Reproduce “The Good, the Bad, and the Ugly of Markov Boundary for Tabular Prediction” 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
Shu Wan, Abhinav Gorantla, Huan Liu, K. Selçuk Candan
The authors test whether a Markov boundary—its target’s parents, children, and spouses—improves regression on wide tables. On SCM3K, a synthetic benchmark of 3,450 tasks, they compare predictors trained on all features with models trained on known boundaries. TabPFN’s median relative test-RMSE reduction is 12%; this oracle gain varies by regressor and does not show estimated boundaries work reliably.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.