<s>
Credal	B-Architecture
networks	I-Architecture
are	O
probabilistic	O
graphical	O
models	O
based	O
on	O
imprecise	O
probability	O
.	O
</s>
<s>
Credal	B-Architecture
networks	I-Architecture
can	O
be	O
regarded	O
as	O
an	O
extension	O
of	O
Bayesian	O
networks	O
,	O
where	O
credal	O
sets	O
replace	O
probability	O
mass	O
functions	O
in	O
the	O
specification	O
of	O
the	O
local	O
models	O
for	O
the	O
network	O
variables	O
given	O
their	O
parents	O
.	O
</s>
<s>
As	O
a	O
Bayesian	O
network	O
defines	O
a	O
joint	O
probability	O
mass	O
function	O
over	O
its	O
variables	O
,	O
a	O
credal	B-Architecture
network	I-Architecture
defines	O
a	O
joint	O
credal	O
set	O
.	O
</s>
<s>
Most	O
of	O
the	O
research	O
on	O
credal	B-Architecture
networks	I-Architecture
focused	O
on	O
the	O
case	O
of	O
strong	O
independence	O
.	O
</s>
<s>
Given	O
strong	O
independence	O
the	O
joint	O
credal	O
set	O
associated	O
to	O
a	O
credal	B-Architecture
network	I-Architecture
is	O
called	O
its	O
strong	O
extension	O
.	O
</s>
<s>
If	O
is	O
,	O
for	O
each	O
,	O
a	O
conditional	O
credal	O
set	O
over	O
,	O
then	O
the	O
strong	O
extension	O
of	O
a	O
credal	B-Architecture
network	I-Architecture
is	O
defined	O
as	O
follows	O
:	O
</s>
<s>
Inference	O
on	O
a	O
credal	B-Architecture
network	I-Architecture
is	O
intended	O
as	O
the	O
computation	O
of	O
the	O
bounds	O
of	O
an	O
expectation	O
with	O
respect	O
to	O
its	O
strong	O
extensions	O
.	O
</s>
<s>
Being	O
a	O
generalization	O
of	O
the	O
same	O
problem	O
for	O
Bayesian	O
networks	O
,	O
updating	O
with	O
credal	B-Architecture
networks	I-Architecture
is	O
a	O
NP-hard	O
task	O
.	O
</s>
