<s>
Neuroevolution	B-Algorithm
,	O
or	O
neuro-evolution	O
,	O
is	O
a	O
form	O
of	O
artificial	B-Application
intelligence	I-Application
that	O
uses	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
to	O
generate	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
(	O
ANN	O
)	O
,	O
parameters	O
,	O
and	O
rules	O
.	O
</s>
<s>
It	O
is	O
most	O
commonly	O
applied	O
in	O
artificial	O
life	O
,	O
general	B-Algorithm
game	I-Algorithm
playing	I-Algorithm
and	O
evolutionary	O
robotics	O
.	O
</s>
<s>
The	O
main	O
benefit	O
is	O
that	O
neuroevolution	B-Algorithm
can	O
be	O
applied	O
more	O
widely	O
than	O
supervised	B-General_Concept
learning	I-General_Concept
algorithms	I-General_Concept
,	O
which	O
require	O
a	O
syllabus	O
of	O
correct	O
input-output	O
pairs	O
.	O
</s>
<s>
In	O
contrast	O
,	O
neuroevolution	B-Algorithm
requires	O
only	O
a	O
measure	O
of	O
a	O
network	O
's	O
performance	O
at	O
a	O
task	O
.	O
</s>
<s>
Neuroevolution	B-Algorithm
is	O
commonly	O
used	O
as	O
part	O
of	O
the	O
reinforcement	O
learning	O
paradigm	O
,	O
and	O
it	O
can	O
be	O
contrasted	O
with	O
conventional	O
deep	B-Algorithm
learning	I-Algorithm
techniques	O
that	O
use	O
gradient	B-Algorithm
descent	I-Algorithm
on	O
a	O
neural	B-Architecture
network	I-Architecture
with	O
a	O
fixed	O
topology	O
.	O
</s>
<s>
Many	O
neuroevolution	B-Algorithm
algorithms	O
have	O
been	O
defined	O
.	O
</s>
<s>
One	O
common	O
distinction	O
is	O
between	O
algorithms	O
that	O
evolve	O
only	O
the	O
strength	O
of	O
the	O
connection	O
weights	O
for	O
a	O
fixed	O
network	O
topology	O
(	O
sometimes	O
called	O
conventional	O
neuroevolution	B-Algorithm
)	O
,	O
and	O
algorithms	O
that	O
evolve	O
both	O
the	O
topology	O
of	O
the	O
network	O
and	O
its	O
weights	O
(	O
called	O
TWEANNs	O
,	O
for	O
Topology	O
and	O
Weight	O
Evolving	O
Artificial	B-Architecture
Neural	I-Architecture
Network	I-Architecture
algorithms	O
)	O
.	O
</s>
<s>
A	O
separate	O
distinction	O
can	O
be	O
made	O
between	O
methods	O
that	O
evolve	O
the	O
structure	O
of	O
ANNs	O
in	O
parallel	O
to	O
its	O
parameters	O
(	O
those	O
applying	O
standard	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
)	O
and	O
those	O
that	O
develop	O
them	O
separately	O
(	O
through	O
memetic	B-Algorithm
algorithms	I-Algorithm
)	O
.	O
</s>
<s>
Most	O
neural	B-Architecture
networks	I-Architecture
use	O
gradient	B-Algorithm
descent	I-Algorithm
rather	O
than	O
neuroevolution	B-Algorithm
.	O
</s>
<s>
However	O
,	O
around	O
2017	O
researchers	O
at	O
Uber	B-Application
stated	O
they	O
had	O
found	O
that	O
simple	O
structural	O
neuroevolution	B-Algorithm
algorithms	O
were	O
competitive	O
with	O
sophisticated	O
modern	O
industry-standard	O
gradient-descent	O
deep	B-Algorithm
learning	I-Algorithm
algorithms	O
,	O
in	O
part	O
because	O
neuroevolution	B-Algorithm
was	O
found	O
to	O
be	O
less	O
likely	O
to	O
get	O
stuck	O
in	O
local	O
minima	O
.	O
</s>
<s>
journalist	O
Matthew	O
Hutson	O
speculated	O
that	O
part	O
of	O
the	O
reason	O
neuroevolution	B-Algorithm
is	O
succeeding	O
where	O
it	O
had	O
failed	O
before	O
is	O
due	O
to	O
the	O
increased	O
computational	O
power	O
available	O
in	O
the	O
2010s	O
.	O
</s>
<s>
It	O
can	O
be	O
shown	O
that	O
there	O
is	O
a	O
correspondence	O
between	O
neuroevolution	B-Algorithm
and	O
gradient	B-Algorithm
descent	I-Algorithm
.	O
</s>
<s>
Evolutionary	B-Algorithm
algorithms	I-Algorithm
operate	O
on	O
a	O
population	O
of	O
genotypes	O
(	O
also	O
referred	O
to	O
as	O
genomes	O
)	O
.	O
</s>
<s>
In	O
