu
et
al
.
,
[17].
He
inve
sti
gate
d
the
ef
fects
of
aut
om
at
ic
alg
ori
thm
con
fig
ur
at
io
n
in
re
ga
rd
to
im
pr
ov
i
ng
the
perform
a
nce
of
the
tw
o
well
known
in
exact
so
lve
rs
for
t
he
TSP,
E
AX
an
d
L
KH.
By
u
s
ing
t
his
new
way
of
a
naly
zi
ng
the
em
pirical
scal
ing
of
r
unning
tim
e
as
a
fu
nction
of
pro
blem
instance
siz
e,
dem
on
strat
ed
t
hat
autom
at
ed
config
ur
at
io
n
h
as
im
po
rtant
eff
ect
on
the
scal
ing
beh
a
vior
of E
AX.
Wor
k
on
a
utom
at
ed
al
gorith
m
con
fig
ur
at
io
n
a
nd
par
am
eter
tu
ning
te
ch
ni
qu
es
ha
s
been
al
so
done
by
Fr
a
nzin
a
nd
S
tützl
e
[16]
w
ho
ha
ve
eval
ua
te
d
seve
ral
a
ccepta
nce
c
rite
ria
base
d
on
exp
e
rim
ental
work.
They
fi
rst
m
ade
tun
in
g
of
nu
m
erical
par
am
et
ers
of
t
he
al
gorithm
s
us
ing
autom
at
ic
al
go
rithm
con
fi
gurati
on
appr
oach
f
or
qu
a
drat
ic
assignm
ent
pro
blem
and
a
per
m
utati
on
flo
ws
hop
pro
blem
.
Fr
anzi
n
et
al
.
,
i
n
[
15]
pr
ese
nt
ed
the
eff
ect
of
tra
ns
f
or
m
at
ion
s
of
num
erical
par
am
et
ers
in
autom
atic
alg
ori
thm
con
figurati
on.
The
a
uthors
he
re
stu
died
t
he
i
m
pact
of
al
te
rin
g
the
sam
pling
sp
ace
of
par
am
et
ers
in
autom
at
ed
al
gorith
m
c
o
n
f
i
g
u
r
a
t
i
o
n
s
.
F
r
a
n
z
i
n
a
n
d
S
t
ü
t
z
l
e
[
1
4
]
d
e
s
c
r
i
b
e
d
S
A
a
s
a
n
e
n
s
e
m
b
l
e
o
f
a
l
g
o
r
i
t
h
m
i
c
c
o
m
p
o
n
e
n
t
s
a
n
d
d
e
s
c
r
i
b
e
S
A
v
a
r
i
a
n
t
s
f
r
o
m
t
h
e
l
i
t
e
r
a
t
u
r
e
w
i
t
h
i
n
t
h
e
s
e
c
o
m
p
o
n
e
n
t
s
.
T
h
e
y
h
a
v
e
a
l
s
o
e
x
p
e
r
i
m
e
n
t
a
l
l
y
dem
on
strat
ed
the
pote
ntial
of
this
ne
w
appr
oach
on
three
well
-
known
com
bin
at
or
ia
l
optim
iz
a
ti
on
pro
blem
s
,
su
ch
as
qu
adr
at
i
c
assignm
ent
prob
le
m
and
two
var
ia
nts
of
t
he
per
m
utati
on
flo
w
shop
pro
blem
.
Pagn
oz
zi
and
Stützl
e
[1
2]
pr
ese
nt
ed
a
n
autom
at
ic
design
of
hy
br
i
d
SLS
al
gorith
m
s
fo
r
pe
rm
ut
at
ion
flo
wsho
p
pro
blem
s
(
PFSP).
Her
e
,
they
a
uto
m
atical
ly
gen
erate
d
a
ne
w
sta
te
-
of
-
the
-
art
al
gorithm
fo
r
so
m
e
of
the
m
os
t
widel
y
stud
i
ed
var
ia
nts of t
he PFSP
.
4.4
.
Ap
pli
ca
tio
n
of SLS
algorith
ms
T
his
par
t
giv
e
s
a
br
ie
f
over
vi
ew
of
s
om
e
su
ccess
fu
l
a
ppli
cat
ion
s
of
SL
S
al
gorithm
s
use
d
to
s
olv
e
op
ti
m
iz
ation
and
sea
rch
pro
bl
e
m
s.
W
e
ha
ve
cat
ego
rize
d
th
ese
exam
ples
i
n
two
m
ajo
r
gr
oups
:
ap
plica
ti
on
of
SLS
al
gorithm
s
in
m
achine
l
earn
i
ng
a
nd
da
ta
m
ining
an
d
,
app
li
cat
ion
of
SLS
al
gorith
m
s
in
oth
er
a
r
eas
of
sci
ence a
nd en
gin
ee
rin
g.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Sto
c
hasti
c local
sea
rc
h: A st
ate
-
of
-
t
he
-
ar
t rev
ie
w
(
Muham
et
Kastrati
)
723
4.4.1
.
SLS in
machine
learn
ing
In
this secti
on,
are p
re
sente
d
s
om
e o
f
the
m
os
t
recent trends
o
n
ap
plica
ti
on
of
SLS
in the a
rea o
f data
m
ining
,
gra
ph
m
ining
an
d
m
a
chine
le
anin
g.
Ho
s
sai
n
et
al
.
,
[48
]
pr
ese
nted
the
app
li
cat
io
n
of
SLS
f
or
pa
tt
ern
set
m
ining
.
He
re
auth
or
s
pro
po
s
ed
the
ap
pl
ic
at
ion
of
SL
S
fo
r
s
olv
in
g
th
e
patte
rn
set
m
ining,
pa
rtic
ularly
,
to
t
h
e
t
a
s
k
o
f
c
o
n
c
e
p
t
l
e
a
r
n
i
n
g
.
