Better Uncertainty Calibration via Proper Scores for Classification and Beyond
This work addresses the need for reliable uncertainty calibration in sensitive real-world applications, offering a method to quantify calibration improvements more accurately, though it is incremental as it builds on existing proper score concepts.
The paper tackles the problem of biased and inconsistent estimators for uncertainty calibration errors in deep neural networks by introducing a framework of proper calibration errors, which relates each error to a proper score and provides an upper bound with optimal estimation properties, theoretically and empirically showing improvements over common estimators.
With model trustworthiness being crucial for sensitive real-world applications, practitioners are putting more and more focus on improving the uncertainty calibration of deep neural networks. Calibration errors are designed to quantify the reliability of probabilistic predictions but their estimators are usually biased and inconsistent. In this work, we introduce the framework of proper calibration errors, which relates every calibration error to a proper score and provides a respective upper bound with optimal estimation properties. This relationship can be used to reliably quantify the model calibration improvement. We theoretically and empirically demonstrate the shortcomings of commonly used estimators compared to our approach. Due to the wide applicability of proper scores, this gives a natural extension of recalibration beyond classification.