"The big idea is taking local data and seeing if it patches together."

I wonder if those new techniques are different from the usually used like: weighted averages of random forest, cross entropy of density estimation, minimization of variance between local estimations and the like.