Reproduce “When Shift Happens - Confounding Is to Blame” 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
Abbavaram Gowtham Reddy, Celia Rubio-Madrigal, Rebekka Burkholz, Krikamol Muandet
The authors analyze how hidden confounding can affect out-of-distribution prediction when an unobserved factor influences inputs and outcomes. Theory and experiments show environment information or informative proxy features can help when pooled relationships mislead, without implying invariance or causal features are unhelpful. On the tabular benchmark, XGBoost accuracy with all covariates exceeded accuracy with causal features, though real-world confounding was not observed.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.