<s>
In	O
machine	O
learning	O
the	O
margin	O
of	O
a	O
single	O
data	O
point	O
is	O
defined	O
to	O
be	O
the	O
distance	O
from	O
the	O
data	O
point	O
to	O
a	O
decision	B-General_Concept
boundary	I-General_Concept
.	O
</s>
<s>
Note	O
that	O
there	O
are	O
many	O
distances	O
and	O
decision	B-General_Concept
boundaries	I-General_Concept
that	O
may	O
be	O
appropriate	O
for	O
certain	O
datasets	O
and	O
goals	O
.	O
</s>
<s>
A	O
margin	B-General_Concept
classifier	I-General_Concept
is	O
a	O
classifier	B-General_Concept
that	O
explicitly	O
utilizes	O
the	O
margin	O
of	O
each	O
example	O
while	O
learning	O
a	O
classifier	B-General_Concept
.	I-General_Concept
</s>
<s>
If	O
such	O
a	O
hyperplane	O
exists	O
,	O
it	O
is	O
known	O
as	O
the	O
maximum-margin	O
hyperplane	O
and	O
the	O
linear	O
classifier	B-General_Concept
it	O
defines	O
is	O
known	O
as	O
a	O
maximum	O
margin	B-General_Concept
classifier	I-General_Concept
;	O
or	O
equivalently	O
,	O
the	O
perceptron	B-Algorithm
of	O
optimal	O
stability	O
.	O
</s>
