François Clément (UW)
-
LOW 117
Suppose we have already sampled n points according to some distribution, and want to sample n more according to (potentially) another distribution. How efficiently can we compute the future points and " correct" the distribution? In this talk, I will present a Wasserstein W_1 metric based approach in one dimension that computes the next point in linear time, while being empirically extremely robust. I will also present a continuous version that allows for a better theoretical analysis of the process.