T
h
e
y
a
p
p
l
i
e
d
h
e
r
e
a
n
u
m
b
e
r
S
L
S
a
l
g
o
r
i
t
h
m
s
o
n
a
c
o
n
v
e
n
t
i
o
n
a
l
b
e
n
c
h
m
a
r
k
p
r
o
b
l
e
m
i
n
s
t
a
n
c
e
f
o
r
p
a
t
t
e
r
n
s
e
t
m
i
n
i
n
g
a
n
d
t
h
e
r
e
s
u
l
t
s
o
b
t
a
i
n
e
d
s
h
o
w
e
d
p
r
o
m
i
s
i
n
g
r
e
s
u
l
t
s
f
o
r
f
u
r
t
h
e
r
e
x
p
l
o
r
a
t
i
o
n
.
Brunat
o
a
nd
Ba
tt
it
i
[4
9]
inv
e
sti
gated
SLS
f
or
direct
trai
ni
ng
of
thr
esh
ol
d
net
wor
ks
.
In
this
stud
y
auth
or
s
a
ppli
ed
SLS
al
go
rith
m
s
fo
r
trai
ni
ng
ne
ur
al
ne
tw
orks
by
in
volv
ing
th
res
ho
l
d
act
ivati
on
f
unc
ti
on
s,
a
novel
te
chn
i
que
known
as
bin
ary
le
arn
i
ng
m
achine
(BL
M).
BLM
wo
r
ks
base
d
on
th
e
pr
inciple
of
changi
ng
ind
ivi
du
al
bits
of each
w
ei
gh
t
and sele
ct
ing i
m
pr
ov
in
g
m
oves.
Brunat
o
an
d
Ba
tt
iti
in
[5
0]
pr
ese
nted
a
new
a
lg
or
it
hm
based
on
m
ulti
scal
e
SLS
with
bin
a
ry
represe
ntati
on
for
trai
ni
ng
ne
ur
al
n
et
w
orks.
In
t
his
w
ork
they
pro
po
s
ed
a
te
le
scop
ic
m
ul
ti
scal
e
ver
sion
of
local
searc
h,
w
her
e t
he nu
m
ber
of
bits was
incr
ease
d
i
n
a
n adaptive
w
ay
,
in
this
way le
ad
ing
t
o
a
faster
s
ear
c
h
process
a
nd
to
local
m
ini
m
a
of
bette
r
qual
it
y.
Laachem
i
and
Bo
ughaci
[
51
]
pr
ese
nte
d
an
inte
resti
ng
work
reg
a
rd
i
ng
W
e
b
ser
vice
cl
as
sific
at
ion
.
In
this
pa
pe
r,
a
uth
ors
c
om
bin
ed
SLS
with
s
uppo
rt
vecto
r
m
achine
(S
VM
)
f
or
We
b
ser
vices
cl
assifi
cat
ion
.
T
hi
s
m
et
ho
d
was
perform
ed
in
two
st
e
ps
,
fi
rst
SLS
m
e
ta
-
he
ur
ist
ic
was
us
e
d
for
featu
re
sel
ect
ion
,
an
d
la
te
r
S
VM
al
gori
thm
s
was
ap
pl
ie
d
to
do
t
he
cl
assifi
cat
io
n
ta
s
k.
Additi
on
al
ly
,
t
his
m
et
ho
d,
w
hich
ty
pical
ly
involve
s
SL
S
a
nd
S
VM
f
or
Web
se
rv
ic
e
cl
as
sific
at
ion
was
furth
e
r
validat
ed
b
y a
ut
hors on the
qu
al
it
y of
w
e
b
se
rv
ic
e
(QWS
) D
at
aset
.
Nekkaa
a
nd
B
oughaci
in
[
52]
pro
po
se
d
a
hybr
i
d
sea
rch
m
et
ho
d
that
i
nteg
rates
ha
rm
on
y
sea
rc
h
al
gorithm
s
(H
SA
)
with
SLS
fo
r
feat
ur
e
sel
ect
ion
as
a
com
bin
at
or
ia
l
optim
iz
at
ion
pro
bl
em
in
cl
assifi
cat
ion
.
They
intr
oduc
e
d
a
novel
sel
ect
ion
strat
e
gy
base
d
on
pro
bab
il
it
y
te
chn
i
qu
e
em
plo
ye
d
in
HAS
-
S
LS
,
w
hich
m
anag
es
to
se
le
ct
the
ap
pro
pr
ia
te
s
olu
t
io
ns
by
usi
ng
SL
S
to
filt
er
out
irreleva
nt
or
r
edun
dan
t
featu
res,
by
pro
vid
in
g
a
go
od
tra
de
-
of
f
be
tween
e
xp
l
or
at
ion
a
nd
e
xp
l
oitat
i
on
.
F
ur
the
r
m
or
e,
the
hybri
d
HAS
-
SLS
i
s
then
integrate
d wit
h a
S
VM
cl
assif
ie
r.
Far
hi
an
d
Bo
ughaci
i
n
[5
3
]
pro
posed
us
e
of
SLS
m
et
ho
d
f
or
s
ol
ving
the
f
re
qu
e
nt
s
ubgr
aph
m
ining
(F
S
M)
pro
blem
.
They
intro
duced
her
e
the
no
ti
on
of
di
versi
ficat
ion
that
com
pr
ise
on
e
of
the
m
os
t
prom
inent
appr
oach
es
a
ppli
ed
t
o
s
olv
e
hard
optim
iz
ation
pro
blem
s.
The
im
ple
m
entat
ion
a
nd
eva
luati
on
of
pro
po
s
ed
m
et
ho
ds
was
t
est
ed
on
se
ver
a
l t
ypes o
f data
set
s includ
i
ng
syntheti
c an
d
r
eal
-
w
or
ld
one.
