A view on model misspecification in uncertainty quantification
This work addresses the problem of unreliable uncertainty quantification for researchers and practitioners in machine learning, but it is incremental as it primarily synthesizes existing ideas without new empirical results.
The paper argues that model misspecification, which always exists as approximations to reality, significantly impacts the reliability of uncertainty estimates in machine learning predictions, and calls for increased attention to this issue through thought experiments and literature review.
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.