Using quasirandom sequences to evenly sample discrete distributions.

I show how to construct an  infinite sequence of samples from a set $S$, such that for all $m>0$, the first $m$ terms of the sequence, all subsets of $S$ occur as evenly as possible.