At the
sam
e tim
e this
m
et
ho
d
was
c
om
par
ed
to
oth
er
sta
te
-
of
-
the
-
art
m
et
ho
ds
currently
us
e
d,
incl
ud
i
ng
l
ocal
searc
h,
GA
a
nd
var
ia
ble
nei
ghbor
hood
sea
rc
h
(VNS
)
al
go
r
it
h
m
s.
This
m
et
hod
was
a
bl
e
to
e
ff
ic
ie
ntl
y
disco
ver
dive
rsified
su
bg
raphs
in
t
he
searc
h
s
pa
ce
thr
ou
gh
ex
plorin
g
ne
w
s
olu
ti
ons
by
usi
ng
ra
ndom
nes
s
durin
g
the
s
earch
process
.
The
obta
ined
resu
lt
s
sh
owe
d
that
S
LS
m
e
tho
d
pro
du
ce
c
om
petit
i
ve
res
ults
and
so
luti
ons
f
ound
ha
ve
bette
r
qu
al
it
y com
par
ed
t
o
th
ose
foun
d by LS
, GA
a
nd
VNS
al
gorithm
s.
Ma
farja
et
al
.
,
[54
]
pro
po
se
d
bi
nar
y
va
ria
nts
f
or
the
la
t
est
ver
sio
n
of
gr
ass
hoppe
r
op
ti
m
iz
ation
al
gorithm
s
(G
OA)
us
e
d
f
or
featur
e
sel
ect
i
on.
More
pr
eci
sel
y,
GOA
al
gorithm
s
is
app
li
ed
to
sel
ect
op
tim
al
featur
e
subset
by
rem
ov
in
g
irreleva
nt
fea
tures,
w
hich
l
at
te
r
will
be
use
d
for
cl
assif
ic
at
ion
pur
pos
es
into
the
wr
a
pper
-
ba
sed
fram
ewo
rk
.
A
ut
hors
he
re
introd
uced
tw
o
m
echan
ism
s
i
n
orde
r
to
desi
gn
a
bi
nar
y
ve
rsi
on
of
GOA.
The
f
irst
m
echan
is
m
,
is
on
e
w
hic
h
is
base
d
on
bo
t
h
Sigm
oid
and
V
-
sh
a
pe
d
trans
fer
functi
on
s
,
an
d
al
gorithm
s
are
nam
ed
as
BGO
A
-
S
an
d
B
GOA
-
V,
res
pe
ct
ively
.
The
s
econd
m
echani
s
m
inv
olv
es
a
novel
te
chn
iq
ue
t
hat
integrates
t
he
best
s
olu
ti
on
obta
ined
so
f
ar.
Additi
on
al
l
y,
a
m
utati
on
operato
r
is
use
d
to
enh
a
nce
the
e
xplo
rati
on
phas
e
in
B
GOA
al
gorithm
and
is
nam
ed
as
B
G
OA
-
M.
These
m
et
ho
ds
are
ev
al
uate
d
on
a
c
onside
ra
bly
seve
ral
be
nch
m
ark
data
set
s
an
d
obta
in
ed
res
ult
sho
w
ed
t
hat
the
pe
r
form
ance
achi
eved
by
BGO
A
an
d
BGO
A
-
M
is
superi
or
w
he
n
c
om
par
ed
to
pe
rfor
m
ance
ob
t
ai
ned
by
oth
e
r
si
m
i
la
r
recent
ly
us
ed
te
chn
iq
ues
in
t
he
li
te
ratu
re.
4.5
.
Oth
er
SLS
appl
ic
at
io
n
In
rece
nt
ye
ars
,
m
or
e
an
d
m
o
re
effor
ts
hav
e
been
de
vote
d
to
the
ap
plica
tio
n
of
SL
S
al
gorithm
s
for
so
lvi
ng
of
se
ve
ral
com
pu
ta
ti
on
al
hard
opti
m
i
z
at
ion
and
s
earch
pro
blem
s.
U
nlike
previ
ou
s
sect
i
on,
he
re
are
br
ie
fly
prese
nted
so
m
e
of
th
e
recent
s
ucce
ssfu
l
a
pp
li
cat
ion
of
SLS
in
oth
e
r
sci
ence
and
i
ndus
try
f
ie
lds
,
includi
ng
S
AT
,
MA
X
-
S
AT
,
TSP,
sche
duli
ng
an
d
routin
g
prob
le
m
s.
Boug
haci
et
al
.
,
[22]
com
bin
e
d
SL
S
m
et
a
-
heu
risti
cs
with
the
le
ast
sign
ific
a
nt
bits
(LSB)
te
c
hn
i
que
f
or
im
age
ste
ganogra
ph
y.
The
a
uthors
i
n
this
stud
y,
im
ple
mented
t
hr
ee
m
et
hods
nam
ed
as
LSB
,
LS
B+
LS
an
d
LS
B+
SLS,
w
hich
achiev
ed
sig
nifican
t
resu
lt
s
by
al
so
dem
on
strat
ing
the
ben
e
fits
of
the
pro
po
s
ed
m
e
tho
d
in
i
m
age
ste
gan
ogra
phy.
Dru
ga
n
[20]
pr
es
e
nted
the
use
of
stoc
hastic
par
et
o
local
s
earch
(
SprLS
)
for
m
any
-
obj
e
ct
ive
qu
a
dr
at
ic
assignm
ent
pr
ob
le
m
(MO
QAP)
in
sta
nces.
P
utik
hin
and
Kasc
heev
[25]
pr
e
sente
d
a
heurist
ic
ap
proac
h
f
or
fin
di
ng
init
ia
l
valu
es
for
SLS
i
n
s
olv
in
g SAT
pr
ob
le
m
b
y usi
ng c
on
ti
nuo
us
e
xtensi
ons
of Bo
olean
for
m
ulas.
