Population Predictive Checks

The episode discusses Gemma Moran, David Blei, and Rajesh Ranganath's paper "Population Predictive Checks" which introduces pop-pc to fix the double use of data problem in Bayesian model checking. The hosts explain how pop-pc uses held-out data to provide calibrated p-values that reliably detect overfitting in models like regression and topic models.

Open the full episode

More from AI Radio