Efficiently learning and sampling from multimodal distributions using data-based initialization
Learning to sample is a central task in generative AI: the goal is to generate (infinitely many more) samples from a target distribution $\mu$ given a small number of samples from $\mu.$
