<s>
Random	B-Algorithm
search	I-Algorithm
(	O
RS	O
)	O
is	O
a	O
family	O
of	O
numerical	O
optimization	O
methods	O
that	O
do	B-Algorithm
not	I-Algorithm
require	I-Algorithm
the	I-Algorithm
gradient	I-Algorithm
of	O
the	O
problem	O
to	O
be	O
optimized	O
,	O
and	O
RS	O
can	O
hence	O
be	O
used	O
on	O
functions	O
that	O
are	O
not	O
continuous	O
or	O
differentiable	O
.	O
</s>
<s>
The	O
name	O
"	O
random	B-Algorithm
search	I-Algorithm
"	O
is	O
attributed	O
to	O
Rastrigin	O
who	O
made	O
an	O
early	O
presentation	O
of	O
RS	O
along	O
with	O
basic	O
mathematical	O
analysis	O
.	O
</s>
<s>
The	O
algorithm	O
described	O
herein	O
is	O
a	O
type	O
of	O
local	O
random	B-Algorithm
search	I-Algorithm
,	O
where	O
every	O
iteration	O
is	O
dependent	O
on	O
the	O
prior	O
iteration	O
's	O
candidate	O
solution	O
.	O
</s>
<s>
There	O
are	O
alternative	O
random	B-Algorithm
search	I-Algorithm
methods	O
that	O
sample	O
from	O
the	O
entirety	O
of	O
the	O
search	O
space	O
(	O
for	O
example	O
pure	O
random	B-Algorithm
search	I-Algorithm
or	O
uniform	O
global	O
random	B-Algorithm
search	I-Algorithm
)	O
,	O
but	O
these	O
are	O
not	O
described	O
in	O
this	O
article	O
.	O
</s>
<s>
Random	B-Algorithm
search	I-Algorithm
has	O
been	O
used	O
in	O
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
for	O
hyper-parameter	O
optimization	O
.	O
</s>
<s>
Truly	O
random	B-Algorithm
search	I-Algorithm
is	O
purely	O
by	O
luck	O
and	O
varies	O
from	O
very	O
costive	O
to	O
very	O
lucky	O
,	O
but	O
the	O
structured	O
random	B-Algorithm
search	I-Algorithm
is	O
strategic	O
.	O
</s>
<s>
This	O
example	O
code	O
converges	O
1-2	O
orders	O
of	O
magnitude	O
slower	O
than	O
the	O
Levenberg	B-Algorithm
–	I-Algorithm
Marquardt	I-Algorithm
algorithm	I-Algorithm
,	O
with	O
an	O
example	O
also	O
provided	O
in	O
the	O
GitHub	O
.	O
</s>
<s>
Fixed	O
Step	O
Size	O
Random	B-Algorithm
Search	I-Algorithm
(	O
FSSRS	O
)	O
is	O
Rastrigin	O
's	O
basic	O
algorithm	O
which	O
samples	O
from	O
a	O
hypersphere	O
of	O
fixed	O
radius	O
.	O
</s>
<s>
Optimum	O
Step	O
Size	O
Random	B-Algorithm
Search	I-Algorithm
(	O
OSSRS	O
)	O
by	O
Schumer	O
and	O
Steiglitz	O
is	O
primarily	O
a	O
theoretical	O
study	O
on	O
how	O
to	O
optimally	O
adjust	O
the	O
radius	O
of	O
the	O
hypersphere	O
so	O
as	O
to	O
allow	O
for	O
speedy	O
convergence	O
to	O
the	O
optimum	O
.	O
</s>
<s>
Adaptive	O
Step	O
Size	O
Random	B-Algorithm
Search	I-Algorithm
(	O
ASSRS	O
)	O
by	O
Schumer	O
and	O
Steiglitz	O
attempts	O
to	O
heuristically	O
adapt	O
the	O
hypersphere	O
's	O
radius	O
:	O
two	O
new	O
candidate	O
solutions	O
are	O
generated	O
,	O
one	O
with	O
the	O
current	O
nominal	O
step	O
size	O
and	O
one	O
with	O
a	O
larger	O
step-size	O
.	O
</s>
<s>
Optimized	O
Relative	O
Step	O
Size	O
Random	B-Algorithm
Search	I-Algorithm
(	O
ORSSRS	O
)	O
by	O
Schrack	O
and	O
Choit	O
approximate	O
the	O
optimal	O
step	O
size	O
by	O
a	O
simple	O
exponential	O
decrease	O
.	O
</s>