Wang
et
al
.
,
[23]
pro
po
se
d
a
novel
rob
ust
op
ti
m
isa
ti
on
ap
proac
h
th
at
inv
ol
ves
es
tim
a
ti
on
of
the
distrib
utio
n
al
gorithm
(ED
A
)
an
d
S
LS
for
ta
ckli
ng
uncertai
n
capaci
ta
te
d
ar
c
routing
pro
blem
s.
Evaluation Warning : The document was created with Spire.PDF for Python.
IS
S
N
:
2088
-
8708
In
t J
Elec
&
C
om
p
En
g,
V
ol.
11
, No
.
1,
Febr
uar
y
2021
:
71
5
-
72
7
724
This
m
et
ho
d
com
bin
es
an
E
DA
with
a
no
vel
two
sta
ge
SLS
proce
dur
e
to
perform
m
ini
m
iz
at
ion
of
the
m
axi
m
al
total
cost
over
a
set
of
di
ff
e
ren
t
sce
nar
i
os
.
He
re
,
t
he
SLS
pr
ocedur
e
is
us
e
d
to
avo
i
d
extrem
e
fitness
values
of
unprom
isi
ng
m
ov
es
in
local
sea
rc
h.
T
he
res
ults
ob
ta
ine
d
e
xper
i
m
enta
ll
y
on
two
set
s
of
be
nc
hm
ark
pro
blem
s sh
owed
that t
his alg
or
it
hm
o
ut
perf
or
m
s stat
e
-
of
-
the
-
a
rt r
es
ults.
Zh
ou
a
nd
H
u
in
[18]
pro
pose
d
gradie
nt
-
base
d
ada
ptiv
e
stochastic
s
earch
for
no
n
-
diff
e
re
ntiable
o
p
t
i
m
i
z
a
t
i
o
n
a
l
g
o
r
i
t
h
m
f
o
r
s
o
l
v
i
n
g
g
e
n
e
r
a
l
o
p
t
i
m
i
z
a
t
i
o
n
p
r
o
b
l
e
m
s
w
i
t
h
l
i
t
t
l
e
s
t
r
u
c
t
u
r
e
.
T
h
e
p
r
o
p
o
s
e
d
a
p
p
r
o
a
c
h
u
s
e
r
a
n
d
o
m
s
a
m
p
l
i
n
g
t
e
c
h
n
i
q
u
e
t
o
i
t
e
r
a
t
i
v
e
l
y
f
i
n
d
h
i
g
h
q
u
a
l
i
t
y
s
o
l
u
t
i
o
n
s
f
r
o
m
a
p
a
r
a
m
e
t
e
r
i
z
e
d
di
stribu
ti
on m
od
el
o
ver
the
so
l
ution
s
pace.
C
hu
et
al
.
,
[
26]
pro
pose
d
a
ne
w
S
LS
al
gorithm
cal
le
d
CC
HNV
(
hard
ne
ig
hbori
ng
var
ia
bles
base
d
co
nf
i
gurati
on
chec
king)
f
or
s
olv
in
g
W
PMS
(w
ei
gh
te
d
pa
rtia
l
m
axi
m
u
m
satisfiabili
ty
).
Re
su
lt
s
ob
ta
in
ed
by
a
nu
m
ber
of
e
xp
e
rim
e
nts
on
a
br
oad
ran
ge
of
WPM
S
instances
sh
owe
d
that
CC
H
N
V
perform
ance outpe
rfor
m
s the
sta
te
-
of
-
the
-
art
SLS
al
gorithm
s r
ece
ntly
u
se
d t
o
s
olv
e
WPMS problem
.
Lu
o
[
27
]
pro
pose
d
a
stoc
has
ti
c
it
erati
ve
evo
luti
on
CT
re
const
ru
ct
io
n
al
gorithm
fo
r
li
m
it
ed
-
ang
le
sp
ars
e
pro
j
ect
ion
data.
T
he
w
orkin
g
pri
nci
ple
of
this
al
go
rithm
is
as
fo
ll
ow:
a
stochastic
appro
ac
h
is
a
pp
li
e
d
for
searc
hing,
and
Ma
rko
v
C
hain
is
us
ed
t
o
pre
dict
it
erati
ve
ev
olu
ti
on
m
od
el
and
acc
el
erate
the
pro
po
s
ed
a
l
g
o
r
i
t
h
m
’
s
c
o
n
v
e
r
g
e
n
c
e
.
T
h
e
e
x
p
e
r
i
m
e
n
t
a
l
r
e
s
u
l
t
s
o
b
t
a
i
n
e
d
b
y
t
h
e
p
r
o
p
o
s
e
d
a
l
g
o
r
i
t
h
m
i
n
i
m
a
g
e
r
e
c
o
n
s
t
r
u
c
t
i
o
n
f
r
o
m
l
i
m
i
t
e
d
-
a
n
g
l
e
s
p
a
r
s
e
p
r
o
j
e
c
t
i
o
n
d
a
t
a
a
r
e
p
r
o
m
i
s
i
n
g
a
n
d
r
o
b
u
s
t
n
e
s
s
.
So
m
e
oth
er
s
uc
cessf
ul
ap
plica
ti
on
s
of
S
LS
introd
uced
rec
ently
include
:
analy
sis
of
SL
S
m
e
tho
ds
us
e
d
f
or so
l
ving the
unr
el
at
e
d par
al
le
l m
achine sc
hedulin
g pro
blem
p
resen
te
d
by Sa
nto
s
et
a
l
.
,
[32], im
pro
ved
g
e
n
e
r
i
c
B
E
T
-
AND
-
R
U
N
s
t
r
a
t
e
g
y
w
i
t
h
p
e
r
f
o
r
m
a
n
c
e
p
r
e
d
i
c
t
i
o
n
f
o
r
S
L
S
b
y
W
e
i
s
e
e
t
a
l
.
