<s>
In	O
artificial	O
intelligence	O
,	O
genetic	B-Algorithm
programming	I-Algorithm
(	O
GP	O
)	O
is	O
a	O
technique	O
of	O
evolving	O
programs	O
,	O
starting	O
from	O
a	O
population	O
of	O
unfit	O
(	O
usually	O
random	O
)	O
programs	O
,	O
fit	O
for	O
a	O
particular	O
task	O
by	O
applying	O
operations	O
analogous	O
to	O
natural	O
genetic	O
processes	O
to	O
the	O
population	O
of	O
programs	O
.	O
</s>
<s>
Although	O
the	O
idea	O
of	O
evolving	O
programs	O
,	O
initially	O
in	O
the	O
computer	O
language	O
Lisp	B-Language
,	O
was	O
current	O
amongst	O
John	O
Holland	O
’s	O
students	O
,	O
it	O
was	O
not	O
until	O
they	O
organised	O
the	O
first	O
Genetic	B-Algorithm
Algorithms	I-Algorithm
(	O
GA	O
)	O
conference	O
in	O
Pittsburgh	O
that	O
Nichael	O
Cramer	O
published	O
evolved	O
programs	O
in	O
two	O
specially	O
designed	O
languages	O
,	O
which	O
included	O
the	O
first	O
statement	O
of	O
modern	O
"	O
tree-based	O
"	O
Genetic	B-Algorithm
Programming	I-Algorithm
(	O
that	O
is	O
,	O
procedural	O
languages	O
organized	O
in	O
tree-based	O
structures	O
and	O
operated	O
on	O
by	O
suitably	O
defined	O
GA-operators	O
)	O
.	O
</s>
<s>
Koza	O
followed	O
this	O
with	O
205	O
publications	O
on	O
“	O
Genetic	B-Algorithm
Programming	I-Algorithm
”	O
(	O
GP	O
)	O
,	O
name	O
coined	O
by	O
David	O
Goldberg	O
,	O
also	O
a	O
PhD	O
student	O
of	O
John	O
Holland	O
.	O
</s>
<s>
Subsequently	O
,	O
there	O
was	O
an	O
enormous	O
expansion	O
of	O
the	O
number	O
of	O
publications	O
with	O
the	O
Genetic	B-Algorithm
Programming	I-Algorithm
Bibliography	O
,	O
surpassing	O
10,000	O
entries	O
.	O
</s>
<s>
In	O
2010	O
,	O
Koza	O
listed	O
77	O
results	O
where	O
Genetic	B-Algorithm
Programming	I-Algorithm
was	O
human	O
competitive	O
.	O
</s>
<s>
In	O
1996	O
,	O
Koza	O
started	O
the	O
annual	O
Genetic	B-Algorithm
Programming	I-Algorithm
conference	O
which	O
was	O
followed	O
in	O
1998	O
by	O
the	O
annual	O
EuroGP	O
conference	O
,	O
and	O
the	O
first	O
book	O
in	O
a	O
GP	O
series	O
edited	O
by	O
Koza	O
.	O
</s>
<s>
GP	O
continued	O
to	O
flourish	O
,	O
leading	O
to	O
the	O
first	O
specialist	O
GP	O
journal	O
and	O
three	O
years	O
later	O
(	O
2003	O
)	O
the	O
annual	O
Genetic	B-Algorithm
Programming	I-Algorithm
Theory	O
and	O
Practice	O
(	O
GPTP	O
)	O
workshop	O
was	O
established	O
by	O
Rick	O
Riolo	O
.	O
</s>
<s>
Genetic	B-Algorithm
Programming	I-Algorithm
papers	O
continue	O
to	O
be	O
published	O
at	O
a	O
diversity	O
of	O
conferences	O
and	O
associated	O
journals	O
.	O
</s>
<s>
Early	O
work	O
that	O
set	O
the	O
stage	O
for	O
current	O
genetic	B-Algorithm
programming	I-Algorithm
research	O
topics	O
and	O
applications	O
is	O
diverse	O
,	O
and	O
includes	O
software	B-Application
