<s>
In	O
computer	B-General_Concept
science	I-General_Concept
,	O
a	O
convolutional	B-Algorithm
deep	I-Algorithm
belief	I-Algorithm
network	I-Algorithm
(	O
CDBN	O
)	O
is	O
a	O
type	O
of	O
deep	B-Algorithm
artificial	B-Architecture
neural	I-Architecture
network	I-Architecture
composed	O
of	O
multiple	O
layers	O
of	O
convolutional	B-Architecture
restricted	B-Algorithm
Boltzmann	I-Algorithm
machines	I-Algorithm
stacked	O
together	O
.	O
</s>
<s>
Alternatively	O
,	O
it	O
is	O
a	O
hierarchical	O
generative	O
model	O
for	O
deep	B-Algorithm
learning	I-Algorithm
,	O
which	O
is	O
highly	O
effective	O
in	O
image	B-Algorithm
processing	I-Algorithm
and	O
object	O
recognition	O
,	O
though	O
it	O
has	O
been	O
used	O
in	O
other	O
domains	O
too	O
.	O
</s>
<s>
Training	O
of	O
the	O
network	O
involves	O
a	O
pre-training	O
stage	O
accomplished	O
in	O
a	O
greedy	B-Algorithm
layer-wise	O
manner	O
,	O
similar	O
to	O
other	O
deep	B-Algorithm
belief	I-Algorithm
networks	I-Algorithm
.	O
</s>
<s>
Depending	O
on	O
whether	O
the	O
network	O
is	O
to	O
be	O
used	O
for	O
discrimination	O
or	O
generative	O
tasks	O
,	O
it	O
is	O
then	O
"	O
fine	O
tuned	O
"	O
or	O
trained	O
with	O
either	O
back-propagation	B-Algorithm
or	O
the	O
up	O
–	O
down	O
algorithm	O
(	O
contrastive	O
–	O
divergence	O
)	O
,	O
respectively	O
.	O
</s>
