It is now well established that near-optimal approximation errors can be achieved, in the 
expected L2 sense, using sampling strategies whose size scales linearly with n, the dimension of 
the projection space. In particular, this can be accomplished via greedy algorithms inspired by techniques 
from spectral sparsification theory. These sampling methods also lead to moderate Lebesgue constants 
for sample sets of size n, ensuring the stability of the resulting interpolation schemes. We discuss efficient 
sampling strategies tailored to high-dimensional settings, both in the sampling space and the projection space.