We identify mechanisms through which normalizing flows can simultaneously increase likelihood for all structured images.

We investigate the spectral properties of linear-width feed-forward neural networks, where the sample size is asymptotically proportional to network width.

This framework has the flexibility of deriving distribution-and algorithm-dependent bounds, which are often tighter than VC-related uniform convergence bounds.

Not surprisingly, there are emerging efforts to speed up the inference of text-to-image diffusion models on mobile devices.