Reproduce “Learning Probabilistic Programs” 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
Yura N. Perov, Frank D. Wood
The authors infer executable sampler programs from observed samples by generating candidate program text from a learned prior, executing candidates, and favoring those whose sample summaries match the data. Their leave-one-distribution-out experiments cover six univariate distributions and report an exact Bernoulli sampler among inferred candidates. Other candidates can be inexact, and finite-time inference does not always find a good sampler.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.