Reproduce “SurVAE Flows: Surjections to Bridge the Gap between VAEs and Flows” on OpenResearch
Turn your favorite coding agents into research agents that review literature, develop hypotheses, and run experiments.
Universal app for Apple silicon and Intel. Requires macOS 11 or later.
About this paper
Didrik Nielsen, Priyank Jaini, Emiel Hoogeboom, Ole Winther, Max Welling
SurVAE Flows presents a composable framework combining bijections, stochastic maps, and non-invertible surjections. Surjections represent sorting, slicing, and max pooling; likelihood behavior depends on direction: inference surjections can support exact computation under stated conditions, while generative surjections generally yield lower bounds. On CIFAR-10, MaxPoolFlow has slightly worse bits/dim but better Inception and FID scores than tensor slicing.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.