Reproduce “The Landscape of Causal Discovery Data: Grounding Causal Discovery in Real-World Applications” 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

Philippe Brouillard, Chandler Squires, Jonas Wahl, Konrad P. Kording, Karen Sachs, Alexandre Drouin, Dhanya Sridhar

This meta-analysis critiques current causal discovery research practices, finding an over-reliance on synthetic data and structural metrics. The authors advocate for an application-driven approach grounded in realistic datasets and interventional evaluation to foster more robust and practically relevant causal inference methods.

Key insights

  • Real-world problems, especially in fields like biology, neuroscience, and Earth sciences, act as 'catalysts for innovation' by exposing fundamental assumption violations (e.g., unobserved confounders, cycles, imperfect interventions).
  • Many established causal discovery methods are evaluated on purely synthetic or small-scale pseudo-real datasets that do not reflect the complexity and noise of real-world phenomena.
  • The Sachs dataset in biology and the bnlearn repository for pseudo-real data are disproportionately used, limiting diversity and generalizability of evaluated methods.

Summary by alphaXiv.

Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.