<s>
A	O
hyper-heuristic	B-Algorithm
is	O
a	O
heuristic	B-Algorithm
search	O
method	O
that	O
seeks	O
to	O
automate	O
,	O
often	O
by	O
the	O
incorporation	O
of	O
machine	O
learning	O
techniques	O
,	O
the	O
process	O
of	O
selecting	O
,	O
combining	O
,	O
generating	O
or	O
adapting	O
several	O
simpler	O
heuristics	B-Algorithm
(	O
or	O
components	O
of	O
such	O
heuristics	B-Algorithm
)	O
to	O
efficiently	O
solve	O
computational	O
search	O
problems	O
.	O
</s>
<s>
One	O
of	O
the	O
motivations	O
for	O
studying	O
hyper-heuristics	B-Algorithm
is	O
to	O
build	O
systems	O
which	O
can	O
handle	O
classes	O
of	O
problems	O
rather	O
than	O
solving	O
just	O
one	O
problem	O
.	O
</s>
<s>
There	O
might	O
be	O
multiple	O
heuristics	B-Algorithm
from	O
which	O
one	O
can	O
choose	O
for	O
solving	O
a	O
problem	O
,	O
and	O
each	O
heuristic	B-Algorithm
has	O
its	O
own	O
strength	O
and	O
weakness	O
.	O
</s>
<s>
The	O
idea	O
is	O
to	O
automatically	O
devise	O
algorithms	O
by	O
combining	O
the	O
strength	O
and	O
compensating	O
for	O
the	O
weakness	O
of	O
known	O
heuristics	B-Algorithm
.	O
</s>
<s>
In	O
a	O
typical	O
hyper-heuristic	B-Algorithm
framework	O
there	O
is	O
a	O
high-level	O
methodology	O
and	O
a	O
set	O
of	O
low-level	O
heuristics	B-Algorithm
(	O
either	O
constructive	O
or	O
perturbative	O
heuristics	B-Algorithm
)	O
.	O
</s>
<s>
Given	O
a	O
problem	O
instance	O
,	O
the	O
high-level	O
method	O
selects	O
which	O
low-level	O
heuristic	B-Algorithm
should	O
be	O
applied	O
at	O
any	O
given	O
time	O
,	O
depending	O
upon	O
the	O
current	O
problem	O
state	O
(	O
or	O
search	O
stage	O
)	O
determined	O
by	O
features	O
.	O
</s>
<s>
The	O
fundamental	O
difference	O
between	O
metaheuristics	B-Algorithm
and	O
hyper-heuristics	B-Algorithm
is	O
that	O
most	O
implementations	O
of	O
metaheuristics	B-Algorithm
search	O
within	O
a	O
search	O
space	O
of	O
problem	O
solutions	O
,	O
whereas	O
hyper-heuristics	B-Algorithm
always	O
search	O
within	O
a	O
search	O
space	O
of	O
heuristics	B-Algorithm
.	O
</s>
<s>
Thus	O
,	O
when	O
using	O
hyper-heuristics	B-Algorithm
,	O
we	O
are	O
attempting	O
to	O
find	O
the	O
right	O
method	O
or	O
sequence	O
of	O
heuristics	B-Algorithm
in	O
a	O
given	O
situation	O
rather	O
than	O
trying	O
to	O
solve	O
a	O
problem	O
directly	O
.	O
</s>
<s>
Hyper-heuristics	B-Algorithm
could	O
be	O
regarded	O
as	O
"	O
off-the-peg	O
"	O
methods	O
as	O
opposed	O
to	O
"	O
made-to-measure	O
"	O
metaheuristics	B-Algorithm
.	O
</s>
<s>
They	O
aim	O
to	O
be	O
generic	O
methods	O
,	O
which	O
should	O
produce	O
solutions	O
of	O
acceptable	O
quality	O
,	O
based	O
on	O
a	O
set	O
of	O
easy-to-implement	O
low-level	O
heuristics	B-Algorithm
.	O
</s>
<s>
