Reproduce “Posterior sampling by source-space MCMC via prior-based few-step transport maps” 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

Hoang Phuc Hau Luu, Marcelo Hartmann, Zhongjian Wang

The authors present iMF source-space sampling for image posteriors. It reweights Gaussian starting points by task weight evaluated after a few-step transport map and uses parallel tempering. Under stated regularity assumptions, population training risk bounds transport and posterior error. On one golden-retriever prompt, iMF+SPT+pCN scores 0.3809 versus SiT+SPT+pCN’s 0.3122 in 2.40 versus 10.51 hours; this protocol-specific comparison does not establish exact posterior sampling.

Summary by alphaXiv.

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