Reproduce “Learning Counterfactually Invariant Predictors” 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
Francesco Quinzan, Cecilia Casolo, Krikamol Muandet, Yucen Luo, Niki Kilbertus
CIP trains predictors for counterfactual invariance by penalizing observational conditional dependence identified from a known causal graph. Under adjustment-set and injectivity assumptions, this condition is sufficient for matching prediction distributions across interventions. In a simulated comparison, CIP had lower test error and counterfactual variance than CF2. These results are limited to modeled settings and do not establish real-world fairness.
Summary by alphaXiv.
Tip: change arxiv.org to autoarxiv.org in any arXiv abstract link to open the paper here.