<s>
In	O
programming	O
languages	O
and	O
machine	O
learning	O
,	O
Bayesian	B-Application
program	I-Application
synthesis	I-Application
(	O
BPS	O
)	O
is	O
a	O
program	B-Application
synthesis	I-Application
technique	O
where	O
Bayesian	O
probabilistic	O
programs	O
automatically	O
construct	O
new	O
Bayesian	O
probabilistic	O
programs	O
.	O
</s>
<s>
Bayesian	B-Application
program	I-Application
synthesis	I-Application
(	O
BPS	O
)	O
has	O
been	O
described	O
as	O
a	O
framework	O
related	O
to	O
and	O
utilizing	O
probabilistic	O
programming	O
.	O
</s>
<s>
This	O
framework	O
can	O
be	O
contrasted	O
with	O
the	O
family	O
of	O
automated	O
program	B-Application
synthesis	I-Application
fields	O
,	O
which	O
include	O
programming	B-Application
by	I-Application
example	I-Application
and	O
programming	B-Application
by	I-Application
demonstration	I-Application
.	O
</s>
<s>
In	O
traditional	O
program	B-Application
synthesis	I-Application
,	O
for	O
instance	O
,	O
verification	O
of	O
logical	O
constraints	O
reduce	O
the	O
state	O
space	O
of	O
possible	O
programs	O
,	O
allowing	O
more	O
efficient	O
search	O
to	O
find	O
an	O
optimal	O
program	O
.	O
</s>
<s>
Bayesian	B-Application
program	I-Application
synthesis	I-Application
differs	O
both	O
in	O
that	O
the	O
constraints	O
are	O
probabilistic	O
and	O
the	O
output	O
is	O
itself	O
a	O
distribution	O
over	O
programs	O
that	O
can	O
be	O
further	O
refined	O
.	O
</s>
<s>
Additionally	O
,	O
Bayesian	B-Application
program	I-Application
synthesis	I-Application
can	O
be	O
contrasted	O
to	O
the	O
work	O
on	O
Bayesian	O
program	O
learning	O
,	O
where	O
probabilistic	O
program	O
components	O
are	O
hand-written	O
,	O
pre-trained	O
on	O
data	O
,	O
and	O
then	O
hand	O
assembled	O
in	O
order	O
to	O
recognize	O
handwritten	O
characters	O
.	O
</s>
