<s>
In	O
logic	O
,	O
statistical	O
inference	O
,	O
and	O
supervised	B-General_Concept
learning	I-General_Concept
,	O
</s>
<s>
people	O
may	O
call	O
this	O
an	O
example	O
of	O
the	O
closely	O
related	O
semi-supervised	B-General_Concept
learning	I-General_Concept
,	O
since	O
Vapnik	O
's	O
motivation	O
is	O
quite	O
different	O
.	O
</s>
<s>
An	O
example	O
of	O
an	O
algorithm	O
in	O
this	O
category	O
is	O
the	O
Transductive	B-Algorithm
Support	I-Algorithm
Vector	I-Algorithm
Machine	I-Algorithm
(	O
TSVM	O
)	O
.	O
</s>
<s>
be	O
allowed	O
in	O
semi-supervised	B-General_Concept
learning	I-General_Concept
.	O
</s>
<s>
The	O
following	O
example	O
problem	O
contrasts	O
some	O
of	O
the	O
unique	O
properties	O
of	O
transduction	B-General_Concept
against	O
induction	O
.	O
</s>
<s>
The	O
inductive	O
approach	O
to	O
solving	O
this	O
problem	O
is	O
to	O
use	O
the	O
labeled	O
points	O
to	O
train	O
a	O
supervised	B-General_Concept
learning	I-General_Concept
algorithm	O
,	O
and	O
then	O
have	O
it	O
predict	O
labels	O
for	O
all	O
of	O
the	O
unlabeled	O
points	O
.	O
</s>
<s>
With	O
this	O
problem	O
,	O
however	O
,	O
the	O
supervised	B-General_Concept
learning	I-General_Concept
algorithm	O
will	O
only	O
have	O
five	O
labeled	O
points	O
to	O
use	O
as	O
a	O
basis	O
for	O
building	O
a	O
predictive	O
model	O
.	O
</s>
<s>
Transduction	B-General_Concept
has	O
the	O
advantage	O
of	O
being	O
able	O
to	O
consider	O
all	O
of	O
the	O
points	O
,	O
not	O
just	O
the	O
labeled	O
points	O
,	O
while	O
performing	O
the	O
labeling	O
task	O
.	O
</s>
<s>
An	O
advantage	O
of	O
transduction	B-General_Concept
is	O
that	O
it	O
may	O
be	O
able	O
to	O
make	O
better	O
predictions	O
with	O
fewer	O
labeled	O
points	O
,	O
because	O
it	O
uses	O
the	O
natural	O
breaks	O
found	O
in	O
the	O
unlabeled	O
points	O
.	O
</s>
<s>
One	O
disadvantage	O
of	O
transduction	B-General_Concept
is	O
that	O
it	O
builds	O
no	O
predictive	O
model	O
.	O
</s>
<s>
A	O
supervised	B-General_Concept
learning	I-General_Concept
algorithm	O
,	O
on	O
the	O
other	O
hand	O
,	O
can	O
label	O
new	O
points	O
instantly	O
,	O
with	O
very	O
little	O
computational	O
cost	O
.	O
</s>
<s>
Transduction	B-General_Concept
algorithms	O
can	O
be	O
broadly	O
divided	O
into	O
two	O
categories	O
:	O
those	O
that	O
seek	O
to	O
assign	O
discrete	O
labels	O
to	O
unlabeled	O
points	O
,	O
and	O
those	O
that	O
seek	O
to	O
regress	O
continuous	O
labels	O
for	O
unlabeled	O
points	O
.	O
</s>
<s>
Algorithms	O
that	O
seek	O
to	O
predict	O
discrete	O
labels	O
tend	O
to	O
be	O
derived	O
by	O
adding	O
partial	O
supervision	O
to	O
a	O
clustering	B-Algorithm
algorithm	I-Algorithm
.	O
</s>
<s>
Two	O
classes	O
of	O
algorithms	O
can	O
be	O
used	O
:	O
flat	O
clustering	B-Algorithm
and	O
hierarchical	O
clustering	B-Algorithm
.	O
</s>
<s>
Partitioning	O
transduction	B-General_Concept
can	O
be	O
thought	O
of	O
as	O
top-down	O
transduction	B-General_Concept
.	O
</s>
<s>
It	O
is	O
a	O
semi-supervised	O
extension	O
of	O
partition-based	O
clustering	B-Algorithm
.	O
</s>
<s>
Max	B-Algorithm
flow	I-Algorithm
min	I-Algorithm
cut	I-Algorithm
partitioning	O
schemes	O
are	O
very	O
popular	O
for	O
this	O
purpose	O
.	O
</s>
<s>
Agglomerative	O
transduction	B-General_Concept
can	O
be	O
thought	O
of	O
as	O
bottom-up	O
transduction	B-General_Concept
.	O
</s>
<s>
It	O
is	O
a	O
semi-supervised	O
extension	O
of	O
agglomerative	O
clustering	B-Algorithm
.	O
</s>
<s>
Manifold-learning-based	O
transduction	B-General_Concept
is	O
still	O
a	O
very	O
young	O
field	O
of	O
research	O
.	O
</s>
