Reproduce “Neural Variational Inference and Learning in Belief Networks” 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

Andriy Mnih, Karol Gregor

NVIL trains directed generative models using a feedforward inference network that samples hidden variables from a learned variational distribution. Variance reduction controls noisy gradients as the model and network jointly optimize a variational lower bound. On binarized MNIST, a three-layer SBN reports estimated NLL 96.7 versus wake-sleep’s 102.3; fDARN reports RCV1 perplexity 598, an upper bound, not an exact value.

Summary by alphaXiv.

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