Many	O
researchers	O
from	O
computer	B-General_Concept
science	I-General_Concept
,	O
artificial	B-Application
intelligence	I-Application
and	O
operational	O
research	O
have	O
already	O
acknowledged	O
the	O
need	O
for	O
developing	O
automated	O
systems	O
to	O
replace	O
the	O
role	O
of	O
a	O
human	O
expert	O
in	O
such	O
situations	O
.	O
</s>
<s>
One	O
of	O
the	O
main	O
ideas	O
for	O
automating	O
the	O
design	O
of	O
heuristics	B-Algorithm
requires	O
the	O
incorporation	O
of	O
machine	O
learning	O
mechanisms	O
into	O
algorithms	O
to	O
adaptively	O
guide	O
the	O
search	O
.	O
</s>
<s>
Both	O
learning	O
and	O
adaptation	O
processes	O
can	O
be	O
realised	O
on-line	O
or	O
off-line	O
,	O
and	O
be	O
based	O
on	O
constructive	O
or	O
perturbative	O
heuristics	B-Algorithm
.	O
</s>
<s>
A	O
hyper-heuristic	B-Algorithm
usually	O
aims	O
at	O
reducing	O
the	O
amount	O
of	O
domain	O
knowledge	O
in	O
the	O
search	O
methodology	O
.	O
</s>
<s>
The	O
resulting	O
approach	O
should	O
be	O
cheap	O
and	O
fast	O
to	O
implement	O
,	O
requiring	O
less	O
expertise	O
in	O
either	O
the	O
problem	O
domain	O
or	O
heuristic	B-Algorithm
methods	O
,	O
and	O
(	O
ideally	O
)	O
it	O
would	O
be	O
robust	O
enough	O
to	O
effectively	O
handle	O
a	O
range	O
of	O
problem	O
instances	O
from	O
a	O
variety	O
of	O
domains	O
.	O
</s>
<s>
The	O
goal	O
is	O
to	O
raise	O
the	O
level	O
of	O
generality	O
of	O
decision	O
support	O
methodology	O
perhaps	O
at	O
the	O
expense	O
of	O
reduced	O
-	O
but	O
still	O
acceptable	O
-	O
solution	O
quality	O
when	O
compared	O
to	O
tailor-made	O
metaheuristic	B-Algorithm
approaches	O
.	O
</s>
<s>
In	O
order	O
to	O
reduce	O
the	O
gap	O
between	O
tailor-made	O
schemes	O
and	O
hyperheuristic-based	O
strategies	O
,	O
parallel	O
hyperheuristics	B-Algorithm
have	O
been	O
proposed	O
.	O
</s>
<s>
The	O
term	O
"	O
hyperheuristics	B-Algorithm
"	O
was	O
first	O
coined	O
in	O
a	O
2000	O
publication	O
by	O
Cowling	O
and	O
Soubeiga	O
,	O
who	O
used	O
it	O
to	O
describe	O
the	O
idea	O
of	O
"	O
heuristics	B-Algorithm
to	O
choose	O
heuristics	B-Algorithm
"	O
.	O
</s>
<s>
They	O
used	O
a	O
"	O
choice	O
function	O
"	O
machine	O
learning	O
approach	O
which	O
trades	O
off	O
exploitation	O
and	O
exploration	O
in	O
choosing	O
the	O
next	O
heuristic	B-Algorithm
to	O
use	O
.	O
</s>
<s>
Subsequently	O
,	O
Cowling	O
,	O
Soubeiga	O
,	O
Kendall	O
,	O
Han	O
,	O
Ross	O
and	O
other	O
authors	O
investigated	O
and	O
extended	O
this	O
idea	O
in	O
areas	O
such	O
as	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
,	O
and	O
pathological	O
low	O
level	O
heuristics	B-Algorithm
.	O
</s>
<s>
Although	O
the	O
term	O
was	O
not	O
then	O
in	O
use	O
,	O
this	O
was	O
the	O