,
[
3
3
]
.
O
n
t
h
e
o
t
h
e
r
s
i
d
e
,
L
o
r
e
n
z
a
n
d
W
ö
r
z
[
5
5
]
d
e
m
o
n
s
t
r
a
t
e
d
b
e
n
e
f
i
t
s
o
f
u
s
i
n
g
t
h
e
l
e
a
r
n
e
d
c
l
a
u
s
e
s
o
n
S
L
S
,
f
o
l
l
o
w
e
d
b
y
H
e
a
t
a
l
.
,
[56
]
who
pr
ese
nted
his
a
ppr
oach
f
or
s
olv
in
g
floati
ng
-
point
c
onstra
int
by
us
in
g
S
LS,
Ca
ceres
et
al
.
,
[5
7
]
who
pro
pose
d
us
in
g
of
SLS
in
direct
ape
rt
ur
e
op
ti
m
isa
t
i
on
pro
blem
in
intensit
y
m
od
ulate
d
ra
diati
on
thera
py.
S
usa
n
an
d
Bhu
ta
ni
[
58
]
i
ntr
oduce
d
a
novel
m
e
m
etic
al
gorithm
that
incorp
or
at
es
gr
ee
dy
SL
S
m
uta
ti
on
f
or
cour
se
sche
du
li
ng
a
nd
Chu
et
al
.
,
[59
]
pr
es
ente
d
pr
om
isi
ng
e
m
pirical
inv
est
igati
on
of
usi
ng
SLS
for
so
l
ving
M
AX
-
SA
T
pro
blem
.
5.
CONCL
US
I
O
N
Nowa
days,
t
he
re
is
gr
eat
interest
in
re
search
com
m
un
it
y
an
d
i
ndus
t
ry
to
so
l
ve
c
om
plex
com
bin
at
or
ia
l
op
ti
m
isa
t
ion
a
nd
decisi
on
pr
ob
le
m
s,
as
the
nu
m
ber
of
the
se
is
pr
ese
nt
in
seve
ral
bran
ches
in
sci
ence
an
d
in
du
st
ry
on
wh
ic
h
com
pu
ta
ti
onal
m
e
tho
ds
ha
ve
bee
n
widel
y
app
li
ed.
SL
S
is
ver
y
i
m
po
rtant
and
powe
rful
to
ol
us
e
d
f
or
s
olv
i
ng
ha
r
d
c
om
bi
nato
rial
optim
i
sat
ion
pro
blem
s.
A
s
a
ne
w
a
p
pr
oach
it
is
m
ai
nly
char
act
e
rized
by
ran
dom
iz
ation
durin
g
the
lo
cal
search.
SL
S
al
go
rit
hm
s
h
ave
bee
n
us
e
d
for
a
wh
il
e
to
ta
ckle
hard
c
om
bin
at
or
ia
l
prob
le
m
s
as
the
num
ber
of
t
hese
is
present
i
n
se
veral
br
a
nc
hes
in
sci
ence
a
nd
in
du
st
ry
includi
ng
a
rtif
ic
ia
l
i
ntelligence,
m
achine
le
arn
i
ng,
data
m
ining
,
ope
rati
on
s
re
searc
h,
bio
in
form
at
ics
an
d
el
ect
ro
nic c
omm
erce.
Var
i
ou
s
SLS
m
et
ho
ds
ha
ve
been
em
plo
ye
ed
to
s
olve
c
om
bin
at
or
ia
l
optim
isa
ti
on
prob
le
m
s,
bu
t
us
ua
ll
y
al
l
thes
e
m
et
ho
ds
ha
ve
been
gro
upe
d
in
th
ree
m
ajor
cl
asses:
the
fi
rst
cl
ass
is
known
as
"
sim
ple"
SLS
m
et
ho
ds,
as
th
e
te
rm
den
otes
they
are
m
a
inly
based
on
sim
ple
searc
h
te
ch
niques,
the
sec
ond
cl
ass
c
omprises
of
hybri
d
SLS
m
et
ho
ds,
w
hic
h
inte
gr
at
e
se
ve
ral
searc
hing
te
chn
iq
ues
an
d
he
ur
ist
ic
s
an
d
the
thir
d
cl
as
s
wh
ic
h
is
m
or
e
com
plex
an
d
is
know
n
as
popula
ti
on
-
base
d
SL
S
m
et
ho
ds,
w
hic
h
pr
oceed
s
wit
h
a
set
or
popu
la
ti
on
of
cand
i
date
so
lut
ion
s
.
These
S
L
S
m
et
ho
ds
c
om
pr
ise
s
a
nu
m
ber
of
well
-
kn
own
te
ch
nique
s
m
ai
nly
insp
ir
ed
by
natu
ral phe
nom
ena,
w
hich
have
been wi
del
y
app
li
ed
to
v
a
rio
us
tas
ks
.
Exce
pt
the
fa
ct
that
these
al
gorithm
s
are
widely
us
ed
and
are
powe
rful
to
ol
in
ta
ckling
ha
r
d
com
bin
at
or
ia
l
pro
blem
s,
m
anu
al
co
nf
ig
urat
ion
of
t
hese
al
gorithm
s
is
com
plex
and
ti
m
e
con
su
m
ing
ta
sk
.
In
order
t
o
ov
erp
as
s
these
pro
blem
s
a
lot
of
researc
h
w
ork
ha
s
bee
n
done
with
re
ga
rd
t
o
the
a
uto
m
at
ed
al
gorithm
config
ur
at
io
n
a
nd
par
am
et
er
tu
ning
te
ch
niqu
es.
