Predictions from machine learning algorithms can vary across randomseeds, inducing instability in downstreamdebiasedmachinelearningestimators.Weformalizerandomseedstabilityviaa concentration condition and prove that subbagging guarantees stability for any bounded-outcome regression algorithm. We introduce a new cross-fitting procedure, adaptive cross-bagging, which simultaneously eliminates seed dependence frombothnuisanceestimationandsamplesplitting in debiased machine learning. Numerical experiments confirm that the method achieves the targeted level of stability whereas alternatives do not. Our method incurs a small computational penalty relative to standard practice whereas alternative methods incur large penalties.
Some key words: Algorithmic Stability, Reproducibility, Bagging, Debiased Machine Learning