Reproduce “Diffusion models for probabilistic programming” 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
Simon Dirmeier, Fernando Perez-Cruz
DMVI uses a diffusion model as a variational guide to approximate posteriors in probabilistic programs. It models parameters and diffusion variables jointly, training the guide with a new bound for the multistep process. On a synthetic hierarchical model with 100 observations, one setting reports MSE 0.39, versus 0.62 for NFVI and 11.77 for ADVI; evidence is limited to synthetic tests.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.