A first attempt : k-center

• Randomly select R= C x k points

• expected C points from the “largest cluster”

• Consider all C-points subsets of R

• (at least one will be “purely” from largest cluster)

• Centroid of this will approximate cluster centre