A somewhat particular case of this, which is missing in the examples, is the IRLS [0] algorithm for minimizing sums of distances. I say "somewhat" because the

A somewhat particular case of this, which is missing in the examples, is the IRLS [0] algorithm for minimizing sums of distances. I say "somewhat" because the random part is missing, you just update all the weights at each iteration. The idea is that optimizing a weighted sum of squared distances is trivial: the optimum is just the weighted average. Thus you set up the weights, iteratively, to the inverse distances, so that you end up minimizing the actual sum of distances. [0] Iteratively Re-weighted Least Squares