generalfactneutralApproximate leave-one-out estimators can significantly accelerate conformal prediction while maintaining asymptotic coverage guaranteesMachine Learning (Statistics)27 Jul 2026http://arxiv.org/abs/2606.31915v1