Reproduce “Logical Guidance of Stochastic Interpolants” 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
Gerrit Gerhartz, Francesco Alesiani, Jonathan Warrell, Henrik Christiansen, Matheus Ferraz, Mathias Niepert
LOGSI guides source-to-target sampling toward Boolean constraints by transporting samples with a composed velocity proxy, then applying Langevin correction steps using a composed conditional score. Score composition is exact under stated probabilistic-circuit assumptions, while velocity composition is exact on Gaussian paths and for selected formulas. On EMNIST-to-CMNIST AND tasks, conformity improves by up to 14 percentage points over compute-matched CompSI; finite correction remains approximate.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.