To my Problem: From the paper on Masked Autoregressive Flows and the accompanying talk I thought that one point for using MAFs is its ability to fit multimodal distributions
