Reproduce “Beyond Distributional Fidelity: Causal-Penalized Diffusion for Synthetic Tabular Data” 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
Lan Tao, Yongxian He, Shirong Xu, Yidong Ouyang, Guang Cheng
The authors train a fully generative tabular diffusion model to preserve treatment-effect inference as well as imitate data. A causal penalty compares average treatment effect (ATE) posterior distributions from real and generated tables using a fixed inference procedure. On IHDP, causal-penalized TabDDPM lowers posterior Wasserstein-1 distance from 0.133 ± 0.059 to 0.097 ± 0.048 versus vanilla TabDDPM; this measures agreement with that procedure, not guaranteed truth.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.