<s>
Evolutionary	B-Algorithm
acquisition	I-Algorithm
of	I-Algorithm
neural	I-Algorithm
topologies	I-Algorithm
(	O
EANT/EANT2	O
)	O
is	O
an	O
evolutionary	O
reinforcement	O
learning	O
method	O
that	O
evolves	O
both	O
the	O
topology	O
and	O
weights	O
of	O
artificial	B-Architecture
neural	I-Architecture
networks	I-Architecture
.	O
</s>
<s>
Like	O
the	O
work	O
of	O
Angeline	O
et	O
al.	O
,	O
the	O
method	O
uses	O
a	O
type	O
of	O
parametric	O
mutation	O
that	O
comes	O
from	O
evolution	B-Algorithm
strategies	I-Algorithm
and	O
evolutionary	B-Algorithm
programming	I-Algorithm
(	O
now	O
using	O
the	O
most	O
advanced	O
form	O
of	O
the	O
evolution	B-Algorithm
strategies	I-Algorithm
CMA-ES	B-Algorithm
in	O
EANT2	O
)	O
,	O
in	O
which	O
adaptive	O
step	O
sizes	O
are	O
used	O
for	O
optimizing	O
the	O
weights	O
of	O
the	O
neural	B-Architecture
networks	I-Architecture
.	O
</s>
<s>
Similar	O
to	O
the	O
work	O
of	O
Stanley	O
(	O
NEAT	B-Algorithm
)	O
,	O
the	O
method	O
starts	O
with	O
minimal	O
structures	O
which	O
gain	O
complexity	O
along	O
the	O
evolution	O
path	O
.	O
</s>
<s>
Despite	O
sharing	O
these	O
two	O
properties	O
,	O
the	O
method	O
has	O
the	O
following	O
important	O
features	O
which	O
distinguish	O
it	O
from	O
previous	O
works	O
in	O
neuroevolution	B-Algorithm
.	O
</s>
<s>
It	O
introduces	O
a	O
genetic	O
encoding	O
called	O
common	O
genetic	O
encoding	O
(	O
CGE	O
)	O
that	O
handles	O
both	O
direct	O
and	O
indirect	O
encoding	O
of	O
neural	B-Architecture
networks	I-Architecture
within	O
the	O
same	O
theoretical	O
framework	O
.	O
</s>
<s>
The	O
encoding	O
has	O
important	O
properties	O
that	O
makes	O
it	O
suitable	O
for	O
evolving	O
neural	B-Architecture
networks	I-Architecture
:	O
</s>
<s>
For	O
evolving	O
the	O
structure	O
and	O
weights	O
of	O
neural	B-Architecture
networks	I-Architecture
,	O
an	O
evolutionary	O
process	O
is	O
used	O
,	O
where	O
the	O
exploration	O
of	O
structures	O
is	O
executed	O
at	O
a	O
larger	O
timescale	O
(	O
structural	O
exploration	O
)	O
,	O
and	O
the	O
exploitation	O
of	O
existing	O
structures	O
is	O
done	O
at	O
a	O
smaller	O
timescale	O
(	O
structural	O
exploitation	O
)	O
.	O
</s>
<s>
In	O
the	O
structural	O
exploitation	O
phase	O
,	O
the	O
weights	O
of	O
the	O
currently	O
available	O
structures	O
are	O
optimized	O
using	O
an	O
evolution	B-Algorithm
strategy	I-Algorithm
.	O
</s>
<s>
Moreover	O
,	O
a	O
newer	O
version	O
of	O
EANT	O
,	O
called	O
EANT2	O
,	O
was	O
tested	O
on	O
a	O
visual	O
servoing	O
task	O
and	O
found	O
to	O
outperform	O
NEAT	B-Algorithm
and	O
the	O
traditional	O
iterative	O
Gauss	B-Algorithm
–	I-Algorithm
Newton	I-Algorithm
method	I-Algorithm
.	O
</s>