A
ddit
ion
al
ly
,
so
m
e
of
t
he
la
te
st
su
cc
essfu
l
app
li
cat
io
n
of
SLS
al
gorithm
s
in
m
achine
lear
ni
ng,
data
m
i
ning
an
d
oth
er
fiel
ds
in
sci
e
nc
e
and
i
ndus
try
hav
e
b
e
e
n
p
r
e
s
e
n
t
e
d
.
W
e
h
o
p
e
t
h
a
t
i
s
s
u
e
s
d
i
s
c
u
s
s
e
d
i
n
t
h
i
s
p
a
p
e
r
w
i
l
l
p
u
s
h
f
o
r
w
a
r
d
t
h
e
d
i
s
c
u
s
s
i
o
n
i
n
t
h
e
a
r
e
a
S
L
S
r
e
s
e
a
r
c
h
,
o
n
t
h
e
s
a
m
e
t
i
m
e
i
t
m
a
y
s
e
r
v
e
a
s
c
o
m
p
l
e
m
e
n
t
a
r
y
m
a
t
e
r
i
a
l
f
o
r
o
t
h
e
r
r
e
s
e
a
r
c
h
e
r
s
i
n
t
e
r
e
s
t
e
d
i
n
a
r
e
a
o
f
S
L
S
.
REFERE
NC
ES
[1]
Holger
H
.
Hoos
and
Thomas Stüt
zl
e
,
“
Stocha
st
ic l
oca
l
sea
r
c
h:
Fou
ndat
ions
and app
li
c
at
ions,
”
El
sev
i
er
,
2004
.
[2]
Shen
Li
n
and
Br
i
an
W
.
Kerni
gh
a
n,
“
An
eff
ec
t
ive
heur
isti
c
al
gor
ithm
for
the
tra
v
eling
-
sale
sm
an
pr
oble
m
,
”
Informs
,
vol.
21
,
no
.
2
,
pp
.
498
-
516
,
1973
.
[3]
Thomas
Bac
k
,
“
Evol
uti
on
ar
y
algorithms
in
th
eo
r
y
and
p
racti
c
e:
evol
u
ti
on
str
ate
gie
s,
evol
ut
iona
r
y
progra
m
m
ing,
gene
t
ic
al
gor
it
h
m
s,”
Oxford
uni
ve
rs
it
y
press
,
19
96.
Evaluation Warning : The document was created with Spire.PDF for Python.
In
t J
Elec
&
C
om
p
En
g
IS
S
N: 20
88
-
8708
Sto
c
hasti
c local
sea
rc
h: A st
ate
-
of
-
t
he
-
ar
t rev
ie
w
(
Muham
et
Kastrati
)
725
[4]
Scott
Kirkpa
tric
k,
C
.
Daniel
G
el
a
tt
,
and
Mar
i
o
P
.
Vec
chi
,
“
Opti
m
iz
at
io
n
b
y
sim
ula
te
d
ann
ea
l
ing,
”
Scienc
e
,
vol.
220
,
no
.
459
8,
pp
.
671
-
680
,
1983.
[5]
Ala
in
Her
tz
and
Dom
ini
que
de
W
err
a,
“
Us
ing
t
abu
sea
r
c
h
tech
nique
s
for
g
rap
h
col
or
ing,”
Co
mputing
,
vo
l.
39
,
no.
4
,
pp
.
345
-
3
51,
1987
.
[6]
Bart
Selman,
H
e
ct
or
J
.
L
eve
sque
,
David
G
.
Mit
ch
el
l
,
et
al.
,
“
A
ne
w m
et
hod
for
solving
har
d
sat
isfi
a
bil
i
t
y
p
r
o
b
l
e
m
s
,
”
In
P
r
o
c
e
e
d
i
n
g
s
T
e
n
t
h
N
a
t
i
o
n
a
l
C
o
n
f
e
r
e
n
c
e
o
n
A
r
t
i
f
i
c
i
a
l
I
n
t
e
l
l
i
g
e
n
c
e
(
A
A
A
I
-
92)
,
v
o
l
.
9
2
,
p
p
.
4
4
0
-
4
4
6
,
1
9
9
2
.
[7]
Bart
Selman,
H
e
nr
y
A
.
Kau
tz,
an
d
Bram
Cohen,
“
Noise
strat
eg
ie
s for
improvin
g
lo
ca
l
se
arc
h
,
”
In
A
AA
I
-
94
,
vo
l.
94
,
pp.
337
-
343
,
19
94.
[8]
Frank
Hutte
r
,
H
olge
r
H
.
Hoos
,
Kevin
L
e
y
ton
-
B
rown,
and
Thomas
Stütz
le,
“
P
ara
m
il
s:
an
aut
o
m
at
ic
al
gori
thm
conf
iguration fra
m
ework,
”
Journ
al
of
Artificia
l
In
te
lligen
ce R
ese
a
rch,
vol
.
36
,
pp
.
267
-
306,
2009
.
[9]
Holger
H
.
Hoos
,
“
Autom
at
ed
a
lg
orit
hm
con
figur
a
ti
on
and
p
ara
m
eter
tun
ing,”
Au
to
nomous
search
,
pp.
37
-
71
,
2011
.
[10]
Marie
-
E
lé
onor
e M
armion,
Franc
o
Mascia,
Manu
el
L
óp
ez
-
Ib
ánez, a
nd
Thomas Stü
tz
l
e,
“
Autom
at
ic de
sign o
f
h
y
brid
s
t
o
c
h
a
s
t
i
c
l
o
c
a
l
s
e
a
r
c
h
a
l
g
o
r
i
t
h
m
s
,
”
I
n
I
n
t
e
r
n
a
t
i
o
n
a
l
W
o
r
k
s
h
o
p
o
n
H
y
b
r
i
d
M
e
t
a
h
e
u
r
i
s
t
i
c
s
,
v
o
l
.
