Does Predictive Performance Translate to Causal Performance? Evaluating tabular foundation models as nuisance learners for causal effect estimation

30秒で要点

何が起きたか

Researchers evaluated whether tabular foundation models called PFNs can estimate causal effects reliably, finding they outperform conventional methods in complex settings.

なぜ重要か

Data scientists building causal inference pipelines should care when choosing nuisance learners for propensity score and outcome modeling in doubly robust estimation.

注意点

PFNs showed propensity score overfitting at small sample sizes without cross-fitting, and performance gains were most pronounced in complex settings, not simple parametric cases.