Prediction uncertainty validation for computational chemists
Résumé
Validation of prediction uncertainty (PU) is becoming an essential task formodern computational chemistry. Designed to quantify the reliability ofpredictions in meteorology, the calibration-sharpness (CS) framework is nowwidely used to optimize and validate uncertainty-aware machine learning (ML)methods. However, its application is not limited to ML and it can serve as aprincipled framework for any PU validation. The present article is intended asa step-by-step introduction to the concepts and techniques of PU validation inthe CS framework, adapted to the specifics of computational chemistry. Thepresented methods range from elementary graphical checks to more sophisticatedones based on local calibration statistics. The concept of tightness, isintroduced. The methods are illustrated on synthetic datasets and applied touncertainty quantification data extracted from the computational chemistryliterature.
Origine : Accord explicite pour ce dépôt
Licence : Copyright (Tous droits réservés)
Licence : Copyright (Tous droits réservés)