Random sampling techniques have been developed
in for geometric/combinatorial optimization problems;
see, e.g.,
\cite{clarkson:focs88,clarkson:jacm95,adler-shamir:mathprog93,gartner-welzl:draft99}.
In this note,
we apply one of these techniques
for obtaining (hopefully) efficient
support vector machine training algorithm.
In particular,
we propose one way to find ``outliers''
by using the sampling technique.