7
9
1
9
,
p
p
.
1
4
4
-
1
5
8
,
2
0
1
3
.
[11]
Franc
o
Masci
a,
Manue
l
Lóp
ez
-
I
báñe
z
,
Jér
émie
Dub
ois
-
La
coste,
and
Thomas
St
ütz
l
e,
“
Gram
m
ar
-
base
d
gen
era
t
io
n
of
stocha
sti
c
lo
c
al
se
arc
h
h
eur
ist
ic
s
through
aut
o
m
at
ic
al
gori
thm
conf
ig
ur
at
ion
t
ools
,”
Comput
er
s
&
operati
ons
research
,
vo
l. 51
,
pp
.
190
-
199
,
2
014.
[12]
Feder
i
co
Pagno
zz
i
and
Thomas
Stütz
le,
“
Aut
om
at
ic
d
esign
of
h
y
br
id
stoch
asti
c
local
se
ar
ch
al
go
rit
hm
s
fo
r
per
m
uta
ti
on
flo
ws
hop
proble
m
s,”
European Jou
rnal
of
Operat
io
nal
R
ese
arch
,
vo
l.
276
,
no
.
2
,
pp
.
409
-
421,
2019
.
[13]
Franc
o
Masci
a,
Manue
l
Lóp
ez
-
I
báñe
z
,
Jéré
m
i
e
Dubois
-
La
coste,
Marie
-
E
lé
on
or
e
Marm
ion,
and
Thomas
Stütz
le,
“
Algorit
hm
compari
son
b
y
aut
o
m
at
ic
a
lly
conf
ig
ura
ble
stocha
st
i
c
lo
ca
l
sea
r
ch
f
rameworks:
A
c
ase
stud
y
usin
g
flow
-
shop
sche
d
uli
ng
prob
le
m
s,”
In
Int
ernati
onal
Workshop
on
H
ybrid
Me
tahe
uri
stic
s
,
vo
l.
8457,
pp.
30
-
44
,
2014
.
[14]
Alb
ert
o
Franz
in
and
Thomas
Stüt
zl
e
,
“
Revi
sit
ing
sim
ula
te
d
ann
eal
ing:
A
compone
nt
-
base
d
an
aly
si
s,”
Computers
&
Operations
Re
se
arch
,
vol
.
104
,
p
p.
191
-
206
,
201
9.
[15]
Alber
to
Franz
in
,
Le
slie
Pére
z
Các
ere
s,
and
Thom
as
Stütz
le,
“
Eff
ec
t
of
tra
nsform
ations
of
nu
m
eri
c
al
par
amet
ers
in
au
tomatic
al
gori
t
hm
conf
igura
t
io
n,
”
Opt
imizati
on
Letters
,
vol
.
12
,
no.
8
,
pp
.
1741
-
1753,
2018
.
[16]
Alb
ert
o
Franz
in
and
Thomas
Stütz
l
e,
“
Com
p
ari
son
of
acce
pt
a
nce
c
riter
ia
in
ran
dom
iz
ed
local
sea
r
che
s,
”
In
Inte
rnational
Co
nfe
renc
e
on
Artif
ic
ial E
vol
ut
ion
(
Ev
olution
Arti
f
iciel
l
e)
,
vol
.
1076
4,
pp
.
16
-
29
,
20
17.
[17]
Zongxu
Mu,
Ho
lge
r
H.
Hoos
,
and
Thomas
Stütz
l
e,
“
The
impact
of
aut
om
ate
d
al
gori
thm
conf
igura
t
ion
on
t
h
e
sca
li
ng
beh
avi
o
ur
of
stat
e
-
of
-
th
e
-
art
in
exact
tsp
solvers,
”
In
Int
ernati
onal
Conf
ere
nce
on
Learning
and
Inte
llige
nt
Optimizati
on
,
vo
l.
10079
,
pp
.
157
-
172
,
2016
.
[18]
Enl
u
Zhou
and
Jiaqi
ao
Hu
,
“
Gradi
en
t
-
base
d
adaptive
sto
cha
st
ic
sea
rch
for
non
-
d
iffe
ren
ti
ab
le
optim
iz
at
ion
,
”
I
EEE
Tr
ansacti
ons o
n
Aut
omatic Cont
r
ol
,
vo
l. 59, no. 7, pp. 1818
-
1832,
2014.
[19]
Christophe
r
D
.
Rosi
n,
“
Unw
ei
ghte
d
stocha
st
ic
l
oca
l
sea
r
ch
ca
n
be
eff
e
c
ti
v
e
for
ran
dom
csp
benc
hm
ark
s
,”
arXiv
pre
print
arXi
v:1
411.
7480
,
2014.
[20]
Mada
li
n
a
M.
Drugan,
“
Stoch
astic
par
et
o
local
se
arc
h
for
m
an
y
o
bje
c
ti
ve
quadr
at
i
c
assignm
ent
pr
oble
m
insta
n
ce
s
,
”
2015
IEEE
Con
gress
on
Ev
olu
tionar
y
Computation (
CEC)
,
Sendai
,
pp
.
1754
-
1761
,
2015
.
[21]
Andrea
s
Fröhlic
h,
Arm
in
Bie
re,
Christoph
M
.
Wi
nte
rstei
g
er,
a
nd
Youssef
Ham
adi
,
“
Stocha
sti
c
loc
al
sea
rch
f
or
sati
sfiability
m
o
dul
o
the
ori
es
,”
Proce
ed
ings
of
the
Tw
ent
y
-
N
i
nth
AA
AI
Con
f
ere
nce
on
Arti
f
ic
ial
In
te
l
li
g
ence
,
pp.
1136
-
1143
,
2015.
