2000 character limit reached
Robust importance-weighted cross-validation under sample selection bias (1710.06514v3)
Published 17 Oct 2017 in cs.LG and stat.ML
Abstract: Cross-validation under sample selection bias can, in principle, be done by importance-weighting the empirical risk. However, the importance-weighted risk estimator produces sub-optimal hyperparameter estimates in problem settings where large weights arise with high probability. We study its sampling variance as a function of the training data distribution and introduce a control variate to increase its robustness to problematically large weights.