Reproduce “Learning $ω$-Powered Conditional Distributions with Importance-Weighted Flow Matching” 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
Maciej Falkiewicz, Frantzeska Lavda, Jason Ramapuram, Alexandros Kalousis
IW-CFM learns a corrector for powered conditional image distributions using density-ratio weights. On synthetic distributions, it follows analytical target contours while matched CFG departs. On ImageNet, its PRDC Recall is 0.63 versus 0.14 for matched CFG at power 32, against a regularized empirical reference rather than a proven exact powered distribution.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.