A view on model misspecification in uncertainty quantification

TL;DR

This paper discusses the impact of model misspecification on the reliability of uncertainty estimates in machine learning, emphasizing the need for more focus on this issue.

Contribution

It highlights the importance of considering model misspecification in uncertainty quantification and provides thought experiments to illustrate its effects.

Findings

Model misspecification affects uncertainty estimate reliability

Current literature underestimates the impact of misspecification

More research needed on robustness of uncertainty estimates

Abstract

Estimating uncertainty of machine learning models is essential to assess the quality of the predictions that these models provide. However, there are several factors that influence the quality of uncertainty estimates, one of which is the amount of model misspecification. Model misspecification always exists as models are mere simplifications or approximations to reality. The question arises whether the estimated uncertainty under model misspecification is reliable or not. In this paper, we argue that model misspecification should receive more attention, by providing thought experiments and contextualizing these with relevant literature.