This story has moved to /story/optimal-multiscale-learning-of-linear-operators/.