Reviews: Masked Autoregressive Flow for Density Estimation

This paper is generally well written and I enjoy reading it. It introduces an expressive density model called masked autoregressive flow (MAF) that stacks multiple MADE layers to form a normalizing flow. Although it seems a bit incremental since the techniques involved have been studied in IAF and MADE, this paper does a good job elaborating on different types of generative modeling and providing guidelines for their use cases. It also makes a connection between MAF and IAF. Only a few comments/questions below: * It'd be helpful to motivate a bit more on the advantage of density models.