neuroevolution	B-Algorithm
,	O
a	O
genotype	O
is	O
mapped	O
to	O
a	O
neural	B-Architecture
network	I-Architecture
phenotype	O
that	O
is	O
evaluated	O
on	O
some	O
task	O
to	O
derive	O
its	O
fitness	O
.	O
</s>
<s>
That	O
is	O
,	O
every	O
neuron	O
and	O
connection	O
in	O
the	O
neural	B-Architecture
network	I-Architecture
is	O
specified	O
directly	O
and	O
explicitly	O
in	O
the	O
genotype	O
.	O
</s>
<s>
Traditionally	O
indirect	O
encodings	O
that	O
employ	O
artificial	B-Algorithm
embryogeny	I-Algorithm
(	O
also	O
known	O
as	O
artificial	B-Algorithm
development	I-Algorithm
)	O
have	O
been	O
categorised	O
along	O
the	O
lines	O
of	O
a	O
grammatical	O
approach	O
versus	O
a	O
cell	O
chemistry	O
approach	O
.	O
</s>
<s>
Complexification	O
:	O
the	O
ability	O
of	O
the	O
system	O
(	O
including	O
evolutionary	B-Algorithm
algorithm	I-Algorithm
and	O
genotype	O
to	O
phenotype	O
mapping	O
)	O
to	O
allow	O
complexification	O
of	O
the	O
genome	O
(	O
and	O
hence	O
phenotype	O
)	O
over	O
time	O
.	O
</s>
<s>
Examples	O
of	O
neuroevolution	B-Algorithm
methods	O
(	O
those	O
with	O
direct	O
encodings	O
are	O
necessarily	O
non-embryogenic	O
)	O
:	O
</s>
<s>
Method	O
Encoding	O
Evolutionary	B-Algorithm
algorithm	I-Algorithm
Aspects	O
evolved	O
Neuro-genetic	O
evolution	O
by	O
E	O
.	O
Ronald	O
,	O
1994	O
Direct	O
Genetic	B-Algorithm
algorithm	I-Algorithm
Network	O
Weights	O
Cellular	O
Encoding	O
(	O
CE	O
)	O
by	O
F	O
.	O
Gruau	O
,	O
1994	O
Indirect	O
,	O
embryogenic	O
(	O
grammar	O
tree	O
using	O
S-expressions	B-Protocol
)	O
Genetic	B-Algorithm
programming	I-Algorithm
Structure	O
and	O
parameters	O
(	O
simultaneous	O
,	O
complexification	O
)	O
GNARL	O
by	O
Angeline	O
et	O
al.	O
,	O
1994	O
Direct	O
Evolutionary	B-Algorithm
programming	I-Algorithm
Structure	O
and	O
parameters	O
(	O
simultaneous	O
,	O
complexification	O
)	O
EPNet	O
by	O
Yao	O
and	O
Liu	O
,	O
1997	O
Direct	O
Evolutionary	B-Algorithm
programming	I-Algorithm
(	O
combined	O
with	O
backpropagation	B-Algorithm
and	O
simulated	B-Algorithm
annealing	I-Algorithm
)	O
Structure	O
and	O
parameters	O
(	O
mixed	O
,	O
complexification	O
and	O
simplification	O
)	O
NeuroEvolution	B-Algorithm
of	I-Algorithm
Augmenting	I-Algorithm
Topologies	I-Algorithm
(	O
NEAT	O
)	O
by	O
Stanley	O
and	O
Miikkulainen	O
,	O
2002	O
Direct	O
Genetic	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Structure	O
and	O
parameters	O
Hypercube-based	B-Algorithm
NeuroEvolution	I-Algorithm
of	I-Algorithm
Augmenting	I-Algorithm
Topologies	I-Algorithm
(	O
HyperNEAT	B-Algorithm
)	O
by	O
Stanley	O
,	O
D'Ambrosio	O
,	O
Gauci	O
,	O
2008	O
Indirect	O
,	O
non-embryogenic	O
(	O
spatial	O
patterns	O
generated	O
by	O
a	O
Compositional	B-Algorithm
pattern-producing	I-Algorithm
network	I-Algorithm
(	O
CPPN	B-Algorithm
)	O
within	O
a	O
hypercube	B-Operating_System
are	O
interpreted	O
as	O
connectivity	O
patterns	O
in	O
a	O
lower-dimensional	O
space	O
)	O
Genetic	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
The	O
NEAT	O
algorithm	O
(	O
above	O
)	O
is	O
used	O
to	O
evolve	O
the	O
CPPN	B-Algorithm
.	O
</s>
<s>
Parameters	O
,	O
structure	O
fixed	O
(	O
functionally	O
fully	O
connected	O
)	O
Evolvable	O
Substrate	O
Hypercube-based	B-Algorithm
NeuroEvolution	I-Algorithm
of	I-Algorithm
Augmenting	I-Algorithm
Topologies	I-Algorithm
(	O
ES-HyperNEAT	O
)	O
by	O
Risi	O
,	O
Stanley	O
2012	O
Indirect	O
,	O
non-embryogenic	O
(	O
spatial	O
patterns	O
generated	O
by	O
a	O
Compositional	B-Algorithm
pattern-producing	I-Algorithm
network	I-Algorithm
