Reproduce “Optimization of Annealed Importance Sampling Hyperparameters” 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
Shirin Goshtasbpour, Fernando Perez-Cruz
The authors introduce OP-AIS, which learns a neural path for annealed importance sampling to estimate marginal likelihoods. The path bridges a proposal and target through shared-parameter intermediate distributions. On binarized MNIST, its 16-transition VAE estimate was 193 ± 31.9, versus 215.6 ± 35.4 for vanilla AIS with 1,024 transitions; this comparison does not establish a match to the tuned reference.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.