On double-descent in uncertainty quantification in overparametrized models (c...

Uncertainty quantification is a central challenge in reliable and trustworthy machine learning. Naive measures such as last-layer scores are well-known to yield overconfident...