(	O
CPPN	B-Algorithm
)	O
within	O
a	O
hypercube	B-Operating_System
are	O
interpreted	O
as	O
connectivity	O
patterns	O
in	O
a	O
lower-dimensional	O
space	O
)	O
Genetic	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
The	O
NEAT	O
algorithm	O
(	O
above	O
)	O
is	O
used	O
to	O
evolve	O
the	O
CPPN	B-Algorithm
.	O
</s>
<s>
Parameters	O
and	O
network	O
structure	O
Evolutionary	B-Algorithm
Acquisition	I-Algorithm
of	I-Algorithm
Neural	I-Algorithm
Topologies	I-Algorithm
(	O
EANT/EANT2	O
)	O
by	O
Kassahun	O
and	O
Sommer	O
,	O
2005	O
/	O
Siebel	O
and	O
Sommer	O
,	O
2007	O
Direct	O
and	O
indirect	O
,	O
potentially	O
embryogenic	O
(	O
Common	O
Genetic	O
Encoding	O
)	O
Evolutionary	O
programming/Evolution	O
strategies	O
Structure	O
and	O
parameters	O
(	O
separately	O
,	O
complexification	O
)	O
Interactively	O
Constrained	O
Neuro-Evolution	O
(	O
ICONE	O
)	O
by	O
Rempis	O
,	O
2012	O
Direct	O
,	O
includes	O
constraint	O
masks	O
to	O
restrict	O
the	O
search	O
to	O
specific	O
topology	O
/	O
parameter	O
manifolds	O
.	O
</s>
<s>
Evolutionary	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Structure	O
and	O
parameters	O
(	O
separately	O
,	O
complexification	O
,	O
interactive	O
)	O
Deus	O
Ex	O
Neural	B-Architecture
Network	I-Architecture
(	O
DXNN	O
)	O
by	O
Gene	O
Sher	O
,	O
2012	O
Direct/Indirect	O
,	O
includes	O
constraints	O
,	O
local	O
tuning	O
,	O
and	O
allows	O
for	O
evolution	O
to	O
integrate	O
new	O
sensors	O
and	O
actuators	O
.	O
</s>
<s>
Memetic	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Structure	O
and	O
parameters	O
(	O
separately	O
,	O
complexification	O
,	O
interactive	O
)	O
Spectrum-diverse	O
Unified	O
Neuroevolution	B-Algorithm
Architecture	O
(	O
SUNA	O
)	O
by	O
Danilo	O
Vasconcellos	O
Vargas	O
,	O
Junichi	O
Murata	O
(	O
Download	O
code	O
)	O
Direct	O
,	O
introduces	O
the	O
Unified	O
Neural	O
Representation	O
(	O
representation	O
integrating	O
most	O
of	O
the	O
neural	B-Architecture
network	I-Architecture
features	O
from	O
the	O
literature	O
)	O
.	O
</s>
<s>
Genetic	B-Algorithm
Algorithm	I-Algorithm
with	O
a	O
diversity	O
preserving	O
mechanism	O
called	O
Spectrum-diversity	O
that	O
scales	O
well	O
with	O
chromosome	O
size	O
,	O
is	O
problem	O
independent	O
and	O
focus	O
more	O
on	O
obtaining	O
diversity	O
of	O
high	O
level	O
behaviours/approaches	O
.	O
</s>
<s>
(	O
Download	O
code	O
)	O
Direct	O
or	O
indirect	O
encoding	O
of	O
Markov	O
networks	O
,	O
Neural	B-Architecture
Networks	I-Architecture
,	O
genetic	B-Algorithm
programming	I-Algorithm
,	O
and	O
other	O
arbitrarily	O
customizable	O
controllers	O
.	O
</s>
<s>
Provides	O
evolutionary	B-Algorithm
algorithms	I-Algorithm
,	O
genetic	B-Algorithm
programming	I-Algorithm
algorithms	O
,	O
and	O
allows	O
customized	O
algorithms	O
,	O
along	O
with	O
specification	O
of	O
arbitrary	O
constraints	O
.	O
</s>
<s>
Evolvable	O
aspects	O
include	O
the	O
neural	O
model	O
and	O
allows	O
for	O
the	O
evolution	O
of	O
morphology	O
and	O
sexual	O
selection	O
among	O
others.Covariance	O
Matrix	O
Adaptation	O
with	O
Hypervolume	O
Sorted	O
Adaptive	O
Grid	O
Algorithm	O
(	O
CMA-HAGA	O
)	O
by	O
Shahin	O
Rostami	O
,	O
and	O
others.Direct	O
,	O
includes	O
an	O
atavism	O
feature	O
which	O
enables	O
traits	O
to	O
disappear	O
and	O
re-appear	O
at	O
different	O
generations.Multi-Objective	O
Evolution	B-Algorithm
Strategy	I-Algorithm
with	O
Preference	O
Articulation	O
(	O
Computational	B-Application
Steering	I-Application
)	O
Structure	O
,	O
weights	O
,	O
and	O
biases	O
.	O
</s>