first	O
"	O
hyper-heuristic	B-Algorithm
"	O
paper	O
.	O
</s>
<s>
Another	O
root	O
inspiring	O
the	O
concept	O
of	O
hyper-heuristics	B-Algorithm
comes	O
from	O
the	O
field	O
of	O
artificial	B-Application
intelligence	I-Application
.	I-Application
</s>
<s>
More	O
specifically	O
,	O
it	O
comes	O
from	O
work	O
on	O
automated	B-Application
planning	I-Application
systems	O
,	O
and	O
its	O
eventual	O
focus	O
towards	O
the	O
problem	O
of	O
learning	O
control	O
knowledge	O
.	O
</s>
<s>
The	O
system	O
can	O
be	O
characterized	O
as	O
a	O
hill-climbing	B-Algorithm
search	O
in	O
the	O
space	O
of	O
possible	O
control	O
strategies	O
.	O
</s>
<s>
Hyper-heuristic	B-Algorithm
approaches	O
so	O
far	O
can	O
be	O
classified	O
into	O
two	O
main	O
categories	O
.	O
</s>
<s>
In	O
the	O
first	O
class	O
,	O
captured	O
by	O
the	O
phrase	O
heuristics	B-Algorithm
to	O
choose	O
heuristics	B-Algorithm
,	O
the	O
hyper-heuristic	B-Algorithm
framework	O
is	O
provided	O
with	O
a	O
set	O
of	O
pre-existing	O
,	O
generally	O
widely	O
known	O
heuristics	B-Algorithm
for	O
solving	O
the	O
target	O
problem	O
.	O
</s>
<s>
The	O
task	O
is	O
to	O
discover	O
a	O
good	O
sequence	O
of	O
applications	O
of	O
these	O
heuristics	B-Algorithm
(	O
also	O
known	O
as	O
low-level	O
heuristics	B-Algorithm
within	O
the	O
domain	O
of	O
hyper-heuristics	B-Algorithm
)	O
for	O
efficiently	O
solving	O
the	O
problem	O
.	O
</s>
<s>
At	O
each	O
decision	O
stage	O
,	O
a	O
heuristic	B-Algorithm
is	O
selected	O
through	O
a	O
component	O
called	O
selection	O
mechanism	O
and	O
applied	O
to	O
an	O
incumbent	O
solution	O
.	O
</s>
<s>
The	O
new	O
solution	O
produced	O
from	O
the	O
application	O
of	O
the	O
selected	O
heuristic	B-Algorithm
is	O
accepted/rejected	O
based	O
on	O
another	O
component	O
called	O
acceptance	O
criterion	O
.	O
</s>
<s>
In	O
the	O
second	O
class	O
,	O
heuristics	B-Algorithm
to	O
generate	O
heuristics	B-Algorithm
,	O
the	O
key	O
idea	O
is	O
to	O
"	O
evolve	O
new	O
heuristics	B-Algorithm
by	O
making	O
use	O
of	O
the	O
components	O
of	O
known	O
heuristics.	O
"	O
</s>
<s>
The	O
process	O
requires	O
,	O
as	O
in	O
the	O
first	O
class	O
of	O
hyper-heuristics	B-Algorithm
,	O
the	O
selection	O
of	O
a	O
suitable	O
set	O
of	O
heuristics	B-Algorithm
known	O
to	O
be	O
useful	O
in	O
solving	O
the	O
target	O
problem	O
.	O
</s>
<s>
However	O
,	O
instead	O
of	O
supplying	O
these	O
directly	O
to	O
the	O
framework	O
,	O
the	O
heuristics	B-Algorithm
are	O
first	O
decomposed	O
into	O
their	O
basic	O
components	O
.	O
</s>
<s>
additional	O
orthogonal	O
classification	O
of	O
hyper-heuristics	B-Algorithm
considers	O
the	O
source	O
providing	O
feedback	O
during	O
the	O
learning	O
