Reproduce “Learning Likelihoods with Conditional Normalizing 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
Christina Winkler, Daniel Worrall, Emiel Hoogeboom, Max Welling
The authors introduce a conditional normalizing flow that learns a distribution over structured outputs, conditioning both its latent prior and invertible transformations on the input. They adapt variational dequantization to binary segmentation labels. On ImageNet64 super-resolution, it achieves 2.90 bpd versus 3.61 for a factorized baseline; on DRIVE, its maximum F-score is close to DRIU’s, not higher.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.