Training-conditional coverage for distribution-free predictive inference

Michael Bian, Rina Foygel Barber

The field of distribution-free predictive inference provides tools for provably valid prediction without any assumptions on the distribution of the data, which can be paired with a…