Reproduce “Scalable Discrete Sampling as a Multi-Armed Bandit Problem” 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
Yutian Chen, Zoubin Ghahramani
The authors introduce an approximate discrete sampler that adaptively evaluates factor contributions to choose among candidate states. It treats sampling as a race, eliminating unlikely winners using partial evidence, with a total-variation error guarantee when best-arm identification succeeds. In coreference, it used five times fewer factor evaluations near convergence than exact Gibbs, with similar B³ F1 by iteration—not runtime.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.