Reproduce “Sampling Interventional Distributions with Orthogonal MMD Flow” 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
Bariscan Bozkurt, Hugh Dance, Zonghao Chen, Alexandre Galashov, Arthur Gretton, Houssam Zenati
IMMD-Flow estimates an interventional target embedding from observational data using conditional outcome models and treatment reweighting, then moves particles toward that fixed target rather than training a separate sampler. On ColorMNIST, its pixel-space sliced 2-Wasserstein score is 9.30 ± 0.30 versus 11.37 ± 0.78 for DecFM–Plugin (table values scaled by 10³); this benchmark supports distributional rebalancing, not individual counterfactual claims.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.