[22]
Dali
l
a
Boughaci,
Abdelha
f
id
Ke
m
ouche
,
and
Ho
ci
ne
L
ac
h
ibi
,
“
St
ocha
sti
c
local
se
arc
h
combined
with
lsb
te
chn
iq
ue
for
image
stega
n
ogra
ph
y
,”
2016
13th
Learning
a
nd
Technol
og
y Confe
renc
e
(
L
&
T)
,
Jedda
h
,
pp
.
1
-
9,
2016
.
[23]
Juan
W
ang,
Ke
Ta
ng,
Jos
e
A
.
L
oza
no,
and
Xin
Yao,
“
Esti
m
at
io
n
of
the
d
istri
bu
ti
on
a
lgori
thm
with
a
stoch
astic
loc
a
l
sea
rch
for
unce
rt
ai
n
c
apaci
t
at
ed
arc
rout
i
ng
proble
m
s,”
I
EE
E
Tr
ansacti
o
ns
on
Ev
olut
ion
ary
Computati
o
n
,
vol.
20
,
no
.
1
,
pp
.
96
-
109
,
2016
.
[24]
Abde
ll
ah
Rez
ou
g
and
Da
li
l
a
Bo
ughac
i
,
”
A
self
-
ada
pt
ive
h
armony
se
arc
h
combin
ed
with
a
sto
chas
ti
c
lo
cal
se
arch
for
the
0
-
1
m
ult
i
dimensional
kna
psac
k
probl
em,
”
Inte
rnational
Jo
urnal
of
Bi
o
-
Ins
pired
Computation
,
vol.
8
,
no.
4
,
pp
.
234
-
239
,
20
16.
[25]
Nikit
a
Putikh
in
and
Nikolay
K
a
sche
ev,
“
A
heur
isti
c
to
find
ini
t
i
al
va
lue
s
for
sto
cha
sti
c
local
se
a
rch
in
sa
t
using
cont
inuous
exte
n
sions
of
boole
an
form
ula
s,”
2017
IEE
E
East
-
W
est
Design
&
Te
st
Symposium
(
EWD
TS)
,
Novi
Sad
,
pp.
1
-
4
,
2017
.
[26]
Yi
Chu,
Chuan
Luo,
W
enxua
n
Huan
g,
Haih
a
ng
You,
and
D
ongrui
Fan,
“
Hard
nei
ghbor
ing
var
ia
b
le
s
base
d
conf
iguration
ch
ec
king
in
sto
ch
a
stic
local
sea
r
ch
for
weight
ed
p
art
i
al
m
axi
m
um
sati
sfiability
,
”
2017
IEE
E
29th
Inte
rnational
Co
nfe
renc
e
on
Tool
s wit
h
Arti
f
ic
ia
l
I
nte
lligen
ce (
ICT
AI)
,
Boston,
MA
,
pp
.
139
-
146
,
2
017.
[27]
La
n
Luo,
Hong
xia
Gao
,
Yingh
ao
Luo
,
and
Yongfei
Ch
en,
“
A
stocha
sti
c
i
terati
v
e
evo
lut
ion
ct
r
ec
onstru
ct
i
on
al
gorit
hm
for
lim
it
ed
-
angle
sparse
proje
ction
dat
a
,
”
2017
IEEE
Inte
rnational
C
onfe
renc
e
on
Bioinformatic
s
and
Bi
omedi
ci
ne
(
BIB
M)
,
Kansas Ci
t
y
,
MO
,
pp
.
740
-
744
,
2017
.
[28]
Tong
Yu,
Branis
la
v
Kveton,
an
d
Ole
J
.
M
engsh
oel
,
“
Thomps
on
sam
pli
ng
for
op
t
imizi
ng
stoch
asti
c
lo
ca
l
sea
r
ch,”
In
J
o
i
n
t
E
u
r
o
p
e
a
n
C
o
n
f
e
r
e
n
c
e
o
n
M
a
c
h
i
n
e
L
e
a
r
n
i
n
g
a
n
d
K
n
o
w
l
e
d
g
e
D
i
s
c
o
v
e
r
y
i
n
D
a
t
a
b
a
s
e
s
,
v
o
l
.
1
0
5
3
4
,
p
p
.
4
9
3
-
510
,
2
0
1
7
.
[29]
Sabrina
Olive
i
ra
,
Moham
ed
Saiful
la
h
Hus
sin,
Andrea
Roli
,
M
arc
o
Dorigo,
an
d
Thomas
Stütz
le
,
“
Anal
y
s
is
of
the
popu
la
t
ion
-
base
d
an
t
co
lon
y
op
ti
m
izati
on
al
gorit
hm
for
t
he
ts
p
and
th
e
qap,”
2017
IE
EE
Congress
o
n
Ev
olutionar
y
Co
mputati
on
(
CEC
)
,
San
Sebastian
,
p
p.
1734
-
1741,
2017.
[30]
Lui
s
Paquete
and
Thomas
Stütz
l
e,
“
Stoch
asti
c
lo
cal
sea
r
c
h
al
gorit
hm
s
for
m
ult
iobj
e
c
tive
combinat
or
i
al
opti
m
iz
ation,
”
Handbook
of
Approxi
mation
Al
gorithms
and
Me
tahe
uristi
cs:
Me
thol
ogi
es
and
Tr
adit
ional
Appl
ic
a
ti
ons
,
vo
l.
1
,
2018
.
[31]
Dangd
ang
Niu,
Le
i
Li
u
,
and
Shuai
Lü,
“
New
stocha
stic
local
sea
r
ch
app
roa
ch
es
for
computing
pre
fer
red
extensions
of
abstract
arg
u
m
ent
at
ion
,
”
AI
Comm
unic
ati
ons
,
vol.
31
,
no
.
5439
,
pp
.
1
-
14
,
2016
.
Evaluation Warning : The document was created with Spire.PDF for Python.