<s>
The	O
original	O
application	O
for	O
the	O
algorithm	O
was	O
the	O
task	O
of	O
Named	B-General_Concept
Entity	I-General_Concept
Classification	I-General_Concept
using	O
very	O
weak	B-Algorithm
learners	I-Algorithm
.	O
</s>
<s>
It	O
may	O
be	O
seen	O
as	O
a	O
combination	O
of	O
co-training	B-Algorithm
and	O
boosting	B-Algorithm
.	O
</s>
<s>
Each	O
example	O
is	O
available	O
in	O
two	O
views	O
(	O
subsections	O
of	O
the	O
feature	O
set	O
)	O
,	O
and	O
boosting	B-Algorithm
is	O
applied	O
iteratively	O
in	O
alternation	O
with	O
each	O
view	O
using	O
predicted	O
labels	O
produced	O
in	O
the	O
alternate	O
view	O
on	O
the	O
previous	O
iteration	O
.	O
</s>
<s>
CoBoosting	B-Algorithm
is	O
not	O
a	O
valid	O
boosting	B-Algorithm
algorithm	O
in	O
the	O
PAC	O
learning	O
sense	O
.	O
</s>
<s>
CoBoosting	B-Algorithm
was	O
an	O
attempt	O
by	O
Collins	O
and	O
Singer	O
to	O
improve	O
on	O
previous	O
attempts	O
to	O
leverage	O
redundancy	O
in	O
features	O
for	O
training	O
classifiers	O
in	O
a	O
semi-supervised	O
fashion	O
.	O
</s>
<s>
The	O
advantage	O
of	O
CoBoosting	B-Algorithm
to	O
CoTraining	O
is	O
that	O
it	O
generalizes	O
the	O
CoTraining	O
pattern	O
so	O
that	O
it	O
could	O
be	O
used	O
with	O
any	O
classifier	O
.	O
</s>
<s>
CoBoosting	B-Algorithm
accomplishes	O
this	O
feat	O
by	O
borrowing	O
concepts	O
from	O
AdaBoost	B-Algorithm
.	O
</s>
<s>
CoBoosting	B-Algorithm
builds	O
on	O
the	O
AdaBoost	B-Algorithm
algorithm	O
,	O
which	O
gives	O
CoBoosting	B-Algorithm
its	O
generalization	O
ability	O
since	O
AdaBoost	B-Algorithm
can	O
be	O
used	O
in	O
conjunction	O
with	O
many	O
other	O
learning	O
algorithms	O
.	O
</s>
<s>
In	O
the	O
AdaBoost	B-Algorithm
framework	O
,	O
weak	B-Algorithm
classifiers	I-Algorithm
are	O
generated	O
in	O
series	O
as	O
well	O
as	O
a	O
distribution	O
over	O
examples	O
in	O
the	O
training	O
set	O
.	O
</s>
<s>
Each	O
weak	B-Algorithm
classifier	I-Algorithm
is	O
given	O
a	O
weight	O
and	O
the	O
final	O
strong	O
classifier	O
is	O
defined	O
as	O
the	O
sign	O
of	O
the	O
sum	O
of	O
the	O
weak	B-Algorithm
classifiers	I-Algorithm
weighted	O
by	O
their	O
assigned	O
weight	O
.	O
</s>
<s>
(	O
See	O
AdaBoost	B-Algorithm
Wikipedia	O
page	O
for	O
notation	O
)	O
.	O
</s>
<s>
In	O
the	O
AdaBoost	B-Algorithm
framework	O
Schapire	O
and	O
Singer	O
have	O
shown	O
that	O
the	O
training	O
error	O
is	O
bounded	O
by	O
the	O
following	O
equation	O
:	O
</s>
<s>
CoBoosting	B-Algorithm
extends	O
this	O
framework	O
in	O
the	O
case	O
where	O
one	O
has	O
a	O
labeled	O
training	O
set	O
(	O
examples	O
from	O
)	O
and	O
an	O
unlabeled	O
training	O
set	O
(	O
from	O
)	O
,	O
as	O
well	O
as	O
satisfy	O
the	O
conditions	O
of	O
redundancy	O
in	O
features	O
in	O
the	O
form	O
of	O
.	O
</s>
<s>
The	O
algorithm	O
trains	O
two	O
classifiers	O
in	O
the	O
same	O
fashion	O
as	O
AdaBoost	B-Algorithm
that	O
agree	O
on	O
the	O
labeled	O
training	O
sets	O
correct	O
labels	O
and	O
maximizes	O
the	O
agreement	O
between	O
the	O
two	O
classifiers	O
on	O
the	O
unlabeled	O
training	O
set	O
.	O
</s>
<s>
Which	O
is	O
identical	O
to	O
the	O
equation	O
in	O
AdaBoost	B-Algorithm
.	O
</s>
<s>
Thus	O
the	O
same	O
process	O
can	O
be	O
used	O
to	O
update	O
the	O
values	O
of	O
as	O
in	O
AdaBoost	B-Algorithm
using	O
and	O
.	O
</s>
