Tag
This paper introduces Heckman-corrected epistemic uncertainty to address selection on unobservables in machine learning, demonstrating that importance weighting fails when selection depends on unobservables correlated with outcomes. The method restores calibration in controlled experiments and real data, outperforming standard UQ baselines.
This paper investigates the position bias phenomenon in online distillation, finding that early tokens provide more useful supervision signals, and proposes the importance-weighted IW-OPD method to improve OPD training.
This paper proposes DRIFT, a framework that combines offline trajectories with importance-weighted supervised fine-tuning to efficiently achieve multi-turn interactive learning performance comparable to reinforcement learning.
TILT introduces a novel objective for unsupervised domain adaptation under covariate shift that penalizes an auxiliary component on unlabeled target data, implicitly achieving self-localized importance weighting with bounded estimands. Theoretical guarantees and experiments on shifted CIFAR-100 show improved target performance over baselines.