Reproduce “VAEBM: A Symbiosis between Variational Autoencoders and Energy-based Models” 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

Zhisheng Xiao, Karsten Kreis, Jan Kautz, Arash Vahdat

VAEBM combines a pretrained VAE with an image-space energy model, then performs MCMC updates in standard-normal noise coordinates for both latent variables and generated images. This gives image and latent updates a common scale without removing sampling computation. On unconditional CIFAR-10, it reports FID 12.96 versus 29.32 for an NVAE-initialized image-only EBM; complete coverage is limited to StackedMNIST.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.