process	O
,	O
which	O
can	O
be	O
either	O
one	O
instance	O
(	O
on-line	O
learning	O
)	O
or	O
many	O
instances	O
of	O
the	O
underlying	O
problem	O
studied	O
(	O
off-line	O
learning	O
)	O
.	O
</s>
<s>
Discover	O
good	O
combinations	O
of	O
fixed	O
,	O
human-designed	O
,	O
well-known	O
low-level	O
heuristics	B-Algorithm
.	O
</s>
<s>
Generate	O
new	O
heuristic	B-Algorithm
methods	O
using	O
basic	O
components	O
of	O
previously	O
existing	O
heuristic	B-Algorithm
methods	O
.	O
</s>
<s>
The	O
learning	O
takes	O
place	O
while	O
the	O
algorithm	O
is	O
solving	O
an	O
instance	O
of	O
a	O
problem	O
,	O
therefore	O
,	O
task-dependent	O
local	O
properties	O
can	O
be	O
used	O
by	O
the	O
high-level	O
strategy	O
to	O
determine	O
the	O
appropriate	O
low-level	O
heuristic	B-Algorithm
to	O
apply	O
.	O
</s>
<s>
Examples	O
of	O
on-line	O
learning	O
approaches	O
within	O
hyper-heuristics	B-Algorithm
are	O
:	O
the	O
use	O
of	O
reinforcement	O
learning	O
for	O
heuristic	B-Algorithm
selection	O
,	O
and	O
generally	O
the	O
use	O
of	O
metaheuristics	B-Algorithm
as	O
high-level	O
search	O
strategies	O
over	O
a	O
search	O
space	O
of	O
heuristics	B-Algorithm
.	O
</s>
<s>
within	O
hyper-heuristics	B-Algorithm
are	O
:	O
learning	B-Algorithm
classifier	I-Algorithm
systems	I-Algorithm
,	O
case-base	O
reasoning	O
and	O
genetic	B-Algorithm
programming	I-Algorithm
.	O
</s>
<s>
An	O
extended	O
classification	O
of	O
selection	O
hyper-heuristics	B-Algorithm
was	O
provided	O
in	O
2020	O
,	O
to	O
provide	O
a	O
more	O
comprehensive	O
categorisation	O
of	O
contemporary	O
selection	O
hyper-heuristic	B-Algorithm
methods	O
.	O
</s>
<s>
Hyper-heuristics	B-Algorithm
have	O
been	O
applied	O
across	O
many	O
different	O
problems	O
.	O
</s>
<s>
Indeed	O
,	O
one	O
of	O
the	O
motivations	O
of	O
hyper-heuristics	B-Algorithm
is	O
to	O
be	O
able	O
to	O
operate	O
across	O
different	O
problem	O
types	O
.	O
</s>
<s>
The	O
following	O
list	O
is	O
a	O
non-exhaustive	O
selection	O
of	O
some	O
of	O
the	O
problems	O
and	O
fields	O
in	O
which	O
hyper-heuristics	B-Algorithm
have	O
been	O
explored	O
:	O
</s>
<s>
Hyper-heuristics	B-Algorithm
are	O
not	O
the	O
only	O
approach	O
being	O
investigated	O
in	O
the	O
quest	O
for	O
more	O
general	O
and	O
applicable	O
search	O
methodologies	O
.	O
</s>
<s>
Many	O
researchers	O
from	O
computer	B-General_Concept
science	I-General_Concept
,	O
artificial	B-Application
intelligence	I-Application
and	O
operational	O
research	O
have	O
already	O
acknowledged	O
the	O
need	O
for	O
developing	O
automated	O
systems	O
to	O
replace	O
the	O
role	O
of	O
a	O
human	O
expert	O
in	O
the	O
process	O
of	O
tuning	O
and	O
adapting	O
search	O
methodologies	O
.	O
</s>
