<s>
Structured	B-General_Concept
prediction	I-General_Concept
or	O
structured	O
(	O
output	O
)	O
learning	O
is	O
an	O
umbrella	O
term	O
for	O
supervised	B-General_Concept
machine	I-General_Concept
learning	I-General_Concept
techniques	O
that	O
involves	O
predicting	O
structured	O
objects	O
,	O
rather	O
than	O
scalar	O
discrete	B-General_Concept
or	O
real	O
values	O
.	O
</s>
<s>
Similar	O
to	O
commonly	O
used	O
supervised	B-General_Concept
learning	I-General_Concept
techniques	O
,	O
structured	B-General_Concept
prediction	I-General_Concept
models	O
are	O
typically	O
trained	O
by	O
means	O
of	O
observed	O
data	O
in	O
which	O
the	O
true	O
prediction	O
value	O
is	O
used	O
to	O
adjust	O
model	O
parameters	O
.	O
</s>
<s>
Due	O
to	O
the	O
complexity	O
of	O
the	O
model	O
and	O
the	O
interrelations	O
of	O
predicted	O
variables	O
the	O
process	O
of	O
prediction	O
using	O
a	O
trained	O
model	O
and	O
of	O
training	O
itself	O
is	O
often	O
computationally	O
infeasible	O
and	O
approximate	B-General_Concept
inference	I-General_Concept
and	O
learning	O
methods	O
are	O
used	O
.	O
</s>
<s>
For	O
example	O
,	O
the	O
problem	O
of	O
translating	O
a	O
natural	O
language	O
sentence	O
into	O
a	O
syntactic	O
representation	O
such	O
as	O
a	O
parse	O
tree	O
can	O
be	O
seen	O
as	O
a	O
structured	B-General_Concept
prediction	I-General_Concept
problem	O
in	O
which	O
the	O
structured	O
output	O
domain	O
is	O
the	O
set	O
of	O
all	O
possible	O
parse	O
trees	O
.	O
</s>
<s>
Structured	B-General_Concept
prediction	I-General_Concept
is	O
also	O
used	O
in	O
a	O
wide	O
variety	O
of	O
application	O
domains	O
including	O
bioinformatics	O
,	O
natural	B-Language
language	I-Language
processing	I-Language
,	O
speech	B-Application
recognition	I-Application
,	O
and	O
computer	B-Application
vision	I-Application
.	O
</s>
<s>
Sequence	O
tagging	O
is	O
a	O
class	O
of	O
problems	O
prevalent	O
in	O
natural	B-Language
language	I-Language
processing	I-Language
,	O
where	O
input	O
data	O
are	O
often	O
sequences	O
(	O
e.g.	O
</s>
<s>
part-of-speech	O
tagging	O
and	O
named	B-General_Concept
entity	I-General_Concept
recognition	I-General_Concept
.	O
</s>
<s>
While	O
this	O
problem	O
can	O
be	O
solved	O
by	O
simply	O
performing	O
classification	B-General_Concept
of	O
individual	O
tokens	O
,	O
that	O
approach	O
does	O
not	O
take	O
into	O
account	O
the	O
empirical	O
fact	O
that	O
tags	O
do	O
not	O
occur	O
independently	O
;	O
instead	O
,	O
each	O
tag	O
displays	O
a	O
strong	O
conditional	O
dependence	O
on	O
the	O
tag	O
of	O
the	O
previous	O
word	O
.	O
</s>
<s>
This	O
fact	O
can	O
be	O
exploited	O
in	O
a	O
sequence	O
model	O
such	O
as	O
a	O
hidden	O
Markov	O
model	O
or	O
conditional	B-General_Concept
random	I-General_Concept
field	I-General_Concept
that	O
predicts	O
the	O
entire	O
tag	O
sequence	O
for	O
a	O
sentence	O
,	O
rather	O
than	O
just	O
individual	O
tags	O
,	O
by	O
means	O
of	O
the	O
Viterbi	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Probabilistic	O
graphical	O
models	O
form	O
a	O
large	O
class	O
of	O
structured	B-General_Concept
prediction	I-General_Concept
models	O
.	O
</s>
<s>
Other	O
algorithms	O
and	O
models	O
for	O
structured	B-General_Concept
prediction	I-General_Concept
include	O
inductive	B-Application
logic	I-Application
programming	I-Application
,	O
case-based	O
reasoning	O
,	O
structured	B-Algorithm
SVMs	I-Algorithm
,	O
Markov	O
logic	O
networks	O
,	O
Probabilistic	O
Soft	O
Logic	O
,	O
and	O
constrained	B-General_Concept
conditional	I-General_Concept
models	I-General_Concept
.	O
</s>
<s>
One	O
of	O
the	O
easiest	O
ways	O
to	O
understand	O
algorithms	O
for	O
general	O
structured	B-General_Concept
prediction	I-General_Concept
is	O
the	O
structured	O
perceptron	B-Algorithm
of	O
Collins	O
.	O
</s>
<s>
This	O
algorithm	O
combines	O
the	O
perceptron	B-Algorithm
algorithm	I-Algorithm
for	O
learning	O
linear	B-General_Concept
classifiers	I-General_Concept
with	O
an	O
inference	O
algorithm	O
(	O
classically	O
the	O
Viterbi	B-Algorithm
algorithm	I-Algorithm
when	O
used	O
on	O
sequence	O
data	O
)	O
and	O
can	O
be	O
described	O
abstractly	O
as	O
follows	O
.	O
</s>
<s>
In	O
practice	O
,	O
finding	O
the	O
argmax	O
over	O
will	O
be	O
done	O
using	O
an	O
algorithm	O
such	O
as	O
Viterbi	O
or	O
an	O
algorithm	O
such	O
as	O
max-sum	O
,	O
rather	O
than	O
an	O
exhaustive	B-Algorithm
search	I-Algorithm
through	O
an	O
exponentially	O
large	O
set	O
of	O
candidates	O
.	O
</s>
<s>
The	O
idea	O
of	O
learning	O
is	O
similar	O
to	O
multiclass	O
perceptron	B-Algorithm
.	O
</s>
