Conformal Prediction via Transported Beta Laws
The episode discusses Thiago R. Ramos and collaborators' paper, "Conformal Prediction via Transported Beta Laws." The hosts explain how this framework analyzes split conformal prediction by focusing on the law of realized calibration-conditional coverage instead of just the marginal guarantee. They conclude that using Wasserstein distances provides a direct way to bound coverage gaps and probabilities of bad calibration events in non-i.i.d. settings.