synthesis	I-Application
and	O
repair	O
,	O
predictive	O
modeling	O
,	O
data	O
mining	O
,	O
financial	O
modeling	O
,	O
soft	O
sensors	O
,	O
design	O
,	O
and	O
image	O
processing	O
.	O
</s>
<s>
GP	O
evolves	O
computer	O
programs	O
,	O
traditionally	O
represented	O
in	O
memory	O
as	O
tree	B-Data_Structure
structures	I-Data_Structure
.	O
</s>
<s>
Thus	O
traditionally	O
GP	O
favors	O
the	O
use	O
of	O
programming	O
languages	O
that	O
naturally	O
embody	O
tree	B-Data_Structure
structures	I-Data_Structure
(	O
for	O
example	O
,	O
Lisp	B-Language
;	O
other	O
functional	B-Language
programming	I-Language
languages	I-Language
are	O
also	O
suitable	O
)	O
.	O
</s>
<s>
Non-tree	O
representations	O
have	O
been	O
suggested	O
and	O
successfully	O
implemented	O
,	O
such	O
as	O
linear	B-Algorithm
genetic	I-Algorithm
programming	I-Algorithm
which	O
suits	O
the	O
more	O
traditional	O
imperative	B-Application
languages	I-Application
[	O
see	O
,	O
for	O
example	O
,	O
Banzhaf	O
et	O
al	O
.	O
</s>
<s>
µGP	O
uses	O
directed	O
multigraphs	O
to	O
generate	O
programs	O
that	O
fully	O
exploit	O
the	O
syntax	O
of	O
a	O
given	O
assembly	B-Language
language	I-Language
.	O
</s>
<s>
Multi	B-Algorithm
expression	I-Algorithm
programming	I-Algorithm
uses	O
Three-address	B-Application
code	I-Application
for	O
encoding	O
solutions	O
.	O
</s>
<s>
Cartesian	B-Algorithm
genetic	I-Algorithm
programming	I-Algorithm
is	O
another	O
form	O
of	O
GP	O
,	O
which	O
uses	O
a	O
graph	O
representation	O
instead	O
of	O
the	O
usual	O
tree	O
based	O
representation	O
to	O
encode	O
computer	O
programs	O
.	O
</s>
<s>
In	O
Genetic	B-Algorithm
Programming	I-Algorithm
two	O
fit	O
individuals	O
are	O
chosen	O
from	O
the	O
population	O
to	O
be	O
parents	O
for	O
one	O
or	O
two	O
children	O
.	O
</s>
<s>
In	O
tree	O
genetic	B-Algorithm
programming	I-Algorithm
,	O
these	O
parents	O
are	O
represented	O
as	O
inverted	O
lisp	B-Language
like	O
trees	O
,	O
with	O
their	O
root	B-Application
nodes	I-Application
at	O
the	O
top	O
.	O
</s>
<s>
In	O
subtree	B-Application
crossover	O
in	O
each	O
parent	O
a	O
subtree	B-Application
is	O
randomly	O
chosen	O
.	O
</s>
<s>
In	O
the	O
root	O
donating	O
parent	O
(	O
in	O
the	O
animation	O
on	O
the	O
left	O
)	O
the	O
chosen	O
subtree	B-Application
is	O
removed	O
and	O
replaced	O
with	O
a	O
copy	O
of	O
the	O
randomly	O
chosen	O
subtree	B-Application
from	O
the	O
other	O
parent	O
,	O
to	O
give	O
a	O
new	O
child	O
tree	O
.	O
</s>
<s>
Sometimes	O
two	O
child	O
crossover	O
is	O
used	O
,	O
in	O
which	O
case	O
the	O
removed	O
subtree	B-Application
(	O
in	O
the	O
animation	O
on	O
the	O
left	O
)	O
is	O
not	O
simply	O
deleted	O
but	O
is	O
copied	O
to	O
a	O
copy	O
of	O
the	O
second	O
parent	O
(	O
