Conformal Prediction as Bayesian Quadrature
This work addresses uncertainty quantification for machine learning practitioners in high-stakes applications, offering a novel Bayesian approach that improves upon existing frequentist methods.
The authors tackled the problem of uncertainty quantification for black-box predictive models in high-stakes situations by proposing a Bayesian alternative to frequentist conformal prediction, resulting in a method that provides interpretable guarantees and a richer representation of likely test-time losses.
As machine learning-based prediction systems are increasingly used in high-stakes situations, it is important to understand how such predictive models will perform upon deployment. Distribution-free uncertainty quantification techniques such as conformal prediction provide guarantees about the loss black-box models will incur even when the details of the models are hidden. However, such methods are based on frequentist probability, which unduly limits their applicability. We revisit the central aspects of conformal prediction from a Bayesian perspective and thereby illuminate the shortcomings of frequentist guarantees. We propose a practical alternative based on Bayesian quadrature that provides interpretable guarantees and offers a richer representation of the likely range of losses to be observed at test time.