here	O
on	O
the	O
right	O
)	O
replacing	O
(	O
in	O
the	O
copy	O
)	O
its	O
randomly	O
chosen	O
subtree	B-Application
.	O
</s>
<s>
Thus	O
this	O
type	O
of	O
subtree	B-Application
crossover	O
takes	O
two	O
fit	O
trees	O
and	O
generates	O
two	O
child	O
trees	O
.	O
</s>
<s>
There	O
are	O
many	O
types	O
of	O
mutation	O
in	O
genetic	B-Algorithm
programming	I-Algorithm
.	O
</s>
<s>
a	O
subtree	B-Application
is	O
randomly	O
chosen	O
(	O
highlighted	O
by	O
yellow	O
)	O
.	O
</s>
<s>
It	O
is	O
removed	O
and	O
replaced	O
by	O
a	O
randomly	O
generated	O
subtree	B-Application
.	O
</s>
<s>
Other	O
mutation	O
operators	O
select	O
a	O
leaf	O
(	O
external	B-Application
node	I-Application
)	O
of	O
the	O
tree	O
and	O
replace	O
it	O
with	O
a	O
randomly	O
chosen	O
leaf	O
.	O
</s>
<s>
Hoist	O
mutation	O
randomly	O
chooses	O
a	O
subtree	B-Application
and	O
replaces	O
it	O
with	O
a	O
subtree	B-Application
within	O
itself	O
.	O
</s>
<s>
Whereas	O
subtree	B-Application
mutation	O
(	O
in	O
the	O
animation	O
)	O
may	O
,	O
depending	O
upon	O
the	O
function	O
and	O
terminal	O
sets	O
,	O
have	O
a	O
bias	O
to	O
either	O
increase	O
or	O
decrease	O
the	O
tree	O
size	O
.	O
</s>
<s>
Other	O
subtree	B-Application
based	O
mutations	O
try	O
to	O
carefully	O
control	O
the	O
size	O
of	O
the	O
replacement	O
subtree	B-Application
and	O
thus	O
the	O
size	O
of	O
the	O
child	O
tree	O
.	O
</s>
<s>
Similarly	O
there	O
are	O
many	O
types	O
of	O
linear	B-Algorithm
genetic	I-Algorithm
programming	I-Algorithm
mutation	O
,	O
each	O
of	O
which	O
tries	O
to	O
ensure	O
the	O
mutated	O
child	O
is	O
still	O
syntactically	O
correct	O
.	O
</s>
<s>
GP	O
has	O
been	O
successfully	O
used	O
as	O
an	O
automatic	B-Application
programming	I-Application
tool	O
,	O
a	O
machine	O
learning	O
tool	O
and	O
an	O
automatic	O
problem-solving	O
engine	O
.	O
</s>
<s>
Some	O
of	O
the	O
applications	O
of	O
GP	O
are	O
curve	O
fitting	O
,	O
data	O
modeling	O
,	O
symbolic	B-Algorithm
regression	I-Algorithm
,	O
feature	B-General_Concept
selection	I-General_Concept
,	O
classification	O
,	O
etc	O
.	O
</s>
<s>
John	O
R	O
.	O
Koza	O
mentions	O
76	O
instances	O
where	O
Genetic	B-Algorithm
Programming	I-Algorithm
has	O
been	O
able	O
to	O
produce	O
results	O
that	O
are	O
competitive	O
with	O
human-produced	O
results	O
(	O
called	O
Human-competitive	O
results	O
)	O
.	O
</s>
<s>
Meta-genetic	O
programming	O
is	O
the	O
proposed	O
meta-learning	B-General_Concept
technique	O
of	O
evolving	O
a	O
genetic	B-Algorithm
programming	I-Algorithm
system	O
using	O
genetic	B-Algorithm
programming	I-Algorithm
itself	O
.	O
</s>
<s>
Doug	O
Lenat	O
's	O
Eurisko	B-Algorithm
is	O
an	O
earlier	O
effort	O
that	O
may	O
be	O
the	O
same	O
technique	O
.	O
</s>
